技术领域Technical field
本发明涉及用于诊断技术设备的设备电池的方法，特别是用于通过异常检测对设备电池进行预见性诊断的方法。The present invention relates to a method for diagnosing a device battery of a technical device, in particular to a method for predictive diagnosis of a device battery by anomaly detection.
背景技术Background technique
独立于电网运行的电设备和机器(例如可电驱动的机动车辆)的能量供应通常使用设备电池或车辆电池来进行。这些电池提供电能以运行所述设备。Electrical installations and machines that operate independently of the grid, such as electrically drivable motor vehicles, are often supplied with energy using installation batteries or vehicle batteries. These batteries provide electrical energy to run the device.
设备电池在其使用寿命期间根据其负荷或使用情况而退化。这种所谓的老化导致最大性能或存储容量连续下降。老化状态对应于用于说明储能器老化的度量。按照惯例，新设备电池的老化状态(关于其容量，SOH-C)可以为100％，在新设备电池的使用寿命期间该老化状态显著下降。设备电池老化的度量(老化状态随时间的变化)取决于设备电池的个体负荷，即在机动车辆的车辆电池情况下取决于驾驶员的使用行为、外部环境条件和车辆电池类型。Device batteries degrade over their lifetime based on their load or usage. This so-called aging results in a continuous decrease in maximum performance or storage capacity. The aging state corresponds to a measure describing the aging of the energy storage device. By convention, a new device battery may have an aging state (with respect to its capacity, SOH-C) of 100%, which decreases significantly during the service life of the new device battery. The measure of the aging of a device battery (change in aging state over time) depends on the individual load on the device battery, that is, in the case of a vehicle battery of a motor vehicle, on the driver's usage behavior, external environmental conditions and the vehicle battery type.
为了监视来自大量设备的设备电池，通常连续采集运行变量数据并作为运行变量变化过程逐块地传输到设备外部的中央单元。为了评估运行变量数据，特别是在基于微分方程的物理或电化学的电池模型中，将运行变量数据作为变化过程以例如介于1和100Hz之间的相对较高时间分辨率(采样率)来采样，并且使用时间积分法从中确定老化状态。In order to monitor the device battery from a large number of devices, operating variable data are usually collected continuously and transmitted block by block as changes in the operating variables to a central unit external to the device. In order to evaluate the operating variable data, in particular in a physical or electrochemical battery model based on differential equations, the operating variable data are acquired as a variation process at a relatively high time resolution (sampling rate), for example between 1 and 100 Hz. Samples are taken and the aging state is determined from them using the time integration method.
为了评估运行变量数据，特别是为了确定确定老化状态的电池状态，使用基于具有多个非线性微分方程的微分方程组的电化学电池模型。这些运行变量数据使得可以借助于时间积分方法对当前电池状态建模。这种电化学电池模型例如由印刷文献US 2020/150185A、WO2022017984A1、EP3919925A1、DE102020206915B3和US20210373082A1公知。In order to evaluate the operating variable data and in particular to determine the battery state that determines the aging state, an electrochemical cell model based on a system of differential equations with several nonlinear differential equations is used. These operating variable data make it possible to model the current battery state with the help of time integration methods. Such electrochemical cell models are known, for example, from the printed documents US 2020/150185A, WO2022017984A1, EP3919925A1, DE102020206915B3 and US20210373082A1.
在所述中央单元中提供运行变量变化过程使得电化学电池模型能够用于和适配于具有相同类型的电池单池或具有相同类型单池化学成分的单池的大量设备电池。借助于微分方程组计算电池状态在计算上是复杂的，从而可以通过外包给所述中央单元来减少设备内部计算装置中的计算负荷。Providing operating variable variations in the central unit enables the electrochemical cell model to be used and adapted to a large number of device batteries having cells of the same type or cells having the same type of cell chemistry. The calculation of the battery state using a system of differential equations is computationally complex, so that the calculation load in the device-internal computing device can be reduced by outsourcing it to the central unit.
在电池运行的技术设备中，出于安全原因必须定期监视所使用设备电池的按规定运行是否存在故障，特别是在高能量密度的情况下。如果电池单池、由多个电池单池组成的单元或整个设备电池失效，则根据出现的故障，技术设备可能会变得无法工作，并且可能在出现导致温度急剧升高的故障时也会损害技术设备和用户的安全。In battery-operated technical equipment, the intended operation of the batteries used in the equipment must be regularly monitored for faults for safety reasons, especially in the case of high energy densities. If a battery cell, a unit consisting of several battery cells, or an entire device battery fails, depending on the fault that occurs, the technical device may become inoperable and may also be damaged in the event of a fault that causes a sharp increase in temperature Safety of technical equipment and users.
然而，由于基于规则的异常检测，到目前为止只有在超过或低于对诸如单池电压、模块温度、电流值或充电状态值以及老化状态值的运行变量应用的故障阈值时才识别出设备电池中的故障。However, due to rule-based anomaly detection, device batteries have so far only been identified when fault thresholds are exceeded or fell below applied to operating variables such as cell voltage, module temperature, current value or state-of-charge value, and aging state value fault in.
印刷文献DE102019208372A1公开了一种用于识别技术系统中异常的计算机实现的方法，具有以下步骤：采集说明所述技术系统的运行状态并且包括一定数量的运行状态变量的运行变量向量，其中所述运行状态变量包括说明所述技术系统运行所处的环境条件的至少一个环境状态变量，以及说明所述技术系统的内部系统状态的系统状态变量；提供环境状态模型和异常检测模型，其中所述环境状态模型根据所述环境状态变量中的至少一个来说明使用异常检测模型对所述运行变量向量关于是否存在异常的可验证性，并且其中所述异常检测模型根据所述运行变量向量说明预期异常的存在，根据基于所述环境状态模型对所述运行变量向量的至少一个环境状态变量的评估以及根据基于所述异常检测模型对所述运行变量向量的评估来发信号通知存在异常或不存在异常。Printed document DE 10 2019 208 372 A1 discloses a computer-implemented method for identifying anomalies in a technical system, which has the following steps: collecting an operating variable vector that describes the operating state of the technical system and includes a certain number of operating state variables, wherein said operating The state variables include at least one environmental state variable that describes the environmental conditions in which the technical system operates, and a system state variable that describes the internal system state of the technical system; an environmental state model and an anomaly detection model are provided, wherein the environmental state The model accounts for the verifiability of the vector of operating variables with respect to the presence or absence of an anomaly using an anomaly detection model in terms of at least one of the environmental state variables, and wherein the anomaly detection model accounts for the presence of an expected anomaly in terms of the vector of operating variables. , signaling the presence or absence of an anomaly based on an evaluation of at least one environmental state variable of the operating variable vector based on the environmental state model and based on an evaluation of the operating variable vector based on the anomaly detection model.
此外，印刷文献K.Park、Y.Choi、W.J.Choi、H.-Y.Ryu和H.Kim的“LSTM-BasedBattery Remaining Useful Life Prediction WithMulti-Channel ChargingProfiles”，IEEE Access，vol.8，第20786-20798页，2020年，doi：10.1109/ACCESS.2020.2968939公开了借助于LSTM模型针对不同充电配置预测电池的剩余寿命。In addition, the printed document "LSTM-BasedBattery Remaining Useful Life Prediction WithMulti-Channel ChargingProfiles" by K. Park, Y. Choi, W. J. Choi, H.-Y. Ryu, and H. Kim, IEEE Access, vol. 8, pp. 20786- Page 20798, 2020, doi: 10.1109/ACCESS.2020.2968939 discloses the prediction of the remaining life of the battery for different charging configurations with the help of LSTM model.
发明内容Contents of the invention
根据本发明，设置了一种根据权利要求1的用于诊断技术设备的具有一个或多个电池单池的设备电池的方法以及根据并列独立权利要求的一种设备和一种电池系统。According to the invention, a method for diagnosing a device battery of a technical device having one or more battery cells is provided according to claim 1 as well as a device and a battery system according to the independent independent claims.
进一步的设计在从属权利要求中说明。Further developments are stated in the dependent claims.
根据第一方面，设置了一种用于监视设备电池以预测性地识别技术设备中设备电池的故障的方法，具有以下步骤：According to a first aspect, a method for monitoring a device battery to predictively identify failures of a device battery in a technical device is provided, having the following steps:
-提供特定设备电池的多个运行变量的随时间的运行变量变化过程；-Provide the changing process of operating variables over time for multiple operating variables of a specific equipment battery;
-确定输入变量向量的时间序列，每个输入变量向量具有对于一个时间步骤包括一个或多个运行变量和/或从其导出的一个或多个变量的元素；- determining a time series of input variable vectors, each input variable vector having elements including for one time step one or more operating variables and/or one or more variables derived therefrom;
-评估基于数据的异常预测模型，所述异常预测模型包括基于数据的时间序列变换器模型和基于数据的预测模型，其中所述异常预测模型基于训练数据集被训练为分类模型，每个训练数据集向输入变量向量的时间序列分配在所述输入变量向量的时间序列的最后一个时间步骤后的特定持续时间后所述设备电池的特定故障的出现概率，- Evaluate a data-based anomaly prediction model, which includes a data-based time series transformer model and a data-based prediction model, wherein the anomaly prediction model is trained as a classification model based on a training data set, each training data The set assigns to the time series of input variable vectors the probability of occurrence of a specific fault of the device battery after a specific duration after the last time step of the time series of said input variable vector,
-根据所述输入变量向量的时间序列，特别是直到当前时间点为止，基于对所述异常预测模型的评估预测性识别在特定持续时间之后所述设备电池的特定故障的出现。- Predictively identifying the occurrence of a specific fault of the device battery after a specific duration based on the evaluation of the anomaly prediction model based on the time series of the input variable vectors, in particular up to the current point in time.
虽然通常用于异常检测的基于自动编码器的方案能够将当前电池状态分类为正常或异常，但不可能可靠地预测未来何时以一定概率出现失效事件。然而，诸如热失控事件或设备电池完全失效(Sudden Death，突然死亡)的安全关键事件是通过电池状态事先宣布的，从而原则上应当可以预测这样的关键事件。While autoencoder-based schemes commonly used for anomaly detection are able to classify the current battery status as normal or abnormal, it is not possible to reliably predict when a failure event will occur with a certain probability in the future. However, safety-critical events such as thermal runaway events or complete device battery failure (Sudden Death) are announced in advance through the battery status, so that such critical events should in principle be predictable.
上述用于预测性识别设备电池中的异常的基于变换器模型的方法设置在与大量设备电池通信连接的设备外部中央单元中，以评估运行变量变化过程。The above-described converter model-based method for predictive identification of anomalies in equipment batteries is provided in a equipment-external central unit communicatively connected to a large number of equipment batteries to evaluate operating variable change processes.
变换器模型例如由Qingsong Wen等人的“Transformers in Time Series：ASurvey”，arXiv：2202.07125公开。The transformer model is disclosed, for example, by Qingsong Wen et al., "Transformers in Time Series: ASurvey", arXiv: 2202.07125.
变换器模型通常用于语音识别领域。变换器模型基于多头自注意力机制，在这种机制中输入变量向量时间序列中的每个输入变量向量与该时间序列的每个其他输入变量向量进行比较，以便以权重或分数的形式学习不同时间点的输入变量向量之间的动态上下文信息。与递归神经网络(尤其是LSTM)相比，变换器模型的优点是它们还适合于预测更远的预测时空的状态，并且可以回忆过去任何时间点并且是该概念的一部分。例如，如果电池或单个单池在BMS侧在不利的负载或使用范围内运行一一这会对健康状态产生长期影响或可能影响异常的概率，则这是相关的。因此，这些效应由变换器建模采集、映射并可用于预测性诊断。Transformer models are commonly used in the field of speech recognition. The transformer model is based on a multi-head self-attention mechanism in which each input variable vector in a time series of input variable vectors is compared with every other input variable vector of that time series in order to learn differences in the form of weights or scores. Dynamic contextual information between input variable vectors at time points. The advantage of transformer models compared to recurrent neural networks (especially LSTMs) is that they are also suitable for predicting states further into the predicted space-time, and can recall any point in time in the past and be part of the concept. This is relevant, for example, if a battery or individual cell on the BMS side is operating within an adverse load or usage range - this can have a long-term impact on the state of health or may affect the probability of anomalies. Therefore, these effects are captured by transducer modeling, mapped and can be used for predictive diagnosis.
出于容量原因，在设备外部中央单元中确定大量设备的设备电池的内部电池状态。为此，这些设备将设备电池的随时间的运行变量变化过程以运行变量时间序列的形式传送到中央单元，所述运行变量例如是电池电流、电池温度、充电状态和/或电池电压，其中在所述中央单元中确定当前电化学内部电池状态和/或老化状态。通过评估运行变量变化过程，基于为所述设备电池的对应电池类型提供的电化学电池模型，可以计算/确定设备特定的内部电池状态以及必要时的其他变量，例如老化状态。评估可以涉及整个设备电池、各个电池单池或由多个电池单池组成的单元/模块。For capacity reasons, the internal battery status of the device battery of a large number of devices is determined in a central unit external to the device. For this purpose, the devices transmit the course of the operating variables of the device battery over time to the central unit in the form of a time series of operating variables, such as battery current, battery temperature, charge state and/or battery voltage, wherein The current electrochemical internal cell status and/or aging status is determined in the central unit. By evaluating the course of the operating variables, the device-specific internal battery state and, if necessary, other variables, such as the aging state, can be calculated/determined based on the electrochemical cell model provided for the corresponding cell type of the device battery. The assessment can involve the entire device battery, individual battery cells, or a unit/module consisting of multiple battery cells.
电化学电池模型包括微分方程组，所述微分方程组基于通过模型参数参数化的微分方程并借助于时间积分方法来对内部电池状态(特别是平衡状态和必要时动力学状态)进行建模并且提供设备电池的电池单池的运行变量(即设备电池的电池电流、电池电压、电池温度和充电状态)之间的关系。这种电化学电池模型例如由印刷文献US2020/150185A、WO2022017984A1、EP3919925A1、DE102020206915B3和US20210373082A1公知。相应设备电池的老化状态可以近似地被确定为内部状态的线性组合。可以在中央处理单元或控制设备中为个体设备电池评估该电化学电池模型，以确定内部电池状态。The electrochemical cell model includes a system of differential equations that model the internal cell states (in particular the equilibrium state and optionally the kinetic state) based on differential equations parameterized by the model parameters and by means of a time integration method and Provides relationships between operating variables of a device battery's battery cells (i.e., battery current, battery voltage, battery temperature, and state of charge of the device battery). Such electrochemical cell models are known, for example, from the printed documents US2020/150185A, WO2022017984A1, EP3919925A1, DE102020206915B3 and US20210373082A1. The aging state of the corresponding device battery can be approximately determined as a linear combination of internal states. This electrochemical cell model can be evaluated for individual device cells in a central processing unit or control device to determine internal cell status.
所述电化学电池模型的模型参数可以在中央单元中基于静止阶段中短时间段(几分钟到几小时)内采集的大量相同类型设备电池的运行变量变化过程在有限时间段内来加以拟合或参数化(通过将误差平方最小化来适配所述模型参数)，其中可以推导出电化学的平衡参数、动力学的模型参数，这些参数例如可以包括电解质浓度、反应速率、层厚度、孔隙度等。该参数化可以基于设备电池的老化状态的高精度测量进行。The model parameters of the electrochemical cell model can be fitted in a central unit within a limited time period based on the changing processes of operating variables of a large number of batteries of the same type of equipment collected in a short period of time (minutes to hours) during the stationary phase. or parameterization (adapting the model parameters by minimizing the squared error), where electrochemical equilibrium parameters, kinetic model parameters can be derived, which may include, for example, electrolyte concentration, reaction rate, layer thickness, pores Degree etc. This parameterization can be based on highly accurate measurements of the aging state of the device's battery.
此外，可以基于运行变量变化过程来拟合电池模型的一个或多个模型参数，所述模型参数同样可以说明电池状态。此外，随时间的运行变量变化过程可以作为统计变量或累积变量聚合为运行特征，以表征设备电池的周期性负荷。In addition, one or more model parameters of the battery model can be fitted based on the change process of the operating variables, and the model parameters can also explain the battery status. In addition, the change process of operating variables over time can be aggregated into operating characteristics as statistical variables or cumulative variables to characterize the periodic load of the equipment battery.
从而例如可以借助于老化状态模型基于所述运行变量变化过程来确定设备电池的老化状态。Thus, for example, the aging state of the device battery can be determined based on the course of the operating variable using an aging state model.
老化状态(SOH：State of Health)在设备电池的情况下是用于说明剩余电池容量或剩余电池电荷的关键变量。老化状态是设备电池老化的度量。在设备电池或电池模块或电池单池的情况下，老化状态可以作为容量保持率(Capacity Retention Rate，SOH-C)来加以说明。容量保持率SOH-C(即与容量有关的老化状态)作为测量的瞬时容量与完全充电的电池的初始容量之比来加以说明，并且随着老化的增加而下降。替代地，老化状态可以作为相对于设备电池使用寿命开始时的内阻的内阻增加(SOH-R)来加以说明。内阻的相对变化SOH-R随着电池老化的增加而增加。State of Health (SOH), in the case of device batteries, is a key variable for describing the remaining battery capacity or remaining battery charge. Aging status is a measure of the age of a device's battery. In the case of equipment batteries or battery modules or battery cells, the aging state can be described as the Capacity Retention Rate (SOH-C). Capacity retention SOH-C (i.e. capacity-dependent aging state) is stated as the ratio of the measured instantaneous capacity to the initial capacity of a fully charged battery and decreases with increasing age. Alternatively, the aging state may be specified as the internal resistance increase (SOH-R) relative to the internal resistance at the beginning of the device's battery life. The relative change in internal resistance SOH-R increases with battery aging.
可以在中央单元中评估老化状态模型以针对个体设备电池确定老化状态。例如，可以借助于物理老化模型来确定当前老化状态，所述物理老化模型代表电化学电池模型的一种形式并且可以按照对应的方式加以参数化。所述物理老化模型对应于微分方程组并借助于时间积分方法加以评估。The aging state model can be evaluated in a central unit to determine the aging state for individual device batteries. For example, the current aging state can be determined by means of a physical aging model, which represents a form of an electrochemical cell model and can be parameterized in a corresponding manner. The physical aging model corresponds to a system of differential equations and is evaluated by means of a time integration method.
为了提高老化状态模型的准确性，可以以混合老化状态模型，即物理老化模型与基于数据的校正模型的组合的形式提供该老化状态模型。在混合老化状态模型的情况下，可以借助于物理或电化学老化模型来确定物理老化状态，并且可以对所述物理老化状态施加从基于数据的校正模型得出的校正值，特别是通过相加或相乘。所述物理老化模型如上所述基于表征非线性微分方程组的电化学状态的电化学模型方程，连续计算所述电化学模型方程并将其映射为物理老化状态以进行输出，作为SOH-C和/或SOH-R。这些计算典型地可以在云中进行，例如每周一次。In order to improve the accuracy of the aging state model, the aging state model may be provided in the form of a hybrid aging state model, that is, a combination of a physical aging model and a data-based correction model. In the case of a mixed aging state model, the physical aging state can be determined by means of a physical or electrochemical aging model, and correction values derived from a data-based correction model can be applied to said physical aging state, in particular by addition Or multiply. The physical aging model is based on the electrochemical model equation characterizing the electrochemical state of the nonlinear differential equation set as described above. The electrochemical model equation is continuously calculated and mapped to the physical aging state for output as SOH-C and /or SOH-R. These calculations can typically be performed in the cloud, for example on a weekly basis.
此外，基于数据的混合老化状态模型的校正模型可以构造为具有概率的或基于人工智能的回归模型，特别是高斯过程模型，并且可以被训练为校正由物理老化模型获得的老化状态。因此，为此存在用于校正SOH-C的基于数据的老化状态校正模型和/或用于校正SOH-R的至少一个另外的基于数据的老化状态校正模型。高斯过程的可能替代方案是另外的监督学习方法，如基于随机森林模型、AdaBoost模型、支持向量机或贝叶斯神经网络的监督学习方法。Furthermore, the correction model of the data-based hybrid aging state model can be constructed as a probabilistic or artificial intelligence-based regression model, especially a Gaussian process model, and can be trained to correct the aging state obtained by the physical aging model. Therefore, for this purpose there is a data-based aging state correction model for correcting SOH-C and/or at least one further data-based aging state correction model for correcting SOH-R. Possible alternatives to Gaussian processes are additional supervised learning methods, such as those based on random forest models, AdaBoost models, support vector machines or Bayesian neural networks.
校正模型可以使用借助于特征提取(Feature-Extraction)方法从运行变量变化过程中确定的运行特征作为输入变量，其中借助于信号处理技术的运算来计算所述特征。分配给运行变量变化过程的运行特征定义了所涉及储能器的运行特征点，其通过储能器的周期性运行(周期性老化)和日历老化(自投入运行或使用寿命开始以来经过的持续时间)来映射负荷状态。The correction model can use as input variables operating characteristics determined from changes in operating variables by means of a feature-extraction method, wherein said characteristics are calculated by means of operations using signal processing techniques. The operating characteristic assigned to the change process of the operating variable defines the operating characteristic point of the energy storage involved, which is determined by the cyclic operation (cyclic aging) and calendar aging (continuous elapsed time since it was put into operation or the beginning of its service life). time) to map the load status.
运行特征可以包括累积的基于负载的特征或聚合特征和/或在整个先前使用寿命期间确定的统计变量。The operating characteristics may include cumulative load-based characteristics or aggregate characteristics and/or statistical variables determined throughout the previous service life.
特别地，可以将来自根据运行变量的变化过程创建的直方图数据的特征确定为运行特征。从而例如可以创建关于电池电流对电池温度和车辆电池的充电状态的直方图、电池温度对车辆电池的充电状态的直方图、充电电流对电池温度的直方图以及放电电流对电池温度的直方图。此外，可以考虑分别自设备电池投入运行以来的累积的总电荷(Ah)、充电过程中的平均容量增加(特别是对于电荷增加量高于总电池容量的阈值份额(例如20％ΔSOC)的充电过程而言)、充电容量以及在充电状态具有足够大冲程的测量的充电过程期间的平滑微分容量极值(例如局部最大值)(dQ/dU的平滑的变化过程：电荷变化除以电池电压变化)或累计行驶里程作为运行特征。另外的运行特征可以对应于在针对电流或电压信号的充电过程上评估的谱峰度的局部极值、小波变换的一个或多个系数和/或傅立叶变换的一个或多个系数，每个系数针对电流或电压信号的充电过程评估或对应于分配给定义的频带的经过变换的频谱值。In particular, features from histogram data created based on the course of changes in the operating variables can be determined as operating features. Thus, for example, histograms of battery current versus battery temperature and state of charge of the vehicle battery, battery temperature versus state of charge of the vehicle battery, charging current versus battery temperature, and discharge current versus battery temperature can be created. Additionally, one can consider the total charge (Ah) accumulated since the device battery was put into operation, and the average capacity increase during charging (especially for charges where the charge increase is above a threshold share of the total battery capacity (e.g. 20% ΔSOC)). process), charging capacity and smooth differential capacity extremes (e.g. local maxima) during the measured charging process with a sufficiently large stroke in the state of charge (smooth course of dQ/dU: charge change divided by battery voltage change ) or accumulated mileage as operating characteristics. Further operating characteristics may correspond to local extrema of the spectral kurtosis evaluated on the charging process for the current or voltage signal, one or more coefficients of the wavelet transform and/or one or more coefficients of the Fourier transform, each coefficient The charging process evaluation for a current or voltage signal corresponds to the transformed spectral values assigned to a defined frequency band.
因此，可以从关于运行变量的直方图中导出运行特征。可以借助于特征工程或特征提取方法从中提取运行特征，例如直方图的平均值、标准偏差以及诸如分布的平均值、中值、最小值、最大值、矩等多维统计值。Therefore, operating characteristics can be derived from histograms with respect to operating variables. Running features can be extracted from them with the help of feature engineering or feature extraction methods, such as the mean, standard deviation of histograms, and multidimensional statistical values such as mean, median, minimum, maximum, moments, etc. of distributions.
此外，可以在中央单元中基于运行变量变化过程来评估电池性能模型，以提供等效电路图参数，例如电池等效电路图的内阻和容量。Furthermore, the battery performance model can be evaluated in the central unit based on the course of operating variables to provide equivalent circuit diagram parameters such as the internal resistance and capacity of the battery equivalent circuit diagram.
此外，导出的一个或多个变量可以包括从运行变量变化过程中导出的一个或多个运行特征和/或从运行变量变化过程中导出的老化状态和/或从运行变量变化过程中导出的一个或多个内部电池状态和/或针对运行变量变化过程拟合的电池模型的一个或多个模型参数。Furthermore, the derived one or more variables may include one or more operating characteristics derived from the operating variable changes and/or an aging state derived from the operating variable changes and/or an operating characteristic derived from the operating variable changes. or multiple internal battery states and/or one or more model parameters of a battery model fitted to a process of changing operating variables.
基于所提供的运行变量变化过程——其被提供为电池电流、电池电压、充电状态和电池温度的时间序列，基于借助于老化状态模型从中导出的老化状态，基于根据电化学老化模型确定的一个或多个内部电池状态作为通过微分方程组以及必要时已经借助于特征提取块为校正模型确定的运行特征(所述运行特征从运行变量变化过程中作为聚合变量导出，例如基于直方图的信号)评估的结果，训练或评估异常预测模型。Based on the provided course of the operating variables, which are provided as time series of battery current, battery voltage, state of charge and battery temperature, based on the aging state derived therefrom by means of the aging state model, based on an or a plurality of internal battery states as an operating characteristic determined for the correction model by means of a system of differential equations and optionally using a feature extraction block (the operating characteristic is derived from the course of the operating variable as an aggregate variable, for example a histogram-based signal) Evaluate the results of training or evaluating anomaly prediction models.
所述异常预测模型被构造为将输入变量向量的时间序列映射为未来时间点发生特定异常事件的概率。如上所述，这些时间序列的输入变量向量均由运行变量变化过程、运行特征、从中导出的老化状态以及从中导出的内部电池状态组成。The anomaly prediction model is constructed to map a time series of input variable vectors into the probability of a specific abnormal event occurring at a future time point. As mentioned above, the input variable vectors of these time series are composed of the operating variable change process, the operating characteristics, the aging state derived therefrom, and the internal battery state derived therefrom.
所述异常预测模型具有时间序列变换器模型，该模型将过去的时间步骤的输入变量向量{xt1，xt2，...xtn}的时间序列映射为所产生的状态变量向量集合{z1，...，zN}。借助于基于数据的预测模型将所产生的状态变量向量集合{z1，...，zN}转换为未来时间点发生关键事件的概率。The anomaly prediction model has a time series transformer model that maps a time series of input variable vectors {xt1 , xt2 , ...xtn } of past time steps into a resulting set of state variable vectors {z1 ,...,zN }. The generated set of state variable vectors {z1 , ..., zN } is converted into the probability of a key event occurring at a future time point with the help of a data-based prediction model.
可以规定，所述时间序列变换器模型具有预处理块，以提供第一状态变量向量的集合，其中所述第一状态变量向量的集合由多头自注意力模块的串行序列处理为所产生的进一步状态变量向量的集合，其中将所产生的进一步状态变量向量的集合在预测块中分配给故障类别，其中所述故障类别说明故障类型以及持续时间，在该持续时间之后出现所述故障类型的故障。It may be provided that the time series transformer model has a preprocessing block to provide a set of first state variable vectors, wherein the set of first state variable vectors is processed by a serial sequence of multi-head self-attention modules to produce a set of further state variable vectors, wherein the resulting set of further state variable vectors is assigned in the prediction block to a fault class, wherein the fault class specifies the fault type and the duration after which the fault type occurs Fault.
现在将考虑的时间步骤t1...tN的输入变量向量{xt1，xt2，...xtn}编码为每个时间步骤的状态变量向量的第一集合z0。The input variable vectors {xt1 , xt2 , ...xtn } of the considered time steps t1...tN are now encoded as a first set z0 of state variable vectors for each time step.
这是通过时间特征编码器和数据特征编码器来实现的。所述时间特征编码器可以简单地说明与过去时间的相对时间差，例如ftime(tN)＝0，ftime(tN-1)＝tN-tN-1等。这在输入变量向量时间序列的输入变量向量之间的时间步骤变化的情况下尤其重要。This is achieved through a temporal feature encoder and a data feature encoder. The time feature encoder can simply describe the relative time difference from the past time, such as ftime (tN )=0, ftime (tN-1 )=tN -tN-1 , etc. This is particularly important in the case of input variable vector time series where time steps vary between input variable vectors.
此外，可以为数据特征编码器设置神经网络，以进行数据特征fdata的特征提取。对每个时间步骤t1...tN都得出状态向量由此可以改变状态变量向量z的维度，从而可以减少整个网络的参数数量。In addition, a neural network can be set up for the data feature encoder to perform feature extraction of the data feature fdata . The state vector is obtained for each time step t1...tN This can change the dimension of the state variable vector z, thereby reducing the number of parameters of the entire network.
在初始创建状态向量的第一集合后，由一个或多个串联布置的多头自注意力模块进行进一步处理。多头自注意力模块的每个输入向量和输出向量都具有相同的维度，从而可以设置任意深度和数量的串行多头自注意力模块。Create a state vector initially After the first set, further processing is performed by one or more multi-head self-attention modules arranged in series. Each input vector and output vector of the multi-head self-attention module have the same dimension, so that any depth and number of serial multi-head self-attention modules can be set up.
多头自注意力模块具有M个并行的自注意力单元(头)。每个自注意力单元可以处理输入变量向量的时间序列的不同特征。由于M个并行的自注意力单元各自提供与对应的输入变量向量具有相同维度的输出向量，因此可以将多个自注意力单元的输出向量组合为后续的状态变量向量，通常使用基于数据的模型，例如神经网络进行组合。The multi-head self-attention module has M parallel self-attention units (heads). Each self-attention unit can process different features of the time series of input variable vectors. Since the M parallel self-attention units each provide an output vector with the same dimensions as the corresponding input variable vector, the output vectors of multiple self-attention units can be combined into subsequent state variable vectors, usually using a data-based model , such as neural networks for combination.
每个自注意力单元以本身已知的方式利用查询键值三元组处理状态变量向量z。然后计算自注意力分数，该自注意力分数分别说明特定时间步骤的每个状态变量向量依赖于其他时间步骤的其他状态向量的程度。Each self-attention unit processes the state variable vector z using query key-value triples in a manner known to itself. A self-attention score is then calculated, which separately accounts for the extent to which each state variable vector at a specific time step depends on other state vectors at other time steps.
由于状态变量向量的序列(状态变量向量的索引)是对应于时间步骤t1...tN布置的，因此可以使用掩蔽，使得一个时间步骤的状态变量向量不能与未来时间步骤的状态变量向量相关。因此，所产生的(每个状态变量向量相对于另一个状态变量向量的分数的)自注意力分数矩阵使用掩蔽函数加以掩蔽，即(zi到zj)＝0自注意力分数矩阵S被计算为Since the sequence of state variable vectors (indexes of state variable vectors) is arranged corresponding to time steps t1...tN, masking can be used so that the state variable vector of one time step cannot be related to the state variable vector of a future time step. Therefore, the resulting matrix of self-attention scores (the fraction of each state variable vector relative to the other state variable vector) is masked using a masking function, i.e. (zi to zj ) = 0 The self-attention score matrix S is calculated as
S＝softmax(QKT+掩蔽)S＝softmax(QKT + mask)
矩阵Q、K和V由时间序列的输入变量向量确定。一般来说，矩阵Q对所讨论的输入变量向量涉及哪些其他输入变量向量进行编码，而矩阵K对输入向量代表什么进行编码。通过相乘获得自注意力分数S。矩阵Q被确定为Q＝T_q*Z，其形式为维度为dim x N的矩阵。dim对应于输入侧状态变量向量的大小，并且N对应于时间序列的长度。T_q对应于大小为dim xdim_input的经过学习的模型参数，并且Z对应于维度为dim_input x N的输入变量向量x的矩阵。The matrices Q, K and V are determined by the input variable vectors of the time series. In general, the matrix Q encodes what other input variable vectors the input variable vector in question refers to, while the matrix K encodes what the input vector represents. The self-attention score S is obtained by multiplying. The matrix Q is determined as Q=T_q*Z, and its form is a matrix with dimensions dim x N. dim corresponds to the size of the input-side state variable vector, and N corresponds to the length of the time series. T_q corresponds to the learned model parameters of size dim xdim_input, and Z corresponds to the matrix of input variable vectors x of dimension dim_input x N.
类似地，矩阵K被计算为K＝T_k*Z，其形式为维度尺寸为dim x N的矩阵。dim对应于输入侧状态变量向量的大小，并且N对应于时间序列的长度。T_k对应于大小为dim xdim_input的经过学习的模型参数，并且Z对应于维度为dim_input x N的输入变量向量x的矩阵。以相同的方式确定矩阵V。Similarly, matrix K is calculated as K=T_k*Z, which is in the form of a matrix with dimensions dim x N. dim corresponds to the size of the input-side state variable vector, and N corresponds to the length of the time series. T_k corresponds to the learned model parameters of size dim xdim_input, and Z corresponds to a matrix of input variable vectors x of dimension dim_input x N. Determine the matrix V in the same way.
在训练异常预测模型时训练模型参数T_q、T_k、T_v。The model parameters T_q, T_k, and T_v are trained when training the anomaly prediction model.
例如，针对其中发生了待预测的故障类型的故障(例如突然死亡、热失控、过快老化等)的设备电池形成训练数据集。所述训练数据集将故障分配给故障类别，所述故障类别说明该故障对应于哪种故障类型以及在最后考虑的输入变量向量之后何时可能出现该故障。选择输入变量向量的时间序列，使得所述时间序列中的最近时间点位于故障事件的时间点之前分配给所述故障类别的持续时间附近。For example, a training data set is formed for device batteries in which a failure of the type of failure to be predicted occurs (eg, sudden death, thermal runaway, excessive aging, etc.). The training data set assigns faults to fault categories that state which fault type the fault corresponds to and when the fault is likely to occur after the last considered vector of input variables. The time series of the input variable vectors is chosen such that the most recent time point in said time series is located close to the duration assigned to said fault category before the time point of the fault event.
异常预测模型的输入变量向量的时间序列用于训练，该训练根据这样定义的故障类别来预测故障事件到特定未来事件的预期概率。为此，所述预测模型可以被构造为传统的分类模型。The time series of the input variable vectors of the anomaly prediction model are used for training that predicts the expected probability of a fault event to a specific future event based on the fault category defined in this way. To this end, the prediction model can be constructed as a traditional classification model.
向所述预测模型的模型输出的每个故障类别分配特定的故障类型以及特定故障类型的故障预计在输入变量向量的时间序列中的最近时间点之后出现的持续时间。为此，将现场设备电池的异常事件用作训练数据，并将异常(即特定故障类型)的对应出现概率标记为1，同时为该训练数据集考虑的输入变量向量的时间序列是结束时间点对应于以下时间点的时间序列，该时间点位于确定所述特定故障类型的故障出现之前分配给对应故障类别的持续时间附近。Each fault category of the model output of the predictive model is assigned a specific fault type and the duration within which a fault of the specific fault type is expected to occur after the most recent time point in the time series of the input variable vector. For this purpose, abnormal events of field device batteries are used as training data, and the corresponding occurrence probability of anomalies (i.e., specific fault types) is marked as 1, while the time series of input variable vectors considered for this training data set is the end time point A time sequence corresponding to time points located in the vicinity of the duration assigned to the corresponding fault category before the occurrence of a fault of that particular fault type was determined to occur.
可以使用二元交叉熵损失函数作为用于训练的损失函数：You can use the binary cross-entropy loss function as the loss function for training:
其中yt对应于故障类别的类别标签，并且B是训练数据集的数量。where yt corresponds to the class label of the fault class, and B is the number of training data sets.
所述训练可以基于其中已确定了异常事件的训练数据集和其中未确定异常事件的训练数据两者。这里然后将所有故障类别的对应标签设置为“0”。可以通过损失函数中合适的加权因子来考虑异常事件和常规电池功能的不同数量的训练数据集。The training may be based on both training data sets in which anomalous events have been identified and training data in which no anomalous events have been identified. Here the corresponding labels of all fault categories are then set to "0". Different numbers of training data sets for abnormal events and regular battery functions can be taken into account through suitable weighting factors in the loss function.
在训练异常预测模型之后，可以将对应的模型参数传输到技术设备中，从而可以在那里执行异常预测模型，以便能够对应于预给定的预测时空立即预测性地确定对应设备电池中的异常。After training the anomaly prediction model, the corresponding model parameters can be transferred to the technical device so that the anomaly prediction model can be executed there so that anomalies in the battery of the corresponding device can be immediately and predictively determined corresponding to a predetermined prediction time and space.
替代地，异常预测模型也可以在中央单元中执行。Alternatively, the anomaly prediction model can also be executed in a central unit.
可以规定，如果对异常预测模型的评估得出对应故障类别的概率高于预给定阈值，则根据输入变量向量的时间序列识别出在特定的持续时间后设备电池中特定故障的出现，其中在确定了出现所述特定故障时用以下方式发信号通知，特别是向用户输出警告。It can be provided that if the evaluation of the anomaly prediction model leads to the probability of the corresponding fault category being higher than a predetermined threshold, then the occurrence of a specific fault in the device battery after a specific duration is identified based on the time series of the input variable vector, where The occurrence of the specified fault is determined to be signaled in the following manner, in particular by outputting a warning to the user.
如果基于借助于输入变量向量的时间序列的评估确定在未来的特定时间点将以特定概率出现异常，即特定故障，则可以向技术设备的用户输出警告。如果基于分配给故障类别的故障类型确定了安全关键的异常，则可以在设备外部通知该异常和/或设备电池可以例如通过快速放电进入安全状态，以导出电池中存储的能量。If it is determined based on the evaluation of the time series with the aid of the input variable vector that an anomaly, ie a specific fault, will occur at a specific point in time in the future with a specific probability, a warning can be output to the user of the technical device. If a safety-critical anomaly is determined based on the fault type assigned to the fault category, this anomaly can be notified externally to the device and/or the device battery can be brought into a safe state, for example by rapid discharge, to derive the energy stored in the battery.
所述异常预测模型可以在定期的时间点重新加以训练。特别地，如果与中央单元连接的多个设备电池之一中出现了新的异常事件，则可以重新训练所述异常预测模型。The anomaly prediction model can be retrained at regular time points. In particular, if a new abnormal event occurs in one of the multiple device batteries connected to the central unit, the abnormality prediction model can be retrained.
附图说明Description of drawings
下面基于附图更详细地解释实施方式。Embodiments are explained in more detail below based on the drawings.
图1示出了用于在中央单元中提供驾驶员特定和车辆特定的运行变量以识别车辆电池的预测异常的系统的示意图；Figure 1 shows a schematic diagram of a system for providing driver-specific and vehicle-specific operating variables in a central unit for identifying predicted anomalies of a vehicle battery;
图2示出了用于预测性识别设备电池中的故障的系统的功能结构的示意图；以及Figure 2 shows a schematic diagram of the functional structure of a system for predictive identification of faults in a device battery; and
图3示出了自注意力单元的功能结构的示意图。Figure 3 shows a schematic diagram of the functional structure of the self-attention unit.
具体实施方式Detailed ways
下面基于车辆电池描述根据本发明的方法，所述车辆电池作为在作为相同类型的设备的大量机动车辆中的设备电池。为此，在中央单元中基于运行变量变化过程对一个或多个电化学电池模型进行评估和参数化。在所述中央单元中训练异常预测模型，并且将异常预测模型的模型参数传输到车队中的车辆的控制设备中，使得在那里可以通过对异常预测模型的连续评估来识别异常。The method according to the invention is described below on the basis of a vehicle battery as an equipment battery in a large number of motor vehicles as equipment of the same type. For this purpose, one or more electrochemical cell models are evaluated and parameterized in a central unit based on the evolution of the operating variables. An anomaly prediction model is trained in said central unit and the model parameters of the anomaly prediction model are transmitted to the control devices of the vehicles in the fleet, so that anomalies can be identified there by continuous evaluation of the anomaly prediction model.
上述示例代表了大量具有独立于电网的能量供应的固定或移动设备，例如车辆(电动车辆、电动自行车等)、设施、机床、家用电器、物联网设备等，它们经由对应的通信连接(例如LAN、互联网)与设备外部的中央单元(云)连接。The above examples represent a large number of fixed or mobile devices with a grid-independent energy supply, such as vehicles (electric vehicles, e-bikes, etc.), facilities, machine tools, household appliances, IoT devices, etc., which are connected via corresponding communication connections (e.g. LAN , Internet) is connected to a central unit (the cloud) external to the device.
图1示出了用于在中央单元2中收集车队数据以创建、运行和评估分别用于对车辆电池的内部电池状态进行建模的电化学电池模型和电池性能模型和用于确定机动车辆中车辆电池的老化状态的老化状态模型的系统1。Figure 1 illustrates the method used to collect fleet data in a central unit 2 for the creation, operation and evaluation of an electrochemical cell model and a battery performance model, respectively, for modeling the internal battery state of a vehicle battery and for determining the internal battery state of a vehicle battery. System 1 of the aging state model of vehicle battery aging state.
图1示出了具有多个机动车辆4的车队3。机动车辆4中的一个在图1中更详细地示出。机动车辆4分别具有带有电池单池45的车辆电池41、电驱动马达42和控制单元43。控制单元43与通信装置44连接，该通信装置44适用于在相应机动车辆4和中央单元2(所谓的云)之间传输数据。FIG. 1 shows a fleet 3 with a plurality of motor vehicles 4 . One of the motor vehicles 4 is shown in greater detail in FIG. 1 . The motor vehicle 4 each has a vehicle battery 41 with a battery cell 45 , an electric drive motor 42 and a control unit 43 . The control unit 43 is connected to a communication device 44 which is suitable for transmitting data between the respective motor vehicle 4 and the central unit 2 (the so-called cloud).
控制单元43特别是被构造为以高时间分辨率(例如在1和50Hz之间，例如10Hz)采集借助于电池管理系统46采集的车辆电池41的运行变量，并经由通信装置44将所述运行变量传送到中央单元2。The control unit 43 is configured in particular to acquire operating variables of the vehicle battery 41 acquired by means of the battery management system 46 with a high temporal resolution (for example between 1 and 50 Hz, for example 10 Hz) and to transmit these operating variables via the communication device 44 The variables are transferred to central unit 2.
机动车辆4将运行变量F发送到中央单元2，所述运行变量至少说明影响车辆电池41的老化状态或者被车辆电池41的老化状态影响并且是确定内部电池状态、老化状态和参数化电化学电池模型所需要的变量。在车辆电池的情况下，运行变量F可以说明在电池包层面、模块层面和/或单池层面的瞬时电池电流、瞬时电池电压、瞬时电池温度和瞬时充电状态(SOC：State of Charge)。运行变量F以0.1Hz至50Hz的快速时间栅格采集为运行变量变化过程并且可以以未压缩和/或压缩的形式定期传输到中央单元2。例如，使用压缩算法以最小化至中央单元2的数据流量，可以以10分钟到数个小时的间隔将时间序列逐块地传输到中央单元2。The motor vehicle 4 transmits operating variables F to the central unit 2 , which at least describe or are affected by the aging state of the vehicle battery 41 and are responsible for determining the internal battery state, the aging state and parameterizing the electrochemical cell. variables required by the model. In the case of vehicle batteries, operating variables F can account for instantaneous battery current, instantaneous battery voltage, instantaneous battery temperature and instantaneous state of charge (SOC: State of Charge) at the battery pack level, module level and/or cell level. The operating variable F is acquired as an operating variable profile on a fast time grid from 0.1 Hz to 50 Hz and can be transmitted periodically to the central unit 2 in uncompressed and/or compressed form. For example, the time series can be transmitted to the central unit 2 block by block at intervals ranging from 10 minutes to several hours, using a compression algorithm to minimize data traffic to the central unit 2 .
中央单元2具有数据处理单元21和数据库22，在数据处理单元21中可以执行下面描述的方法的一部分，而数据库22用于存储数据点、模型参数、状态等。The central unit 2 has a data processing unit 21 in which part of the method described below can be carried out, and a database 22 for storing data points, model parameters, states, etc.
中央单元2被构造为接收运行变量变化过程。中央单元2可以从相应车辆电池41的运行变量变化过程中以本身已知的方式确定当前老化状态(例如借助于老化状态模型)、一个或多个内部电池状态(例如借助于电化学电池模型和/或作为电池性能模型的模型参数)和/或作为聚合或累积的变量或基于直方图的变量的一个或多个运行特征作为导出的变量。The central unit 2 is designed to receive operating variable changes. The central unit 2 can determine the current aging state (for example using an aging state model), one or more internal battery states (for example using an electrochemical cell model and /or as a model parameter of a battery performance model) and/or as an aggregated or accumulated variable or one or more operating characteristics of a histogram-based variable as a derived variable.
图2以示例的方式示意性地示出了用于预测性识别车辆电池中的异常的系统10的功能结构。该系统在中央单元2中被实现为软件或硬件，并且评估说明诸如电池电流、电池电压、充电状态和电池温度的运行变量变化过程的运行变量的时间序列。FIG. 2 schematically shows by way of example the functional structure of a system 10 for predictive identification of anomalies in a vehicle battery. The system is implemented as software or hardware in the central unit 2 and evaluates time series of operating variables describing the course of operating variables such as battery current, battery voltage, state of charge and battery temperature.
从运行变量变化过程F中形成离散时间步骤t1...tN处的输入变量向量xt1...xtN的时间序列。为此，可以借助于一个或多个电池模型和特征提取模型来预处理运行变量变化过程，以提供老化状态和/或一个或多个内部电池状态和/或一个或多个运行特征。可以包括一个或多个内部电池状态、一个或多个运行特征以及在时间步骤处的一个或多个运行变量的老化状态可以形成在时间步骤处的输入变量向量。A time series of input variable vectors xt1 ...x tN at discrete time steps t1 ...tN is formed from the operating variable change process F. For this purpose, the operating variable profile can be preprocessed by means of one or more battery models and feature extraction models to provide an aging state and/or one or more internal battery states and/or one or more operating characteristics. An aging state that may include one or more internal battery states, one or more operating characteristics, and one or more operating variables at a time step may form an input variable vector at a time step.
电池模型可以包括以下模型中的一个或多个：老化状态模型11、电化学电池模型12和电化学性能模型13。The battery model may include one or more of the following models: aging state model 11, electrochemical cell model 12, and electrochemical performance model 13.
从而可以在中央单元2中实现老化状态模型11，该老化状态模型作为混合模型部分地基于数据。老化状态模型11可以定期地，即例如在相应的评估持续时间结束后，用于基于运行变量的时间变化过程(分别自相应的车辆电池投入运行以来或基于已知电池状态的状态)和由此确定的运行特征M确定相关联车队3的所涉及车辆电池41的瞬时老化状态。It is thus possible to implement an aging state model 11 in the central unit 2 which is partly based on data as a hybrid model. The aging state model 11 can be used periodically, ie, for example after the end of the respective evaluation period, based on the time course of the operating variables (in each case since the respective vehicle battery was put into operation or on the basis of a known battery state) and thereby The determined operating characteristic M determines the instantaneous aging state of the vehicle battery 41 in question of the associated fleet 3 .
老化状态模型11包括物理老化模型11a和校正模型11b。物理老化模型11a是基于微分方程并通过时间积分方法来计算物理老化状态的非线性数学模型。特别是自设备电池的使用寿命开始以来，对具有运行变量变化过程的老化状态模型11的物理老化模型11a的评估导致产生物理微分方程的方程组的内部状态，该内部状态对应于设备电池的物理内部状态。由于物理老化模型11a基于物理和电化学定律，因此物理老化模型的模型参数是说明物理属性的变量。The aging state model 11 includes a physical aging model 11a and a correction model 11b. The physical aging model 11a is a nonlinear mathematical model based on differential equations and uses a time integration method to calculate the physical aging state. In particular since the beginning of the service life of the device battery, the evaluation of the physical aging model 11a with the aging state model 11 of the operating variable evolution process leads to the generation of an internal state of the system of physical differential equations, which internal state corresponds to the physical state of the device battery. internal state. Since the physical aging model 11a is based on the laws of physics and electrochemistry, the model parameters of the physical aging model are variables that describe physical properties.
因此，车辆电池41的运行变量F的时间序列直接进入物理老化状态模型11a，该物理老化状态模型优选被设计为电化学模型并且借助于非线性微分方程和多维状态向量建模对应的内部电化学电池状态，如层厚度(例如SEI厚度)、由于阳极/阴极副反应引起的可环化锂的变化、电解质的快速消耗、电解质的缓慢消耗、阳极中活性材料的损失、阴极中活性材料的损失等。因此，物理老化模型对应于电化学电池模型的变体。Therefore, the time series of the operating variables F of the vehicle battery 41 directly enters the physical aging state model 11a, which is preferably designed as an electrochemical model and models the corresponding internal electrochemistry by means of nonlinear differential equations and multidimensional state vectors. Battery status such as layer thickness (e.g. SEI thickness), changes in cyclizable lithium due to anode/cathode side reactions, rapid electrolyte consumption, slow electrolyte consumption, loss of active material in the anode, loss of active material in the cathode wait. Therefore, the physical aging model corresponds to a variant of the electrochemical cell model.
然而，物理老化状态的由物理老化模型11a提供的模型值在特定状况下是不准确的，并且因此可以规定使用校正变量来校正这些模型值。校正变量由基于数据的校正模型11b提供，该校正模型借助于来自车队3的车辆4的训练数据集和/或借助于实验室数据加以训练。特别地，物理老化状态和校正变量在求和块中相加或者在其他情况下也相乘(未示出)，以输出老化状态作为车辆电池的状态变量。However, the model values of the physical aging state provided by the physical aging model 11a are inaccurate in certain situations, and therefore the use of correction variables can be provided to correct these model values. The correction variables are provided by a data-based correction model 11 b which is trained with the aid of training data sets from the vehicles 4 of the fleet 3 and/or with the aid of laboratory data. In particular, the physical aging state and the correction variable are added in a summation block or otherwise also multiplied (not shown) to output the aging state as a state variable of the vehicle battery.
校正模型11b在输入侧获得运行特征M，所述运行特征是借助于特征提取块14从运行变量/单池运行变量F的变化过程中确定的，并且还可以包括物理模型的微分方程组的一个或多个内部电化学状态。此外，校正模型11b可以在输入侧获得从物理老化模型11a获得的物理老化状态。The calibration model 11b obtains on the input side an operating characteristic M which is determined from the course of the operating variable/single-pool operating variable F by means of a feature extraction block 14 and may also comprise one of the system of differential equations of the physical model or multiple internal electrochemical states. Furthermore, the correction model 11b can obtain the physical aging state obtained from the physical aging model 11a on the input side.
当前评估时间段的运行特征M可以在特征提取块14中基于随时间的运行变量变化过程F产生。运行特征M还包括来自电化学物理老化模型11a的状态向量的内部状态并且有利地包括物理老化状态。The operating characteristics M of the current evaluation time period can be generated in the feature extraction block 14 based on the operating variable variation process F over time. The operating characteristic M also includes internal states from the state vector of the electrochemical-physical aging model 11a and advantageously includes physical aging states.
特征提取模型14使得能够将运行变量变化过程聚合为运行特征，例如上面关于混合老化状态模型所描述的。特别地，运行特征可以包括状态特性和基于直方图的特征。The feature extraction model 14 enables the aggregation of operating variable variation processes into operating features, such as described above with respect to the hybrid aging state model. In particular, operational features may include state characteristics and histogram-based features.
运行特征M可以包括例如涉及评估时间段的特征和/或在评估时间段期间累积的特征和/或在整个迄今为止的使用寿命期间确定的统计变量。特别地，运行特征可以包括例如：电化学状态(例如SEI层厚度、由于阳极/阴极副反应引起的可循环锂的变化、电解质溶剂的快速吸收、电解质溶剂的缓慢吸收、锂沉积、阳极活性材料的损失和阴极活性材料的损失)、关于阻抗或内阻的信息、直方图特征(例如温度对充电状态、充电电流对温度和放电电流对温度，特别是关于电池温度分布对充电状态、充电电流分布对温度和/或放电电流分布对温度的多维直方图数据)、以安培小时为单位的电流吞吐量、累积的总电荷(Ah)、充电过程中的平均容量增加(特别是对于电荷增加量高于总电池容量的阈值份额(例如20％ΔSOC)的充电过程而言)、充电容量以及在充电状态具有足够大冲程的测量的充电过程期间的微分容量极值(例如最大值)(dQ/dU的平滑的变化过程：电荷变化除以电池电压变化)或累计行驶里程。这些变量优选被换算为，使得所述变量尽可能最好地表征真实的使用行为并在特征空间中进行归一化。运行特征M可以全部或仅部分用于校正模型11b。The operating characteristics M may include, for example, characteristics related to the evaluation period and/or characteristics accumulated during the evaluation period and/or statistical variables determined over the entire service life to date. In particular, operating characteristics may include, for example: electrochemical state (e.g., SEI layer thickness), changes in recyclable lithium due to anode/cathode side reactions, rapid uptake of electrolyte solvent, slow uptake of electrolyte solvent, lithium deposition, anode active material loss and loss of cathode active material), information about impedance or internal resistance, histogram characteristics (such as temperature versus charge state, charge current versus temperature, and discharge current versus temperature, especially regarding battery temperature distribution versus charge state, charge current Multidimensional histogram data of distribution versus temperature and/or discharge current distribution versus temperature), current throughput in ampere hours, total charge accumulated (Ah), average capacity increase during charging (especially for charge increase (dQ/ Smooth change process of dU: charge change divided by battery voltage change) or accumulated mileage. These variables are preferably scaled in such a way that they characterize the real usage behavior as best as possible and are normalized in the feature space. The operating characteristics M can be used entirely or only partially to calibrate the model 11b.
在中央单元2中训练混合老化状态模型。为此，定义将单池运行变量变化过程分配给作为标签的凭经验确定的或基于模型确定的老化状态的训练数据集。这些训练数据集用于拟合物理老化模型的参数并基于剩余残留物训练校正模型。The hybrid aging state model is trained in central unit 2. For this purpose, a training data set is defined that assigns single-pool operating variable change processes to empirically determined or model-determined aging states as labels. These training data sets were used to fit the parameters of the physical aging model and train the calibration model based on the remaining residues.
作为标签的老化状态的确定可以以本身已知的方式通过在产生标签的定义的负载和环境条件下在车辆中或中央单元2中使用附加的老化模型评估运行变量变化过程进行，例如在车间中、在测试台上或在诊断或标签产生模式下，所述诊断或标签产生模式是一种运行模式并且保证符合车辆电池的预定运行条件，例如恒温、恒流等。例如，可以通过用于确定车辆电池的剩余总容量的哥伦计数来确定老化状态。The determination of the aging state of a tag can be carried out in a manner known per se by using an additional aging model in the vehicle or in the central unit 2 to evaluate the evolution of the operating variables under defined load and ambient conditions for which the tag was generated, for example in a workshop. , on a test bench or in a diagnostic or tag generation mode, which is an operating mode and guaranteed to comply with the predetermined operating conditions of the vehicle battery, such as constant temperature, constant current, etc. For example, the aging state may be determined by the Colon count used to determine the total remaining capacity of the vehicle battery.
此外，可以在中央单元2中使用电化学电池模型以对内部电池状态进行建模。所述电化学电池模型基于具有多个非线性微分方程的微分方程组。运行变量数据使得能够借助于时间积分方法对当前电池状态进行建模。这种电化学电池模型例如由印刷文献US 2020/150185A、WO2022017984A1、EP3919925A1、DE102020206915B3和US20210373082A1公知。Furthermore, an electrochemical cell model can be used in the central unit 2 to model the internal cell states. The electrochemical cell model is based on a system of differential equations with multiple nonlinear differential equations. The operating variable data enable modeling of the current battery state with the help of time integration methods. Such electrochemical cell models are known, for example, from the printed documents US 2020/150185A, WO2022017984A1, EP3919925A1, DE102020206915B3 and US20210373082A1.
此外，可以以电化学电池模型的形式设置电池性能模型13，其通过其模型参数来表征并且对电池单池的内部状态进行建模。电化学电池模型基于由模型参数参数化的电化学模型方程，这些电化学模型方程表征非线性微分方程组的电化学状态并且可以根据时间积分方法加以连续评估。电化学电池性能模型通常对应于将电池电流和电池温度分配给电池电压的观测器模型，其目标是描述电池的动态性。Furthermore, the battery performance model 13 can be provided in the form of an electrochemical cell model, which is characterized by its model parameters and models the internal states of the battery cells. Electrochemical cell models are based on electrochemical model equations parameterized by model parameters, which characterize the electrochemical state of a system of nonlinear differential equations and can be continuously evaluated according to time integration methods. Electrochemical cell performance models typically correspond to observer models that assign cell current and cell temperature to cell voltage, with the goal of describing the dynamics of the cell.
电化学电池性能模型13可以在静止阶段针对运行变量变化过程拟合，例如借助于最小二乘法等。基于电池性能模型13，可以针对电池单池基于拟合的电化学模型参数来确定老化状态。电化学电池性能模型14可以对电池状态进行建模并且通过模型参数(特别是平衡参数和动力学参数)来描述。模型参数可以通过拟合以规律的间隔重新参数化，特别是如果在至少几个(例如三个)小时的定义时间段内存在具有高采样率的运行变量变化过程的话。电池性能模型14的这种模型参数可以被解释为电池状态。The electrochemical cell performance model 13 can be fitted to the changing process of operating variables during the stationary phase, for example by means of the least squares method or the like. Based on the battery performance model 13, the aging state can be determined for the battery cell based on the fitted electrochemical model parameters. The electrochemical cell performance model 14 may model the cell state and describe it by model parameters, in particular equilibrium parameters and kinetic parameters. The model parameters can be reparameterized at regular intervals by fitting, especially if there is a process of operating variable variation with a high sampling rate over a defined time period of at least several (e.g. three) hours. Such model parameters of the battery performance model 14 can be interpreted as battery status.
从在特定时间点/时间步骤对上述模型的评估产生异常预测模型15的输入变量，所述输入变量包括以下变量中的一个或多个：老化状态、电化学电池模型的一个或多个内部电池状态、电池性能模型的一个或多个模型参数和/或一个或多个运行特征。此外，异常预测模型15的输入变量包括运行状态变化过程的运行变量，特别是在两个彼此相继的时间步骤之间的持续时间段上平均的运行变量。The input variables of the anomaly prediction model 15 are generated from the evaluation of the above model at specific time points/time steps, said input variables including one or more of the following variables: aging state, one or more internal cells of the electrochemical cell model state, one or more model parameters of the battery performance model, and/or one or more operating characteristics. Furthermore, the input variables of the anomaly prediction model 15 include operating variables of the operating state change process, in particular operating variables averaged over the time period between two successive time steps.
异常预测模型15包括时间序列变换器模型16和基于数据的预测模型17。The anomaly prediction model 15 includes a time series transformer model 16 and a data-based prediction model 17 .
以彼此相继的时间步骤的输入变量向量xt1、xt2、...xtn的时间序列{xt1，xt2，...xtn}的形式提供异常预测模型15的输入变量。The input variables of the anomaly prediction model 15 are provided in the form of a time series {xt1 , xt2 , ...xtn } of input variable vectors xt1 , xt2 , ...xtn of successive time steps.
在时间序列变换器模型16中，首先在预处理块16a中将输入变量向量的时间序列{xt1，xt2，...xtn}处理为N个状态变量向量的第一集合。In the time series transformer model 16, the time series of input variable vectors {xt1 , xt2 , ... xtn } is first processed into N state variable vectors in a preprocessing block 16 a of the first collection.
这是通过时间特征编码器ZM和数据特征编码器DM来实现的。时间特征编码器ZM可以简单地说明过去时间的相对时间差，例如ftime(tN)＝0，(tN-1)＝tN-tN-1等。This is achieved through the temporal feature encoder ZM and the data feature encoder DM. The temporal feature encoder ZM can simply illustrate the relative time difference in the past time, such as ftime (tN )=0, (tN-1 )=tN -tN-1 , etc.
数据特征编码器DM可以被设置为基于数据的模型，例如神经网络，以执行数据特征fdata的特征提取。所述特征提取例如可以用于减小维度。在每个时间步骤t1...tN为每个输入变量向量{xt1，xt2，...xtn}产生第一状态变量向量由此可以改变状态向量z的维度，从而可以减少整个网络的参数数量。The data feature encoder DM can be set as a data-based model, such as a neural network, to perform feature extraction of the data feature fdata . The feature extraction can be used, for example, to reduce dimensions. At each time step t1...tN generates a first state variable vector for each input variable vector {xt1 , xt2 ,...xtn } This can change the dimension of the state vector z, thereby reducing the number of parameters of the entire network.
第一状态变量向量的该集合由一个或多个(数量S)多头自注意力模块16b串行处理。每个多头自注意力模块16b将第s级的相应状态变量向量/>处理为输出侧相同维度的状态变量向量/>使得多头自注意力模块16b可以以任意数量S并排排列。first state variable vector This set of is processed serially by one or more (number S) multi-head self-attention modules 16b. Each multi-head self-attention module 16b converts the corresponding state variable vector of level s/> Processed as a state variable vector of the same dimension on the output side/> This allows the multi-head self-attention modules 16b to be arranged side by side in any number S.
每个多头自注意力模块16b具有M个并行的自注意力单元16c，所述自注意力单元考虑相应输入侧状态变量向量的集合的不同特征。每个自注意力单元16c的输入向量和输出向量具有相同的维度，所述维度可以借助于基于数据的模型(例如深度神经网络)形式的简化模型16d减少为状态变量向量/>的维度。Each multi-head self-attention module 16b has M parallel self-attention units 16c, which consider the corresponding input-side state variable vectors. collection of different characteristics. The input and output vectors of each self-attention unit 16c have the same dimensions, which can be reduced to state variable vectors by means of a simplified model 16d in the form of a data-based model (eg a deep neural network). dimensions.
如图3中所示，每个自注意力单元16c将状态变量向量的集合处理为矩阵Q，以获得QKT形式的矩阵。As shown in Figure 3, each self-attention unit 16c converts the state variable vector The set of is processed as a matrix Q to obtain a matrix of the form QKT.
每个自注意力单元16c以本身已知的方式利用Query(Q)-Key(K)-Value(V)三元组来处理状态变量向量Each self-attention unit 16c processes the state variable vector using Query(Q)-Key(K)-Value(V) triples in a manner known per se
矩阵Q、K和V由时间序列的输入变量向量(维度dim_input)确定。一般来说，矩阵Q对所讨论的输入变量向量涉及哪些其他输入变量向量进行编码，而矩阵K对输入变量向量的表示进行编码。通过相乘获得自注意力分数S。矩阵Q被确定为Q＝T_q*Z，其形式为维度为dim x N的矩阵。dim对应于输入侧状态变量向量的大小，并且N对应于时间序列的长度。T_q对应于大小为dim x dim_input的经过学习的模型参数，并且Z对应于维度为dim_input xN的输入变量向量x的矩阵。The matrices Q, K and V are determined by the input variable vector (dimension dim_input) of the time series. In general, the matrix Q encodes which other input variable vectors are involved in the input variable vector in question, while the matrix K encodes the representation of the input variable vector. The self-attention score S is obtained by multiplying. The matrix Q is determined as Q=T_q*Z, and its form is a matrix with dimensions dim x N. dim corresponds to the size of the input-side state variable vector, and N corresponds to the length of the time series. T_q corresponds to the learned model parameters of size dim x dim_input, and Z corresponds to a matrix of input variable vectors x of dimension dim_input xN.
类似地，矩阵K被计算为K＝T_k*Z，其形式为维度为dim x N的矩阵。dim对应于输入侧状态变量向量的大小，并且N对应于时间序列的长度。T_k对应于大小为dim x dim_input的经过学习的模型参数，并且Z对应于维度为dim_input x N的输入变量向量x的矩阵。Similarly, the matrix K is calculated as K=T_k*Z, which is in the form of a matrix of dimension dim x N. dim corresponds to the size of the input-side state variable vector, and N corresponds to the length of the time series. T_k corresponds to the learned model parameters of size dim x dim_input, and Z corresponds to a matrix of input variable vectors x of dimension dim_input x N.
矩阵V被确定为V＝T_v*Z，其形式为维度为dim×N的矩阵。dim对应于输入侧状态变量向量的大小，并且N对应于时间序列的长度。T_v对应于大小为dim x dim_input的经过学习的模型参数，并且Z对应于维度为dim_input x N的输入变量向量x的矩阵。The matrix V is determined as V=T_v*Z, and its form is a matrix with dimensions dim×N. dim corresponds to the size of the input-side state variable vector, and N corresponds to the length of the time series. T_v corresponds to the learned model parameters of size dim x dim_input, and Z corresponds to a matrix of input variable vectors x of dimension dim_input x N.
在训练异常预测模型15时，训练模型参数T_q、T_k、T_v。When training the anomaly prediction model 15, model parameters T_q, T_k, and T_v are trained.
然后以自注意力分数矩阵SAS的形式计算自注意力分数，该自注意力分数分别说明特定时间步骤的每个状态变量向量依赖于其他时间步骤的其他状态向量的程度。The self-attention scores are then calculated in the form of a self-attention score matrix SAS, which separately accounts for each state variable vector at a specific time step. other state vectors that depend on other time steps Degree.
为此，对所产生的矩阵进行掩蔽。为此，使用通过掩蔽矩阵MASK的掩蔽，该掩蔽矩阵考虑了与时间步骤有关的矩阵元素，使得当前状态变量向量的评估度量/分数不能依赖于未来的状态变量向量。为此，向该矩阵的对应元素分配-∞。To do this, the resulting matrix is masked. For this purpose, masking by the masking matrix MASK is used, which takes into account the matrix elements related to the time step such that the evaluation measure/score of the current state variable vector cannot depend on the future state variable vector. To do this, assign -∞ to the corresponding element of this matrix.
然后将softmax函数16e逐行应用于经过掩蔽的矩阵的元素，以获得自注意力分数矩阵SAS。The softmax function 16e is then applied row-wise to the elements of the masked matrix to obtain the self-attention score matrix SAS.
所产生的自注意力分数矩阵SAS通过Value矩阵V加权并且可选地通过特征提取模型16f进一步处理。The resulting self-attention score matrix SAS is weighted by the Value matrix V and optionally further processed by the feature extraction model 16f.
这些步骤对应于用于执行变换器模型的常规步骤，例如从Qingsong Wen等人的“Transformers in Time Series：A Survey”，arXiv：2202.07125中公知的。These steps correspond to conventional steps for performing transformer models, known for example from Qingsong Wen et al. "Transformers in Time Series: A Survey", arXiv:2202.07125.
在一个或多个多头自注意力模块16b的输出处获得的状态变量向量的集合现在借助于预测模型17转换为分类结果，所述分类结果说明了未来特定时间点出现特定故障类型的故障的预测。The state variable vector obtained at the output of one or more multi-head self-attention modules 16b The set of is now converted by means of the prediction model 17 into a classification result which illustrates the prediction of the occurrence of a fault of a specific fault type at a specific point in time in the future.
在创建了异常预测模型15之后，可以将模型参数传送到车辆4，使得可以在车辆4上执行异常预测模型15。可以在每次在中央单元2中更新了异常预测模型15之后在车辆4中更新模型参数。After the anomaly prediction model 15 is created, the model parameters can be transferred to the vehicle 4 so that the anomaly prediction model 15 can be executed on the vehicle 4 . The model parameters can be updated in the vehicle 4 each time the anomaly prediction model 15 is updated in the central unit 2 .
对车辆中的异常预测模型15的评估通过故障类别中的模型输出值得出关于在未来的预定时间点出现异常或特定故障类型的故障的概率或失效概率的说明。如果对于特定故障类别确定了高于特定阈值的概率，则可以例如以光信号或声信号的形式发出警告。替代地，如果预测的失效概率高于预给定阈值并且故障类别说明关键故障以及该故障很可能在短持续时间之后出现，则车辆电池41可以进入安全状态。The evaluation of the anomaly prediction model 15 in the vehicle leads to an indication of the probability or failure probability of an anomaly or a fault of a specific fault type occurring at a predetermined time point in the future through the model output values in the fault category. If a probability above a certain threshold value is determined for a certain fault category, a warning can be issued, for example, in the form of a light or acoustic signal. Alternatively, the vehicle battery 41 can enter a safe state if the predicted probability of failure is above a predetermined threshold and the fault category indicates a critical fault and that this fault is likely to occur after a short duration.
一般而言，可以基于训练数据集来训练异常预测模型15。这些训练数据集可以包括针对车队3中出现的故障情况的训练数据集。如果出现故障情况，则将“1”作为标签分配给说明特定的故障类型以及在其之后应当出现特定故障的持续时间的故障类别。将该标签分配给输入变量向量的时间序列，其最后一个时间步骤位于出现故障的时间点之前所述持续时间附近。此外，可以为按规定的电池创建训练数据集，即将标签“0”分配给所有故障类别。借助于基于梯度的训练方法(例如反向传播)以本身已知的方式训练异常预测模型15。In general, anomaly prediction models can be trained based on training data sets15. These training data sets may include training data sets for fault situations occurring in Fleet 3. If a fault condition occurs, a "1" is assigned as a label to a fault class that describes the specific fault type and the duration after which the specific fault should occur. Assign this label to a time series of input variable vectors whose last time step is located near the stated duration before the time point at which the failure occurs. Additionally, a training dataset can be created for batteries as specified, i.e. assigning the label "0" to all fault categories. The anomaly prediction model 15 is trained in a manner known per se by means of gradient-based training methods (eg backpropagation).
每当与中央单元2通信连接的车辆的车辆电池41中出现一个或多个新的异常时，可以重新训练异常预测模型15。与此无关地，附加地可以以规律的时间间隔(例如每六个月)重新训练异常预测模型15，以考虑整个车队的电池老化的新信息。The anomaly prediction model 15 may be retrained each time one or more new anomalies occur in the vehicle battery 41 of a vehicle communicatively connected to the central unit 2 . Independently of this, the anomaly prediction model 15 can additionally be retrained at regular intervals (eg every six months) to take into account new information on battery aging of the entire fleet.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| DE102022208932.4ADE102022208932A1 (en) | 2022-08-29 | 2022-08-29 | Method and device for the predictive diagnosis of a device battery of a technical device using a transformer model |
| DE102022208932.4 | 2022-08-29 |
| Publication Number | Publication Date |
|---|---|
| CN117630672Atrue CN117630672A (en) | 2024-03-01 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311091325.0APendingCN117630672A (en) | 2022-08-29 | 2023-08-28 | Method and device for predicting a device battery of a diagnostic technology device by means of a converter model |
| Country | Link |
|---|---|
| CN (1) | CN117630672A (en) |
| DE (1) | DE102022208932A1 (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119716576A (en)* | 2025-02-26 | 2025-03-28 | 广州市凯捷电源实业有限公司 | Battery production charging and discharging status monitoring and early warning method and system |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117951585A (en)* | 2024-03-27 | 2024-04-30 | 国网山东省电力公司曲阜市供电公司 | Real-time detection method and system for operation state of power equipment |
| CN118362802B (en)* | 2024-04-17 | 2025-02-18 | 河北承捷电力设备制造有限公司 | A method and system for monitoring abnormal operation of transformer |
| CN118425790B (en)* | 2024-07-04 | 2024-09-20 | 成都赛力斯科技有限公司 | Battery state of charge estimation method, training method and device of model |
| CN118625156A (en)* | 2024-08-14 | 2024-09-10 | 长安绿电科技有限公司 | A lithium battery status monitoring method, device, medium and equipment |
| CN118940221B (en)* | 2024-10-15 | 2025-02-07 | 西安交通大学 | Multi-parameter fusion early warning method and system for thermal runaway of lithium iron phosphate battery |
| CN118965035B (en)* | 2024-10-15 | 2025-01-28 | 国网浙江省电力有限公司乐清市供电公司 | A photovoltaic string aging fault degree assessment method and system |
| CN119482866B (en)* | 2025-01-13 | 2025-07-18 | 深圳市氮化镓电子科技有限公司 | Charging method, device and charger based on real-time feedback control |
| CN119619810B (en)* | 2025-02-13 | 2025-06-03 | 中国安全生产科学研究院 | Method and device for identifying operation status of crane circuit |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11243258B2 (en) | 2018-11-13 | 2022-02-08 | Robert Bosch Gmbh | Method for approximating algorithms for fast charging li-ion batteries based on electrochemical battery models |
| DE102019208372A1 (en) | 2019-06-07 | 2020-12-10 | Robert Bosch Gmbh | Method and device for anomaly detection in technical systems |
| DE102020206592A1 (en) | 2020-05-27 | 2021-12-02 | Robert Bosch Gesellschaft mit beschränkter Haftung | Method and device for operating an electrically drivable motor vehicle as a function of a predicted state of aging of an electrical energy store |
| DE102020206915B3 (en) | 2020-06-03 | 2021-07-22 | Robert Bosch Gesellschaft mit beschränkter Haftung | Method for determining an approximation and / or prognosis for the true degradation state of a rechargeable battery and method for training a hidden Markov model for use in the method for determining an approximation and / or prognosis |
| DE102020209339A1 (en) | 2020-07-24 | 2022-01-27 | Robert Bosch Gesellschaft mit beschränkter Haftung | Method for predicting a remaining lifetime of vehicle batteries of a fleet of electric vehicles |
| CN114924192B (en)* | 2022-05-20 | 2024-09-06 | 重庆理工大学 | Parallel battery pack safety early warning method based on neural network |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN119716576A (en)* | 2025-02-26 | 2025-03-28 | 广州市凯捷电源实业有限公司 | Battery production charging and discharging status monitoring and early warning method and system |
| Publication number | Publication date |
|---|---|
| DE102022208932A1 (en) | 2024-02-29 |
| Publication | Publication Date | Title |
|---|---|---|
| US20240168095A1 (en) | Method and Apparatus for Predictive Diagnosis of a Device Battery of a Technical Device Using a Multivariate Transformer Model | |
| CN117630672A (en) | Method and device for predicting a device battery of a diagnostic technology device by means of a converter model | |
| US20220099743A1 (en) | Method and Apparatus for Operating a System for Providing Predicted States of Health of Electrical Energy Stores for a Device Using Machine Learning Methods | |
| Chen et al. | Remaining useful life prediction of battery using a novel indicator and framework with fractional grey model and unscented particle filter | |
| Renold et al. | Comprehensive review of machine learning, deep learning, and digital twin data-driven approaches in battery health prediction of electric vehicles | |
| CN114609524A (en) | Method and device for operating a system for providing a predictive aging state of an electrical energy store | |
| CN109604186A (en) | Power battery performance flexibility assesses method for separating | |
| CN115219932A (en) | Method and device for evaluating the relative aging state of a battery of a device | |
| CN114462259B (en) | SOC-based battery remaining life prediction method and system, automobile and medium | |
| KR20230105316A (en) | Method and system for efficiently monitoring battery cells of a device battery in a device-external central unit using a digital twin | |
| US20230305073A1 (en) | Method and apparatus for providing a predicted aging state of a device battery based on a predicted usage pattern | |
| US20230280406A1 (en) | Verfahren und Vorrichtung zum Erkennen einer kritischen Anomalie in einer Gerätebatterie basierend auf maschinellen Lernverfahren | |
| CN115718258B (en) | Battery fault detection | |
| CN115701545A (en) | Method and device for providing a calculated and predicted state of aging of an electrical energy store | |
| US11733313B2 (en) | Method and apparatus for operating a system for providing states of health of electrical energy stores for a multiplicity of devices with the aid of machine learning methods | |
| CN119749240B (en) | Power battery abnormality diagnosis method and device for new energy automobile and storage medium | |
| US20230324463A1 (en) | Method and Apparatus for Operating a System for Detecting an Anomaly of an Electrical Energy Store for a Device by Means of Machine Learning Methods | |
| CN116643178A (en) | SOC estimation method and related device of battery management system | |
| CN114879070B (en) | A battery status evaluation method and related equipment | |
| CN119669949A (en) | Method and apparatus for creating an anomaly recognition model using an extended autoencoder to identify current or impending anomalies of a device battery | |
| CN116256640A (en) | Computer-implemented method for providing an aging state model to determine a current and predicted aging state of an electrical accumulator | |
| CN118051852A (en) | Method and apparatus for creating an anomaly recognition model that recognizes anomalies in a device battery pack | |
| CN118393354A (en) | Method for predictive diagnosis of device battery using hybrid differential equations | |
| Degla et al. | An enhanced neural network application for a lithium-ion battery pack state-of-health estimator | |
| Sui et al. | Early prediction of lithium-ion batteries lifetime via few-shot learning |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |