（一）技术领域(1) Technical field
本发明涉及设备故障监测技术领域，特别涉及一种设备故障预警及状态监测方法。The invention relates to the technical field of equipment failure monitoring, in particular to an equipment failure early warning and state monitoring method.
（二）背景技术(2) Background technology
设备故障预警和状态监测根据设备运行规律或观测得到的可能性前兆，在设备真正发生故障之前，及时预报设备的异常状况，采取相应的措施，从而最大程度的降低设备故障所造成的损失。随着设备装置和工程控制系统的规模和复杂性日益增大，为保证生产过程的安全平稳,通过可靠的状态监控技术及时有效的监测和诊断过程异常就显得尤为迫切和重要。Equipment failure early warning and status monitoring According to equipment operation rules or observed possible precursors, before the equipment actually fails, timely forecast the abnormal condition of the equipment and take corresponding measures to minimize the loss caused by equipment failure. With the increasing scale and complexity of equipment and engineering control systems, in order to ensure the safety and stability of the production process, it is particularly urgent and important to monitor and diagnose process abnormalities in a timely and effective manner through reliable state monitoring technology.
现有的设备故障预警技术主要分为三大类：基于机理模型的方法、基于知识的方法和基于数据驱动的方法。Existing equipment failure early warning technologies are mainly divided into three categories: method based on mechanism model, method based on knowledge and method based on data drive.
基于机理模型的方法是发展最早也最为深入的故障预警和状态监测方法，它主要包括两个阶段：（1）残差产生阶段：通过设备运行机理建立精确的数学模型来估计系统输出，并将之与实际测量值比较，获得残差，这个阶段构建的模型又叫残差产生器；（2）残差评价阶段：对残差进行分析以确定过程是否发生故障，并进一步辨识故障类型。该类方法与控制理论紧密结合，主要采用参数估计、状态估计和等价空间三类具体的方法来实现残差序列的构建，其中状态估计方法最为常用，可使用观测器或卡尔曼滤波器实现。The method based on the mechanism model is the earliest and most in-depth fault warning and condition monitoring method. It mainly includes two stages: (1) Residual error generation stage: establish an accurate mathematical model through the operation mechanism of the equipment to estimate the system output, and The model constructed in this stage is also called the residual generator; (2) Residual evaluation stage: analyze the residual to determine whether the process is faulty, and further identify the type of fault. This type of method is closely combined with control theory, mainly using three specific methods of parameter estimation, state estimation and equivalent space to realize the construction of residual sequence, among which the state estimation method is the most commonly used, which can be realized by using an observer or a Kalman filter .
基于知识的方法主要以相关专家和操作人员的启发性经验知识为基础，定性或定量描述过程中各单元之间的连接关系、故障传播模式等，在设备出现异常征兆后通过推理、演绎等方式模拟过程专家在监测上的推理能力，从而自动完成设备故障预警和设备监测。该类方法无需精确的数学模型，但对专家知识有较强的依赖性，常用的方法主要包括专家系统、故障决策树、有向图、模糊逻辑等。The knowledge-based method is mainly based on the heuristic experience knowledge of relevant experts and operators, qualitatively or quantitatively describes the connection relationship between the various units in the process, the fault propagation mode, etc., and uses reasoning, deduction, etc. Simulate the reasoning ability of process experts in monitoring, so as to automatically complete equipment failure warning and equipment monitoring. This type of method does not require precise mathematical models, but has a strong dependence on expert knowledge. Commonly used methods mainly include expert systems, fault decision trees, directed graphs, and fuzzy logic.
基于数据驱动的方法通过挖掘过程数据中的内在信息建立数学模型和表达过程状态，根据模型来实施过程的有效监测。随着智能化仪表和计算机存储技术的广泛应用，海量的过程数据得以有效地监测、收集和存储，而该类方法正是基于这样的海量数据，在监测和预警算法上它又可以分为基于信号处理、粗糙集、机器学习、信息融合和多元统计这五大类算法，其中机器学习算法是在理论和实践中发展最为活跃的分支，它包括贝叶斯分类器，神经网络，支持向量机，k最近邻算法，聚类算法，主成分分析等算法。The data-driven method establishes a mathematical model and expresses the process state by mining the inherent information in the process data, and implements effective monitoring of the process according to the model. With the wide application of intelligent instruments and computer storage technology, massive process data can be effectively monitored, collected and stored, and this type of method is based on such massive data, and it can be divided into monitoring and early warning algorithms based on Signal processing, rough sets, machine learning, information fusion and multivariate statistics are five categories of algorithms, among which machine learning algorithms are the most active branch in theory and practice, including Bayesian classifiers, neural networks, support vector machines, K nearest neighbor algorithm, clustering algorithm, principal component analysis and other algorithms.
基于机理模型的监控方法能够把物理认识与监控系统结合起来，通过分析残差来进行故障预警的方式也更有利于专业人员的理解，但由于多数机理模型均为简化的线性系统，因此当面对非线性、自由度较高以及多变量耦合的复杂系统时，其使用效果并不理想；另外，对复杂系统建立机理模型可能要付出巨大的成本；再者，实际工业过程中的噪音影响，环境因素的变化等都提高了模型失效的风险。以上原因都使得基于机理模型的监控方法检测效果不佳，应用范围不广。The monitoring method based on the mechanism model can combine the physical understanding with the monitoring system, and the method of fault warning by analyzing the residual is also more conducive to the understanding of professionals. However, since most mechanism models are simplified linear systems, it is difficult to For complex systems with nonlinear, high degrees of freedom, and multivariate coupling, the effect is not ideal; in addition, it may cost a lot to establish a mechanism model for complex systems; moreover, the impact of noise in actual industrial processes, Changes in environmental factors have increased the risk of model failure. For the above reasons, the detection effect of the monitoring method based on the mechanism model is not good, and the application range is not wide.
基于知识的监控方法使用定性的模型实现预警和监测,当被监控对象较为简单，工艺知识和生产经验较为充足时，其性能较为优良。但需要注意的是该类方法的预警准确度对知识库中专家知识的丰富程度和专家知识水平的高低具有很强的依赖性；同时，部分专家实际操作经验很难用一种合理的形式化表达方式进行描述,当系统较为复杂时还有可能出现“冲突消解”、“组合爆炸”等问题；另外，这类方法的通用性较差,且先验知识的完整性一般难以保证。The knowledge-based monitoring method uses a qualitative model to realize early warning and monitoring. When the monitored object is relatively simple, and the process knowledge and production experience are sufficient, its performance is relatively good. However, it should be noted that the early warning accuracy of this type of method has a strong dependence on the richness of expert knowledge in the knowledge base and the level of expert knowledge; at the same time, it is difficult for some experts to use a reasonable formalization When the system is more complex, there may be problems such as "conflict resolution" and "combination explosion"; in addition, this kind of method has poor versatility, and the integrity of prior knowledge is generally difficult to guarantee.
基于数据驱动的故障预警和状态监测技术直接通过系统的历史数据建立故障预警模型，不需要知道系统精确的机理模型，因此其通用性和自适应能力都较强。但是由于这类方法并不明确系统的内部结构和机理信息,所以对预警结果的分析和解释则相对比较困难；另外，机器学习等基于数据驱动的算法主要都是应用于故障诊断，而基于数据的故障预警技术还处于起步阶段,可靠有效的方法也相对较少；再者，由于数据量较大且基于数据驱动的算法的时间复杂度都普遍较高，所以如何提高监测算法的效率也是亟待解决的问题。Based on the data-driven fault early warning and condition monitoring technology, the fault early warning model is established directly through the historical data of the system, without knowing the precise mechanism model of the system, so its versatility and adaptability are strong. However, since this type of method does not clarify the internal structure and mechanism information of the system, it is relatively difficult to analyze and interpret the early warning results; in addition, data-driven algorithms such as machine learning are mainly used in fault diagnosis, while data-based The fault warning technology is still in its infancy, and there are relatively few reliable and effective methods; moreover, due to the large amount of data and the generally high time complexity of data-driven algorithms, how to improve the efficiency of monitoring algorithms is also an urgent need. solved problem.
（三）发明内容(3) Contents of the invention
本发明为了弥补现有技术的不足，提供了一种既适用于复杂非线性系统，又便于专业人员分析理解的设备故障预警及状态监测方法。In order to make up for the deficiencies of the prior art, the present invention provides an equipment fault early warning and state monitoring method that is applicable to complex nonlinear systems and is convenient for professionals to analyze and understand.
本发明是通过如下技术方案实现的：The present invention is achieved through the following technical solutions:
一种设备故障预警及状态监测方法，其特征在于：包括建立模型和运行模型两个过程；建立模型过程的步骤为先获取训练数据，对训练数据进行数据预处理工作，再采用非参数学习算法选取记忆矩阵，训练残差产生器，获取各参数残差阈值；运行模型的步骤为先获取实时数据，对实时数据进行数据预处理工作，再计算实时数据各参数残差，分析残差以判断设备状态是否正常，并进一步定位故障原因。A device failure early warning and state monitoring method, characterized in that: it includes two processes of building a model and running the model; the steps of building the model are firstly obtaining training data, performing data preprocessing on the training data, and then using a non-parametric learning algorithm Select the memory matrix, train the residual generator, and obtain the residual threshold of each parameter; the steps to run the model are to obtain real-time data first, perform data preprocessing on the real-time data, then calculate the residual of each parameter of the real-time data, and analyze the residual to judge Check whether the equipment status is normal, and further locate the cause of the failure.
本发明基于设备数据，采用非参数学习算法(Non-Parametric LearningAlgorithm)与支持向量回归机（SVR）相结合的方式构建传统机理模型中的残差产生器，并对残差进行分析以达到故障预警的目的。Based on the equipment data, the present invention uses the combination of Non-Parametric Learning Algorithm and Support Vector Regression (SVR) to construct the residual generator in the traditional mechanism model, and analyzes the residual to achieve fault warning the goal of.
本发明的更优技术方案为：The more optimal technical scheme of the present invention is:
所述建立模型过程中，选取与设备安全运行相关的关键参数，并对设备历史数据进行筛选，以设备正常运行状态下的历史健康数据作为训练数据，然后对训练数据进行删除无效数据、归一化的预处理。In the process of building the model, key parameters related to the safe operation of the equipment are selected, and the historical data of the equipment is screened, and the historical health data under the normal operation of the equipment is used as the training data, and then the training data is deleted and invalid data is normalized. optimized preprocessing.
所述建立模型过程中，采用非参数学习算法选取记忆矩阵，首先计算训练数据中每个观测向量的范数，并计算范数的取值范围Nrange，然后按照范数取值范围Nrange分为h份，再以Nrange/h为步距从训练矩阵中挑选出若干个符合要求的观测向量加入到记忆矩阵D中，训练数据中去除记忆矩阵之后的剩余数据作为剩余矩阵进行保存，以备训练模型的残差阈值时使用。In the process of building the model, a non-parametric learning algorithm is used to select the memory matrix, firstly calculate the norm of each observation vector in the training data, and calculate the value range Nrange of the norm, and then divide according to the value range Nrange of the norm Then select a number of observation vectors that meet the requirements from the training matrix and add them to the memory matrix D at a step of Nrange /h, and save the remaining data after removing the memory matrix from the training data as the remaining matrix. Used when preparing the residual threshold for the trained model.
所述建立模型过程中，再采用非参数学习算法提取的记忆矩阵基础上，采用非线性回归算法支持向量回归机训练设备的估计值计算模型，令回归机的输出为记忆矩阵中某一观测向量的一个参数，输入为该向量的其他参数，即用观测向量中的其他参数来拟合一个参数，得到估计值计算模型之后，即可获得残差产生器。In the process of establishing the model, on the basis of the memory matrix extracted by the non-parametric learning algorithm, the estimated value calculation model of the training equipment of the non-linear regression algorithm support vector regression machine is adopted, so that the output of the regression machine is a certain observation vector in the memory matrix A parameter of , the input is other parameters of the vector, that is, other parameters in the observation vector are used to fit a parameter, and after the estimated value calculation model is obtained, the residual generator can be obtained.
所述建立模型过程中，将剩余矩阵带入残差产生器，即可获得正常运行状态下，数据的残差范围，其中各参数残差的上下限即可作为各参数的残差阈值。In the process of building the model, the residual matrix is brought into the residual generator to obtain the residual range of the data in the normal operation state, wherein the upper and lower limits of the residual of each parameter can be used as the residual threshold of each parameter.
所述运行模型过程中，对实时数据的预处理为对其进行归一化处理，使各参数值全部映射到[0，1]的区间内，然后将归一化的实时数据代入残差产生器，获得各参数对应的残差。In the process of running the model, the preprocessing of the real-time data is to normalize it, so that all parameter values are mapped to the interval [0, 1], and then the normalized real-time data is substituted into the residual to generate device to obtain the residual corresponding to each parameter.
所述运行模型过程中，将实时数据的各参数残差与训练所得的各参数残差阈值进行对比，如果实时数据残差超过了设定的上下限则认为设备出现异常，而对应的超限参数即为设备可能发生异常的位置，从而进一步定位故障原因。In the process of running the model, the parameter residuals of the real-time data are compared with the parameter residual thresholds obtained from training. If the real-time data residuals exceed the set upper and lower limits, it is considered that the equipment is abnormal, and the corresponding over-limit The parameter is the position where the device may be abnormal, so as to further locate the cause of the fault.
本发明是数据驱动的故障预警方法，同时又有效利用了特定诊断对象的先验信息和历史信息，相比传统利用残差产生器的故障预警方法，本发明不需要建立精确的数学机理模型，因此对复杂的非线性系统具有很强的通用性。The present invention is a data-driven fault early warning method, and at the same time effectively utilizes the prior information and historical information of a specific diagnostic object. Compared with the traditional fault early warning method using a residual generator, the present invention does not need to establish an accurate mathematical mechanism model, Therefore, it has strong versatility for complex nonlinear systems.
本发明通过展示设备关键参数实际值与系统估计值残差的方式，展示设备的运行状况，因此相比一般的基于数据驱动的故障预警方法，又具有更直观，更易对预警结果进行分析和解释的特点。The present invention shows the operating status of the equipment by displaying the actual value of the key parameter of the equipment and the residual of the system estimated value, so compared with the general data-driven fault early warning method, it is more intuitive and easier to analyze and explain the early warning results specialty.
本发明建立的预警模型为基于多参数的非线性模型，充分考虑了多参数之间的相互影响，而非基于单一参数的故障预警，能更多的揭示参数之间隐含的复杂因果关系和条件关系，从而更准确的预警设备故障。The early warning model established by the present invention is a nonlinear model based on multiple parameters, which fully considers the mutual influence between multiple parameters, rather than a fault early warning based on a single parameter, and can reveal more hidden complex causal relationships between parameters and Conditional relationship, so as to more accurately warn of equipment failure.
本发明建立的预警系统由估计值计算和残差阈值分析组成，实际上构成了一个基于动态阈值的预警系统，相比于传统固定阈值的预警系统，具有更好的预警效果，减少了虚报误报，有效提高了预警的准确率。The early warning system established by the present invention is composed of estimated value calculation and residual threshold analysis, and actually constitutes an early warning system based on dynamic threshold. Compared with the traditional early warning system with fixed threshold, it has better early warning effect and reduces false alarms and errors. Effectively improve the accuracy of early warning.
本发明建立的预警系统基于非参数学习算法，改变了传统学习算法中模型参数数量固定，一旦建模后模型固定不变等特点，因此具有更强的自适应能力，且是缓解过拟合/欠拟合问题的有效手段。The early warning system established by the present invention is based on a non-parametric learning algorithm, which changes the characteristics of the fixed number of model parameters in the traditional learning algorithm. An effective tool for underfitting problems.
本发明所构造的基于支持向量回归机建立的估计值计算模型，是高维甚至是无限维的非线性模型，相比传统的多项式拟合或神经网络回归算法等，具有更高的准确率和运算速度。The estimated value calculation model based on the support vector regression machine constructed by the present invention is a high-dimensional or even infinite-dimensional nonlinear model, which has higher accuracy and higher accuracy than traditional polynomial fitting or neural network regression algorithms. calculating speed.
（四）附图说明(4) Description of drawings
下面结合附图对本发明作进一步的说明。The present invention will be further described below in conjunction with the accompanying drawings.
图1为本发明模型建立过程的流程示意图；Fig. 1 is a schematic flow chart of the model building process of the present invention;
图2为本发明非参数学习算法记忆矩阵提取过程的算法流程图；Fig. 2 is the algorithm flowchart of non-parametric learning algorithm memory matrix extraction process of the present invention;
图3为本发明构建的支持向量回归机的结构示意图；Fig. 3 is the structural representation of the support vector regression machine that the present invention builds;
图4为本发明模型运行过程的流程示意图；Fig. 4 is a schematic flow chart of the model operation process of the present invention;
图5为本发明实施例轴承震动估计值与残差曲线示意图；Fig. 5 is a schematic diagram of a bearing vibration estimation value and a residual curve according to an embodiment of the present invention;
图6为本发明实施例电机线圈温度估计值与残差曲线示意图；Fig. 6 is a schematic diagram of the estimated value of the motor coil temperature and the residual error curve according to the embodiment of the present invention;
图7为本发明实施例驱动端瓦温估计值与残差曲线示意图。FIG. 7 is a schematic diagram of an estimated value of the tile temperature of the driving end and a residual curve according to an embodiment of the present invention.
（五）具体实施方式(5) Specific implementation methods
下面结合附图及实施例，对本发明作进一步的详细阐述：Below in conjunction with accompanying drawing and embodiment, the present invention is described in further detail:
本发明是一种适用于多元非线性复杂系统的故障预警及状态监测方法，它基于设备数据，结合了非参数学习算法(Non-Parametric Learning Algorithm)和支持向量回归机以构建传统机理模型中的残差产生器，并通过对设备实时数据的残差进行分析以达到设备状态监测和故障自动预警的功能。该方法主要包括建立模型和运行模型两个过程。The present invention is a fault early warning and state monitoring method suitable for multivariate nonlinear complex systems. It is based on equipment data and combines non-parametric learning algorithm (Non-Parametric Learning Algorithm) and support vector regression machine to construct traditional mechanism model. Residual error generator, and through the analysis of the residual error of real-time equipment data to achieve the function of equipment status monitoring and automatic fault warning. This method mainly includes two processes of building a model and running the model.
图1为本发明建立模型的流程图，整个建模过程主要包括以下步骤：Fig. 1 is the flow chart of modeling of the present invention, and whole modeling process mainly comprises the following steps:
步骤1：获取训练数据Step 1: Get training data
选取与设备安全运行相关的关键参数，并对设备历史数据进行筛选，以设备正常运行状态下的历史健康数据作为训练数据。需要在针对设备的监控参数和指标中，选取与设备安全运行相关的关键参数，对它们进行建模以检测设备运行状况。假设某设备有n个这样的监控参数，则在某一时刻检测到的这n个指标即可构成一个n维的观测向量：Select the key parameters related to the safe operation of the equipment, and filter the historical data of the equipment, and use the historical health data of the equipment in the normal operation state as the training data. It is necessary to select the key parameters related to the safe operation of the equipment among the monitoring parameters and indicators for the equipment, and model them to detect the operation status of the equipment. Assuming that a certain device has n such monitoring parameters, the n indicators detected at a certain moment can constitute an n-dimensional observation vector:
。 .
因此该设备的历史数据即可视为由上述观测向量所构成的数据观测矩阵，接着需要对数据观测矩阵中的观测向量进行筛选，去除掉历史数据中的异常部分，将涵盖系统正常运行状态的数据作为模型训练数据。常见的异常类型主要包括突变异常，超出限值，频率异常等，经过筛选后的训练数据应当满足以下要求：(1):涵盖尽可能完整的设备正常运行状况；(2)每个观测向量都代表了设备的一个正常运行状态；(3)每个观测向量中的各个参数值应为同一时刻的采样值。Therefore, the historical data of the device can be regarded as a data observation matrix composed of the above-mentioned observation vectors, and then the observation vectors in the data observation matrix need to be screened to remove abnormal parts in the historical data, which will cover the normal operating state of the system. data as model training data. Common anomaly types mainly include mutation anomalies, exceeding limits, frequency anomalies, etc. After screening, the training data should meet the following requirements: (1): Cover as complete as possible the normal operating conditions of equipment; (2) Each observation vector is Represents a normal operating state of the equipment; (3) Each parameter value in each observation vector should be the sampling value at the same time.
步骤2：对训练数据进行数据预处理工作Step 2: Perform data preprocessing on the training data
需要对训练数据进行删除无效数据，归一化等预处理措施。由于数据获取过程中可能存在的问题，最初获得的原始训练数据可能存在空数据等无效数据，需要将包含无效数据的观测向量进行删除。另外，由于设备模型相关参数的量纲不同，且不同参数数据绝对值相差很大，为保证使用非线性算子正确衡量不同观测向量之间的距离，需要对各个参数的测量值根据各自的极值进行归一化处理。可采用如1式所示的线性归一化方式，对各参数数据进行预处理，将各参数值全部映射到[0 1]区间内：It is necessary to perform preprocessing measures such as deleting invalid data and normalizing the training data. Due to possible problems in the data acquisition process, the original training data obtained initially may contain invalid data such as empty data, and the observation vectors containing invalid data need to be deleted. In addition, due to the different dimensions of the relevant parameters of the equipment model, and the great difference in the absolute values of different parameter data, in order to ensure that the nonlinear operator is used to correctly measure the distance between different observation vectors, it is necessary to measure the measured values of each parameter according to their respective extremes. Values are normalized. The linear normalization method shown in formula 1 can be used to preprocess the data of each parameter, and map all parameter values to the [0 1] interval:
。 .
步骤3：采用非参数学习算法选取记忆矩阵Step 3: Select memory matrix using non-parametric learning algorithm
由于训练数据可能包含了数量巨大的观测向量，因此为了大幅提高模型训练和运行的效率，本发明采用非参数学习算法从训练数据中自动提取记忆矩阵，并在记忆矩阵的基础上进行残差产生器的训练。Since the training data may contain a huge number of observation vectors, in order to greatly improve the efficiency of model training and operation, the present invention uses a non-parametric learning algorithm to automatically extract the memory matrix from the training data, and generates residuals based on the memory matrix device training.
非参数学习算法改变了传统学习算法中模型参数数量固定，一旦建模后模型固定不变等特点，其模型参数的个数随训练数据的大小而变化，且在建模之后模型也会根据相应的数据的不同而发生变化，因此非参数学习算法具有更强的自适应能力，且是有效缓解过拟合/欠拟合问题的手段。The non-parametric learning algorithm changes the fixed number of model parameters in the traditional learning algorithm. Once the model is fixed after modeling, the number of model parameters changes with the size of the training data, and the model will also be based on the corresponding model parameters after modeling. Therefore, the non-parametric learning algorithm has a stronger adaptive ability and is an effective means to alleviate the problem of overfitting/underfitting.
通过训练数据获取的记忆矩阵是典型的非参数学习算法，记忆矩阵应当涵盖尽量完整的设备正常运行情况，因此采取如下的方法提取记忆矩阵：首先计算训练数据中每个观测向量的范数，并计算范数的取值范围Nrange，然后按照范数取值范围Nrange分为h份，再以Nrange/h为步距从训练矩阵中挑选出若干个符合要求的观测向量加入到记忆矩阵D中，具体的记忆矩阵提取方法参见图2所示的算法流程图，其中m为训练矩阵中观测向量的数目，δ为一小的正数用来控制符合要求的观测向量数量。采用此方法构造的记忆矩阵，能够将训练数据中有代表性的观测向量选入记忆矩阵中，既保证不重复录入又能较好地覆盖设备的正常工作空间。The memory matrix obtained from the training data is a typical non-parametric learning algorithm. The memory matrix should cover the normal operation of the equipment as complete as possible. Therefore, the following method is used to extract the memory matrix: first calculate the norm of each observation vector in the training data, and Calculate the value range Nrange of the norm, and then divide it into h parts according to the value range Nrange of the norm, and then select a number of observation vectors that meet the requirements from the training matrix with a step of Nrange /h and add them to the memory matrix In D, refer to the algorithm flowchart shown in Figure 2 for the specific memory matrix extraction method, where m is the number of observation vectors in the training matrix, and δ is a small positive number used to control the number of observation vectors that meet the requirements. The memory matrix constructed by this method can select representative observation vectors in the training data into the memory matrix, which not only ensures no repeated entry but also better covers the normal working space of the equipment.
训练数据中去除记忆矩阵之后的剩余数据作为剩余矩阵进行保存，以备训练模型的残差阈值时使用。The remaining data after removing the memory matrix in the training data is saved as a residual matrix for use when training the residual threshold of the model.
步骤4：训练残差产生器Step 4: Train the Residual Generator
在之前采用非参数学习算法提取的记忆矩阵基础上，采用非线性回归算法支持向量回归机训练设备的估计值计算模型，该模型用来估计系统中每个参数的正常输出。与传统的估计值计算模型不同，该模型基于数据，无须建立精确的数学机理模型。On the basis of the memory matrix extracted by the previous non-parametric learning algorithm, the estimated value calculation model of the equipment is trained by the non-linear regression algorithm support vector regression machine, which is used to estimate the normal output of each parameter in the system. Different from the traditional estimated value calculation model, this model is based on data and does not need to establish an accurate mathematical mechanism model.
本发明所采用的非线性支持向量回归机的主要原理是，通过引入核函数，将原空间的向量非线性映射到一个特征空间，在该特征空间中原问题是一个线性可分的，并可求解此问题的最优分界面，此分界面即为原空间中的一个非线性分界面，支持向量回归机的本质是求解下面的优化问题，其中目标函数中的C为惩罚因子，ξi,ξi*分别为松弛变量的上下限，。The main principle of the nonlinear support vector regression machine adopted in the present invention is that by introducing a kernel function, the vector of the original space is nonlinearly mapped to a feature space. In this feature space, the original problem is a linear separable and can be solved The optimal interface of this problem, this interface is a nonlinear interface in the original space, the essence of the support vector regression machine is to solve the following optimization problem, where C in the objective function is the penalty factor, ξi , ξi* are the upper and lower bounds of the slack variable respectively, .
支持向量回归机的结构示意图可以表示为图3中的形式，其中模型的输入为左侧的多个参数值，中间经过核函数K(x, xi)的转化，最后构造回归函数输出估计值。从图3中可以看到，支持向量回归机模型是一个多输入单输出的模型，因此在训练估计值计算模型时，应当令回归机的输出为记忆矩阵中某一观测向量的一个参数，输入为该向量的其他参数，即用观测向量中的其他参数来拟合一个参数。这样，训练一个n维的估计值计算模型，需要训练与各个参数一一对应的n个支持向量回归机。The structure schematic diagram of the support vector regression machine can be expressed as the form in Figure 3, where the input of the model is a plurality of parameter values on the left, and the kernel function K(x, xi ) is transformed in the middle, and finally the regression function is constructed to output the estimated value . It can be seen from Figure 3 that the support vector regression machine model is a model with multiple inputs and single output, so when training the estimated value calculation model, the output of the regression machine should be a parameter of an observation vector in the memory matrix, and the input For the other parameters of the vector, that is, to fit a parameter with other parameters in the observation vector. In this way, to train an n-dimensional estimation value calculation model, it is necessary to train n support vector regression machines corresponding to each parameter one by one.
在使用记忆矩阵数据训练得到估计值计算模型之后，即可获得残差产生器，残差的计算公式为：残差=实测数据-估计值。After using the memory matrix data to train the estimated value calculation model, the residual generator can be obtained. The residual calculation formula is: residual = measured data - estimated value.
步骤5：获取各参数残差阈值Step 5: Get the residual threshold of each parameter
由于剩余矩阵为除去记忆矩阵之后的训练数据，因此其中的观测向量也代表了设备的各个正常运行状态。将剩余矩阵代入残差产生器，即可获得正常运行状态下，数据的残差范围，其中各参数残差的上下限即可作为各参数的残差阈值。当实时数据的某参数残差超过该参数的残差阈值时，则认为设备出现异常。Since the remaining matrix is the training data after removing the memory matrix, the observation vectors in it also represent the normal operating states of the equipment. By substituting the residual matrix into the residual generator, the residual range of the data under normal operation can be obtained, and the upper and lower limits of the residual of each parameter can be used as the residual threshold of each parameter. When the residual of a certain parameter of real-time data exceeds the residual threshold of this parameter, it is considered that the device is abnormal.
图4为本发明运行模型的流程图，主要包括以下步骤：Fig. 4 is the flowchart of operation model of the present invention, mainly comprises the following steps:
步骤1：获取实时数据Step 1: Get real-time data
实时数据为设备实时运行时在线监测到的数据，它与训练数据中的观测向量一致也由与设备安全运行相关的n个关键参数的观测值构成：The real-time data is the data monitored online during the real-time operation of the equipment. It is consistent with the observation vector in the training data and is composed of the observed values of n key parameters related to the safe operation of the equipment:
。 .
步骤2：对实时数据进行数据预处理工作Step 2: Perform data preprocessing on real-time data
对实时数据同样需要进行归一化处理，以将各参数值全部映射到[0 1]的区间内，归一化所采用的公式如1式所示，需要注意的是式中的ximin和ximax与训练数据归一化时对应参数所用的值相同。Real-time data also needs to be normalized to map all parameter values to the interval [0 1]. The formula used for normalization is shown in formula 1. It should be noted that ximin and ximax is the same as the value used for the corresponding parameter when normalizing the training data.
步骤3：计算实时数据各参数残差Step 3: Calculate the residual of each parameter of the real-time data
将归一化后的实时数据代入残差产生器，即可获得各参数对应的残差。需要注意的是残差产生器是由n个估计值计算模型构成的，每个计算模型的输入均为一个n-1维的实时数据观测向量，计算的是剩余一个参数的估计值。残差通过实时数据与估计值的差值表示。Substituting the normalized real-time data into the residual generator, the residual corresponding to each parameter can be obtained. It should be noted that the residual generator is composed of n estimated value calculation models. The input of each calculation model is an n-1 dimensional real-time data observation vector, and the estimated value of the remaining parameter is calculated. Residuals are expressed as the difference between live data and estimated values.
步骤4：分析残差并对故障进行预警Step 4: Analyze residuals and give early warning of failures
该步骤为残差评价阶段，对各参数的残差进行分析以判断设备的运行是否存在风险，是否需要进行故障预警。具体方法：将实时数据的各参数残差与训练所得的各参数残差阈值进行对比，如果实时数据残差超过了设定的上下限则认为设备出现异常，而对应的超限参数即为设备可能发生异常的位置，从而进一步定位故障原因。This step is the residual error evaluation stage, and the residual error of each parameter is analyzed to determine whether there is risk in the operation of the equipment and whether a fault warning is required. Specific method: compare the residual of each parameter of the real-time data with the residual threshold of each parameter obtained from training. If the residual of the real-time data exceeds the set upper and lower limits, it is considered that the device is abnormal, and the corresponding overrun parameter is the device The position where the exception may occur, so as to further locate the cause of the fault.
实施例：Example:
本实施例以北方某火力发电厂1#机组的一次风机为监测对象，一次风机是电厂重要的辅机设备，它结构复杂，影响因素多，难以建立精确数学机理模型，且易发多发故障，符合本发明所针对的多元非线性系统的特点。通过本实施例的详细阐述，进一步说明本发明的实施过程。This embodiment takes the primary fan of unit 1# of a thermal power plant in the north as the monitoring object. The primary fan is an important auxiliary equipment of the power plant. It has a complex structure and many influencing factors. It is difficult to establish an accurate mathematical mechanism model, and it is prone to frequent failures. It conforms to the characteristics of the multivariate nonlinear system targeted by the present invention. Through the detailed elaboration of this embodiment, the implementation process of the present invention is further described.
本发明实施例对某电厂一次风机设备的故障预警和状态监测的实施步骤如下：The implementation steps of the fault warning and status monitoring of the primary fan equipment of a certain power plant in the embodiment of the present invention are as follows:
1.设备故障预警与状态检测系统建模过程1. Modeling process of equipment failure early warning and status detection system
（1）获取训练数据(1) Obtain training data
与该一次风机安全运行相关的关键参数有28个，包括实发功率(MW)，风机出口压力(kPa)，轴承x向振动(mm/s)等，因此该设备的观测向量为28维的向量：There are 28 key parameters related to the safe operation of the primary fan, including real power (MW), fan outlet pressure (kPa), bearing x-direction vibration (mm/s), etc., so the observation vector of the device is 28-dimensional vector:
。 .
获取的历史数据为该设备2012年8月1日至2013年2月1日半年的数据，剔除历史数据中的异常数据，剩下的正常历史数据即为模型的训练数据。The acquired historical data is the half-year data of the device from August 1, 2012 to February 1, 2013. The abnormal data in the historical data is removed, and the remaining normal historical data is the training data of the model.
（2）对训练数据进行数据预处理工作(2) Perform data preprocessing on the training data
对挑选出的训练数据进行删除无效数据和归一化等预处理措施。按照1式进行线性归一化方式，将各参数值全部映射到[0 1]的区间内。Perform preprocessing measures such as deleting invalid data and normalizing the selected training data. According to the linear normalization method of formula 1, all the parameter values are mapped to the interval of [0 1].
（3）采用非参数学习算法选取记忆矩阵(3) Use non-parametric learning algorithm to select memory matrix
采用非参数学习算法从训练数据中自动提取记忆矩阵，本实例中记忆矩阵的提取方法参见图2所示的算法流程图。A non-parametric learning algorithm is used to automatically extract the memory matrix from the training data. For the method of extracting the memory matrix in this example, refer to the algorithm flow chart shown in Figure 2.
（4）训练残差产生器(4) Training residual generator
由于本实例涉及的设备有28个关键参数，因此需要训练28个支持向量回归机来分别计算这些参数的估计值，其中每个回归机的输入为27维向量，输出为1维向量，支持向量回归机的结构如图3所示。在使用记忆矩阵数据训练得到估计值计算模型之后，即可获得残差产生器。Since the equipment involved in this example has 28 key parameters, it is necessary to train 28 support vector regression machines to calculate the estimated values of these parameters respectively, where the input of each regression machine is a 27-dimensional vector, the output is a 1-dimensional vector, and the support vector The structure of the regression machine is shown in Figure 3. After using the memory matrix data to train the estimated value calculation model, the residual generator can be obtained.
（5）获取各参数残差阈值(5) Get the residual threshold of each parameter
将剩余矩阵代入残差产生器，即可获得正常运行状态下，数据的残差范围，其中各参数残差的上下限即可作为各参数的残差阈值。By substituting the residual matrix into the residual generator, the residual range of the data under normal operation can be obtained, and the upper and lower limits of the residual of each parameter can be used as the residual threshold of each parameter.
2.设备故障预警与状态检测系统模型运行过程2. The operation process of the equipment failure early warning and status detection system model
（1）采集实时数据(1) Collect real-time data
本实例中实时数据的采样周期为1分钟，与训练数据中的观测向量一致，实时数据由28个关键参数的观测值构成。The sampling period of the real-time data in this example is 1 minute, which is consistent with the observation vector in the training data, and the real-time data consists of the observed values of 28 key parameters.
（2）对实时数据进行数据预处理工作(2) Perform data preprocessing on real-time data
将采集的实时数据进行归一化，映射到[0 1]的区间内，归一化所采用的公式如1式所示，式中的 ximin和ximax与训练数据归一化时采用的值相同。Normalize the collected real-time data and map it to theinterval of [0 1]. same value.
（3）计算实时数据各参数残差(3) Calculate the residual of each parameter of the real-time data
将归一化后的实时数据代入残差产生器，即可获得各参数对应的残差，残差通过实时数据与估计值的差值表示。Substituting the normalized real-time data into the residual generator, the residual corresponding to each parameter can be obtained, and the residual is represented by the difference between the real-time data and the estimated value.
（4）分析残差并对故障进行预警(4) Analyze residuals and give early warning of failures
将实时数据的各参数残差与训练所得的各参数残差阈值进行对比，若实时残差超过了残差阈值则发出故障预警，并进一步辨识故障原因。Compare the residual of each parameter of the real-time data with the residual threshold of each parameter obtained from training. If the real-time residual exceeds the residual threshold, a fault warning will be issued and the cause of the fault will be further identified.
本实例以某火力发电厂一次风机2012年8月1日至2013年2月1日的历史数据进行建模，并自2013年2月2日开始采集实时数据进行状态监测，图5至图7展示了本实例的实施效果，如图5所示，在设备正常运行一段时间之后，关键参数轴承x向振动的实际值与估计值发生了明显的偏离，图5中的残差曲线充分表明了该设备的劣化趋势，可以看到轴承振动残差逐渐增大并超过了阈值上限，从而触发了预警，而传统的预警方法在振动信号大于4.6mm/s时才会触发报警，事实表明，在本发明设计的预警系统发出告警约6天之后，该设备由于x向振动过于剧烈发生了意外停机的严重故障，本发明成功预测了这次故障。图6和图7为另外两个关键参数电极线圈温度(℃)和驱动端瓦温(℃)的运行曲线图，可以看到这两个参数的实际值与估计值并没有发生明显偏离，它们的残差也没有超出残差阈值，因此这两个没有发生故障的参数并没有触发预警，因此本实例说明本发明不仅可以成功预警设备故障，也可以帮助快速定位故障产生的原因。This example is based on the historical data of a thermal power plant's primary fan from August 1, 2012 to February 1, 2013, and real-time data has been collected since February 2, 2013 for status monitoring, as shown in Figures 5 to 7 The implementation effect of this example is shown. As shown in Figure 5, after a period of normal operation of the equipment, the actual value of the key parameter bearing x-vibration deviates significantly from the estimated value, and the residual curve in Figure 5 fully shows that The deterioration trend of the equipment shows that the bearing vibration residual gradually increases and exceeds the upper threshold, which triggers an early warning. However, the traditional early warning method only triggers an alarm when the vibration signal is greater than 4.6mm/s. Facts show that in About 6 days after the early warning system designed in the present invention issued an alarm, the equipment suffered a serious fault of unexpected shutdown due to excessive vibration in the x direction, and the present invention successfully predicted this fault. Figures 6 and 7 are the operating curves of the other two key parameters, the electrode coil temperature (°C) and the drive end tile temperature (°C). It can be seen that the actual values of these two parameters do not deviate significantly from the estimated values. The residual error does not exceed the residual error threshold, so these two parameters without failure did not trigger an early warning. Therefore, this example shows that the present invention can not only successfully warn equipment failures, but also help quickly locate the cause of the failure.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410322115.2ACN104102773B (en) | 2014-07-05 | 2014-07-05 | A kind of equipment fault early-warning and state monitoring method |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410322115.2ACN104102773B (en) | 2014-07-05 | 2014-07-05 | A kind of equipment fault early-warning and state monitoring method |
| Publication Number | Publication Date |
|---|---|
| CN104102773A CN104102773A (en) | 2014-10-15 |
| CN104102773Btrue CN104102773B (en) | 2017-06-06 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410322115.2AActiveCN104102773B (en) | 2014-07-05 | 2014-07-05 | A kind of equipment fault early-warning and state monitoring method |
| Country | Link |
|---|---|
| CN (1) | CN104102773B (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105302848B (en)* | 2014-10-11 | 2018-11-13 | 山东鲁能软件技术有限公司 | A kind of assessed value calibration method of device intelligence early warning system |
| CN104317778A (en)* | 2014-10-30 | 2015-01-28 | 国家电网公司 | Massive monitoring data based substation equipment fault diagnosis method |
| CN104462846B (en)* | 2014-12-22 | 2017-11-10 | 山东鲁能软件技术有限公司 | A kind of equipment fault intelligent diagnosing method based on SVMs |
| CN104742895B (en)* | 2015-02-09 | 2018-01-05 | 中国计量学院 | A kind of car air braking system fault detection method based on analytic modell analytical model |
| CN104935464B (en)* | 2015-06-12 | 2018-07-06 | 北京奇虎科技有限公司 | The fault early warning method and device of a kind of web station system |
| CN105424224B (en)* | 2015-12-14 | 2018-04-06 | 国家电网公司 | A kind of disconnecting switch state monitoring method and device |
| CN105548764B (en)* | 2015-12-29 | 2018-11-06 | 山东鲁能软件技术有限公司 | A kind of Fault Diagnosis for Electrical Equipment method |
| CN105574284B (en)* | 2015-12-29 | 2019-06-14 | 山东鲁能软件技术有限公司 | A kind of Fault Diagnosis for Electrical Equipment method based on trend feature point |
| CN105721194B (en)* | 2016-01-13 | 2017-11-21 | 广州衡昊数据科技有限公司 | Mobile network potential faults intelligent positioning system |
| CN105928611A (en)* | 2016-04-14 | 2016-09-07 | 中国神华能源股份有限公司 | Fault early-warning system and method of mechanical device |
| KR101827108B1 (en)* | 2016-05-04 | 2018-02-07 | 두산중공업 주식회사 | Plant fault detection learning method and system |
| CN106649727B (en)* | 2016-12-23 | 2019-12-24 | 南京航空航天大学 | A database construction method for fault detection of UAV flight control system |
| CN106709662B (en)* | 2016-12-30 | 2021-07-02 | 山东鲁能软件技术有限公司 | Power equipment operation condition division method |
| JP6848656B2 (en)* | 2017-04-28 | 2021-03-24 | 横河電機株式会社 | Display device, display method and program |
| CN107505884B (en)* | 2017-07-31 | 2021-03-12 | 新奥泛能网络科技股份有限公司 | Diagnosis method of universal energy station equipment, cloud server and system |
| CN107701468B (en)* | 2017-09-27 | 2019-07-05 | 郑州大学 | On-line comprehensive monitoring method and device for mixed-flow pump |
| CN111066003A (en)* | 2017-10-10 | 2020-04-24 | 西门子股份公司 | Method, device and medium for condition monitoring of equipment in process industry |
| CN109766138A (en) | 2017-11-08 | 2019-05-17 | 广东欧珀移动通信有限公司 | Application program prediction model establishing and preloading method, device, medium and terminal |
| CN109784127B (en)* | 2017-11-10 | 2023-08-01 | 郑州大学 | Equipment health state early warning method and system thereof |
| CN109814936A (en)* | 2017-11-20 | 2019-05-28 | 广东欧珀移动通信有限公司 | Application program prediction model establishing and preloading method, device, medium and terminal |
| CN109814937A (en)* | 2017-11-20 | 2019-05-28 | 广东欧珀移动通信有限公司 | Application predictive model establishment, preloading method, device, medium and terminal |
| CN107967489A (en)* | 2017-11-29 | 2018-04-27 | 中国科学院空间应用工程与技术中心 | A kind of method for detecting abnormality and system |
| CN108037387B (en)* | 2017-12-05 | 2020-06-19 | 北京能源集团有限责任公司 | Equipment fault analysis method and device based on clustering |
| CN108133326A (en)* | 2017-12-22 | 2018-06-08 | 华润电力（菏泽）有限公司 | A kind of status early warning method and system based on thermal power generating equipment |
| CN108199795B (en)* | 2017-12-29 | 2019-05-10 | 北京百分点信息科技有限公司 | Method and device for monitoring equipment status |
| CN108108852B (en)* | 2018-01-05 | 2020-04-21 | 广东电科院能源技术有限责任公司 | Thermal power generating unit short-term reliability evaluation method and device based on fault early warning technology |
| CN108168548B (en)* | 2018-02-13 | 2022-03-15 | 南京师范大学 | Pedestrian inertial navigation system and method assisted by machine learning algorithm and model |
| CN108460207A (en)* | 2018-02-28 | 2018-08-28 | 上海华电电力发展有限公司 | A kind of fault early warning method of the generating set based on operation data model |
| CN108460144B (en)* | 2018-03-14 | 2021-11-12 | 西安华光信息技术有限责任公司 | Coal equipment fault early warning system and method based on machine learning |
| CN108446734A (en)* | 2018-03-20 | 2018-08-24 | 中科边缘智慧信息科技（苏州）有限公司 | Disk failure automatic prediction method based on artificial intelligence |
| CN108664696B (en)* | 2018-04-02 | 2023-04-07 | 国家计算机网络与信息安全管理中心 | Method and device for evaluating running state of water chiller |
| CN108596229B (en)* | 2018-04-13 | 2021-09-10 | 北京华电智慧科技产业有限公司 | Method and system for monitoring and diagnosing online abnormity |
| CN109344976A (en)* | 2018-08-24 | 2019-02-15 | 华能国际电力股份有限公司海门电厂 | A kind of electrical system operating status intelligent analysis method based on Keras |
| CN110968069B (en)* | 2018-09-28 | 2022-01-25 | 新疆金风科技股份有限公司 | Fault prediction method of wind generating set, corresponding device and electronic equipment |
| CN109660419B (en)* | 2018-10-08 | 2022-06-17 | 平安科技（深圳）有限公司 | Method, device, equipment and storage medium for predicting abnormity of network equipment |
| CN109524139B (en)* | 2018-10-23 | 2023-06-13 | 中核核电运行管理有限公司 | Real-time equipment performance monitoring method based on equipment working condition change |
| CN109586239B (en)* | 2018-12-10 | 2020-03-31 | 国网四川省电力公司电力科学研究院 | Real-time diagnosis and fault early warning method of intelligent substation |
| CN109657982B (en)* | 2018-12-20 | 2022-02-11 | 三一重能有限公司 | Fault early warning method and device |
| CN109726463A (en)* | 2018-12-25 | 2019-05-07 | 中铁隧道局集团有限公司 | A kind of shield TBM fault early warning method based on SVM algorithm |
| CN110046182A (en)* | 2019-03-21 | 2019-07-23 | 华能澜沧江水电股份有限公司 | A kind of huge hydroelectric power plant's intelligent alarm threshold setting method and system |
| KR102398307B1 (en)* | 2019-03-26 | 2022-05-16 | 도시바 미쓰비시덴키 산교시스템 가부시키가이샤 | Abnormality judgment support device |
| CN110173453A (en)* | 2019-04-04 | 2019-08-27 | 上海发电设备成套设计研究院有限责任公司 | A kind of online assessment method of power plant pressure fan state |
| CN110046146A (en)* | 2019-04-16 | 2019-07-23 | 中国联合网络通信集团有限公司 | The monitoring method and device of industrial equipment based on mobile edge calculations |
| CN110059968A (en)* | 2019-04-23 | 2019-07-26 | 深圳市华星光电技术有限公司 | Process data monitoring method and process data monitoring system |
| CN110322049B (en)* | 2019-06-03 | 2023-06-09 | 浙江图灵软件技术有限公司 | Public security big data early warning method |
| CN110298455B (en)* | 2019-06-28 | 2023-06-02 | 西安因联信息科技有限公司 | Mechanical equipment fault intelligent early warning method based on multivariate estimation prediction |
| CN110533294B (en)* | 2019-07-30 | 2024-04-16 | 中国核电工程有限公司 | Nuclear power plant operation fault alarm method based on artificial intelligence technology |
| CN110414154B (en)* | 2019-07-31 | 2022-09-30 | 北京天泽智云科技有限公司 | Fan component temperature abnormity detection and alarm method with double measuring points |
| CN111913443A (en)* | 2019-08-24 | 2020-11-10 | 南京鸿雁讯通信息科技有限公司 | Industrial equipment fault early warning method based on similarity |
| CN110501154B (en)* | 2019-09-05 | 2020-12-29 | 国网河北省电力有限公司电力科学研究院 | A GIS equipment fault detection and location method based on MOSVR and boxplot analysis |
| CN110689203A (en)* | 2019-09-30 | 2020-01-14 | 国网山东省电力公司电力科学研究院 | Fault early warning method for primary fan in thermal power plant based on self-encoder |
| CN110764474B (en)* | 2019-10-16 | 2023-01-31 | 上海电气集团股份有限公司 | Method and system for monitoring running state of equipment |
| CN112747416B (en)* | 2019-10-31 | 2022-04-05 | 北京国双科技有限公司 | Energy consumption prediction method and device for air conditioning system |
| CN110991666B (en)* | 2019-11-25 | 2023-09-15 | 远景智能国际私人投资有限公司 | Fault detection method, training device, training equipment and training equipment for model, and storage medium |
| CN110794799A (en)* | 2019-11-28 | 2020-02-14 | 桂林电子科技大学 | Big data system with fault diagnosis function applied to industrial production |
| CN110806743A (en)* | 2019-12-05 | 2020-02-18 | 成都天玙兴科技有限公司 | Equipment fault detection and early warning system and method based on artificial intelligence |
| CN113027696B (en)* | 2019-12-24 | 2022-11-15 | 新疆金风科技股份有限公司 | Fault diagnosis method and device of hydraulic variable pitch system |
| CN111275331A (en)* | 2020-01-20 | 2020-06-12 | 张家口卷烟厂有限责任公司 | Monitoring method and device for production system |
| CN111563685B (en)* | 2020-05-09 | 2022-03-08 | 国网江苏省电力有限公司 | Power generation equipment state early warning method based on auto-associative kernel regression algorithm |
| CN111814848B (en)* | 2020-06-22 | 2024-04-09 | 浙江大学 | An adaptive early warning strategy design method for temperature faults in wind turbines |
| CN111814849B (en)* | 2020-06-22 | 2024-02-06 | 浙江大学 | A fault early warning method for key components of wind turbines based on DA-RNN |
| CN111783486B (en)* | 2020-06-24 | 2023-08-22 | 中国工商银行股份有限公司 | Maintenance early warning method and device for card reader equipment |
| CN111767275B (en)* | 2020-06-28 | 2024-04-19 | 北京林克富华技术开发有限公司 | Data processing method and device and data processing system |
| CN112067335A (en)* | 2020-09-03 | 2020-12-11 | 华能国际电力股份有限公司玉环电厂 | Power plant blower fault early warning method based on multivariate state estimation |
| CN112036089A (en)* | 2020-09-03 | 2020-12-04 | 华能国际电力股份有限公司玉环电厂 | A coal mill fault early warning method based on DPC-MND and multivariate state estimation |
| CN112231849B (en)* | 2020-11-09 | 2023-03-10 | 北京国信会视科技有限公司 | Axle temperature fault detection method based on NEST and SPRT fusion algorithm |
| CN112629585A (en)* | 2020-12-02 | 2021-04-09 | 三门核电有限公司 | Equipment on-line monitoring method and device based on multi-dimensional parameter estimation |
| CN112633561B (en)* | 2020-12-09 | 2024-10-18 | 北京名道恒通信息技术有限公司 | Intelligent production risk prediction and early warning method based on industrial big data |
| CN112580784A (en)* | 2020-12-16 | 2021-03-30 | 哈尔滨电站设备成套设计研究所有限公司 | Intelligent early warning method for equipment based on multi-input multi-output convolutional neural network |
| CN112611971B (en)* | 2020-12-23 | 2021-10-15 | 东北大学 | A data-driven networked DC motor anomaly detection method |
| CN112686389A (en)* | 2020-12-25 | 2021-04-20 | 中能融安(北京)科技有限公司 | Estimation method and estimation device for optimal value of equipment parameter |
| CN112668200B (en)* | 2021-01-06 | 2023-08-29 | 西安理工大学 | Special equipment safety analysis method and system |
| CN113011656B (en)* | 2021-03-22 | 2022-08-02 | 内蒙古电力(集团)有限责任公司内蒙古电力科学研究院分公司 | Method and system for early warning of power station auxiliary machine fault |
| CN112948163B (en)* | 2021-03-26 | 2023-09-19 | 中国航空无线电电子研究所 | Method for evaluating influence of equipment on functional failure based on BP neural network |
| CN113219939A (en)* | 2021-04-07 | 2021-08-06 | 山东润一智能科技有限公司 | Equipment fault prediction method and system based on residual autoregression |
| CN113420954A (en)* | 2021-05-08 | 2021-09-21 | 中国电建集团华东勘测设计研究院有限公司 | Engineering management service informatization method based on mechanism model |
| CN113110402B (en)* | 2021-05-24 | 2022-04-01 | 浙江大学 | Knowledge and data driven large-scale industrial system distributed state monitoring method |
| CN113722813B (en)* | 2021-07-20 | 2024-07-05 | 中科云谷科技有限公司 | Pump truck arm support condition monitoring method and device and pump truck |
| CN113657454B (en)* | 2021-07-23 | 2024-02-23 | 杭州安脉盛智能技术有限公司 | Nuclear power rotating machinery state monitoring method based on autoregressive BiGRU |
| CN113777908B (en)* | 2021-08-09 | 2023-11-21 | 杭州集益科技有限公司 | Control signal optimization method for multiple redundancy uncertain measurement parameters |
| CN113433917A (en)* | 2021-08-11 | 2021-09-24 | 内蒙古京隆发电有限责任公司 | DCS (distributed control system) for power plant and control method |
| CN113673600B (en)* | 2021-08-23 | 2024-02-02 | 中海石油气电集团有限责任公司 | Industrial signal abnormality early warning method, system, storage medium and computing device |
| CN113743531B (en)* | 2021-09-14 | 2024-08-23 | 深圳市出新知识产权管理有限公司 | Equipment working condition early warning method and related device |
| CN113915153A (en)* | 2021-09-30 | 2022-01-11 | 山东浪潮通软信息科技有限公司 | A method, system, equipment and storage medium for abnormality detection of a mining fan |
| CN113984114B (en)* | 2021-10-18 | 2022-12-06 | 大连理工大学 | A method for diagnosing abnormalities of underwater structures of marine floating platforms |
| CN113878214B (en)* | 2021-12-08 | 2022-03-25 | 苏芯物联技术(南京)有限公司 | Welding quality real-time detection method and system based on LSTM and residual distribution |
| CN114446027A (en)* | 2021-12-17 | 2022-05-06 | 广州盛原成科技有限公司 | A device fault alarm method, system, device and medium based on the Internet of Things |
| CN114152880B (en)* | 2022-02-08 | 2022-04-12 | 湖南大学 | Soft package battery sensor fault online detection method |
| CN115059634B (en)* | 2022-06-14 | 2023-04-18 | 重庆英科铸数网络科技有限公司 | Fault diagnosis method and device for air blowing equipment and electronic equipment |
| CN115169650B (en)* | 2022-06-20 | 2022-11-22 | 四川观想科技股份有限公司 | Equipment health prediction method for big data analysis |
| CN115524556A (en)* | 2022-09-22 | 2022-12-27 | 新源智储能源发展(北京)有限公司 | An abnormal identification method for DC capacitors of electrochemical energy storage converters |
| CN115585842B (en)* | 2022-09-30 | 2024-07-23 | 浙江浙能绍兴滨海热电有限责任公司 | Comprehensive monitoring algorithm of urea hydrolysis system based on power generation production environment-friendly specialty |
| CN115795999B (en)* | 2022-10-26 | 2023-08-01 | 国网新源控股有限公司 | Early warning method for abnormal performance of long-term service pumped storage unit |
| CN116049654B (en)* | 2023-02-07 | 2023-10-13 | 北京奥优石化机械有限公司 | Safety monitoring and early warning method and system for coal preparation equipment |
| CN116861164A (en)* | 2023-05-08 | 2023-10-10 | 华电电力科学研究院有限公司 | Turbine operation fault monitoring system |
| CN116384980B (en)* | 2023-05-25 | 2023-08-25 | 杭州青橄榄网络技术有限公司 | Repair reporting method and system |
| CN119882705A (en)* | 2025-03-27 | 2025-04-25 | 深圳中科精能源技术有限公司 | Operation monitoring and early warning method of intelligent energy-saving control equipment |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8301406B2 (en)* | 2008-07-24 | 2012-10-30 | University Of Cincinnati | Methods for prognosing mechanical systems |
| CN101930486B (en)* | 2010-07-12 | 2012-10-17 | 沈阳工业大学 | A device and method for predicting load index of wind turbines in a wind farm |
| CN102331772B (en)* | 2011-03-30 | 2013-03-27 | 浙江省电力试验研究院 | Method for carrying out early warning of abnormal superheated steam temperature and fault diagnosis on direct current megawatt unit |
| CN103324834B (en)* | 2013-04-07 | 2016-08-17 | 北京航空航天大学 | Mechatronic Systems under a kind of completely cut off data qualification and critical component life-span prediction method thereof |
| CN103488091A (en)* | 2013-09-27 | 2014-01-01 | 上海交通大学 | Data-driving control process monitoring method based on dynamic component analysis |
| Publication number | Publication date |
|---|---|
| CN104102773A (en) | 2014-10-15 |
| Publication | Publication Date | Title |
|---|---|---|
| CN104102773B (en) | A kind of equipment fault early-warning and state monitoring method | |
| CN109146246B (en) | Fault detection method based on automatic encoder and Bayesian network | |
| CN103499921B (en) | Structure changes fuzzy system sensor fault diagnosis method | |
| CN111505424A (en) | Large experimental device power equipment fault diagnosis method based on deep convolutional neural network | |
| CN109492790A (en) | Wind turbines health control method based on neural network and data mining | |
| CN114519923B (en) | Intelligent diagnosis and early warning method and system for power plant | |
| KR102108975B1 (en) | Apparatus and method for condition based maintenance support of naval ship equipment | |
| CN118035814A (en) | State fault early warning method and system based on digital twin field data analysis equipment | |
| CN117350377B (en) | A method and device for equipment fault diagnosis based on knowledge graph drive | |
| CN117032165A (en) | Industrial equipment fault diagnosis method | |
| Li et al. | Application of Big Data Technology in Evaluation of Operating State of High-Pressure Hydrogenation Heat Exchanger | |
| CN115081647A (en) | Industrial intelligent instrument fault pre-diagnosis method based on Bayesian network model | |
| CN113313365A (en) | Degradation early warning method and device for primary air fan | |
| Zhang et al. | A flexible monitoring framework via dynamic-multilayer graph convolution network | |
| CN118965247B (en) | Power plant data management method and system based on multi-source data | |
| CN114386603A (en) | Early warning method of generator fault in thermal power plant based on time series causal network | |
| CN118934496A (en) | Wind turbine monitoring method, device and medium based on SCADA data | |
| CN119444185A (en) | A self-service water filling machine intelligent operation and maintenance management method and platform | |
| CN118296313A (en) | Underwater robot fault diagnosis method, device, medium and system | |
| Dash et al. | A comparison of model-based and machine learning techniques for fault diagnosis | |
| CN117354168A (en) | Hybrid ensemble method for predictive modeling of Internet of things | |
| CN118210787A (en) | Real-time vehicle condition automatic detection method based on multidimensional vehicle big data and vehicle condition intelligent defect inference technology | |
| CN116011982A (en) | A method and system for on-line monitoring of coal mill grinding roller breakage | |
| CN118361650B (en) | Intelligent water-repellent method based on big data model | |
| Ouyang et al. | Fault diagnosis of process systems based on graph neural networks |
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| ASS | Succession or assignment of patent right | Owner name:ELECTRIC POWER RESEARCH INSTITUTE, STATE GRID SHAN Effective date:20141117 | |
| C41 | Transfer of patent application or patent right or utility model | ||
| C53 | Correction of patent for invention or patent application | ||
| CB03 | Change of inventor or designer information | Inventor after:Ding Shugeng Inventor after:Xu Yang Inventor after:Li Haibin Inventor after:An Baijing Inventor before:Xu Yang Inventor before:Li Haibin Inventor before:An Baijing | |
| COR | Change of bibliographic data | Free format text:CORRECT: INVENTOR; FROM: XU YANG LI HAIBIN AN BAIJING TO: DING SHUGENG XU YANG LI HAIBIN AN BAIJING | |
| TA01 | Transfer of patent application right | Effective date of registration:20141117 Address after:Xinluo Avenue high tech Zone of Ji'nan City, Shandong province 250101 silver bearing No. 2008 building B block 5 layer Applicant after:Shandong Luneng Software Technology Co., Ltd. Applicant after:Electric Power Research Institute of State Grid Shandong Electric Power Company Address before:Xinluo Avenue high tech Zone of Ji'nan City, Shandong province 250101 silver bearing No. 2008 building B block 5 layer Applicant before:Shandong Luneng Software Technology Co., Ltd. | |
| C53 | Correction of patent for invention or patent application | ||
| CB03 | Change of inventor or designer information | Inventor after:Ding Shugeng Inventor after:Xu Yang Inventor after:Li Haibin Inventor after:An Baijing Inventor before:Ding Shugeng Inventor before:Xu Yang Inventor before:Li Haibin Inventor before:An Baijing | |
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CP01 | Change in the name or title of a patent holder | ||
| CP01 | Change in the name or title of a patent holder | Address after:250101 5th floor, block B, Yinhe building, 2008 Xinluo street, high tech Zone, Jinan City, Shandong Province Patentee after:Shandong luruan Digital Technology Co.,Ltd. Patentee after:ELECTRIC POWER RESEARCH INSTITUTE OF STATE GRID SHANDONG ELECTRIC POWER Co. Address before:250101 5th floor, block B, Yinhe building, 2008 Xinluo street, high tech Zone, Jinan City, Shandong Province Patentee before:SHANDONG LUNENG SOFTWARE TECHNOLOGY Co.,Ltd. Patentee before:ELECTRIC POWER RESEARCH INSTITUTE OF STATE GRID SHANDONG ELECTRIC POWER Co. |