CN112910288A - Over-temperature early warning method based on inverter radiator temperature prediction - Google Patents
Over-temperature early warning method based on inverter radiator temperature predictionDownload PDFInfo
- Publication number
- CN112910288A CN112910288ACN202011442899.4ACN202011442899ACN112910288ACN 112910288 ACN112910288 ACN 112910288ACN 202011442899 ACN202011442899 ACN 202011442899ACN 112910288 ACN112910288 ACN 112910288A
- Authority
- CN
- China
- Prior art keywords
- temperature
- inverter
- warning
- distribution
- over
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02M—APPARATUS FOR CONVERSION BETWEEN AC AND AC, BETWEEN AC AND DC, OR BETWEEN DC AND DC, AND FOR USE WITH MAINS OR SIMILAR POWER SUPPLY SYSTEMS; CONVERSION OF DC OR AC INPUT POWER INTO SURGE OUTPUT POWER; CONTROL OR REGULATION THEREOF
- H02M7/00—Conversion of AC power input into DC power output; Conversion of DC power input into AC power output
- H02M7/42—Conversion of DC power input into AC power output without possibility of reversal
- H02M7/44—Conversion of DC power input into AC power output without possibility of reversal by static converters
- H02M7/48—Conversion of DC power input into AC power output without possibility of reversal by static converters using discharge tubes with control electrode or semiconductor devices with control electrode
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Power Engineering (AREA)
- Inverter Devices (AREA)
Abstract
Translated fromChinese
Description
Translated fromChinese技术领域technical field
本发明涉及电力电子技术领域，尤其涉及一种逆变器过温预警方法。The invention relates to the technical field of power electronics, in particular to an inverter over-temperature warning method.
背景技术Background technique
随着经济的快速发展，化石能源消耗量持续增加，全球正面临着日益严重的能源短缺和环境破坏问题。光伏发电作为环境友好型的可再生能源发电技术,是目前可再生能源中最具规模化开发条件和商业化发展前景的发电方式之一,正受到越来越多的关注。由于光伏电站常常建于戈壁、海岛等自然地理环境恶劣的偏远地区,确保核心器件光伏并网逆变器可靠稳定运行尤其重要。过温故障是逆变器经常出现的一种故障类型。引发这一故障的原因有很多，例如环境温度上升、逆变器散热风机故障、逆变器发电功率过高、逆变器老化、逆变器电流电压超限等。逆变器过温会导致发电功率降额，严重时会直接停机，造成巨大的发电经济损失。因此,快速准确地预测逆变器散热器温度，对逆变器提前发出过温预警十分必要，有助于适时调整供电计划，提升光伏电站运营的经济效益。With the rapid economic development, the consumption of fossil energy continues to increase, and the world is facing increasingly serious problems of energy shortage and environmental damage. Photovoltaic power generation, as an environment-friendly renewable energy power generation technology, is one of the power generation methods with the most large-scale development conditions and commercial development prospects in renewable energy, and is receiving more and more attention. Since photovoltaic power plants are often built in remote areas with harsh natural geographical environments such as the Gobi and islands, it is particularly important to ensure the reliable and stable operation of the photovoltaic grid-connected inverters for core components. Over-temperature fault is a type of fault that often occurs in inverters. There are many reasons for this failure, such as rising ambient temperature, failure of the inverter cooling fan, excessive power generation of the inverter, aging of the inverter, over-limit of the current and voltage of the inverter, etc. Over-temperature of the inverter will lead to derating of power generation, and in severe cases, it will directly shut down, resulting in huge economic losses in power generation. Therefore, it is necessary to quickly and accurately predict the temperature of the radiator of the inverter, and it is necessary to issue an over-temperature warning to the inverter in advance, which will help to adjust the power supply plan in a timely manner and improve the economic benefits of the operation of the photovoltaic power station.
现有关于逆变器温度预测的发明主要是基于逆变器的内部物理参数进行机理建模。这类方案一般会首先建立光伏逆变器元件温度预测方程，预测方程所需的参数信息包括环境温度、散热器温升、逆变器元件温升以及逆变器元件温度。其中环境温度是利用温度传感器或气象监测仪测量得到的。接下来，此类方案会通过建立光伏逆变器散热器的热平衡状态方程，计算逆变器散热器温升。之后，利用绝缘栅双极型晶体管(Insulated GateBipolar Transistor，IGBT)散热系数和功率消耗建立IGBT在稳定状态下和散热器之间的温度差方程，也即逆变器元件温升。最后，根据建立的光伏逆变器元件温度预测方程，结合求得的环境温度、散热器温升、逆变器元件温升以及光伏逆变器元件温度，计算出光伏逆变器元件温度的预测值。The existing inventions about inverter temperature prediction are mainly based on the mechanism modeling based on the internal physical parameters of the inverter. This type of scheme generally first establishes a photovoltaic inverter element temperature prediction equation. The parameter information required for the prediction equation includes ambient temperature, radiator temperature rise, inverter element temperature rise, and inverter element temperature. The ambient temperature is measured by a temperature sensor or a weather monitor. Next, such a scheme calculates the temperature rise of the inverter radiator by establishing the thermal equilibrium state equation of the photovoltaic inverter radiator. Then, the temperature difference equation between the IGBT and the heat sink in a steady state, that is, the temperature rise of the inverter element, is established by using the heat dissipation coefficient and power consumption of an insulated gate bipolar transistor (IGBT). Finally, according to the established photovoltaic inverter element temperature prediction equation, combined with the obtained ambient temperature, radiator temperature rise, inverter element temperature rise and photovoltaic inverter element temperature, the prediction of the photovoltaic inverter element temperature is calculated. value.
现有发明的另一种方案是根据大量数据通过数理统计方法建立一种气象相关的光伏组件工作温度预测方法。该方案首先建立了光伏组件工作温度与环境温度、辐射强度和风速的非线性模型。之后，根据能量守恒定律得到含参数的方程，最后，根据大量数据通过数理统计的方法进行线性拟合，最终得到线性温度预测模型。Another solution of the existing invention is to establish a weather-related photovoltaic module operating temperature prediction method through a mathematical statistical method according to a large amount of data. The scheme firstly establishes the nonlinear model of photovoltaic module operating temperature and ambient temperature, radiation intensity and wind speed. After that, the equation with parameters is obtained according to the law of conservation of energy. Finally, the linear fitting is carried out by mathematical statistics method according to a large amount of data, and finally a linear temperature prediction model is obtained.
现有发明中也有基于神经网络的逆变器温度预测方案。该方案首先搭建了一个用于IGBT结温预测的反向传播(Back Propagation，BP)神经网络，其中输入层的个数为1，包括三个神经元，分别用于输入相电流峰值、开关频率和环境温度；输出层的个数为1，包括一个神经元，用于输出IGBT结温。之后，采用ANSYS Icepak软件构建逆变器的3D热仿真模型，通过改变环境信息以及其它参数信息，采集了多组IGBT结温以及相应的结温特征，其中，结温特征包括：相电流峰值、开关频率和环境温度，这些数据经过预处理后被用作BP神经网络训练样本，通过划分训练集和测试集，训练得到了基于BP神经网络的IGBT结温预测模型。最后，将采集到的实际结温特征输入到预训练好的IGBT结温预测模型，得到待测IGBT的结温。There are also inverter temperature prediction schemes based on neural networks in the existing inventions. The scheme first builds a Back Propagation (BP) neural network for IGBT junction temperature prediction, in which the number of input layers is 1, including three neurons, which are used to input phase current peak value and switching frequency respectively. and ambient temperature; the number of output layers is 1, including one neuron for outputting the IGBT junction temperature. After that, ANSYS Icepak software was used to build the 3D thermal simulation model of the inverter. By changing the environmental information and other parameter information, several groups of IGBT junction temperatures and corresponding junction temperature characteristics were collected. The junction temperature characteristics include: phase current peak value, Switching frequency and ambient temperature, these data are preprocessed and used as BP neural network training samples. By dividing training set and test set, the IGBT junction temperature prediction model based on BP neural network is obtained by training. Finally, the collected actual junction temperature characteristics are input into the pre-trained IGBT junction temperature prediction model to obtain the junction temperature of the IGBT to be measured.
现有发明方法都将逆变器的散热器温度视为已知量，或者假定这个量可以通过温度传感器测得，然而实际现场安装的逆变器出于成本考虑，很少具备散热器测温功能。同时，现有发明方案忽略了对现场采集数据的预处理问题，例如三相电流和三相电压数据中包含有许多干扰噪声，会对预测结果的精度造成较大的影响。此外，现有基于神经网络的预测方案采用的是最基础的BP神经网络，没有考虑深度学习下多层网络预测方法，而且仅仅是针对逆变器各个时刻的温度值进行点预测，没有利用逆变器温度的后验分布信息。The existing invention methods all regard the radiator temperature of the inverter as a known quantity, or assume that this quantity can be measured by a temperature sensor. However, due to cost considerations, the inverters actually installed on site rarely have radiator temperature measurement. Function. At the same time, the existing invention scheme ignores the preprocessing of the data collected on site. For example, the three-phase current and three-phase voltage data contain a lot of interference noise, which will greatly affect the accuracy of the prediction result. In addition, the existing neural network-based prediction scheme uses the most basic BP neural network, without considering the multi-layer network prediction method under deep learning, and only performs point prediction for the temperature value of the inverter at each moment, without using the inverse Information on the posterior distribution of the transformer temperature.
因此，基于以上分析，在大型地面光伏电站中，亟需一种有效的数据预处理和准确的逆变器温度预测方法，最小化系统的计算成本，满足光伏传感器数据被实时存储、处理的时延要求，同时根据逆变器温度预测值及时有效地把握逆变器散热器的温度变化趋势，实现逆变器过温预警。Therefore, based on the above analysis, in large-scale ground photovoltaic power plants, an effective data preprocessing and accurate inverter temperature prediction method is urgently needed to minimize the calculation cost of the system and meet the needs of real-time storage and processing of photovoltaic sensor data. At the same time, according to the predicted value of the inverter temperature, the temperature change trend of the inverter radiator can be grasped in a timely and effective manner to realize the inverter over-temperature warning.
为此，本领域的技术人员致力于开发一种大型地面光伏电站中基于经验模态分解(Empirical Mode Decomposition,EMD)和贝叶斯长短时记忆网络(Bayesian Long Short-Term Memory，BLSTM)的逆变器过温预警方法，从而可以对逆变器提前发出过温预警，对于避免逆变器过温降额和停机造成巨大的发电经济损失和重大安全事故具有重要意义。To this end, those skilled in the art are committed to developing an inverse method based on Empirical Mode Decomposition (EMD) and Bayesian Long Short-Term Memory (BLSTM) in large-scale ground photovoltaic power plants. The inverter over-temperature warning method can issue an over-temperature warning to the inverter in advance, which is of great significance to avoid huge power generation economic losses and major safety accidents caused by over-temperature derating and shutdown of the inverter.
发明内容SUMMARY OF THE INVENTION
有鉴于现有技术的上述缺陷，本发明所要解决的技术问题是如何通过数据预处理以及采用多层神经网络挖掘逆变器预测数据的时序性信息和后验分布信息，快速准确地预测逆变器散热器温度，并设计逆变器过温预警机制，从而实现逆变器过温预警功能，避免大型地面光伏电站由于逆变器过温降额和停机造成的巨大发电经济损失，以及严重安全事故的发生。In view of the above-mentioned defects of the prior art, the technical problem to be solved by the present invention is how to quickly and accurately predict the inverter through data preprocessing and using a multi-layer neural network to mine the time series information and posterior distribution information of the inverter prediction data. The temperature of the radiator of the inverter is controlled, and the inverter over-temperature warning mechanism is designed, so as to realize the inverter over-temperature warning function, and avoid the huge power generation economic loss caused by the inverter over-temperature derating and shutdown of the large-scale ground photovoltaic power station, as well as the serious safety the occurrence of the accident.
为实现上述目的，本发明提供了一种基于逆变器散热器温度预测的过温预警方法，首先通过引入经验模态分解(Empirical Mode Decomposition,EMD)方法，去除了数据采集过程中产生的噪声；然后基于贝叶斯长短时记忆网络(Bayesian Long Short-TermMemory，BLSTM)，提出了一种逆变器散热器温度预测模型，包括：(1)构建传统的长短时记忆网络(Long Short-Term Memory,LSTM)，提取数据时序性特征；(2)引入贝叶斯思想，使用蒙特卡洛dropout方法来进行近似推断，通过最小化网络权重的近似分布和后验分布之间KL散度学习网络权重；(3)根据逆变器温度预测结果的分布情况，分别采用平方马氏距离和局部密度比这两种度量方式来计算模型预测值与实际值之间的偏差情况，调整网络权重。In order to achieve the above purpose, the present invention provides an over-temperature warning method based on the temperature prediction of the inverter radiator. First, by introducing an empirical mode decomposition (Empirical Mode Decomposition, EMD) method, the noise generated in the data acquisition process is removed. ; Then based on the Bayesian Long Short-Term Memory (BLSTM), a temperature prediction model of the inverter radiator is proposed, including: (1) Constructing a traditional Long Short-Term Memory network (Long Short-Term Memory). Memory, LSTM), extracting the temporal features of the data; (2) Introducing the Bayesian idea, using the Monte Carlo dropout method for approximate inference, and learning the network by minimizing the KL divergence between the approximate distribution of the network weight and the posterior distribution (3) According to the distribution of the inverter temperature prediction results, the square Mahalanobis distance and the local density ratio are used to calculate the deviation between the model predicted value and the actual value, and adjust the network weight.
进一步地，所述基于逆变器散热器温度预测的过温预警方法包括如下步骤：Further, the over-temperature warning method based on the temperature prediction of the inverter radiator includes the following steps:
步骤a、数据清洗，实际逆变器采集的三相电流和三相电压数据包含较多的噪声，本发明选用经验模态分解对原始数据进行清洗；Step a, data cleaning, the three-phase current and three-phase voltage data collected by the actual inverter contain a lot of noise, and the present invention selects empirical mode decomposition to clean the original data;
步骤b、数据预处理，将数据转化为神经网络可以处理的有效数据；Step b, data preprocessing, convert the data into valid data that can be processed by the neural network;
步骤c、建立LSTM；Step c, establish LSTM;
步骤d、建立BLSTM；Step d, establish BLSTM;
步骤e、基于变分dropout实现BLSTM；Step e. Implement BLSTM based on variational dropout;
步骤f、多步预测，对逆变器建立超短时、短时、长时等各种时间尺度下的预警机制，所设计BLSTM的输出维度是可调节的；Step f, multi-step prediction, establish an early warning mechanism for the inverter under various time scales such as ultra-short-term, short-term, and long-term, and the output dimension of the designed BLSTM is adjustable;
步骤g、量化近似后验分布；Step g, quantify the approximate posterior distribution;
步骤h、逆变器过温预警。Step h, the inverter over-temperature warning.
进一步地，所述步骤a还包括：Further, the step a also includes:
步骤a1、信号分解，利用所述经验模态方法将原始所述三相电流和所述三相电压数据分别分解为14个本征模函数(Intrinsic Mode Function，IMF)分量；Step a1, signal decomposition, using the empirical mode method to decompose the original three-phase current and the three-phase voltage data into 14 eigenmode function (Intrinsic Mode Function, IMF) components respectively;
步骤a2、信号滤波，然后将分解得到的14个所述IMF分量按频率从低到高排序，最后四个高频所述IMF分量被视为伪分量，也即高频噪声，将其剔除；Step a2, signal filtering, then the 14 IMF components obtained by decomposition are sorted from low to high frequency, and the last four high-frequency IMF components are regarded as pseudo components, that is, high-frequency noise, and are eliminated;
步骤a3、信号重构，最后将信号滤波后所述三相电流和所述三相电压各自的10个有效所述IMF分量分别叠加，得到重构后的所述三相电流和所述三相电压数据；Step a3, signal reconstruction, and finally superimpose the respective 10 valid IMF components of the three-phase current and the three-phase voltage after signal filtering, to obtain the reconstructed three-phase current and the three-phase current. voltage data;
进一步地，所述步骤b还包括：Further, the step b also includes:
步骤b1、数据集划分，首先对训练集和测试集的比例划分参数进行设置，将所述逆变器历史运行数据量的70％划分为所述训练集，用于训练逆变器散热器温度预测模型，另外30％用于模型性能测试，作为所述测试集；Step b1, data set division, first set the proportional division parameters of the training set and the test set, and divide 70% of the historical operation data of the inverter into the training set, which is used to train the temperature of the inverter radiator Predictive models, and another 30% for model performance testing, as the test set;
步骤b2、数据归一化，选用MinMaxScaler方法对数据进行处理，数学表达式如下：Step b2, normalize the data, select the MinMaxScaler method to process the data, and the mathematical expression is as follows:
其中xmax代表输入样本中的最大值，xmin代表样本中的最小值，X是标准化后的结果，范围在0到1之间。where xmax represents the maximum value in the input sample, xmin represents the minimum value in the sample, and X is the normalized result, ranging from 0 to 1.
进一步地，所述步骤c中，为了学习所述LSTM网络的参数，损失函数通常选择为均方误差损失：Further, in the step c, in order to learn the parameters of the LSTM network, the loss function is usually selected as the mean square error loss:
或者选取交叉熵损失函数：Or choose the cross-entropy loss function:
其中Θ表示所述LSTM的参数集合，
L(Θ)＝J(Θ)+λ(||Wf||2+||Wi||2+||Wc||2+||Wo||2)L(Θ)=J(Θ)+λ(||Wf ||2 +||Wi ||2 +||Wc ||2 +||Wo ||2 )
其中λ是正则化参数。where λ is the regularization parameter.
进一步地，所述BLSTM通过整合所述LSTM网络参数的统计建模，得到了一个输入输出映射的概率模型；所述概率模型求解方法包括：Further, the BLSTM obtains a probability model of input-output mapping by integrating the statistical modeling of the LSTM network parameters; the method for solving the probability model includes:
步骤1、参数ω＝{Wf,Wi,Wc,Wo,bf,by,bc,bo}作为先验分布p(ω)的随机变量，因此，所述LSTM的细胞状态和输出可以重新表示为：
Ct＝fiω(xt,ht-1)Ct =fiω (xt ,ht-1 )
其中下标i和o分别表示隐藏层和输出层节点的索引，fiω和
每个数据点输出的可能性为：The probability of each data point output is:
其中τ是反映数据固有噪声的精度参数；where τ is the precision parameter reflecting the inherent noise of the data;
步骤2、给定一个包含X(所述逆变器的所述三相电流、所述三相电压以及所述逆变器机内空气温度、变压器温度)和Y(光伏系统中所述逆变器散热器的实际温度)的训练数据集(大型地面光伏电站存储的所述逆变器历史运行数据)，在参数空间上学习需要估计后验分布p(ω∣X,Y)；利用更新后的分布，通过积分得到逆变器散热器温度的预测输出y*的分布：Step 2. Given a set containing X (the three-phase current of the inverter, the three-phase voltage, the air temperature in the inverter, and the temperature of the transformer) and Y (the inverter in the photovoltaic system) (the actual temperature of the radiator) of the training data set (the historical operating data of the inverter stored in the large-scale ground photovoltaic power station), the posterior distribution p(ω∣X,Y) needs to be estimated for learning in the parameter space; after using the updated , the distribution of the predicted output y* of the inverter radiator temperature is obtained by integrating:
p(y*∣x*,X,Y)＝∫p(y*∣x*,ω)p(ω∣X,Y)dωp(y* ∣x* ,X,Y)＝∫p(y* ∣x* ,ω)p(ω∣X,Y)dω
其中，x*表示一个新的观测值，对于先验分布，通常选择权重矩阵p(W)上的标准零均值高斯先验，预测的不确定性将直接反映在后验分布p(y*∣x*,X,Y)。Among them, x* represents a new observation value. For the prior distribution, the standard zero-mean Gaussian prior on the weight matrix p(W) is usually selected, and the uncertainty of the prediction will be directly reflected in the posterior distribution p(y* ∣ x* ,X,Y).
进一步地，所述步骤e中，首先用简单的参数化分布q(ω)近似难处理的后验分布p(ω∣X,Y)，然后，使用所述q(ω)的蒙特卡罗(MC)积分来近似，具体包括以下步骤：Further, in the step e, first approximate the intractable posterior distribution p(ω∣X,Y) with a simple parameterized distribution q(ω), and then use the Monte Carlo ( MC) integral to approximate, including the following steps:
步骤e1、通过权重矩阵分解得到近似分布，对于wk的每一行，变分dropout会施加一个变化分布，也即近似分布q(ω)是由两个具有小方差的高斯分布混合得到：In step e1, the approximate distribution is obtained by decomposing the weight matrix. For each row of wk , the variational dropout will impose a changing distribution, that is, the approximate distribution q(ω) is obtained by mixing two Gaussian distributions with small variance:
其中p是预先定义的dropout概率，σ2是一个小精度参数，mk是一个变分参数；where p is a pre-defined dropout probability,σ2 is a small precision parameter, andmk is a variational parameter;
步骤e2、通过最小化近似分布与后验分布之间的KL散度来学习网络的权值，使得变分推论中的近似分布q(ω)尽可能接近真实的后验分布p(ω∣X,Y)，具体来说，即最小化下面的目标函数：Step e2, learn the weights of the network by minimizing the KL divergence between the approximate distribution and the posterior distribution, so that the approximate distribution q(ω) in the variational inference is as close as possible to the real posterior distribution p(ω∣X , Y), specifically, minimizing the following objective function:
KL(q(ω)||p(ω∣X,Y)))KL(q(ω)||p(ω∣X,Y)))
进一步地，所述步骤f是通过调整所述步骤d中所述BLSTM的输出神经元个数来实现多步预测的，其中每个所述神经元对应于一个预测步长。Further, in the step f, multi-step prediction is realized by adjusting the number of output neurons of the BLSTM in the step d, wherein each neuron corresponds to a prediction step size.
进一步地，所述步骤g引入了两种方法来量化实际逆变器散热器温度值与其对应预测分布的偏差大小,具体包括：Further, two methods are introduced in the step g to quantify the deviation between the actual inverter radiator temperature value and its corresponding predicted distribution, specifically including:
(1)适用于高斯预测分布的平方马氏距离方法：如果蒙特卡洛方法得到的预测分布是高斯分布，或者近似高斯分布，则可以使用平方马氏距离来表征预测值与实际值的偏差大小。首先，使用预测分布在t时刻的蒙特卡洛样本
当观测真值Xt可用时，平方马氏距离由下式确定：When the observed ground truth Xt is available, the squared Mahalanobis distance is determined by:
(2)适用于非高斯分布的局部密度比方法：如果预测分布不能很好地描述为高斯分布，那么就需要利用非参数方法来量化每个观测值的异常。针对这些情况，引入一种与局部离群因子密切相关的局部密度比(LDR)方法；LDR统计量利用其周围最近的k个邻居(k-NNs)观测值的密度的估计值，量化每个新观测值相于其预测分布的偏离情况；基于k-NNs的局部密度估计值
其中，
然后，观测值xt的局部密度比定义为：Then, the local density ratio of observations xt is defined as:
也即，xt周围最近的k个预测值的局部密度求平均值后除以xt的局部密度。That is, the local density of the k nearest predicted values around xt is averaged and divided by the local density of xt .
进一步地，所述步骤h中，按预测温度占逆变器温度上限的比例，将逆变器过温预警分为轻度预警(70％～80％)、中度预警(80％～90％)和重度预警(90％及以上)三个等级。同时，针对不同的预警等级，本发明设计了相应的响应机制，可以分为以下几种情况处理：Further, in the step h, according to the proportion of the predicted temperature to the upper limit of the inverter temperature, the inverter over-temperature warning is divided into mild warning (70%-80%) and moderate warning (80%-90%). ) and severe warning (90% and above) three levels. At the same time, for different warning levels, the present invention designs a corresponding response mechanism, which can be divided into the following situations:
S1，出现重度预警，任何时候，当某次温度预测结果达到重度预警时，系统会立即发出过温预警，提示电站运维人员对逆变器采取相应的检修维护措施；S1, a severe warning occurs. At any time, when a certain temperature prediction result reaches a severe warning, the system will immediately issue an over-temperature warning, prompting the power station operation and maintenance personnel to take corresponding maintenance measures for the inverter;
S2，出现中度预警，当某次温度预测结果为中度预警时，系统并不会马上发出过温警告，而是会继续追踪未来连续2次(也即未来5分钟和10分钟)的预测结果，如果预测结果均为中度预警且呈温度上升趋势(10分钟时温度预测值高于5分钟时温度预测值)，系统才会发出过温警告。特殊的，当未来两次预测结果中出现重度预警时，会立即告警；S2, there is a moderate warning. When a certain temperature prediction result is a moderate warning, the system will not immediately issue an over-temperature warning, but will continue to track the forecast for 2 consecutive times in the future (that is, 5 minutes and 10 minutes in the future). As a result, the system will only issue an over-temperature warning if the forecast results are all moderate warnings and the temperature is on a rising trend (the temperature forecast at 10 minutes is higher than the temperature forecast at 5 minutes). Specially, when there is a severe warning in the next two forecast results, the warning will be given immediately;
S3，出现轻度预警，当某次温度预测结果为轻度预警时，系统并不会马上发出过温警告，而是会继续追踪未来连续5次(也即未来5分钟、10分钟、15分钟、20分钟和25分钟)的预测结果，如果预测结果均为轻度预警及以上且温度呈上升趋势，系统才会发出过温警告。特殊的，当未来5次预测结果中出现重度预警时，会立即告警；S3, there is a mild warning. When a certain temperature prediction result is a mild warning, the system will not issue an over-temperature warning immediately, but will continue to track 5 consecutive times in the future (that is, 5 minutes, 10 minutes, 15 minutes in the future). , 20 minutes and 25 minutes), if the forecast results are all mild warnings and above and the temperature is on the rise, the system will issue an over-temperature warning. Specially, when there is a severe warning in the next 5 forecast results, the warning will be issued immediately;
S4，出现轻度预警与中度预警跳变，当预测温度从轻度预警上升为中度预警时，温度呈上升趋势，应按照中度预警规则进行后续判断。当预测温度从中度预警降低为轻度预警，温度呈下降趋势，应按照轻度预警规则进行后续判断。特殊的，当预测结果中出现重度预警时，会立即告警。S4, there is a jump between mild early warning and moderate early warning. When the predicted temperature rises from mild early warning to moderate early warning, the temperature is on the rise, and follow-up judgments should be made according to the moderate early warning rules. When the predicted temperature decreases from a moderate early warning to a mild early warning, and the temperature shows a downward trend, follow-up judgments should be made according to the mild early warning rules. Specially, when there is a severe warning in the forecast result, it will immediately give an alarm.
本发明针对大型地面光伏电站逆变器难以通过机理建模测量及预测内部模块温度、容易发生过温故障导致发电功率降额停机造成巨大发电经济损失的问题，利用EMD方法有效提升了原始数据质量，基于贝叶斯长短时记忆网络实现了逆变器散热器温度的快速准确预测，有效地把握了逆变器散热器温度变化趋势，实现了逆变器过温预警。Aiming at the problems that large-scale ground photovoltaic power station inverters are difficult to measure and predict the temperature of internal modules through mechanism modeling, and are prone to over-temperature faults, resulting in power generation derating and shutdown, resulting in huge power generation economic losses, the EMD method is used to effectively improve the quality of the original data , based on the Bayesian long-short-term memory network, the rapid and accurate prediction of the temperature of the inverter radiator is realized, the temperature change trend of the inverter radiator is effectively grasped, and the over-temperature warning of the inverter is realized.
以下将结合附图对本发明的构思、具体结构及产生的技术效果作进一步说明，以充分地了解本发明的目的、特征和效果。The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings, so as to fully understand the purpose, characteristics and effects of the present invention.
附图说明Description of drawings
图1是本发明的一个较佳实施例的整体流程图；Fig. 1 is the overall flow chart of a preferred embodiment of the present invention;
图2是本发明的一个较佳实施例的LSTM整体运行机制；Fig. 2 is the LSTM overall operation mechanism of a preferred embodiment of the present invention;
图3是本发明的一个较佳实施例的BLSTM结构图；Fig. 3 is the BLSTM structure diagram of a preferred embodiment of the present invention;
图4是本发明的一个较佳实施例的基于云边协同的逆变器过温预警机制。FIG. 4 is an inverter over-temperature warning mechanism based on cloud-edge collaboration according to a preferred embodiment of the present invention.
具体实施方式Detailed ways
以下参考说明书附图介绍本发明的多个优选实施例，使其技术内容更加清楚和便于理解。本发明可以通过许多不同形式的实施例来得以体现，本发明的保护范围并非仅限于文中提到的实施例。The following describes several preferred embodiments of the present invention with reference to the accompanying drawings, so as to make its technical content clearer and easier to understand. The present invention can be embodied in many different forms of embodiments, and the protection scope of the present invention is not limited to the embodiments mentioned herein.
如图4所示，本发明针对大型地面光伏电站提出了一种基于经验模态分解和贝叶斯长短时记忆网络的逆变器过温预警系统。As shown in FIG. 4 , the present invention proposes an inverter over-temperature warning system based on empirical mode decomposition and Bayesian long-short-term memory network for large-scale ground photovoltaic power plants.
系统在每个采样时刻从逆变器获取实时数据信息，利用EMD方法对数据中的三相电压和三相电流数据进行清洗，剔除其中夹杂的高频噪声。之后，本地计算设备利用事先训练好的逆变器散热器温度预测模块，对清洗后的数据进行处理，输出逆变器散热器温度的预测区间和点预测值。最后，根据本发明所建立的多级过温预警机制，可以有效避免漏判和误判。The system obtains real-time data information from the inverter at each sampling time, uses the EMD method to clean the three-phase voltage and three-phase current data in the data, and removes the high-frequency noise. After that, the local computing device uses the pre-trained inverter radiator temperature prediction module to process the cleaned data, and outputs the predicted interval and point predicted value of the inverter radiator temperature. Finally, according to the multi-level over-temperature warning mechanism established by the present invention, missed judgment and misjudgment can be effectively avoided.
下面结合附图1，对本发明作进一步说明。Below in conjunction with accompanying drawing 1, the present invention will be further described.
本发明针对大型地面光伏电站逆变器难以通过机理建模测量及预测内部模块温度、容易发生过温故障导致发电功率降额停机造成巨大发电经济损失的问题，利用EMD方法有效提升了原始数据质量，基于贝叶斯长短时记忆网络实现了逆变器散热器温度的快速准确预测，有效地把握了逆变器散热器温度变化趋势，实现了逆变器过温预警，具体包括以下步骤：Aiming at the problems that large-scale ground photovoltaic power station inverters are difficult to measure and predict the temperature of internal modules through mechanism modeling, and are prone to over-temperature faults, resulting in power generation derating and shutdown, resulting in huge power generation economic losses, the EMD method is used to effectively improve the quality of the original data , based on the Bayesian long-short-term memory network, the rapid and accurate prediction of the temperature of the inverter radiator is realized, the temperature change trend of the inverter radiator is effectively grasped, and the over-temperature warning of the inverter is realized, which includes the following steps:
1、数据清洗：考虑到实际逆变器采集的三相电流和三相电压数据包含较多的噪声，本发明选用经验模态分解对原始数据进行清洗，提升数据质量。首先是信号分解，本发明利用EMD将原始三相电流和三相电压数据分别分解为14个本征模函数(Intrinsic ModeFunction，IMF)分量。其次是信号滤波，本发明分解得到的14个IMF分量按频率从低到高排序，最后四个高频IMF分量被视为伪分量，也即高频噪声，将其剔除。最后是信号重构，本发明将信号滤波后三相电流和三相电压各自的10个有效IMF分量分别叠加，得到重构后的三相电流和三相电压数据。1. Data cleaning: Considering that the three-phase current and three-phase voltage data collected by the actual inverter contain more noise, the present invention selects empirical mode decomposition to clean the original data to improve the data quality. The first is signal decomposition. The present invention uses EMD to decompose the original three-phase current and three-phase voltage data into 14 eigenmode function (Intrinsic ModeFunction, IMF) components respectively. The second is signal filtering. The 14 IMF components decomposed by the present invention are sorted from low to high frequency, and the last four high-frequency IMF components are regarded as pseudo components, that is, high-frequency noise, and are eliminated. The last is signal reconstruction. The present invention superimposes the respective 10 effective IMF components of the three-phase current and three-phase voltage after signal filtering to obtain reconstructed three-phase current and three-phase voltage data.
2、数据预处理：数据清洗之后，为了将数据转化为神经网络可以处理的有效数据，本发明进一步对数据进行了预处理。首先是数据集划分，本发明对训练集和测试集的比例划分参数进行设置，该参数的大小可在0～1之间(需保证大于0小于1)。本发明选取的数据集划分比例为0.7，也即将逆变器历史运行数据量的70％划分为训练集用于训练逆变器散热器温度预测模型，另外30％用于模型性能测试。之后是数据归一化，逆变器记录的历史数据包含许多不同的类型，具有不同的数据尺度，例如三相电压和三相电流的数值都在三四百，但温度数值只有几十，逆变器效率为小数，因此需要对多源异构数据进行标准化处理。本发明选用的是MinMaxScaler方法对数据进行归一化处理，数学表达式如下：2. Data preprocessing: After data cleaning, in order to convert the data into valid data that can be processed by the neural network, the present invention further preprocesses the data. The first is the division of the data set. The present invention sets the proportional division parameter of the training set and the test set, and the size of the parameter can be between 0 and 1 (it must be ensured that it is greater than 0 and less than 1). The data set division ratio selected by the present invention is 0.7, that is, 70% of the inverter historical operation data is divided into a training set for training the inverter radiator temperature prediction model, and the other 30% is used for model performance testing. After that is data normalization. The historical data recorded by the inverter contains many different types with different data scales. For example, the three-phase voltage and three-phase current value are in three or four hundred, but the temperature value is only a few dozen. The efficiency of the transformer is decimal, so it is necessary to standardize the multi-source heterogeneous data. The present invention selects the MinMaxScaler method to normalize the data, and the mathematical expression is as follows:
其中xmax代表输入样本中的最大值，xmin代表样本中的最小值，X是标准化后的结果，范围在0到1之间。where xmax represents the maximum value in the input sample, xmin represents the minimum value in the sample, and X is the normalized result, ranging from 0 to 1.
3、建立长短时记忆网络：LSTM通过精心设计的被称为“门”的结构来去除或者增加信息到细胞状态中。门是一种决定信息是否通过的方法，包含一个Sigmoid神经网络层和一个按位的乘法操作。其中，Sigmoid函数的数学公式为：3. Build a long-term and short-term memory network: LSTM removes or adds information to the cell state through carefully designed structures called "gates". A gate is a method of deciding whether information passes through, and consists of a sigmoid neural network layer and a bitwise multiplication operation. Among them, the mathematical formula of the sigmoid function is:
其中x表示输入，Sigmoid函数的输出值y是一个0到1之间的数值，0代表“不许任何量通过”，1就指“允许任意量通过”。LSTM拥有三个门，即输入门、遗忘门和输出门，用于保护和控制细胞状态。Where x represents the input, and the output value of the sigmoid function, y, is a value between 0 and 1, where 0 means "do not allow any amount to pass through", and 1 means "allow any amount to pass through". LSTM has three gates, namely input gate, forget gate and output gate, which are used to protect and control the cell state.
其中，遗忘门会读取细胞上一时刻的输出值ht-1和当前时刻的输入值xt，输出一个在0到1之间的数值ft给细胞状态Ct-1。Among them, the forget gate will read the output value ht-1 of the cell at the previous moment and the input value xt of the current moment, and output a value ft between 0 and 1 to the cell state Ct-1 .
ft＝σ(Wf×[Ct-1,ht-1,xt]+bf)ft =σ(Wf ×[Ct-1 ,ht-1 ,xt ]+bf )
其中Wf和bf都是网络的权重参数，σ是Sigmoid激活函数。where Wf and bf are the weight parameters of the network, and σ is the sigmoid activation function.
输入门(input gate)用于更新细胞信息。The input gate is used to update cell information.
it＝σ(Wi·[ht-1,xt]+bi)it =σ(Wi ·[ht-1 ,xt ]+bi )
另一方面，构建一个候选值向量(cell)：
其中，Wi、bi、XC、bC是网络的权重参数，tanh是激活函数。Among them, Wi ,bi , XC , and bC are the weight parameters of the network, and tanh is the activation function.
之后，LSTM更新细胞状态Ct-1为Ct，ft点乘Ct-1代表掉要丢弃遗忘的信息。
最后，LSTM利用输出门决定网络的输出量ht，得到最终的输出值。Finally, LSTM uses the output gate to determine the output ht of the network to obtain the final output value.
ot＝σ(Wo[ht-1,xt]+bo)ot =σ(Wo [ht-1 ,xt ]+bo )
ht＝ot*tanh(Ct)ht =ot *tanh(Ct )
其中Wo、bo是网络权重参数。whereWo andbo are network weight parameters.
为了优化LSTM网络的参数，通常选择均方误差作为损失函数：In order to optimize the parameters of the LSTM network, the mean squared error is usually chosen as the loss function:
或者选取交叉熵损失函数：Or choose the cross-entropy loss function:
其中Θ表示LSTM模型的参数集合，
L(Θ)＝J(Θ)+λ(||Wf||2+||Wi||2+||Wc||2+||Wo||2)L(Θ)=J(Θ)+λ(||Wf ||2 +||Wi ||2 +||Wc ||2 +||Wo ||2 )
其中λ是正则化参数。where λ is the regularization parameter.
4、建立贝叶斯长短时记忆网络：将模型参数ω＝{Wf,Wi,Wc,Wo,bf,by,bc,bo}作为先验分布p(ω)的随机变量。因此，LSTM网络的细胞状态和输出可以重新表示为：4. Establish a Bayesian long-term memory network: take the model parameters ω={Wf ,Wi ,Wc ,Wo ,bf ,by ,bc ,bo } as thevalue of the prior distribution p(ω) Random Variables. Therefore, the cell state and output of the LSTM network can be re-expressed as:
Ct＝fiω(xt,ht-1)Ct =fiω (xt ,ht-1 )
其中下标i和o分别表示隐藏层和输出层节点的索引，fiω和
每个数据点输出的可能性为。The probability of each data point output is .
其中τ是反映数据固有噪声的精度参数，在本发明中，为了便于计算，假定似然函数满足正态分布，似然函数会通过LSTM网络向前传递。Among them, τ is a precision parameter reflecting the inherent noise of the data. In the present invention, in order to facilitate the calculation, it is assumed that the likelihood function satisfies the normal distribution, and the likelihood function will be forwarded through the LSTM network.
然后，给定一个包含X(本发明中具体指逆变器的三相电流、三相电压等电气参数信息以及逆变器机内空气温度、变压器温度等环境信息)和Y(本发明中具体指光伏系统中逆变器散热器的实际温度)的训练数据集(大型地面光伏电站存储的逆变器历史运行数据)，在参数空间上学习需要估计后验分布p(ω∣X,Y)。利用更新后的分布，可以通过积分得到逆变器散热器温度的预测输出y*的分布：Then, a given file contains X (in the present invention, it specifically refers to the electrical parameter information such as the three-phase current and three-phase voltage of the inverter, and environmental information such as the air temperature in the inverter and the temperature of the transformer) and Y (in the present invention, the specific information is Refers to the training data set of the actual temperature of the inverter radiator in the photovoltaic system (the historical operation data of the inverter stored in the large-scale ground photovoltaic power station). Learning in the parameter space requires estimating the posterior distribution p(ω∣X,Y) . Using the updated distribution, the distribution of the predicted output y* of the inverter heatsink temperature can be obtained by integrating:
p(y*∣x*,X,Y)＝∫p(y*∣x*,ω)p(ω∣X,Y)dωp(y* ∣x* ,X,Y)＝∫p(y* ∣x* ,ω)p(ω∣X,Y)dω
其中，x*表示一个新的观测值，此处忽略了对精度参数、隐藏层状态和过去输入的依赖性。对于先验分布，通常选择权重矩阵p(W)上的标准零均值高斯先验，预测的不确定性将直接反映在后验分布p(y*∣x*,X,Y)。where x* represents a new observation, ignoring dependencies on precision parameters, hidden layer states, and past inputs. For the prior distribution, a standard zero-mean Gaussian prior on the weight matrix p(W) is usually chosen, and the prediction uncertainty will be directly reflected in the posterior distribution p(y* ∣x* ,X,Y).
5、引入蒙特卡洛(MC)dropout技术：本发明将变分dropout用于贝叶斯长短时记忆网络的变分推理。变分推论是一种利用简单的参数化分布q(ω)去近似难以处理的后验分布p(ω∣X,Y)的技术。此时，积分项可以通过q(ω)的蒙特卡罗积分进行近似。具体而言，近似分布是通过权重矩阵分解得到的。对于wk的每一行，变分dropout都会施加一个变化分布，也即近似分布q(ω)可以由两个具有小方差的高斯分布整合得到：5. Introducing Monte Carlo (MC) dropout technology: The present invention uses variational dropout for variational reasoning of Bayesian long-short-term memory networks. Variational inference is a technique for approximating an intractable posterior distribution p(ω∣X,Y) using a simple parametric distribution q(ω). In this case, the integral term can be approximated by Monte Carlo integration of q(ω). Specifically, the approximate distribution is obtained by decomposing the weight matrix. For each row of wk , variational dropout imposes a varying distribution, that is, the approximate distribution q(ω) can be obtained by integrating two Gaussian distributions with small variance:
其中p是预定义的dropout概率，σ2是小精度参数，mk是变分参数。为了使变分推论中的近似分布q(ω)最大程度地接近真实的后验分布p(ω∣X,Y)，本发明方法通过最小化近似分布和后验分布之间的KL散度来学习网络的权值，具体来说，即最小化下面的目标函数：where p is the predefined dropout probability,σ2 is the small precision parameter, andmk is the variational parameter. In order to make the approximate distribution q(ω) in variational inference as close as possible to the true posterior distribution p(ω∣X,Y), the method of the present invention solves the problem by minimizing the KL divergence between the approximate distribution and the posterior distribution. Learn the weights of the network, specifically, minimize the following objective function:
KL(q(ω)||p(ω∣X,Y)))KL(q(ω)||p(ω∣X,Y)))
值得注意的是，变分长短时记忆网络在每个时间步长使用固定的dropout掩码，包括循环层。在每个时间步长随意地丢弃输入、输出和循环连接。这与现有技术形成对比，在现有技术中，不同神经网络单元会在不同的时间步长被丢弃，而且不会对全连接层进行丢弃。Notably, Variational Length Short-Term Memory Networks use a fixed dropout mask at each time step, including recurrent layers. Input, output, and recurrent connections are arbitrarily dropped at each time step. This is in contrast to the prior art, where different neural network units are dropped at different time steps, and fully connected layers are not dropped.
测试过程中所使用变分dropout方法可以看作是后验预测分布p(ω∣X,Y)蒙特卡洛样本的近似值。给定一个新的观测值x*，通过N个随机模型样本前向传递，可以收集N个近似预测后验的样本
其中，τ可以估计为
6、多步预测：本发明考虑到实际光伏现场可能需要对逆变器建立超短时、短时、长时等各种时间尺度下的预警机制，所设计贝叶斯长短时记忆网络的输出维度是可调节的。6. Multi-step prediction: The present invention takes into account that the actual photovoltaic field may need to establish an early warning mechanism under various time scales such as ultra-short-term, short-term, and long-term for the inverter, and the output of the designed Bayesian long-term memory network is designed. Dimensions are adjustable.
本发明将逆变器实际数据的采样时间间隔视为一个步长，例如逆变器数据每5分钟采集一次，那么单步长预测就表示预测未来一次采样(5分钟后)的逆变器散热器温度，六步长预测就表示预测未来六次采样(5分钟后、10分钟后、15分钟后、20分钟后、25分钟后和30分钟后)的逆变器散热器温度。In the present invention, the sampling time interval of the actual data of the inverter is regarded as a step size. For example, the inverter data is collected every 5 minutes, and the single-step prediction means predicting the heat dissipation of the inverter for a sampling in the future (after 5 minutes). Inverter temperature, the six-step prediction means predicting the inverter radiator temperature for the next six samples (after 5 minutes, after 10 minutes, after 15 minutes, after 20 minutes, after 25 minutes, and after 30 minutes).
具体来说，本发明是通过调整步骤四中贝叶斯长短时记忆网络的输出神经元个数来实现多步预测的，其中每个神经元对应于一个预测步长。例如，单步长预测时设置网络的输出神经元个数为1，六步长时设置输出神经元个数为6。在本发明中，调整贝叶斯长短时记忆网络的输出神经元个数仅需对极少部分模型参数进行修改，之后利用历史数据对模型进行重新训练即可部署到实际应用现场，便可以高效地实现任意时间尺度下的逆变器过温预警Specifically, the present invention realizes multi-step prediction by adjusting the number of output neurons of the Bayesian long-term memory network in step 4, wherein each neuron corresponds to a prediction step. For example, the number of output neurons of the network is set to 1 for single-step prediction, and the number of output neurons is set to 6 for six-step prediction. In the present invention, adjusting the number of output neurons of the Bayesian long-short-term memory network only needs to modify a very small number of model parameters, and then use historical data to retrain the model and deploy it to the actual application site, which can efficiently Realize inverter over-temperature warning at any time scale
7、量化近似后验分布，优化网络权重：本发明引入了两种方法来量化实际逆变器散热器温度预测值与实际值的偏差大小。7. Quantify approximate posterior distribution and optimize network weights: The present invention introduces two methods to quantify the deviation between the predicted value of the actual inverter radiator temperature and the actual value.
(1)适用于高斯预测分布的平方马氏距离方法：如果预测分布是高斯分布，或者近似高斯分布，则可以使用平方马氏距离来表征预测值与实际值的偏差大小。首先，使用预测分布在t时刻的蒙特卡洛样本
当观测真值Xt可用时，平方马氏距离由下式确定：When the observed ground truth Xt is available, the squared Mahalanobis distance is determined by:
马氏距离大表明实际观测值与预测后验分布偏差大，表明逆变器散热器温度预测模型还不够准确，模型中的参数需要进一步调整。A large Mahalanobis distance indicates that the actual observed value has a large deviation from the predicted posterior distribution, indicating that the inverter radiator temperature prediction model is not accurate enough, and the parameters in the model need to be further adjusted.
(2)适用于非高斯分布的局部密度比方法：如果预测分布不能很好地描述为高斯分布，那么就利用非参数化的局部密度比(LDR)方法来量化每个观测值的异常状况。LDR统计量利用其周围最近的k个邻居(k-NNs)观测值的密度的估计值，量化每个新观测值相对于其预测分布的偏离情况。(2) Local density ratio method for non-Gaussian distributions: If the predicted distribution cannot be well described as a Gaussian distribution, then a nonparametric local density ratio (LDR) method is used to quantify the abnormality of each observation. The LDR statistic quantifies how much each new observation deviates from its predicted distribution using an estimate of the density of observations from its k nearest neighbors (k-NNs).
基于k-NNs的局部密度估计值
其中，
然后，观测值xt的局部密度比定义为：Then, the local density ratio of observations xt is defined as:
也即，xt周围最近的k个预测值的局部密度求平均值后除以xt的局部密度。当LDR较大时，表明后验分布与预测值的偏差大，逆变器散热器温度预测模型还不够准确，模型中的参数需要进一步调整。同时，需要选取合适的k值，较小的k会导致模型的预测值波动较大，较大的k会使得模型的预测精度降低降低。That is, the local density of the k nearest predicted values around xt is averaged and divided by the local density of xt . When the LDR is large, it indicates that the deviation between the posterior distribution and the predicted value is large, the temperature prediction model of the inverter radiator is not accurate enough, and the parameters in the model need to be further adjusted. At the same time, it is necessary to select an appropriate value of k. A smaller k will cause the prediction value of the model to fluctuate greatly, and a larger k will reduce the prediction accuracy of the model.
9、逆变器过温预警：考虑到不同光伏系统中逆变器的运行工况不同，且不同逆变器的工作温度上限不同，本发明没有基于绝对温度(比如温度超过某个具体数值就发出过温预警)设计逆变器过温预警机制，而是按温度比例设计了多级过温预警机制。9. Inverter over-temperature warning: Considering that the operating conditions of inverters in different photovoltaic systems are different, and the upper limit of operating temperature of different inverters is different, the present invention is not based on absolute temperature (for example, if the temperature exceeds a certain value, the Issue over-temperature warning) to design the inverter over-temperature warning mechanism, but design a multi-level over-temperature warning mechanism according to the temperature ratio.
本发明按预测温度占逆变器温度上限的比例，将逆变器过温预警分为轻度预警(70％～80％)、中度预警(80％～90％)和重度预警(90％及以上)三个等级。同时，针对不同的预警等级，本发明设计了相应的响应机制，有效避免误判和漏判。According to the proportion of the predicted temperature to the upper limit of the inverter temperature, the invention divides the inverter over-temperature warning into mild warning (70%-80%), moderate warning (80%-90%) and severe warning (90%). and above) three levels. At the same time, for different warning levels, the present invention designs a corresponding response mechanism to effectively avoid misjudgment and missed judgment.
具体来说，可以分为以下几种情况处理：Specifically, it can be divided into the following situations:
(1)出现重度预警的情况，任何时候，当某次温度预测结果达到重度预警时，系统会立即发出过温预警，提示电站运维人员对逆变器采取相应的检修维护措施。(1) In the event of a severe warning, at any time, when a certain temperature prediction result reaches a severe warning, the system will immediately issue an over-temperature warning, prompting the power station operation and maintenance personnel to take corresponding maintenance measures for the inverter.
(2)出现中度预警的情况，当某次温度预测结果为中度预警时，系统并不会马上发出过温警告，而是会继续追踪未来连续2次(也即未来5分钟和10分钟)的预测结果，如果预测结果均为中度预警且呈温度上升趋势(10分钟时温度预测值高于5分钟时温度预测值)，系统才会发出过温警告。特殊的，当未来两次预测结果中出现重度预警时，会立即告警。(2) In the case of a moderate warning, when a certain temperature prediction result is a moderate warning, the system will not issue an over-temperature warning immediately, but will continue to track 2 consecutive times in the future (that is, 5 minutes and 10 minutes in the future). ), if the forecast results are all moderate warnings and the temperature is rising (the temperature forecast value at 10 minutes is higher than the temperature forecast value at 5 minutes), the system will issue an over-temperature warning. In particular, when a severe warning appears in the next two forecast results, an immediate warning will be issued.
(3)出现轻度预警的情况，当某次温度预测结果为轻度预警时，系统并不会马上发出过温警告，而是会继续追踪未来连续5次(也即未来5分钟、10分钟、15分钟、20分钟和25分钟)的预测结果，如果预测结果均为轻度预警及以上且温度呈上升趋势，系统才会发出过温警告。特殊的，当未来5次预测结果中出现重度预警时，会立即告警。(3) In the case of a mild warning, when a certain temperature prediction result is a mild warning, the system will not issue an over-temperature warning immediately, but will continue to track 5 consecutive times in the future (that is, 5 minutes and 10 minutes in the future). , 15 minutes, 20 minutes and 25 minutes), if the forecast results are all mild warnings and above and the temperature is on the rise, the system will issue an over-temperature warning. In particular, when a severe warning occurs in the next five forecast results, an immediate warning will be issued.
(4)出现轻度预警与中度预警跳变，当预测温度从轻度预警上升为中度预警时，温度呈上升趋势，应按照中度预警规则进行后续判断。当预测温度从中度预警降低为轻度预警，温度呈下降趋势，应按照轻度预警规则进行后续判断。特殊的，当预测结果中出现重度预警时，会立即告警。(4) There is a jump between mild early warning and moderate early warning. When the predicted temperature rises from mild early warning to moderate early warning, the temperature is on the rise, and follow-up judgments should be made according to the moderate early warning rules. When the predicted temperature decreases from a moderate early warning to a mild early warning, and the temperature shows a downward trend, follow-up judgments should be made according to the mild early warning rules. Specially, when there is a severe warning in the forecast result, it will immediately give an alarm.
以上详细描述了本发明的较佳具体实施例。应当理解，本领域的普通技术无需创造性劳动就可以根据本发明的构思作出诸多修改和变化。因此，凡本技术领域中技术人员依本发明的构思在现有技术的基础上通过逻辑分析、推理或者有限的实验可以得到的技术方案，皆应在由权利要求书所确定的保护范围内。The preferred embodiments of the present invention have been described in detail above. It should be understood that many modifications and changes can be made according to the concept of the present invention by those skilled in the art without creative efforts. Therefore, all technical solutions that can be obtained by those skilled in the art through logical analysis, reasoning or limited experiments on the basis of the prior art according to the concept of the present invention shall fall within the protection scope determined by the claims.
Claims (10)
Translated fromChinese
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011442899.4ACN112910288B (en) | 2020-12-08 | 2020-12-08 | Over-temperature early warning method based on inverter radiator temperature prediction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011442899.4ACN112910288B (en) | 2020-12-08 | 2020-12-08 | Over-temperature early warning method based on inverter radiator temperature prediction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112910288Atrue CN112910288A (en) | 2021-06-04 |
| CN112910288B CN112910288B (en) | 2022-08-09 |
Family
ID=76111434
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011442899.4AActiveCN112910288B (en) | 2020-12-08 | 2020-12-08 | Over-temperature early warning method based on inverter radiator temperature prediction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112910288B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114126369A (en)* | 2021-11-26 | 2022-03-01 | 上海明华电力科技有限公司 | Heat dissipation control method of photovoltaic inverter |
| CN114710114A (en)* | 2022-05-23 | 2022-07-05 | 北京华清未来能源技术研究院有限公司 | Photovoltaic inverter fault prediction method |
| CN115700324A (en)* | 2022-10-17 | 2023-02-07 | 出门问问信息科技有限公司 | Tile temperature early warning method and device and electronic equipment |
| CN115758162A (en)* | 2022-12-07 | 2023-03-07 | 广东电网有限责任公司 | Data prediction model training, photovoltaic inverter fault prediction method and device |
| CN115936060A (en)* | 2022-12-28 | 2023-04-07 | 四川物通科技有限公司 | Transformer substation capacitance temperature early warning method based on depth certainty strategy gradient |
| WO2024169681A1 (en)* | 2023-02-15 | 2024-08-22 | 山东科技大学 | Thermal power plant temperature variable alarm prediction method and system based on amplitude change trend |
| US12073668B1 (en) | 2023-06-08 | 2024-08-27 | Mercedes-Benz Group AG | Machine-learned models for electric vehicle component health monitoring |
| CN119247092A (en)* | 2024-12-04 | 2025-01-03 | 青岛中微创芯电子有限公司 | A junction temperature detection method based on RC-IGBT |
| CN119668126A (en)* | 2025-02-24 | 2025-03-21 | 西北工业大学 | Electrical automatic control method and system |
| CN119891731A (en)* | 2024-12-17 | 2025-04-25 | 杭州电子科技大学 | CLLC resonant converter soft start method based on edge neural network prediction |
Citations (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102880170A (en)* | 2012-10-08 | 2013-01-16 | 南京航空航天大学 | System failure early warning method based on baseline model and Bayesian factor |
| CN107729716A (en)* | 2017-11-27 | 2018-02-23 | 西安建筑科技大学 | A kind of mine water inrush Forecasting Methodology based on long Memory Neural Networks in short-term |
| CN107766968A (en)* | 2017-09-26 | 2018-03-06 | 河海大学 | Short-term wind speed forecasting method based on CAPSO RVM built-up patterns |
| EP3407267A1 (en)* | 2017-05-25 | 2018-11-28 | Hitachi, Ltd. | Deep learning network architecture optimization for uncertainty estimation in regression |
| CN109583346A (en)* | 2018-11-21 | 2019-04-05 | 齐鲁工业大学 | EEG feature extraction and classifying identification method based on LSTM-FC |
| CN109948861A (en)* | 2019-03-26 | 2019-06-28 | 西南交通大学 | A short-term passenger flow prediction method for urban rail transit based on modal decomposition and deep learning |
| CN110059844A (en)* | 2019-02-01 | 2019-07-26 | 东华大学 | Energy storage device control method based on set empirical mode decomposition and LSTM |
| US20190340235A1 (en)* | 2018-05-01 | 2019-11-07 | Capital One Services, Llc | Text categorization using natural language processing |
| CN110751199A (en)* | 2019-10-15 | 2020-02-04 | 南京航空航天大学 | A Satellite Anomaly Detection Method Based on Bayesian Neural Network |
| CN110881037A (en)* | 2019-11-19 | 2020-03-13 | 北京工业大学 | Network intrusion detection method and training method and device of model thereof, and server |
| CN111626506A (en)* | 2020-05-27 | 2020-09-04 | 华北电力大学 | Regional photovoltaic power probability prediction method based on federal learning and cooperative regulation and control system thereof |
| WO2020181240A1 (en)* | 2019-03-06 | 2020-09-10 | Gritstone Oncology, Inc. | Identification of neoantigens with mhc class ii model |
| CN111695724A (en)* | 2020-06-01 | 2020-09-22 | 浙江大学 | Wind speed prediction method based on hybrid neural network model |
| WO2020191004A1 (en)* | 2019-03-18 | 2020-09-24 | Georgia Tech Research Corporation | Tracking and alert method and system for worker productivity and safety |
| CN111815065A (en)* | 2020-07-21 | 2020-10-23 | 东北大学 | Short-term power load forecasting method based on long-short-term memory neural network |
| CN111950759A (en)* | 2020-05-18 | 2020-11-17 | 武汉科技大学 | A Short-Term Wind Speed Prediction Method Based on Two-Level Decomposition, LSTM and AT |
| CN112765873A (en)* | 2020-11-30 | 2021-05-07 | 中国南方电网有限责任公司超高压输电公司广州局 | LSTM algorithm-based power transformation equipment heating prediction method |
- 2020
- 2020-12-08CNCN202011442899.4Apatent/CN112910288B/enactiveActive
Patent Citations (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102880170A (en)* | 2012-10-08 | 2013-01-16 | 南京航空航天大学 | System failure early warning method based on baseline model and Bayesian factor |
| EP3407267A1 (en)* | 2017-05-25 | 2018-11-28 | Hitachi, Ltd. | Deep learning network architecture optimization for uncertainty estimation in regression |
| CN107766968A (en)* | 2017-09-26 | 2018-03-06 | 河海大学 | Short-term wind speed forecasting method based on CAPSO RVM built-up patterns |
| CN107729716A (en)* | 2017-11-27 | 2018-02-23 | 西安建筑科技大学 | A kind of mine water inrush Forecasting Methodology based on long Memory Neural Networks in short-term |
| US20190340235A1 (en)* | 2018-05-01 | 2019-11-07 | Capital One Services, Llc | Text categorization using natural language processing |
| CN109583346A (en)* | 2018-11-21 | 2019-04-05 | 齐鲁工业大学 | EEG feature extraction and classifying identification method based on LSTM-FC |
| CN110059844A (en)* | 2019-02-01 | 2019-07-26 | 东华大学 | Energy storage device control method based on set empirical mode decomposition and LSTM |
| WO2020181240A1 (en)* | 2019-03-06 | 2020-09-10 | Gritstone Oncology, Inc. | Identification of neoantigens with mhc class ii model |
| WO2020191004A1 (en)* | 2019-03-18 | 2020-09-24 | Georgia Tech Research Corporation | Tracking and alert method and system for worker productivity and safety |
| CN109948861A (en)* | 2019-03-26 | 2019-06-28 | 西南交通大学 | A short-term passenger flow prediction method for urban rail transit based on modal decomposition and deep learning |
| CN110751199A (en)* | 2019-10-15 | 2020-02-04 | 南京航空航天大学 | A Satellite Anomaly Detection Method Based on Bayesian Neural Network |
| CN110881037A (en)* | 2019-11-19 | 2020-03-13 | 北京工业大学 | Network intrusion detection method and training method and device of model thereof, and server |
| CN111950759A (en)* | 2020-05-18 | 2020-11-17 | 武汉科技大学 | A Short-Term Wind Speed Prediction Method Based on Two-Level Decomposition, LSTM and AT |
| CN111626506A (en)* | 2020-05-27 | 2020-09-04 | 华北电力大学 | Regional photovoltaic power probability prediction method based on federal learning and cooperative regulation and control system thereof |
| CN111695724A (en)* | 2020-06-01 | 2020-09-22 | 浙江大学 | Wind speed prediction method based on hybrid neural network model |
| CN111815065A (en)* | 2020-07-21 | 2020-10-23 | 东北大学 | Short-term power load forecasting method based on long-short-term memory neural network |
| CN112765873A (en)* | 2020-11-30 | 2021-05-07 | 中国南方电网有限责任公司超高压输电公司广州局 | LSTM algorithm-based power transformation equipment heating prediction method |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114126369A (en)* | 2021-11-26 | 2022-03-01 | 上海明华电力科技有限公司 | Heat dissipation control method of photovoltaic inverter |
| CN114710114A (en)* | 2022-05-23 | 2022-07-05 | 北京华清未来能源技术研究院有限公司 | Photovoltaic inverter fault prediction method |
| CN115700324A (en)* | 2022-10-17 | 2023-02-07 | 出门问问信息科技有限公司 | Tile temperature early warning method and device and electronic equipment |
| CN115758162A (en)* | 2022-12-07 | 2023-03-07 | 广东电网有限责任公司 | Data prediction model training, photovoltaic inverter fault prediction method and device |
| CN115936060A (en)* | 2022-12-28 | 2023-04-07 | 四川物通科技有限公司 | Transformer substation capacitance temperature early warning method based on depth certainty strategy gradient |
| CN115936060B (en)* | 2022-12-28 | 2024-03-26 | 四川物通科技有限公司 | Substation capacitance temperature early warning method based on depth deterministic strategy gradient |
| WO2024169681A1 (en)* | 2023-02-15 | 2024-08-22 | 山东科技大学 | Thermal power plant temperature variable alarm prediction method and system based on amplitude change trend |
| US12073668B1 (en) | 2023-06-08 | 2024-08-27 | Mercedes-Benz Group AG | Machine-learned models for electric vehicle component health monitoring |
| CN119247092A (en)* | 2024-12-04 | 2025-01-03 | 青岛中微创芯电子有限公司 | A junction temperature detection method based on RC-IGBT |
| CN119891731A (en)* | 2024-12-17 | 2025-04-25 | 杭州电子科技大学 | CLLC resonant converter soft start method based on edge neural network prediction |
| CN119891731B (en)* | 2024-12-17 | 2025-09-23 | 杭州电子科技大学 | CLLC resonant converter soft start method based on edge neural network prediction |
| CN119668126A (en)* | 2025-02-24 | 2025-03-21 | 西北工业大学 | Electrical automatic control method and system |
| CN119668126B (en)* | 2025-02-24 | 2025-05-02 | 西北工业大学 | Electrical automatic control method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112910288B (en) | 2022-08-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112910288B (en) | Over-temperature early warning method based on inverter radiator temperature prediction | |
| Mei et al. | Day-ahead nonparametric probabilistic forecasting of photovoltaic power generation based on the LSTM-QRA ensemble model | |
| Yao et al. | Multivariate time-series prediction in industrial processes via a deep hybrid network under data uncertainty | |
| CN116167527B (en) | Pure data-driven power system static safety operation risk online assessment method | |
| Wang et al. | A novel wind power prediction model improved with feature enhancement and autoregressive error compensation | |
| CN115130743B (en) | A wind turbine regulation rate prediction method and system based on variational inference | |
| CN119674932B (en) | New energy storage and digestion capacity index prediction method | |
| Liu et al. | A Bayesian deep learning-based probabilistic risk assessment and early-warning model for power systems considering meteorological conditions | |
| Kaplan et al. | Fault diagnosis of smart grids based on deep learning approach | |
| CN116826740A (en) | A method for predicting power quality in photovoltaic grid-connected low-voltage Taiwan areas | |
| Gao et al. | Mechanism enhanced data-driven method for reliability improvement of load margin estimation | |
| Zheng et al. | Technical indicator enhanced ultra‐short‐term wind power forecasting based on long short‐term memory network combined XGBoost algorithm | |
| Liang et al. | WPFormer: A spatial-temporal graph transformer with auto-correlation for wind power forecasting | |
| Liu et al. | Deep learning-based intelligent fault diagnosis for power distribution networks | |
| Zhang | Short‐Term Power Load Forecasting Based on SAPSO‐CNN‐LSTM Model considering Autocorrelated Errors | |
| Niu et al. | Evaluation of wind-solar hybrid power generation system based on Monte Carlo method | |
| Liu et al. | A Bayesian deep learning-based adaptive wind farm power prediction method within the entire life cycle | |
| Yang et al. | A short-term PV resource assessment method with parallel DenseNet201 and BiLSTM under multiple data features | |
| Zhu et al. | Power data preprocessing method of mountain wind farm based on POT-DBSCAN | |
| CN118399378A (en) | Method, apparatus, device, storage medium and program product for predicting generated power | |
| CN117937596A (en) | Photovoltaic operation and maintenance management system based on data cleaning and power prediction | |
| Shen et al. | An interval analysis scheme based on empirical error and MCMC to quantify uncertainty of wind speed | |
| Chen et al. | Gru based time series forecast of oil temperature in power transformer | |
| Yang | [Retracted] Short‐Term Load Monitoring of a Power System Based on Neural Network | |
| Gijón et al. | Integrating Physics and Data-Driven Approaches: An Explainable and Uncertainty-Aware Hybrid Model for Wind Turbine Power Prediction |
Legal Events
| 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |