Method and device for detecting voltage consistency fault of energy storage battery

Abstract

The invention discloses an energy storage battery voltage consistency fault detection method and device, and belongs to the technical field of energy storage equipment, and the detection method comprises the steps: obtaining the voltage online data of a target energy storage battery, and carrying out the preprocessing; inputting the preprocessed voltage online data into a pre-trained voltage prediction network model, and predicting voltage time sequence data at a future moment; inputting the voltage time sequence data at the future moment into a pre-trained anomaly detection algorithm model to obtain a voltage anomaly value; and if the voltage abnormal value is greater than the set threshold value, determining that a voltage consistency fault exists. According to the invention, characteristic quantity extraction of data is carried out through a voltage prediction network model, and a later operation state is predicted; then, a prediction result is sent into an anomaly detection algorithm model, and an abnormal value of the prediction time sequence is calculated; and finally, the abnormal value is compared with a set threshold value, so that the voltage consistency fault is determined, the efficiency, adaptability and accuracy of fault detection are high, and early warning can be performed on the fault in time.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage devices, and in particular, to a method and device for detecting voltage consistency faults of energy storage batteries. Background Art

[0002] With the development of energy storage batteries, due to their long cycle life, low cost and other characteristics, energy storage batteries are widely used in the energy storage market. In actual use, in order to enable the energy storage cabin to store more electricity, multiple monomers are usually combined into a PACK battery pack according to different series-parallel connection methods. However, when the voltage of the battery monomers is inconsistent and continues to be used, faults such as under-voltage will occur, thus triggering a series of battery faults. Therefore, in order to ensure the safety of the battery cabin, it is necessary to give long-term early warning of voltage consistency.

[0003] In view of the above technical problems, the prior art, such as a method and system for diagnosing multiple faults of a battery string based on corrected sample entropy disclosed in Patent Publication No. CN110703109B, the fault diagnosis method includes the following steps: obtaining the battery voltage of the battery string to be diagnosed measured; constructing a battery voltage sequence according to the obtained battery voltage of the battery string to be diagnosed, and calculating the sample entropy value of the battery voltage sequence; setting a correction coefficient for characterizing voltage fluctuation information, and correcting the sample entropy value through the correction coefficient to obtain a corrected sample entropy value; judging and outputting the fault type of the battery string to be diagnosed according to the numerical change of the corrected sample entropy value. It can accurately diagnose the faults of the battery without a model. By setting the correction coefficient, the sample entropy values under different faults can be distinguished, improving the intuitiveness and efficiency of fault judgment, and being able to quickly, accurately and stably diagnose and predict the fault type and time of lithium-ion batteries. This method has a good diagnostic effect on faults causing voltage mutation. However, since the sample entropy is only sensitive to the change of the sequence with increased complexity, this method is ineffective for some special battery faults. For example, for the vast majority of early minor faults of the battery, it is difficult for the above method to detect, and the adaptability to battery fault detection is relatively general. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method and device for detecting voltage consistency faults of energy storage batteries, so as to solve the technical problem that the adaptability of the prior art to battery fault detection is relatively general.

[0005] To achieve the above purpose, the present invention is implemented by adopting the following technical solution:

[0006] In the first aspect, the present invention provides a method for detecting voltage consistency faults of energy storage batteries, including:

[0007] Obtaining the online voltage data of the target energy storage battery and performing preprocessing;

[0008] Input the preprocessed online voltage data into a pre-trained voltage prediction network model to predict the voltage time-series data at future moments;

[0009] Input the voltage time-series data at the future moments into a pre-trained anomaly detection algorithm model to obtain voltage anomaly values;

[0010] If there exists a voltage anomaly value greater than the set threshold, there is a voltage consistency fault.

[0011] The detection method provided by the present invention first obtains the online voltage data of the target energy storage battery from the data platform and preprocesses it; uses a voltage prediction network model to extract the feature quantities of the data and predict the future operating state online; then sends the prediction results into an anomaly detection algorithm model to calculate the anomaly values of the predicted time series; finally, compares the anomaly values with the set threshold to determine the voltage consistency fault and give an early warning of the fault.

[0012] Optionally, the preprocessing includes normalization processing and noise reduction processing.

[0013] Normalization processing and noise reduction processing are two core steps in data preprocessing, aiming at the problems of unifying the data dimension and eliminating noise respectively. They can provide cleaner inputs for subsequent model algorithms, thus ensuring the efficiency and accuracy of model algorithm operations.

[0014] Optionally, the voltage prediction network model is a CNN-LSTM deep learning model combining a CNN convolutional neural network and an LSTM long short-term memory network;

[0015] The CNN convolutional neural network is used to extract features from the input samples. During the one-dimensional convolution process, the convolution kernel slides over each input sample and performs convolution. The output dimension of the convolution layer is determined by the number of convolution kernels, and each input sample is transformed into a feature map :

[0016]

[0017] In the formula, is the weight matrix and bias term of the convolution layer, is the non-linear activation function, is the time window data from the th moment to the th moment, is the length of the time window, is the feature map of the input sample at the th moment;

[0018] The LSTM long short-term memory network is used to capture the feature map The long-term time dependence relationship to obtain the hidden state :

[0019]

[0020]

[0021]

[0022] Wherein, is the forget gate, is the input gate, is the output gate, is the state of the memory cell, are the weight matrix and the bias term;

[0023] A fully connected layer is arranged after the LSTM long short-term memory network, and the fully connected layer generates a prediction result according to the hidden state

[0024] The CNN-LSTM deep learning model combines the advantages of the convolutional neural network (CNN) and the long short-term memory network (LSTM). The local features extracted by the CNN can be used as the input of the LSTM, and the LSTM further models the time dependence relationship of these features. By combining local and global information, the model can capture the complex patterns of the data more accurately and improve the prediction performance. At the same time, it can be accelerated by the GPU to support the efficient processing of large-scale data.

[0025] Optionally, the anomaly detection algorithm model is the LOF algorithm model; the processing process of the LOF algorithm model includes:

[0026] For each data point in the input sample, calculate its k-nearest neighbor distance :

[0027]

[0028] Wherein, is the k-nearest neighbor distance of the data point is the data point the th neighboring data point of is the data point the distance between;

[0029] Calculate the local reachability distance of each data point based on the k-nearest neighbor distance :

[0030]

[0031] Wherein, is the data point​​ The k-nearest neighbor distance, is the local reachability distance of the data point ; , is the set of the first nearest neighboring data points of the data point ;

[0032] Calculate the local reachability density of each data point based on the local reachability distance :

[0033]

[0034] In the formula, is the local reachability density of the data point ;

[0035] Calculate the average value of the local reachability density of each data point based on the local reachability density :

[0036]

[0037] In the formula, is the average value of the local reachability density of the data point , is the local reachability density of the data point ;

[0038] Calculate the local outlier factor of each data point based on the local reachability density and the average value of the local reachability density :

[0039]

[0040] In the formula, is the local outlier factor of the data point .

[0041] The LOF algorithm model determines the degree of abnormality by comparing the local density of the data point with the density difference in its neighborhood. It is an unsupervised learning method that does not require pre-labeling of abnormal or normal points in the data and is suitable for scenarios lacking labels. It does not need to assume the data distribution, can automatically adapt to different types of data structures, and can efficiently and accurately detect abnormal points in the data.

[0042] Optionally, the setting of the set threshold of the voltage anomaly value includes:

[0043] Obtain the voltage historical data of the target energy storage battery;

[0044] Input the voltage historical data into the pre-trained anomaly detection algorithm model to obtain the voltage anomaly value and determine the battery module where the corresponding battery cell is located;

[0045] Calculate the average voltage of the battery module and its adjacent battery modules as the set threshold.

[0046] By using the set threshold to further judge the voltage outliers, the accuracy of fault detection is improved. Compared with directly judging by voltage outliers, misjudgment is reduced and work efficiency is improved.

[0047] In a second aspect, the present invention provides a detection device for the voltage consistency fault of an energy storage battery, including:

[0048] A data acquisition module, configured to acquire the online voltage data of the target energy storage battery and perform preprocessing;

[0049] An online prediction module, configured to input the preprocessed online voltage data into a pre-trained voltage prediction network model to predict the voltage time series data at a future moment;

[0050] An anomaly detection module, configured to input the voltage time series data at the future moment into a pre-trained anomaly detection algorithm model to obtain voltage outliers;

[0051] A fault judgment module, configured to determine that there is a voltage consistency fault if there is a voltage outlier greater than the set threshold.

[0052] Optionally, the voltage prediction network model is a CNN-LSTM deep learning model combining a CNN convolutional neural network and an LSTM long short-term memory network; and / or, the anomaly detection algorithm model is a LOF algorithm model.

[0053] In a third aspect, the present invention provides an electronic device, including a processor and a storage medium;

[0054] The storage medium is used to store instructions;

[0055] The processor is used to operate according to the instructions to execute the steps of the above method.

[0056] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the above method are implemented.

[0057] In a fifth aspect, the present invention provides a computer program product, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the above method are implemented.

[0058] Compared with the prior art, the beneficial effects achieved by the present invention:

[0059] A detection method and device for the voltage consistency fault of an energy storage battery provided by the present invention first obtains voltage historical data from a data platform and pre-trains a voltage prediction network model and an anomaly detection algorithm model in an offline state. In the online stage, voltage online data is obtained in real time, and a voltage prediction network model is used to extract the characteristic quantities of the data and predict the future operating state. Subsequently, the prediction result is sent into the anomaly detection algorithm model to calculate the anomaly value of the predicted time series. Finally, the anomaly value is compared with a set threshold to determine the voltage consistency fault, with high efficiency, adaptability, and accuracy in fault detection, and can give an early warning of the fault immediately. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a schematic flowchart of the detection method for the voltage consistency fault of an energy storage battery provided by an embodiment of the present invention;

[0061] Figure 2 is an execution process diagram of the detection method for the voltage consistency fault of an energy storage battery provided by an embodiment of the present invention;

[0062] Figure 3 is an execution process diagram of the CNN-LSTM deep learning model and the LOF algorithm model provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0063] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and should not be used to limit the protection scope of the present invention.

[0064] Embodiment 1:

[0065] As Figure 1 shown, an embodiment of the present invention provides a detection method for the voltage consistency fault of an energy storage battery, including the following steps:

[0066] Step S1: Obtain the voltage online data of the target energy storage battery and perform preprocessing.

[0067] Specifically, in this embodiment, the preprocessing includes normalization processing and noise reduction processing.

[0068] Normalization processing and noise reduction processing are two core steps in data preprocessing, respectively aiming at the problem of unifying the dimension of data and eliminating noise. It can provide a cleaner input for subsequent model algorithms, thereby ensuring the efficiency and accuracy of model algorithm operations. In other alternative embodiments, those skilled in the art can select other preprocessing methods according to needs.

[0069] Step S2: Input the preprocessed voltage online data into the pre-trained voltage prediction network model to predict the voltage time series data at a future moment.

[0070] Specifically, in this embodiment, the voltage prediction network model is a CNN-LSTM deep learning model that combines a CNN (Convolutional Neural Network) and an LSTM (Long Short-Term Memory Network). The CNN-LSTM deep learning model combines the advantages of the Convolutional Neural Network (CNN) and the Long Short-Term Memory Network (LSTM). The local features extracted by the CNN can be used as the input of the LSTM, and the LSTM further models the temporal dependence of these features. By combining local and global information, the model can capture the complex patterns of the data more accurately and improve the prediction performance. At the same time, it can be accelerated by GPU to support the efficient processing of large-scale data.

[0071] The CNN (Convolutional Neural Network) includes a convolutional layer for feature extraction from input samples. During one-dimensional convolution, the convolutional kernel slides over each input sample and performs convolution. The output dimension of the convolutional layer is determined by the number of convolutional kernels, and each input sample is transformed into a feature map. :

[0072]

[0073] In the formula, is the weight matrix and bias term of the convolutional layer, is the non-linear activation function, is from the th moment to the th moment of the time window data, is the length of the time window, is the th moment of the feature map of the input sample.

[0074] After the convolutional layer, a pooling layer and a flattening layer are usually set up. The feature dimension is compressed through max-pooling or average-pooling to reduce the computational complexity and enhance the anti-noise ability. The flattening layer converts the multi-dimensional feature map output by the convolutional layer into a one-dimensional sequence to adapt to the input format of the LSTM.

[0075] The LSTM (Long Short-Term Memory Network) is used to capture the long-term temporal dependence relationship of the feature map to obtain the hidden state :

[0076]

[0077]

[0078]

[0079] In the formula, is the forget gate, is the input gate, is the output gate, is the state of the memory cell, are the weight matrix and the bias term.

[0080] In the LSTM long short-term memory network, an attention mechanism can usually be set to dynamically allocate weights at different time steps and enhance the attention to key features.

[0081] A fully connected layer is set after the LSTM long short-term memory network. The fully connected layer generates a prediction result based on the hidden state

[0082] Step S3: Input the voltage time series data at future moments into the pre-trained anomaly detection algorithm model to obtain voltage anomaly values.

[0083] Specifically, in this embodiment, the anomaly detection algorithm model is the LOF algorithm model. The LOF algorithm model judges the degree of anomaly by comparing the local density of data points with the density difference in their neighborhoods. It is an unsupervised learning method that does not require pre-labeling of anomaly points or normal points in the data and is suitable for scenarios lacking labels. Without assuming the data distribution, it can automatically adapt to different types of data structures and can efficiently and accurately detect anomaly points in the data.

[0084] The processing process of the LOF algorithm model includes:

[0085] For each data point in the input sample, calculate its k-nearest neighbor distance :

[0086]

[0087] In the formula, is the k-nearest neighbor distance of the data point , is the th nearest neighbor data point of the data point is the distance between the data points ;

[0088] Calculate the local reachability distance of each data point based on the k-nearest neighbor distance :

[0089]

[0090] In the formula, is the k-nearest neighbor distance of the data point , is the local reachability distance of the data point , , is the th nearest neighbor data point set of the data point​

[0091] Calculate the local reachability density of each data point based on the local reachability distance :

[0092]

[0093] In the formula, is the local reachability density of the data point ;

[0094] Calculate the average value of the local reachability density of each data point based on the local reachability density :

[0095]

[0096] In the formula, is the average value of the local reachability density of the data point , is the data point ;

[0097] Calculate the local outlier factor of each data point based on the local reachability density and the average value of the local reachability density :

[0098]

[0099] In the formula, is the local outlier factor of the data point ;

[0100] Step S4, if there is a voltage outlier greater than the set threshold, there is a voltage consistency fault

[0101] In a specific implementation process, all voltage values can be labeled through an anomaly detection algorithm model. The voltage outliers are labeled as +1, and the voltage normal values are labeled as -1. When judging, select the voltage values labeled as +1 and compare them with the set threshold one by one to determine the voltage consistency fault

[0102] The setting of the set threshold for voltage outliers includes:

[0103] Obtain the voltage historical data of the target energy storage battery

[0104] Input the voltage historical data into the pre-trained anomaly detection algorithm model to obtain the voltage outliers and determine the battery module where the corresponding battery cell is located

[0105] Calculate the voltage mean values of the battery module and its adjacent battery modules as the set threshold

[0106] By setting a threshold value to further judge the voltage outliers, the accuracy of fault detection is improved. Compared with directly judging by voltage outliers, misjudgment is reduced and work efficiency is improved.

[0107] In summary, as Figure 2 and Figure 3 shown, the detection method provided by the embodiment of the present invention first obtains the online voltage data of the target energy storage battery from the data platform and preprocesses it; uses a voltage prediction network model based on CNN-LSTM to extract the characteristic quantities of the data and predict the subsequent operating state online; then sends the prediction result into an outlier detection algorithm model based on the LOF algorithm to calculate the outliers of the predicted time series; finally, compares the outliers with the set threshold value to determine the voltage consistency fault and give an early warning of the fault.

[0108] Embodiment 2:

[0109] The embodiment of the present invention provides a detection device for the voltage consistency fault of an energy storage battery, including:

[0110] A data acquisition module configured to acquire the online voltage data of the target energy storage battery and preprocess it;

[0111] An online prediction module configured to input the preprocessed online voltage data into a pre-trained voltage prediction network model to predict the voltage time series data at a future moment;

[0112] An outlier detection module configured to input the voltage time series data at a future moment into a pre-trained outlier detection algorithm model to obtain voltage outliers;

[0113] A fault judgment module configured to determine that there is a voltage consistency fault if there is a voltage outlier greater than the set threshold value.

[0114] Specifically, in this embodiment, the voltage prediction network model is a CNN-LSTM deep learning model combining a CNN convolutional neural network and an LSTM long short-term memory network; and / or, the outlier detection algorithm model is an LOF algorithm model. In other optional embodiments, those skilled in the art can select other algorithm models for operation according to needs.

[0115] Embodiment 3:

[0116] Based on the detection method provided in Embodiment 1, the embodiment of the present invention provides an electronic device, including a processor and a storage medium;

[0117] The storage medium is used to store instructions;

[0118] The processor is used to operate according to the instructions to execute the steps of the above method.

[0119] Example 4:

[0120] Based on the detection method provided in Example 1, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the above method are implemented.

[0121] Example 5:

[0122] Based on the detection method provided in Example 1, an embodiment of the present invention provides a computer program product, including a computer program / instructions, and when the computer program / instructions are executed by a processor, the steps of the above method are implemented.

[0123] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can adopt the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0124] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0125] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0126] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide means for implementing the functions in the processFigure 1 one process or multiple processes and / or boxes Figure 1 steps of functions specified in one box or multiple boxes.

[0127] The above is only the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A detection method for the voltage consistency fault of an energy storage battery, characterized in that Including: Obtain the online voltage data of the target energy storage battery and perform preprocessing; Input the preprocessed online voltage data into a pre-trained voltage prediction network model to predict the voltage time-series data at future moments; Input the voltage time-series data at future moments into a pre-trained anomaly detection algorithm model to obtain voltage anomaly values; If there is a voltage anomaly value greater than the set threshold, there is a voltage consistency fault.

2. The detection method for the voltage consistency fault of the energy storage battery according to claim 1, wherein The preprocessing includes normalization processing and noise reduction processing.

3. The detection method for the voltage consistency fault of the energy storage battery according to claim 1, wherein The voltage prediction network model is a CNN-LSTM deep learning model that combines a CNN (Convolutional Neural Network) and an LSTM (Long Short-Term Memory Network); The CNN convolutional neural network is used to extract features from input samples. During the one-dimensional convolution process, the convolutional kernel slides over each input sample and performs convolution. The output dimension of the convolutional layer is determined by the number of convolutional kernels, and each input sample is transformed into a feature map : Wherein, are the weight matrix and bias term of the convolutional layer, is the non-linear activation function, is the time window data from the th moment to the th moment, is the length of the time window, is the feature map of the input sample at the th moment; The LSTM (Long Short-Term Memory) network is used to capture the long-term temporal dependencies of the feature map and obtain the hidden state : wherein, is the forget gate, is the input gate, is the output gate, is the state of the memory cell, are the weight matrix and the bias term; A fully connected layer is set after the LSTM long short-term memory network, and the fully connected layer generates a prediction result according to the hidden state ​ 4. The detection method for the voltage consistency fault of the energy storage battery according to claim 1, characterized in that The anomaly detection algorithm model is a LOF algorithm model; the processing process of the LOF algorithm model includes: For each data point in the input sample, calculate its k-nearest neighbor distance : wherein, is the k-nearest neighbor distance of the data point , is the -th neighboring data point of the data point , is the distance between the data points ; Calculate the local reachability distance of each data point based on the k-nearest neighbor distance : Wherein, is the k-nearest neighbor distance of the data point , is the local reachability distance of the data point , , is the set of the first nearest neighboring data points of the data point ; Calculate the local reachability density of each data point based on the local reachability distance : In the formula, is the local reachability density of the data point , and Calculate the average of the local reachability density of each data point based on the local reachability density : Wherein, is the average of the local reachability density of the data point , is the local reachability density of the data point ; Calculate the local outlier factor for each data point based on the local reachability density and the average of the local reachability densities : In the formula, is the local outlier factor of the data point .

5. The detection method for the voltage consistency fault of the energy storage battery according to claim 1, wherein The setting of the set threshold for the voltage anomaly value includes: Obtain the historical voltage data of the target energy storage battery; Input the historical voltage data into a pre-trained anomaly detection algorithm model to obtain voltage anomaly values and determine the battery module where the corresponding battery cell is located; Calculate the voltage mean values of the battery module and its adjacent battery modules as the set threshold.

6. A detection device for the voltage consistency fault of an energy storage battery, characterized in that, Including: A data acquisition module configured to obtain the online voltage data of the target energy storage battery and perform preprocessing; An online prediction module configured to input the preprocessed online voltage data into a pre-trained voltage prediction network model to predict the voltage time-series data at future moments; An anomaly detection module configured to input the voltage time-series data at future moments into a pre-trained anomaly detection algorithm model to obtain voltage anomaly values; A fault judgment module configured to determine that there is a voltage consistency fault if there is a voltage anomaly value greater than the set threshold.

7. The detection device for the voltage consistency fault of the energy storage battery according to claim 6, wherein, The voltage prediction network model is a CNN-LSTM deep learning model that combines a CNN (Convolutional Neural Network) and an LSTM (Long Short-Term Memory Network); and / or, the anomaly detection algorithm model is a LOF algorithm model.

8. An electronic device, characterized in that, Including a processor and a storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of the method according to any one of claims 1-5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method according to any one of claims 1-5.

10. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, it implements the steps of the method according to any one of claims 1-5.

Images