Lithium battery abnormity identification method based on multivariate signal reconstruction

Abstract

The invention relates to the technical field of lithium battery abnormity identification, and discloses a lithium battery abnormity identification method based on multivariate signal reconstruction, and the method comprises the steps: firstly building a BMS simulation system, obtaining a lithium battery charging and discharging multivariate signal curve, and carrying out the noise reduction processing; multivariate signal feature data points are obtained through second-order sensitivity and Gaussian distribution weight selection, and after normalization processing, differential compression storage is carried out; constructing a multi-input multi-output LSTM (Long Short Term Memory) model; extracting the stored data for reverse normalization processing to obtain a lithium battery characteristic data set, combining the processed multivariate signal curve to serve as a training sample of an LSTM model, and adopting an alpha evolutionary optimization algorithm to perform optimization processing on the node number, the time sequence length and the initial learning rate of each layer in an LSTM hidden layer to obtain a lithium battery characteristic data set; and finally obtaining a multi-element signal reconstruction model of the battery and reconstructing and predicting a full life cycle signal of the battery, thereby realizing short-term battery abnormal state identification, long-term battery abnormal state identification and backtracking and analysis of a fault process.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery abnormality identification, and in particular to a lithium battery abnormality identification method based on multi-element signal reconstruction. Background Art

[0002] The operation of lithium batteries is often accompanied by various abnormal conditions and failures, which can cause battery life to decline and may even cause fires and other accidents. Electric vehicles use a battery management system (BMS) to collect multiple battery signals (such as voltage, current, and temperature) in real time, detect abnormal battery conditions, and take measures. Battery status abnormalities and failures are often the result of multiple complex factors. Early signs are often not obvious and difficult to distinguish through instantaneous parameters. Therefore, historical data is of great significance for identifying abnormal battery conditions and predicting potential failures.

[0003] However, due to the limitation of hardware cost and system complexity, BMS is usually equipped with small-capacity EEPROM memory, which cannot store all the historical data of the battery, and easily leads to the failure to capture early abnormal signs. This limitation has become a bottleneck for data utilization under the existing technical framework. With the deepening of research, pressure and strain as signals that can characterize the internal state of the battery have become new bases for determining battery abnormalities. The introduction of these new signals has further increased the burden of data collection and storage.

[0004] Most of the existing technologies use fragments to estimate the battery status, but do not provide a basis for selecting fragments, and do not solve the problem of small amount of data stored in BMS. Moreover, most of the existing technologies focus on the use of voltage signals, and do not involve signals such as temperature, pressure and strain, which has certain limitations. Therefore, how to optimize the data storage strategy under limited hardware resources, extract key battery multi-signals, restore the complete operating status through reconstruction technology when necessary, and identify battery abnormalities are problems that need to be solved urgently. Summary of the invention

[0005] The purpose of the present invention is to solve the problem of how to optimize data storage strategies and extract key battery multivariate signals under limited hardware resource conditions, restore the complete operating state through reconstruction technology when necessary, and identify battery abnormalities, and propose a lithium battery abnormality identification method based on multivariate signal reconstruction.

[0006] The technical solution of the present invention to solve the above technical problems is as follows: A lithium battery abnormality recognition method based on multi-signal reconstruction comprises the following steps: S1. Build a BMS simulation system, obtain the multivariate signal curve of lithium battery charging and discharging, and perform noise reduction on the multivariate signal to obtain the processed multivariate signal curve; S2. Obtain the multivariate signal feature data points in the processed multivariate signal curve by selecting through second-order sensitivity and Gaussian distribution weights; S3. After normalizing the multivariate signal feature data points, store them in the EEPROM memory of the BMS simulation system through differential compression; S4. Construct a multi-input multi-output LSTM model; S5. Extract data from the EEPROM memory for denormalization processing to obtain the lithium battery feature data set. Combine it with the processed multivariate signal curve as the training sample of the LSTM model. Use the alpha evolution optimization algorithm to optimize the number of nodes in each layer, the time series length, and the initial learning rate in the LSTM hidden layer, and finally obtain the lithium battery multivariate signal reconstruction model; S6. Use the lithium battery multivariate signal reconstruction model to reconstruct and predict the signals in the entire life cycle of the lithium battery, and realize the recognition of short-term lithium battery abnormal states, the recognition of long-term lithium battery abnormal states, and the backtracking and analysis of the fault process.

[0007] Based on the above technical solutions, the present invention can also be improved as follows.

[0008] Preferably, the specific steps of S2 are as follows: S2.1. Use to represent the multivariate signal, where x i is the sampling point position, f(x i ) is the value corresponding to the signal at the x i position. Then calculate the second-order sensitivity ω i of f(x); S2.2. Use the Gaussian distribution function to weight the second-order sensitivity ω i to obtain W i ; S2.3. Set the threshold T, and select the multivariate signal feature data points according to the judgment formula W i <T.

[0009] Preferably, the multivariate signals include voltage signals, temperature signals, pressure signals, and strain signals.

[0010] Preferably, the specific steps of S3 are as follows: S3.1. Set identifiers for the voltage signal, temperature signal, pressure signal, and strain signal respectively to represent the signal types; S3.2. Normalize the multivariate signal feature data points obtained in S2 according to the signal types respectively; S3.3. Integrate the multivariate signal feature data points together according to the time stamp, and store the data using the differential compression method.

[0011] Preferably, the specific steps of S5 are as follows: S5.1. Extract the differentially compressed data in S3, perform denormalization processing, restore it to multi-source signal feature data points, and arrange them in the order of sampling timestamps to form a lithium battery feature dataset; S5.2. Use the multi-source signal matrix S recorded within the timestamp in the lithium battery feature dataset as the input layer of the LSTM model. The matrix S is: ; where S vol , S tem , S for , S str respectively represent the feature data points of voltage, temperature, pressure, and strain signals; S5.3. Use the lithium battery feature dataset and the processed multi-source signal curve to train the LSTM model, and use the alpha evolution optimization algorithm to optimize the number of nodes in each layer, the time series length, and the initial learning rate in the LSTM hidden layer, and finally obtain a lithium battery multi-source signal reconstruction model.

[0012] Preferably, the calculation formula of the second-order sensitivity ω i in S2.1 is as follows: ; where f ,, (x i ) is the second derivative of f(x i ).

[0013] Preferably, the weighting formula in S2.2 is as follows: ; where μ is the mean of the second-order sensitivity ω i , and σ is the standard deviation of the second-order sensitivity ω i . Preferably, the specific steps of S2.3 are as follows: Set a threshold T. According to Wi < T, select the multi-source signal feature data points. The index idx of the multi-source signal feature data points is: ; Select the marked data points in the index from the multi-source signal respectively, and record the values and timestamps of these data points as the multi-source signal feature data points.

[0014] Preferably, the difference formula in S3.3 is as follows: ; where Δy nor is the difference value, and y nor (k) and y nor(k-1) are the normalized values ​​of the kth and k-1th feature data points respectively.

[0015] Preferably, the specific steps of integrating the multivariate signal feature data points according to timestamps in S3.3 are as follows: The characteristic data points of voltage signal, temperature signal, pressure signal and strain signal are integrated together according to the timestamp, and a timestamp is shared. For a certain type of signal that does not exist in the timestamp, a null character is used instead. The stored data structure is: ; Among them, t l is the timestamp, v l is a collection of multivariate signal data within a timestamp.

[0016] Compared with the prior art, the technical solution of this application has the following beneficial technical effects: 1) The present invention can accurately extract and store data points that are highly relevant to the operating status of lithium batteries, effectively alleviating the limitations of the BMS system in terms of storage resources, enabling it to store and utilize important historical data for a long time; 2) The present invention can realize efficient data utilization under the condition of limited BMS hardware resources, avoid the introduction of large-capacity memory or high-performance computing unit, and significantly reduce the development and operation costs of the system; 3) The present invention predicts the operating status of the lithium battery throughout its life cycle through a lithium battery multivariate signal reconstruction model. By comparison, it can accurately predict potential failures of the lithium battery and effectively reduce safety risks. Even if a small number of failures occur suddenly without prior signs, the method proposed in the present invention can still reconstruct the operating data of the faulty lithium battery and achieve backtracking and analysis of the failure process. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the overall process of the present invention; Figure 2 This is a result diagram of selecting characteristic data points of voltage signal under constant current charge and discharge of the present invention; Figure 3 This is a result diagram of the pressure signal characteristic data point selection under constant current charging and discharging of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] A lithium battery abnormality recognition method based on multi-signal reconstruction comprises the following steps: S1. Build a BMS simulation system, obtain the multivariate signal curve of lithium battery charging and discharging, and perform noise reduction on the multivariate signal to obtain the processed multivariate signal curve; The specific steps of S1 are as follows: In the laboratory environment, a full life cycle experiment of lithium batteries is designed, using a constant current-constant voltage charging and constant current discharging strategy until the lithium battery reaches the end of its life. A BMS simulation system is built using an MCU chip and EEPROM memory. The MCU chip integrates a voltage sensor, a thermocouple, a pressure film sensor, and a strain gauge sensor to collect the voltage, temperature, pressure, and strain signals of the lithium battery in real time. After that, the multivariate signals are subjected to noise reduction processing to obtain a full life cycle data set of lithium batteries. S2, obtaining multivariate signal characteristic data points in the processed multivariate signal curve through second-order sensitivity and Gaussian distribution weight selection; S3, after normalizing the characteristic data points of the multivariate signal, they are stored in the EEPROM memory of the BMS simulation system after differential compression; S4. Build a multi-input and multi-output LSTM model; S5. Extract data from the EEPROM memory and perform denormalization processing to obtain a lithium battery feature data set, combine the processed multivariate signal curve, and use it as a training sample for the LSTM model. Use the alpha evolution optimization algorithm to optimize the number of nodes in each layer of the LSTM hidden layer, the time series length, and the initial learning rate, and finally obtain a lithium battery multivariate signal reconstruction model. S6. Use the lithium battery multivariate signal reconstruction model to reconstruct and predict the full life cycle signal of the lithium battery, realize the short-term lithium battery abnormal state identification, long-term lithium battery abnormal state identification and the backtracking and analysis of the fault process.

[0020] The specific steps of S6 are as follows: S6.1 Use the lithium battery multivariate signal reconstruction model to predict the operating data of the lithium battery in this cycle, and compare it with the operating data collected in real time by the BMS to identify the abnormal state of the short-term lithium battery; S6.2 Use the model to predict the multivariate signal curve of the lithium battery's entire life cycle, and compare it with the multivariate signal curve reconstructed from the characteristic data stored in the BMS to achieve long-term lithium battery abnormal status identification; S6.3 If there is no obvious abnormality before the fault occurs, then the model is used to reconstruct all operating signals of the faulty lithium battery and trace back and analyze the fault.

[0021] The specific steps of S2 are as follows: S2.1 Use represents a multi - variable signal, where x i is the sampling point position, and f(x i ) is the value corresponding to the position signal of x i , and then calculate the second - order sensitivity ω of f(x) i ; S2.2. Use the Gaussian distribution function to weight the second - order sensitivity ω i to obtain W i ; S2.3. Set a threshold T, and according to the judgment formula W i <T, select the multi - variable signal feature data points.

[0022] The multi - variable signal includes a voltage signal, a temperature signal, a pressure signal, and a strain signal. The multi - variable signal includes a voltage signal, a temperature signal, a pressure signal, and a strain signal.

[0023] The specific steps of S3 are as follows: S3.1. Set identifiers for the voltage signal, the temperature signal, the pressure signal, and the strain signal respectively to represent the signal types; S3.2. Perform normalization processing on the multi - variable signal feature data points obtained in S2 according to the signal types respectively; S3.3. Integrate the multi - variable signal feature data points together according to the time stamps, and store the data using the differential compression method.

[0024] The specific steps of S5 are as follows: S5.1. Extract the differentially compressed data in S3, perform denormalization processing, restore it to the multi - variable signal feature data points, arrange them in the order of the sampling time stamps, and form a lithium - battery feature data set; S5.2. Use the multi - variable signal matrix S recorded within the time stamps in the lithium - battery feature data set as the input layer of the LSTM model. The matrix S is: ; where S vol , S tem , S for , S str represent the feature data points of the voltage, temperature, pressure, and strain signals respectively; S5.3. Use the lithium - battery feature data set and the processed multi - variable signal curve to train the LSTM model, and use the alpha - evolution optimization algorithm to optimize the number of nodes in each layer, the time - series length, and the initial learning rate in the LSTM hidden layer, and finally obtain a lithium - battery multi - variable signal reconstruction model.

[0025] The specific steps for optimizing the number of nodes in each layer of the LSTM hidden layer, the length of the time series, and the initial learning rate using the alpha evolution optimization algorithm are as follows: (1. Set the input parameters of the optimization algorithm, including the number of candidate solutions N, the dimension dim of the optimization parameters, the upper and lower bounds [ub, lb] of the variables, and the maximum number of iterations Imax (the candidate solutions correspond to the number of nodes in each layer of the LSTM hidden layer, the length of the time series, and the initial learning rate); (2. Select the root mean square error (RMSE) as the fitness function, and the optimization goal is to minimize the fitness. Generate an initial candidate solution matrix. The candidate solutions include three dimensions: the number of nodes in each layer of the LSTM hidden layer, the length of the time series, and the initial learning rate. Iteratively optimize during the LSTM training process until the maximum number of iterations is reached; (3. Output the best combination of the number of nodes in each layer of the LSTM hidden layer, the length of the time series, and the initial learning rate, and the model training is completed.

[0026] The second-order sensitivity ω in S2.1 i is calculated as follows: ; where f ,, (x i ) is the second derivative of f(x i ).

[0027] The weighting formula in S2.2 is as follows: ; where μ is the mean of the second-order sensitivity ω i , and σ is the standard deviation of the second-order sensitivity ω i .

[0028] The specific steps of S2.3 are as follows: Set a threshold T. According to Wi < T, select the multi-signal feature data points. The index idx of the multi-signal feature data points is: ; Select the marked data points in the index from the multi-signal respectively, and record the values and timestamps of these data points as the multi-signal feature data points.

[0029] The difference formula in S3.3 is as follows: ; where Δy nor is the difference value, and y nor (k) and y nor (k - 1) are the normalized values of the kth and (k - 1)th feature data points respectively.

[0030] The specific steps of integrating the multivariate signal feature data points according to timestamps in S3.3 are as follows: The characteristic data points of voltage signal, temperature signal, pressure signal and strain signal are integrated together according to the timestamp, and a timestamp is shared. For a certain type of signal that does not exist in the timestamp, a null character is used instead. The stored data structure is: ; Among them, t l is the timestamp, v l is a collection of multivariate signal data within a timestamp.

[0031] Combination Figure 2 and Figure 3 By using the second-order sensitivity combined with the Gaussian distribution weight method, more than 200 feature data points can be selected from the original data points of more than 7,000 in length, which can save a lot of space when storing in the BMS. In addition, the trained LSTM can restore the original voltage and pressure curves based on these more than 200 data points for subsequent abnormal lithium battery identification.

[0032] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0033] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A lithium battery abnormality identification method based on multi-signal reconstruction, characterized in that: It includes the following steps: S1. Build a BMS simulation system, obtain the multi-signal curve of lithium battery charging and discharging, and perform noise reduction processing on the multi-signal to obtain the processed multi-signal curve; S2. Select the multi-signal characteristic data points in the processed multi-signal curve through second-order sensitivity and Gaussian distribution weight selection; S3. After normalizing the multi-signal characteristic data points, store them in the EEPROM memory of the BMS simulation system through differential compression; S4. Construct a multi-input multi-output LSTM model; S5. Extract data from the EEPROM memory for inverse normalization processing to obtain the lithium battery characteristic data set. Combine it with the processed multi-signal curve as the training sample of the LSTM model. Use the alpha evolution optimization algorithm to optimize the number of nodes in each layer, the time series length, and the initial learning rate in the LSTM hidden layer, and finally obtain the lithium battery multi-signal reconstruction model; S6. Use the lithium battery multi-signal reconstruction model to reconstruct and predict the full life cycle signal of the lithium battery, and realize the identification of short-term lithium battery abnormal states, the identification of long-term lithium battery abnormal states, and the backtracking and analysis of the fault process.

2. A lithium battery abnormality identification method based on multi-signal reconstruction according to claim 1, characterized in that: The specific steps of S2 are as follows: S2.1 Use represents a multivariate signal, where x i is the sampling point position, f(x i ) is x i The value corresponding to the position signal, and then calculate the second-order sensitivity ω of f(x) i ; S2.2, using Gaussian distribution function to calculate the second-order sensitivity ω i Weighted to get W i ; S2.3, set the threshold T, according to the judgment formula W i <T ， Select the characteristic data points of the multivariate signal.

3. The lithium battery abnormality identification method based on multi-signal reconstruction according to claim 1 is characterized in that: The multi-signal includes voltage signal, temperature signal, pressure signal and strain signal.

4. The lithium battery abnormality identification method based on multi-signal reconstruction according to claim 3 is characterized in that: The specific steps of S3 are as follows: S3.

1. Set identifiers for the voltage signal, temperature signal, pressure signal and strain signal respectively to represent the signal types; S3.

2. Perform normalization processing on the multi-signal characteristic data points obtained in S2 according to the signal types respectively; S3.

3. Integrate the multi-signal characteristic data points together according to the time stamps, and store the data using the differential compression method.

5. The method for identifying lithium battery abnormality based on multi-signal reconstruction according to claim 4, characterized in that: The specific steps of S5 are as follows: S5.

1. Extract the differentially compressed data in S3, perform inverse normalization processing, restore it to the multi-signal characteristic data points, arrange them in the order of sampling time stamps, and form the lithium battery characteristic data set; S5.

2. Use the multi-signal matrix S recorded within the time stamp in the lithium battery characteristic data set as the input layer of the LSTM model. The matrix S is: ； Among them, S vol , S tem , S for , S str Represent the characteristic data points of voltage, temperature, pressure and strain signals respectively; S5.

3. Use the lithium battery characteristic data set and the processed multi-signal curve to train the LSTM model, and use the alpha evolution optimization algorithm to optimize the number of nodes in each layer, the time series length, and the initial learning rate in the LSTM hidden layer, and finally obtain the lithium battery multi-signal reconstruction model.

6. The lithium battery abnormality identification method based on multi-signal reconstruction according to claim 2 is characterized in that: The second-order sensitivity ω in S2.1 i The calculation formula is as follows: ； Among them, f ,, (x i ) is f(x i ) is the second-order derivative of .

7. The lithium battery abnormality identification method based on multi-signal reconstruction according to claim 2 is characterized in that: The weighting formula in S2.2 is as follows: ； Where μ is the second-order sensitivity ω i The mean of i The standard deviation of .

8. The lithium battery abnormality identification method based on multi-signal reconstruction according to claim 2 is characterized in that: The specific steps of S2.3 are as follows: Set a threshold T. According to Wi < T, select the multi-signal characteristic data points. The index idx of the multi-signal characteristic data points is: ； Select the marked data points in the index in the multi-signal respectively, and record the values and time stamps of these data points as the multi-signal characteristic data points.

9. The lithium battery abnormality identification method based on multi-signal reconstruction according to claim 4 is characterized in that: The differential formula in S3.3 is as follows: ； Among them, Δy nor is the difference value, y nor (k) and y nor (k-1) are the normalized values ​​of the kth and k-1th feature data points respectively.

10. A lithium battery abnormality identification method based on multi-signal reconstruction according to any one of claims 4 or 8, characterized in that: The specific steps of integrating the multi-signal characteristic data points together according to the time stamps in S3.3 are as follows: The characteristic data points of voltage signal, temperature signal, pressure signal and strain signal are integrated together according to the timestamp, and a timestamp is shared. For a certain type of signal that does not exist in the timestamp, a null character is used instead. The stored data structure is: ； Among them, t l is the timestamp, v l is a collection of multivariate signal data within a timestamp.

Images