Lithium battery health state monitoring method and system

Abstract

The invention discloses a lithium battery health state monitoring method and system, and the method comprises the steps: carrying out the real-time monitoring through a sensor, obtaining the original data of a battery, carrying out the preprocessing of the original data, obtaining multi-dimensional time series data, and dividing the data into a training set and a test set; the multi-dimensional time series data are optimized by using a generative adversarial network ImputeGAN of multivariate time series interpolation, and the precision and efficiency of multi-dimensional time series data interpolation are improved; a parallel convolutional neural network PCNN is adopted to extract feature vectors of the optimized multi-dimensional time sequence data; improving a whale optimization algorithm WOA algorithm by using a relative learning strategy and a differential sorting variation strategy to obtain an IWOA algorithm; inputting the feature vector extracted by the PCNN into deep extended learning for training in a periodic time sequence prediction model DEPTS, optimizing parameters of the DEPTS by using an IWOA algorithm, and monitoring the health state of the lithium battery by using the trained and optimized DEPTS; according to the invention, lithium battery health state monitoring can be accurately and efficiently carried out.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium battery health status monitoring, and in particular relates to a lithium battery health status monitoring method and system. Background Art

[0002] With the rapid development of new energy vehicles and portable electronic devices, lithium batteries, as the main energy storage device, have become particularly important to monitor their health status. The performance and life of lithium batteries are affected by many factors, including charge and discharge cycles, temperature changes, current and voltage fluctuations, etc. These factors can cause battery capacity decay, increase internal resistance, and even safety accidents. Therefore, real-time monitoring of key parameters of lithium batteries, such as voltage, current, temperature, and surface pressure, is essential for evaluating the health status of batteries, predicting battery life, and avoiding potential safety risks. An effective health status monitoring system can provide key information about battery performance and safety, thereby guiding battery maintenance and replacement, reducing operating costs, and improving energy efficiency.

[0003] In lithium battery health status monitoring, the collected data is usually multidimensional time series data, which is often accompanied by missing values ​​due to sensor failure, data transmission errors, etc. Traditional interpolation methods, such as linear interpolation or mean interpolation, often fail to effectively utilize the intrinsic structural information of time series data, resulting in low accuracy and efficiency of interpolation results. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a lithium battery health status monitoring method and system that can improve monitoring accuracy and efficiency.

[0005] Technical solution: A lithium battery health status monitoring method according to the present invention comprises:

[0006] (1) using sensors for real-time monitoring to obtain raw data of the battery, preprocessing the raw data to obtain multi-dimensional time series data, and dividing the data into a training set and a test set; the raw data includes voltage, current, temperature, surface pressure and current capacity;

[0007] (2) Using ImputeGAN, a generative adversarial network for multidimensional time series interpolation, to optimize multidimensional time series data and improve the accuracy and efficiency of multidimensional time series data interpolation;

[0008] (3) Using parallel convolutional neural network (PCNN) to extract the feature vector of optimized multidimensional time series data;

[0009] (4) The adversarial learning strategy and differential sorting mutation strategy are used to improve the whale optimization algorithm WOA algorithm to obtain the IWOA algorithm;

[0010] (5) The feature vector extracted by PCNN is input into deep extended learning for training the periodic time series prediction model DEPTS. At the same time, the IWOA algorithm is used to optimize the parameters of DEPTS, and the trained and optimized DEPTS is used to monitor the health status of lithium batteries.

[0011] Furthermore, the preprocessing of the raw data in step (1) includes: cleaning the data, removing noise and inconsistency in the data, and detecting and processing outliers.

[0012] Furthermore, in step (2), the loss function of ImputeGAN includes a generator loss function and a discriminator loss function;

[0013] The loss function of the generator is expressed as:

[0014]

[0015] Among them, L G represents the loss function of the generator; E represents the expected value, which represents the average value of a random variable; z represents the noise vector sampled from the prior distribution, which is the input of the generator; p z (z) represents the probability distribution of the noise vector z; G(z) represents the function of the generator, which is used to receive the noise vector z and generate data; D(G(z)) represents the judgment of the discriminator on the false data generated by the generator, and outputs a probability value to represent the probability that the false data is real data;

[0016] The loss function of the discriminator is used to maximize the ability to distinguish between real data and generated data, expressed as:

[0017]

[0018] Among them, L D represents the loss function of the discriminator; E represents the expected value, which represents the average value of a random variable; x represents the distribution of real data p data (x) is the sampled data point; p data (x) is the probability distribution of real data; D(x) is the output of the discriminator for input x, indicating the probability that x is real data.

[0019] Furthermore, the step (3) comprises:

[0020] The convolutional layer of the PCNN is represented as:

[0021] f θ (x) = ReLU(W*x+b)

[0022] Among them, f θ(x) is the output after being processed by the ReLU activation function; W is the convolution kernel weight; b is the bias term; * represents the convolution operation; ReLU is the activation function; x represents a multidimensional time series data point;

[0023] The feature maps of different channels after the convolution layer are fused, and the fusion operation is expressed as:

[0024] P fusion =Pool({f θ1 (x1), f θ1 (x2),…,f θn (x n )})

[0025] Among them, P fusion represents the fused feature representation, which is a feature vector; Pool represents the pooling operation; n is the number of data points, x n is the nth data point.

[0026] Furthermore, the step (4) comprises:

[0027] The adversarial learning strategy is introduced to enhance the search capability of the whale optimization algorithm WOA. The adversarial learning strategy generates an adversarial solution for each solution. The adversarial solution is a solution that is equidistant from the original solution in the search space but in the opposite direction. For a solution y, its adversarial solution y′ can be calculated by the following formula:

[0028] y′=L+Uy

[0029] Among them, L and U are the lower and upper bounds of the search space respectively;

[0030] The differential sorting mutation strategy is used to enhance the diversity of the population. The differences between individuals in the population are calculated, and the individuals are sorted according to these differences, and the mutation operation is applied. The differential sorting mutation is expressed as:

[0031] v i =u i +F·(u i+1 -u i-1 )

[0032] Among them, u i is the current individual, u i+1 and u i-1 are adjacent individuals in the population, and F is the scaling factor.

[0033] Furthermore, in step (5), the feature vector extracted by PCNN is input into deep extended learning for training the periodic time series prediction model DEPTS, including:

[0034] The DEPTS includes an extension module and a periodic module; the extension module is responsible for processing the complex dependencies of time series data, and the periodic module is used to capture the periodic characteristics of time series;

[0035] Among them, the periodic characteristic z t By parameterizing the periodic function g φ (t) means:

[0036]

[0037] Among them, A0 is the baseline term, which represents the average value of the periodic pattern; A k is the amplitude, indicating the intensity of the kth periodic component; F k is the frequency, indicating the periodicity of the kth periodic component; P k is the phase, indicating the starting point of the kth periodic component; K is the total number of periodic components;

[0038] The expansion module f θ Responsible for the observation value X t-L:t and the periodic characteristics z t-L:t+H Make predictions:

[0039] X t:t+H =f θ (X t-L:t , z t-L:t+H )+∈ t:t+H

[0040] Among them, X t:t+H represents the predicted value sequence of the health status of the lithium battery from the current time step t to the future H time steps, including the battery voltage, current, temperature, surface pressure and current capacity data; f θ Represents the function in the DEPTS model, which is a parameterized function used to predict the future health status of lithium batteries based on input characteristics and periodic characteristics; X t-L:t represents the historical lithium battery health status feature vector sequence from tL to t; z t-L:t+H Represents the periodic characteristic sequence from tL to t+H; ∈ t:t+H It represents the prediction error term, which is assumed to be independent and identically distributed Gaussian noise;

[0041] Construct a correlation formula to correlate the voltage, current, temperature, surface pressure of the lithium battery with the battery capacity decay ΔC:

[0042] ΔC=k V ·V t:t+H +k I I t:t+H +k T ·T t:t+H +k P·P t:t+H

[0043] Among them, ΔC represents the battery capacity decay, and the current state C is obtained by combining the initial state C0 of the battery t ; V t:t+H ,I t:t+H 、T t:t+H and P t:t+H are the predicted values ​​of voltage, current, temperature and surface pressure respectively; k V , k I , k T and k P They are the voltage coefficient, current coefficient, temperature coefficient and surface pressure coefficient;

[0044] The battery's state of health (SOH) is defined based on the attenuation of battery capacity:

[0045]

[0046] Used to indicate the remaining percentage of battery capacity relative to the initial capacity.

[0047] Furthermore, in step (5), the parameters of DEPTS are optimized by using the IWOA algorithm, and the health status of the lithium battery is monitored by using the trained and optimized DEPTS, including:

[0048] Define the parameter space of the DEPTS model, where the parameters in the parameter space include the weights and biases of the dilation module and the periodic module; initialize the whale population. In IWOA, each whale represents a set of parameters of the DEPTS model, and randomly initialize the position of the whale population;

[0049] Evaluate the fitness of each whale and define the fitness function as a function of SOH:

[0050] Fitness(θ)=SOH(θ)

[0051] Where θ represents the parameter set represented by the whale. For each set of parameters θ, the DEPTS model is used to predict the current capacity C of the battery. t , to calculate SOH;

[0052] Determining the best whale in the current group, IWOA updates the whale's position by the following formula:

[0053] θ i (d+1)=θ i (d)+A·C·(θ best -θ i (d))+D·X

[0054] Among them, θ i(d) is the position of the ith whale at the dth iteration, θ best is the optimal solution found so far, A and C are acceleration coefficients, D is the step size factor, and X is a random vector;

[0055] The optimal parameter θ found by the IWOA algorithm best , updated to the DEPTS model, and used the updated parameters to evaluate the performance of the DEPTS model on the test set. The deep extended learning after training optimization was used for the periodic time series prediction model to monitor the health status of the lithium battery. The health status of the lithium battery was evaluated according to the SOH value monitored by the DEPTS model. The closer the SOH value is to 100%, the closer the battery is to the initial capacity and the better the health status; if the SOH value is lower, it indicates that the current capacity of the lithium battery has dropped significantly compared with the initial capacity.

[0056] Based on the same inventive concept, the present invention also provides a lithium battery health status monitoring system, comprising:

[0057] An initialization module is used to obtain raw data of the battery by real-time monitoring using sensors, preprocess the raw data to obtain multi-dimensional time series data, and divide the data into a training set and a test set; the raw data includes voltage, current, temperature, surface pressure and current capacity;

[0058] The optimization module is used to optimize multidimensional time series data using ImputeGAN, a generative adversarial network for multidimensional time series interpolation, to improve the accuracy and efficiency of multidimensional time series data interpolation;

[0059] An extraction module, used for extracting the feature vector of the optimized multi-dimensional time series data using a parallel convolutional neural network PCNN;

[0060] The improvement module is used to improve the whale optimization algorithm WOA algorithm by using the adversarial learning strategy and the differential sorting mutation strategy to obtain the IWOA algorithm;

[0061] The monitoring module is used to input the feature vector extracted by PCNN into the deep extended learning for training the periodic time series prediction model DEPTS, and optimize the parameters of DEPTS using the IWOA algorithm, and use the trained and optimized DEPTS to monitor the health status of the lithium battery.

[0062] Based on the same inventive concept, the present invention also provides a computing device, comprising: one or more processors, one or more memories and one or more programs, wherein the programs are stored in the memories and configured to be executed by the processors, and when the programs are loaded into the processors, the steps of the lithium battery health status monitoring method according to any one of the above items are implemented.

[0063] Based on the same inventive concept, the present invention also provides a storage medium, which stores a computer program, and the computer program includes program instructions. When the program instructions are executed by a processor, the processor executes the steps of the lithium battery health status monitoring method described in any one of the above items.

[0064] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: 1. In terms of data integrity and interpolation accuracy, the present invention adopts ImputeGAN technology to effectively solve the missing problem in multidimensional time series data; compared with traditional interpolation methods, ImputeGAN improves the accuracy and efficiency of data interpolation and enhances data integrity by generating adversarial networks; this method can provide more accurate data completion, especially in the case of data missing caused by sensor failure or data transmission errors, thereby ensuring the continuity and stability of health status monitoring; 2. In terms of feature extraction and expression capabilities, the present invention uses parallel convolutional neural networks (PCNN) to extract features from multidimensional time series data. Compared with traditional methods, PCNN can process multiple input channels in parallel, extract richer features, and enhance the feature expression capabilities of data; this improvement enables the system to more accurately capture battery status The changes in the state provide strong data support for the accurate prediction of the health status; 3. In terms of algorithm optimization and search capabilities, by introducing the whale optimization algorithm (IWOA) improved by the adversarial learning strategy and the differential sorting mutation strategy, the present invention has achieved significant improvement in algorithm performance; compared with traditional optimization algorithms, the IWOA algorithm improves the global search capability and convergence speed, optimizes the parameters of the periodic time series prediction model, and thus more effectively serves the optimization needs of the battery management system; 4. In terms of the accuracy of health status prediction, the present invention inputs the feature vector extracted by PCNN into the deep extended learning for the periodic time series prediction model (DEPTS), and uses the IWOA algorithm to optimize the model parameters, thereby achieving accurate prediction of the health status of lithium batteries; compared with traditional prediction methods, this deep learning method can more accurately capture subtle changes in battery performance and improve the accuracy and reliability of predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0066] Figure 2 Schematic diagram of a periodic time series prediction model using deep extended learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0067] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0068] As attached Figure 1 As shown, the lithium battery health status monitoring method of this embodiment includes:

[0069] Step 1: Use sensors to monitor in real time to obtain raw data of the battery, pre-process the raw data to obtain multi-dimensional time series data, and divide it into a training set and a test set; the raw data includes voltage, current, temperature, surface pressure and current capacity;

[0070] Step 2: Use ImputeGAN, a generative adversarial network for multivariate time series interpolation, to optimize multidimensional time series data and improve the accuracy and efficiency of multidimensional time series data interpolation;

[0071] Step 3: Use the parallel convolutional neural network PCNN to extract the feature vector of the optimized multidimensional time series data;

[0072] Step 4: Use the adversarial learning strategy and differential sorting mutation strategy to improve the whale optimization algorithm WOA algorithm to obtain the IWOA algorithm;

[0073] Step 5: Input the feature vector extracted by PCNN into deep extended learning for training the periodic time series prediction model DEPTS. At the same time, use the IWOA algorithm to optimize the parameters of DEPTS, and use the trained and optimized DEPTS to monitor the health status of lithium batteries.

[0074] Specifically, in step 1, sensors are used to monitor key data such as battery voltage, current, temperature, surface pressure and current capacity in real time, and data preprocessing is performed; preprocessing of the raw data includes: cleaning the data to remove noise and inconsistencies in the data set; detecting and processing outliers; after the preprocessing is completed, the data set is then divided into a training set and a test set.

[0075] Step 2: Use the generative adversarial network (ImputeGAN) for multidimensional time series interpolation to effectively improve the accuracy and efficiency of multidimensional time series data interpolation;

[0076] In multidimensional time series data, data missing is inevitable due to sensor failure, data transmission errors, etc. ImputeGAN aims to solve this problem by using Generative Adversarial Networks (GANs) to complete the missing data.

[0077] ImputeGAN consists of two main parts, the generator (G) and the discriminator (D). The goal of the generator is to generate missing data that is similar to the real data distribution, while the discriminator is used to distinguish the generated data from the real data. The two compete with each other during the training process to improve the quality of the generated data.

[0078] The loss function of ImputeGAN consists of two parts, one is the loss of the generator and the other is the loss of the discriminator. The loss function of the generator can be expressed as:

[0079] L G =-E z～pz(z) [log(D(G(z)))]

[0080] Among them, LG represents the loss function of the generator; E represents the expected value, which represents the average value of a random variable; z represents the noise vector sampled from the prior distribution, which is the input of the generator; p z (z) represents the probability distribution of the noise vector z; G(z) represents the function of the generator, which receives the noise vector z and generates data; D(G(z)) represents the discriminator's judgment on the false data generated by the generator, and outputs a probability value, indicating the probability that this data is real data.

[0081] The loss function of the discriminator aims to maximize the ability to distinguish between real data and generated data, which can be expressed as:

[0082] L D =-E x～pdata(x) [log(D(x))]-E z～pz(z) [log(1-D(G(z)))]

[0083] Among them, L D represents the loss function of the discriminator; E represents the expected value, which represents the average value of a random variable; x represents the distribution of real data p data (x) is the sampled data point; p data (x) is the probability distribution of real data; D(x) is the output of the discriminator for input x, indicating the probability that x is real data.

[0084] ImputeGAN adopts an iterative strategy based on complementary result gradients to solve the problem that traditional methods cannot effectively utilize time information or the completion results are unstable when processing time series data. This strategy ensures the generalization ability of the model and the rationality of the completion results.

[0085] Through the above steps, ImputeGAN not only improves the accuracy of multi-dimensional time series data interpolation, but also enhances the generalization ability of the model and the stability of the interpolation results through the adversarial training mechanism, providing an effective data preprocessing method for lithium battery health status monitoring.

[0086] Step 3: Then, a parallel convolutional neural network (PCNN) is used to extract features from multidimensional time series data, which enhances the feature expression ability of the data;

[0087] PCNN is a special convolutional neural network that can process multiple input channels in parallel. In the lithium battery health status monitoring method, each channel represents a different sensor reading, such as the battery's voltage, current, temperature, surface pressure, and current capacity value. The PCNN model extracts the features of each channel through parallel convolutional layers.

[0088] In PCNN, the kernel size, number, and stride of the convolution layer are key design parameters. The kernel size determines the range of input data that can be covered. The number of convolution kernels determines the depth of the feature map. The stride affects the size of the feature map. In general, the convolution layer can be expressed as:

[0089] f θ (x) = ReLU(W*x+b)

[0090] Among them, f θ (x) is the output after ReLU activation function processing; W is the convolution kernel weight; b is the bias term; * represents the convolution operation; ReLU is the activation function; x represents a multidimensional time series data point.

[0091] In PCNN, feature maps of different channels need to be fused after passing through the convolutional layer. This can be achieved through global average pooling or global maximum pooling to reduce the size of the feature map and extract the most important features. The fusion operation can be expressed as:

[0092] P fusion =Pool({f θ1 (x1), f θ1 (x2),…,f θn (x n )})

[0093] Among them, P fusion It represents the fused feature representation, that is, the feature vector; Pool represents the pooling operation.

[0094] Through the above steps, PCNN can extract rich features from multidimensional time series data and provide strong data support for the health status monitoring of lithium batteries.

[0095] Step 4: Improve the Whale Optimization Algorithm (WOA) algorithm using the adversarial learning strategy and the differential sorting mutation strategy to obtain the IWOA algorithm;

[0096] First, the adversarial learning strategy is introduced to enhance the search capability of the Whale Optimization Algorithm (WOA). The adversarial learning strategy increases the diversity of the population by generating an adversarial solution for each solution, that is, a solution that is equidistant from the original solution in the search space but in the opposite direction. For a solution y, its adversarial solution y′ can be calculated by the following formula:

[0097] y′=L+Uy

[0098] Among them, L and U are the lower bound and upper bound of the search space respectively.

[0099] Secondly, a differential sorting mutation strategy is used to further enhance the diversity of the population. This strategy calculates the differences between individuals in the population, sorts the individuals according to these differences, and then applies the mutation operation. The differential sorting mutation can be expressed as:

[0100] v i =u i +F·(u i+1 -u i-1 )

[0101] Among them, u i is the current individual, u i+1 and u i-1 are adjacent individuals in the population, and F is the scaling factor.

[0102] Combining the above two strategies, an improved whale optimization algorithm (IWOA) is formed. Based on WOA, IWOA improves the global search capability and convergence speed of the algorithm by alternately using the adversarial learning strategy and the differential sorting mutation strategy in the iterative process, thereby better serving the optimization needs of the battery management system.

[0103] Step 5: Then input the feature vector extracted by PCNN into Figure 2 The deep extended learning shown is used for training in a periodic time series prediction model (DEPTS), and the IWOA algorithm is used to optimize the parameters of the deep extended learning for periodic time series prediction model. The deep extended learning for periodic time series prediction model after training optimization is used to predict the health status of the lithium battery, and the health status of the lithium battery is evaluated based on the prediction results.

[0104] In the lithium battery health status monitoring method, PCNN is used to extract feature vectors from the battery's operating data, which can represent the battery's current state and behavior pattern. These feature vectors are then used as input to the DEPTS model to predict the battery's future health status.

[0105] The DEPTS model framework consists of two main modules: the extension module and the periodic module. The extension module is responsible for handling the complex dependencies of time series data, while the periodic module focuses on capturing the periodic characteristics of time series.

[0106] Periodicity characteristics t is parameterized by the periodic function g φ (t), which is composed of a series of cosine functions:

[0107]

[0108] Among them, A0 is the baseline term, which represents the average value of the periodic pattern; A k is the amplitude, indicating the intensity of the kth periodic component; F k is the frequency, indicating the periodicity of the kth periodic component; P k is the phase, indicating the starting point of the kth periodic component; K is the total number of periodic components.

[0109] Extension Module θ Responsible for the observation value X t-L:t and the periodic characteristics z t-L:t+H Make predictions:

[0110] X t:t+H =f θ (X t-L:t , z t-L:t+H )+∈ t:t+H

[0111] Among them, X t:t+H represents the predicted value sequence of the health status of the lithium battery from the current time step t to the next H time steps, including key data such as battery voltage, current, temperature, surface pressure and current capacity value; f θ represents the function in the DEPTS model, which is a parameterized function used to predict the future health status of lithium batteries based on input characteristics and periodic characteristics; X tL:t It represents the historical lithium battery health status feature vector sequence from tL to t. tL:t+H represents the periodic characteristic sequence from tL to t+H; ∈ t:t+H represents the prediction error term, which is assumed to be independent and identically distributed Gaussian noise.

[0112] Battery health status (SOH) assessment is a complex process involving multiple battery parameters and historical data. In order to simplify the health status assessment of lithium batteries, it is necessary to first construct a correlation formula to link the voltage, current, temperature, surface pressure of lithium batteries with battery capacity decay (ΔC). The correlation formula is as follows:

[0113] ΔC=k V ·V t:t+H +k I I t:t+H +k T ·T t:t+H +k P ·P t:t+H

[0114] Among them, ΔC represents the battery capacity decay, that is, the battery from the initial state C0 to the current capacity state C tChange (C0-C t );V t:t+H ,I t:t+H 、T t:t+H , P t:t+H are the predicted values ​​of voltage, current, temperature and surface pressure respectively; k V , k I , k T , k P They are the voltage coefficient, current coefficient, temperature coefficient and surface pressure coefficient.

[0115] Therefore, the battery state of health (SOH) evaluation formula can be defined based on the battery capacity decay. The SOH can be calculated by the following formula, which represents the remaining percentage of the battery capacity relative to the initial capacity:

[0116]

[0117] The IWOA algorithm is used to optimize the parameters of the deep extended learning model for periodic time series prediction. The parameters of the model optimized here are the weights and biases of the expansion module.

[0118] First, define the parameter space of the DEPTS model. These parameters include the weights and biases of the expansion module and the periodic module. Initialize the whale group. In IWOA, each whale represents a set of parameters of the DEPTS model, and randomly initialize the position of the whale group.

[0119] After that, the fitness of each whale (parameter set) is evaluated. The fitness function can be defined as a function of SOH, namely:

[0120] Fitness(θ)=SOH(θ)

[0121] Where θ represents the parameter set represented by the whale. For each set of parameters θ, the DEPTS model is used to predict the current capacity C of the battery. t , to calculate SOH.

[0122] Determine the best whale (optimal parameter set) in the current group, and IWOA updates the position of the whale by the following formula:

[0123] θ i (d+1)=θ i (d)+A·C·(θ best -θ i (d))+D·X

[0124] Among them, θ i (t) is the position of the ith whale at the tth iteration, θ best is the optimal solution currently found, A and C are acceleration coefficients, D is the step size factor, and X is a random vector.

[0125] The adversarial learning strategy enhances the search capability of the whale optimization algorithm by generating an adversarial solution for each solution, that is, a solution that is equidistant from the original solution in the search space but in the opposite direction. The difference sorting mutation strategy is used to further enhance the diversity of the population, which calculates the differences between individuals in the population, sorts the individuals according to these differences, and then applies the mutation operation.

[0126] The optimal parameter θ found by the IWOA algorithm best , updated to the DEPTS model. The performance of the DEPTS model is evaluated on the test set using the updated parameters, that is, the deep extended learning after training optimization is used for the periodic time series prediction model to predict the health status of the lithium battery, and the health status of the lithium battery is evaluated according to the SOH value predicted by the DEPTS model. The closer the SOH value is to 100%, the closer the battery is to its initial capacity and the better the health status; if the SOH value is lower, it indicates that the current capacity of the lithium battery has dropped significantly compared to its initial capacity, which is a direct indicator of battery aging and performance degradation. The monitoring of SOH value can be used for predictive maintenance and provide a scientific basis for battery health management.

[0127] Through the above steps, the IWOA algorithm can effectively optimize the parameters of the DEPTS model and improve the accuracy of periodic time series prediction. The introduction of the IWOA algorithm, especially the adversarial learning strategy and the differential sorting mutation strategy, provides a new optimization method for the DEPTS model, which helps to solve complex time series prediction problems.

[0128] The generative adversarial network (GAN) of this embodiment is introduced into the interpolation of time series data. The generative adversarial network (ImputeGAN) for multivariate time series interpolation can effectively complete the missing data and improve the accuracy and efficiency of interpolation through adversarial training of the generator and the discriminator. This method can not only deal with the problem of missing data, but also maintain the continuity and stability of time series data, providing a high-quality data foundation for subsequent feature extraction and model training.

[0129] The application of deep learning in battery health monitoring, especially convolutional neural networks (CNNs), has shown strong feature extraction capabilities in processing multi-dimensional time series data. Parallel convolutional neural networks (PCNNs), as a special type of CNN, can process multiple input channels in parallel, which is an important feature for lithium battery health monitoring systems because it can simultaneously process data from different sensors, such as voltage, current, temperature, etc. PCNN extracts features from each channel through convolutional layers and fuses these features through pooling layers to form feature vectors that can represent the battery state. These feature vectors are then used to train the Deep Extended Learning for Periodic Time Series Prediction Model (DEPTS), which can not only capture the complex dependencies of time series data, but also identify periodic features, thereby accurately predicting the future health state of lithium batteries. The prediction results of the deep learning model can be used to evaluate the health state of the battery, providing a scientific basis for battery health management and maintenance.

[0130] Based on the same inventive concept, this embodiment also provides a lithium battery health status monitoring system, including:

[0131] An initialization module is used to obtain raw data of the battery by real-time monitoring using sensors, preprocess the raw data to obtain multi-dimensional time series data, and divide the data into a training set and a test set; the raw data includes voltage, current, temperature, surface pressure and current capacity;

[0132] The optimization module is used to optimize multidimensional time series data using ImputeGAN, a generative adversarial network for multidimensional time series interpolation, to improve the accuracy and efficiency of multidimensional time series data interpolation;

[0133] An extraction module, used for extracting the feature vector of the optimized multi-dimensional time series data using a parallel convolutional neural network PCNN;

[0134] The improvement module is used to improve the whale optimization algorithm WOA algorithm by using the adversarial learning strategy and the differential sorting mutation strategy to obtain the IWOA algorithm;

[0135] The monitoring module is used to input the feature vector extracted by PCNN into the deep extended learning for training the periodic time series prediction model DEPTS, and optimize the parameters of DEPTS using the IWOA algorithm, and use the trained and optimized DEPTS to monitor the health status of the lithium battery.

[0136] Based on the same inventive concept, this embodiment also provides a computing device, including: one or more processors, one or more memories, and one or more programs, wherein the programs are stored in the memories and configured to be executed by the processors, and when the programs are loaded into the processors, the steps of the lithium battery health status monitoring method described in any one of the above items are implemented.

[0137] Based on the same inventive concept, this embodiment also provides a storage medium, which stores a computer program, and the computer program includes program instructions. When the program instructions are executed by a processor, the processor executes the steps of the lithium battery health status monitoring method described in any one of the above items.

[0138] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with the technology to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent transformation or modification made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for monitoring the health status of a lithium battery, characterized in that: include: (1) using sensors for real-time monitoring to obtain raw data of the battery, preprocessing the raw data to obtain multi-dimensional time series data, and dividing the data into a training set and a test set; the raw data includes voltage, current, temperature, surface pressure and current capacity; (2) Using ImputeGAN, a generative adversarial network for multidimensional time series interpolation, to optimize multidimensional time series data and improve the accuracy and efficiency of multidimensional time series data interpolation; (3) Using parallel convolutional neural network (PCNN) to extract the feature vector of optimized multidimensional time series data; (4) The adversarial learning strategy and differential sorting mutation strategy are used to improve the whale optimization algorithm WOA algorithm to obtain the IWOA algorithm; (5) The feature vector extracted by PCNN is input into deep extended learning for training the periodic time series prediction model DEPTS. At the same time, the IWOA algorithm is used to optimize the parameters of DEPTS, and the trained and optimized DEPTS is used to monitor the health status of lithium batteries.

2. The lithium battery health status monitoring method according to claim 1, characterized in that: The preprocessing of the raw data in step (1) includes: cleaning the data, removing noise and inconsistency in the data, and detecting and processing outliers.

3. The lithium battery health status monitoring method according to claim 1, characterized in that: In the step (2), the loss function of ImputeGAN includes a generator loss function and a discriminator loss function; The loss function of the generator is expressed as: L G =-E z～pz(z) [log(D(G(z)))] Among them, L G represents the loss function of the generator; E represents the expected value, which represents the average value of a random variable; z represents the noise vector sampled from the prior distribution, which is the input of the generator; p z (z) represents the probability distribution of the noise vector z; G(z) represents the function of the generator, which is used to receive the noise vector z and generate data; D(G(z)) represents the judgment of the discriminator on the false data generated by the generator, and outputs a probability value to represent the probability that the false data is real data; The loss function of the discriminator is used to maximize the ability to distinguish between real data and generated data, expressed as: L D =-E x～pdata(x) [log(D(x))]-E z～pz(z) [log(1-D(G(z)))] Among them, L D represents the loss function of the discriminator; E represents the expected value, which represents the average value of a random variable; x represents the distribution of real data p data (x) is the sampled data point; p data (x) is the probability distribution of real data; D(x) is the output of the discriminator for input x, indicating the probability that x is real data.

4. The lithium battery health status monitoring method according to claim 1, characterized in that: The step (3) comprises: The convolutional layer of the PCNN is represented as: f θ (x)=ReLU(W*x+b) Among them, f θ (x) is the output after being processed by the ReLU activation function; W is the convolution kernel weight; b is the bias term; * represents the convolution operation; ReLU is the activation function; x represents a multidimensional time series data point; The feature maps of different channels after the convolution layer are fused, and the fusion operation is expressed as: P fusion =Pool({f θ1 (x1),f θ1 (x2),…,f θn (x n )}) Among them, P fusion represents the fused feature representation, which is a feature vector; Pool represents the pooling operation; n is the number of data points, x n is the nth data point.

5. The lithium battery health status monitoring method according to claim 1, characterized in that: The step (4) comprises: The adversarial learning strategy is introduced to enhance the search capability of the whale optimization algorithm WOA. The adversarial learning strategy generates an adversarial solution for each solution. The adversarial solution is a solution that is equidistant from the original solution in the search space but in the opposite direction. For a solution y, its adversarial solution y′ can be calculated by the following formula: y′=L+Uy Among them, L and U are the lower and upper bounds of the search space respectively; The differential sorting mutation strategy is used to enhance the diversity of the population. The differences between individuals in the population are calculated, and the individuals are sorted according to these differences, and the mutation operation is applied. The differential sorting mutation is expressed as: in i =in i +F·(in i+1 -in i-1 ) Among them, u i is the current individual, u i+1 and u i-1 are adjacent individuals in the population, and F is the scaling factor.

6. The lithium battery health status monitoring method according to claim 1, characterized in that: Step (5) of inputting the feature vector extracted by PCNN into deep extended learning for training in the periodic time series prediction model DEPTS includes: The DEPTS includes an extension module and a periodic module; the extension module is responsible for processing the complex dependencies of time series data, and the periodic module is used to capture the periodic characteristics of time series; Among them, the periodic characteristic z t By parameterizing the periodic function g φ (t) means: Among them, A0 is the baseline term, which represents the average value of the periodic pattern; A k is the amplitude, indicating the intensity of the kth periodic component; F k is the frequency, indicating the periodicity of the kth periodic component; P k is the phase, indicating the starting point of the kth periodic component; K is the total number of periodic components; The expansion module f θ Responsible for the observation value X t-L:t and the periodic characteristics z t-L:t+H Make predictions: X t:t+H =f θ (X t-L:t ,z t-L:t+H )+∈ t:t+H Among them, X t:t+H represents the predicted value sequence of the health status of the lithium battery from the current time step t to the future H time steps, including the battery voltage, current, temperature, surface pressure and current capacity data; f θ Represents the function in the DEPTS model, which is a parameterized function used to predict the future health status of lithium batteries based on input characteristics and periodic characteristics; X t-L:t represents the historical lithium battery health status feature vector sequence from tL to t; z t-L:t+H Represents the periodic characteristic sequence from tL to t+H; ∈ t:t+H It represents the prediction error term, which is assumed to be independent and identically distributed Gaussian noise; Construct a correlation formula to correlate the voltage, current, temperature, surface pressure of the lithium battery with the battery capacity decay ΔC: ΔC=k V ·V t:t+H +k I ·I t:t+H +k T ·T t:t+H +k P ·P t:t+H Among them, ΔC represents the battery capacity decay, and the current state C is obtained by combining the initial state C0 of the battery t ; V t:t+H ,I t:t+H , T t:t+H and P t:t+H are the predicted values ​​of voltage, current, temperature and surface pressure respectively; k V , k I , k T and k P They are the voltage coefficient, current coefficient, temperature coefficient and surface pressure coefficient; The battery's state of health (SOH) is defined based on the attenuation of battery capacity: Used to indicate the remaining percentage of battery capacity relative to the initial capacity.

7. The lithium battery health status monitoring method according to claim 5, characterized in that: In step (5), the parameters of DEPTS are optimized by using the IWOA algorithm, and the health status of the lithium battery is monitored by using the trained and optimized DEPTS, including: Define the parameter space of the DEPTS model, where the parameters in the parameter space include the weights and biases of the dilation module and the periodic module; initialize the whale population. In IWOA, each whale represents a set of parameters of the DEPTS model, and randomly initialize the position of the whale population; Evaluate the fitness of each whale and define the fitness function as a function of SOH: Fitness(θ)=SOH(θ) Where θ represents the parameter set represented by the whale. For each set of parameters θ, the DEPTS model is used to predict the current capacity C of the battery. t , to calculate SOH; Determining the best whale in the current group, IWOA updates the whale's position by the following formula: i i (d+1)=θ i (d)+A·C·(θ best -θ i (d))+D·X Among them, θ i (d) is the position of the ith whale at the dth iteration, θ best is the optimal solution found so far, A and C are acceleration coefficients, D is the step size factor, and X is a random vector; The optimal parameter θ found by the IWOA algorithm best , updated to the DEPTS model, and used the updated parameters to evaluate the performance of the DEPTS model on the test set. The deep extended learning after training optimization was used for the periodic time series prediction model to monitor the health status of the lithium battery. The health status of the lithium battery was evaluated according to the SOH value monitored by the DEPTS model. The closer the SOH value is to 100%, the closer the battery is to the initial capacity and the better the health status; if the SOH value is lower, it indicates that the current capacity of the lithium battery has dropped significantly compared with the initial capacity.

8. A lithium battery health status monitoring system, characterized in that: include: An initialization module is used to obtain raw data of the battery by real-time monitoring using sensors, preprocess the raw data to obtain multi-dimensional time series data, and divide the data into a training set and a test set; the raw data includes voltage, current, temperature, surface pressure and current capacity; The optimization module is used to optimize multidimensional time series data using ImputeGAN, a generative adversarial network for multidimensional time series interpolation, to improve the accuracy and efficiency of multidimensional time series data interpolation; An extraction module, used for extracting the feature vector of the optimized multi-dimensional time series data using a parallel convolutional neural network PCNN; The improvement module is used to improve the whale optimization algorithm WOA algorithm by using the adversarial learning strategy and the differential sorting mutation strategy to obtain the IWOA algorithm; The monitoring module is used to input the feature vector extracted by PCNN into the deep extended learning for training the periodic time series prediction model DEPTS, and optimize the parameters of DEPTS using the IWOA algorithm, and use the trained and optimized DEPTS to monitor the health status of the lithium battery.

9. A computing device, characterized in that include: One or more processors, one or more memories, and one or more programs, wherein the programs are stored in the memories and configured to be executed by the processors, and when the programs are loaded into the processors, the steps of the lithium battery health status monitoring method according to any one of claims 1 to 7 are implemented.

10. A storage medium, characterized in that: The storage medium stores a computer program, which includes program instructions. When the program instructions are executed by a processor, the processor executes the steps of the lithium battery health status monitoring method according to any one of claims 1 to 7.

Images