Electric vehicle battery state detection method and system based on deep learning

Abstract

The invention relates to the technical field of battery state detection, in particular to an electric vehicle battery state detection method and system based on deep learning, and provides the following scheme: firstly, carrying out complementary energy test on an electric vehicle battery, and calculating the state of charge (SOC) of the electric vehicle battery according to a complementary energy test result; the method comprises the following steps: firstly, acquiring an infrared video of a battery discharge process through an infrared imager, processing the acquired infrared video, extracting discharge contour features of an electric vehicle battery, finally training the discharge contour features of the electric vehicle battery through a battery health state recognition network, and outputting the SOH of the electric vehicle battery.

Description

Technical Field

[0001] The present invention relates to the technical field of battery status detection, and in particular to a method and system for detecting the battery status of an electric vehicle based on deep learning. Background Art

[0002] With the rapid growth of the electric vehicle market, the performance and maintenance of batteries, as one of the core components of electric vehicles, are crucial to the reliability and safety of the vehicle. Battery aging, capacity decay and other health problems may lead to performance degradation of electric vehicles and even safety accidents. Therefore, developing an efficient and accurate battery status detection method is of great significance to ensure the safe operation of electric vehicles and extend the battery life. Traditional battery status detection methods often rely on physical testing and chemical analysis, which are time-consuming, costly and may cause damage to the battery. In recent years, with the development of artificial intelligence and deep learning technology, new solutions have been provided for battery status detection. Deep learning can learn the patterns and trends of battery performance degradation through a large amount of historical data, and realize rapid and accurate assessment of battery health.

[0003] For example, a Chinese patent with the authorization announcement number CN114720878B discloses a state detection method for retired batteries, including: using the main program analysis method to analyze the test data of the battery termination state consistency under conventional charging conditions to obtain the characteristic value of the charging curve, clustering the characteristic value using the Canopy algorithm and the PSO-k-means classification algorithm to obtain the final classification result of the battery, using the classified cycle aging test data of the battery model, extracting the characteristic value of the working condition curve, using the characteristic value to establish a battery aging model based on the adaptive PSO-neural network capacity prediction model, using the characteristic value to characterize the attenuation of the battery capacity and internal resistance, extracting the characteristic value of the battery in the core part of the battery cluster, and using the established battery aging model to quickly estimate the actual capacity and impedance of the battery. The method can realize rapid classification and state detection of retired power batteries with unknown remaining power, remaining capacity and impedance status.

[0004] The above patents all have the problems raised by this background technology: the battery needs to be disassembled or carried out under laboratory conditions, which not only increases the detection cost, but may also cause damage to the battery itself, shorten its service life, and cannot achieve rapid detection and instant feedback. To solve the above problems, this application designs an electric vehicle battery status detection method and system based on deep learning. Summary of the invention

[0005] The technical problem to be solved by the present invention is to address the deficiencies in the prior art and provide an electric vehicle battery status detection method and system based on deep learning. First, the electric vehicle battery is subjected to a residual energy test, and the state of charge (SOC) of the electric vehicle battery is calculated based on the residual energy test result. Secondly, an infrared video of the battery discharge process is obtained through an infrared imager, and the obtained infrared video is processed to extract the discharge profile features of the electric vehicle battery. Finally, the discharge profile features of the electric vehicle battery are trained through a battery health status recognition network to output the health status (SOH) of the electric vehicle battery.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The electric vehicle battery status detection method based on deep learning is characterized by comprising the following steps:

[0008] S1: Perform a residual energy test on the electric vehicle battery, extract the capacity loss characteristic parameters, and calculate the state of charge SOC of the electric vehicle battery according to the capacity loss characteristic parameters;

[0009] S2: Discharge the electric vehicle battery at a 2C rate and use an infrared imager to collect infrared video of the first five minutes of the battery discharge process;

[0010] S3: Process the acquired infrared video to extract the discharge profile features of the electric vehicle battery;

[0011] S4: construct a battery health state recognition network, use the discharge profile characteristics of the electric vehicle battery as input parameters of the battery health state recognition network, train the input parameters through the battery health state recognition network, and output the health state SOH of the electric vehicle battery.

[0012] The residual energy test includes:

[0013] S1.1: At standard room temperature, discharge the electric vehicle battery with a current of 1 / 4C. When the electric vehicle battery voltage is equal to the discharge cut-off voltage, stop discharging, record the data and leave it alone.

[0014] S1.2: Under standard room temperature conditions, charge the electric vehicle battery with a current of 1 / 4C. When the electric vehicle battery voltage is equal to the charging cut-off voltage, switch to constant voltage charging and detect the charging current. When the charging current is equal to the charging cut-off current, stop charging, record the data and leave it alone.

[0015] S1.3: Under standard room temperature conditions, discharge the fully charged electric vehicle battery at a current of 1 / 4C. When the electric vehicle battery voltage is equal to the discharge cut-off voltage, stop discharging, record the data and leave it alone;

[0016] S1.4: Under standard room temperature conditions, charge the discharged electric vehicle battery with a current of 1 / 4C. When the electric vehicle battery voltage is equal to the charging cut-off voltage, stop charging, record the data and leave it alone.

[0017] S1.5: Perform curve fitting on the data recorded in S1.1 to S1.4 according to the pattern search nonlinear optimization algorithm, optimize the fitting results through the loss function, and output the capacity loss characteristic parameters. The calculation formula of the loss function is:

[0018]

[0019] Where L represents the loss function, i represents the measurement time point in the residual energy test, J represents the total measurement time point in the residual energy test, E{·} represents the annealing genetic function, and K b represents the Boltzmann constant, y model [·] represents the pattern search output function, U p represents the positive electrode potential, p o Indicates the initial lithium insertion rate of the positive electrode, l i represents the current at the i-th measurement time point, Q p Indicates the initial capacity of the positive electrode, U n represents the negative electrode potential, n o Indicates the initial lithium insertion rate of the negative electrode, Q n represents the initial capacity of the negative electrode, θ represents the model parameter matrix, u i represents the voltage at the i-th measurement time point.

[0020] The capacity loss characteristic parameters include negative electrode material activity loss parameters, internal resistance material activity loss parameters, positive electrode material activity loss parameters and lithium loss parameters. The calculation of the state of charge (SOC) of the electric vehicle battery based on the capacity loss characteristic parameters includes data preprocessing, SOC evaluation model parameter optimization and SOC evaluation model application.

[0021] The data preprocessing is used to preprocess the capacity loss characteristic parameters, divide the preprocessed data into a training set and a test set, and perform normalization processing. The SOC evaluation model parameter optimization optimizes the core parameters and penalty factors of the SOC evaluation model according to the training set through the moth optimization algorithm. The SOC evaluation model application is used to evaluate the test set according to the optimized core parameters and penalty factors, and output the state of charge SOC of the electric vehicle battery.

[0022] The specific steps of S3 are as follows:

[0023] S3.1: Collect infrared video of the first five minutes of the battery discharge process using an infrared imager, perform differential processing on the infrared video, perform denoising on the infrared video differential image, and extract the infrared video inter-frame image;

[0024] S3.2: According to the infrared video inter-frame image, a difference operation is performed on the grayscale values ​​of the corresponding pixels of the continuous infrared video inter-frame images, and the absolute value of the pixel difference is compared with the background threshold. If it is greater than or equal to the background threshold, the pixel is defined as the image foreground; if it is less than the background threshold, the pixel is defined as the image background and deleted, and the pixel of the image foreground is binarized to obtain the key frame binary image;

[0025] S3.3: Perform background subtraction on the key frame binary image after differential denoising, and enhance the key frame binary image through morphological opening operation. The calculation formula for key frame binary image enhancement is:

[0026]

[0027] Among them, EI key represents the enhanced key frame binary image, k represents a single Gaussian distribution of the key frame binary image in the mixed Gaussian model, K represents the total number of Gaussian distributions of the key frame binary image in the mixed Gaussian model, ω k represents the weight value of the kth Gaussian distribution, μ(·) represents the probability density function of the Gaussian distribution, X represents the key frame binary image, M X represents the Gaussian distribution mean of the key frame binary image, V X represents the Gaussian distribution variance of the key frame binary image, ⊙ represents the corrosion operation, represents the expansion operation;

[0028] S3.4: Extract the contour features of the key frame binary image through the edge detection algorithm, perform short-time Fourier transform on the contour features of the key frame binary image, calculate the time domain spectrum features of the contour features, perform logarithmic operation and size reconstruction on the time domain spectrum features, and obtain the discharge contour features of the electric vehicle battery.

[0029] The specific steps of performing denoising on the infrared video differential image are as follows:

[0030] S3.1.1: extract feature information from the infrared video difference image of the current frame and the next frame according to Gaussian difference filtering, and obtain a four-layer two-dimensional Gaussian difference filtering infrared video difference image;

[0031] S3.1.2: Segment the four-layer two-dimensional Gaussian difference filtering infrared video difference image into four directions, filter out high-frequency noise from the segmented four-layer two-dimensional Gaussian difference filtering infrared video difference image through spatial domain filtering, fuse the four-layer two-dimensional Gaussian difference filtering infrared video difference image after denoising in four directions, and obtain the four-layer two-dimensional Gaussian difference filtering infrared video difference image after high-frequency noise removal;

[0032] S3.1.3: Fuse the two-dimensional Gaussian difference filtered infrared video difference images after denoising at each layer to obtain the infrared video difference denoised image.

[0033] The battery health status recognition network includes an input layer, a feature fusion layer and an output layer;

[0034] The input layer is used to perform convolution on the input parameters, expand the plane dimension of the input parameters according to the bilinear difference, and convert the contour features of different levels and scales into a unified scale;

[0035] The feature fusion layer includes a fast branch, a slow branch and a 3DResNet50 fusion network, wherein the fast branch is used to extract temporal features from the input parameters of the feature fusion layer, and the slow branch is used to extract spatial features from the input parameters of the feature fusion layer. The 3DResNet50 fusion network is used to perform feature conversion processing on temporal features and spatial features, and autocorrelation weights are assigned to temporal features and spatial features through an attention mechanism, and temporal features and spatial features are fused according to the assigned weights to obtain fusion features;

[0036] The output layer includes an LSTM model, which processes the temperature change context information of the long time series in the fusion feature through the LSTM model, integrates the output through the fully connected Dropout, and outputs the health status SOH of the electric vehicle battery.

[0037] A power battery state detection system based on deep learning, the system comprising a charge state assessment module and a health state assessment module;

[0038] The state of charge evaluation module is used to perform a remaining energy test on the electric vehicle battery, extract the capacity loss characteristic parameters, and calculate the state of charge SOC of the electric vehicle battery according to the capacity loss characteristic parameters;

[0039] The health status assessment module is used to record the discharge process of the electric vehicle battery through an infrared imager, process the infrared video, extract the discharge profile features of the electric vehicle battery, train the discharge profile features of the electric vehicle battery according to the battery health status recognition network, and output the health status SOH of the electric vehicle battery;

[0040] The state of charge assessment module comprises:

[0041] The residual energy test unit is used to test the residual energy of electric vehicle batteries and extract characteristic parameters of capacity loss;

[0042] A state of charge calculation unit, used to calculate the state of charge SOC of the electric vehicle battery through the capacity loss characteristic parameter according to the SOC evaluation model;

[0043] The health status assessment module includes:

[0044] An infrared video acquisition unit, used for collecting infrared video of the first five minutes of the battery discharge process using an infrared imager;

[0045] An infrared video processing unit, used to process the acquired infrared video and extract the discharge profile features of the electric vehicle;

[0046] The health state calculation unit is used to construct a power battery health state recognition network, take the discharge profile characteristics of the retired power battery as input parameters of the power battery health state recognition network, train the input parameters through the power battery health state recognition network, and output the health state SOH of the retired power battery.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] 1. The present invention establishes the capacity loss characteristics that characterize the loss of active lithium and active materials through residual energy testing, combines the identified characteristic parameters with the support vector regression model, and realizes the rapid and accurate estimation of the residual energy of retired electric vehicle batteries;

[0049] 2. The present invention proposes a method for rapid detection of the health status of lithium-ion batteries based on infrared thermal imaging. According to the different temperature change patterns of different aged batteries during the discharge process, they are converted into differences in battery infrared videos. Rapid detection of the battery health status is achieved through video recognition, which greatly reduces the cost of lithium battery recycling. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0051] Figure 1 This is a flow chart of a method for detecting the battery status of an electric vehicle based on deep learning according to Embodiment 1 of the present invention;

[0052] Figure 2 This is a schematic diagram of a method for estimating the state of charge of a retired power battery according to Embodiment 1 of the present invention;

[0053] Figure 3 This is a schematic diagram of the infrared video differential image denoising steps of Example 1 of the present invention;

[0054] Figure 4 This is a module diagram of a power battery status detection system based on deep learning according to Example 2 of the present invention. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0056] Embodiment 1:

[0057] See also Figure 1 , an embodiment provided by the present invention: a method for detecting the battery status of an electric vehicle based on deep learning, comprising the following steps;

[0058] S1: Perform a residual energy test on the electric vehicle battery, extract the capacity loss characteristic parameters, and calculate the state of charge SOC of the electric vehicle battery according to the capacity loss characteristic parameters;

[0059] S2: Discharge the electric vehicle battery at a 2C rate and use an infrared imager to collect infrared video of the first five minutes of the battery discharge process;

[0060] S3: Process the acquired infrared video to extract the discharge profile features of the electric vehicle battery;

[0061] S4: construct a battery health state recognition network, use the discharge profile characteristics of the electric vehicle battery as input parameters of the battery health state recognition network, train the input parameters through the battery health state recognition network, and output the health state SOH of the electric vehicle battery.

[0062] The residual energy test includes:

[0063] S1.1: At standard room temperature, discharge the electric vehicle battery with a current of 1 / 4C. When the electric vehicle battery voltage is equal to the discharge cut-off voltage, stop discharging, record the data and leave it alone.

[0064] S1.2: Under standard room temperature conditions, charge the electric vehicle battery with a current of 1 / 4C. When the electric vehicle battery voltage is equal to the charging cut-off voltage, switch to constant voltage charging and detect the charging current. When the charging current is equal to the charging cut-off current, stop charging, record the data and leave it alone.

[0065] S1.3: Under standard room temperature conditions, discharge the fully charged electric vehicle battery at a current of 1 / 4C. When the electric vehicle battery voltage is equal to the discharge cut-off voltage, stop discharging, record the data and leave it alone;

[0066] S1.4: Under standard room temperature conditions, charge the discharged electric vehicle battery with a current of 1 / 4C. When the electric vehicle battery voltage is equal to the charging cut-off voltage, stop charging, record the data and leave it alone.

[0067] S1.5: Perform curve fitting on the data recorded in S1.1 to S1.4 according to the pattern search nonlinear optimization algorithm, optimize the fitting results through the loss function, and output the capacity loss characteristic parameters. The calculation formula of the loss function is:

[0068]

[0069] Where L represents the loss function, i represents the measurement time point in the residual energy test, J represents the total measurement time point in the residual energy test, E{·} represents the annealing genetic function, and K b represents the Boltzmann constant, y model [·] represents the pattern search output function, U p represents the positive electrode potential, p o Indicates the initial lithium insertion rate of the positive electrode, l i represents the current at the i-th measurement time point, Q p Indicates the initial capacity of the positive electrode, U n represents the negative electrode potential, n o Indicates the initial lithium insertion rate of the negative electrode, Q n represents the initial capacity of the negative electrode, θ represents the model parameter matrix, u i represents the voltage at the i-th measurement time point;

[0070] The specific steps of the pattern search nonlinear optimization algorithm are as follows:

[0071] S1.5.1: Draw a charge and discharge charge variation curve based on the data recorded in S1.1 to S1.4, and determine the initial model parameter matrix based on the curvature of the curve;

[0072] S1.5.2: Substitute the initial model parameter matrix into the calculation formula of the loss function to calculate the initial loss function;

[0073] S1.5.3: Optimize the parameters through the direction matrix D = [d1, d2, d3, d4], calculate the loss function error value for each of the four directions, and if the minimum error value of the four directions is less than the minimum error value in the previous iteration, update the initial model parameter matrix according to the direction matrix;

[0074] S1.5.4: Repeat steps S1.5.2 to S1.5.3 until the minimum error in the three directions is greater than the error value of the previous iterative process, and the optimization process ends;

[0075] The capacity loss characteristic parameters include negative electrode material activity loss parameters, internal resistance material activity loss parameters, positive electrode material activity loss parameters and lithium loss parameters. The calculation of the state of charge (SOC) of the retired power battery based on the capacity loss characteristic parameters includes data preprocessing, SOC evaluation model parameter optimization and SOC evaluation model application.

[0076] The data preprocessing is used to preprocess the capacity loss characteristic parameters, divide the preprocessed data into a training set and a test set, and perform normalization processing; the SOC evaluation model parameter optimization optimizes the core parameters and penalty factors of the SOC evaluation model according to the training set through the moth optimization algorithm; the SOC evaluation model application is used to evaluate the test set according to the optimized core parameters and penalty factors, and output the state of charge SOC of the retired power battery;

[0077] See also Figure 2 , a schematic diagram of a method for estimating the state of charge of a retired power battery according to an embodiment of the present invention, firstly, the capacity loss characteristic parameters of N groups of retired power batteries are obtained through residual energy testing, and the capacity loss characteristic parameters of the N groups of retired power batteries are used as training sets of the SOC evaluation model, and then the kernel parameters and penalty factors of the SOC evaluation model are optimized according to the training set through the moth optimization algorithm to obtain the optimal kernel parameters and penalty factors, wherein the moth optimization algorithm is a heuristic intelligent optimization algorithm based on a population, and finally, the test set is evaluated according to the optimal kernel parameters and penalty factors, and the state of charge SOC of the retired power battery is output.

[0078] The specific steps of S3 are as follows:

[0079] S3.1: Collect infrared video of the first five minutes of the battery discharge process using an infrared imager, perform differential processing on the infrared video, perform denoising on the infrared video differential image, and extract the infrared video inter-frame image;

[0080] S3.2: According to the infrared video inter-frame image, a difference operation is performed on the grayscale values ​​of the corresponding pixels of the continuous infrared video inter-frame images, and the absolute value of the pixel difference is compared with the background threshold. If it is greater than or equal to the background threshold, the pixel is defined as the image foreground; if it is less than the background threshold, the pixel is defined as the image background and deleted, and the pixel of the image foreground is binarized to obtain the key frame binary image;

[0081] S3.3: Perform background subtraction on the key frame binary image after differential denoising, and enhance the key frame binary image through morphological opening operation. The calculation formula for key frame binary image enhancement is:

[0082]

[0083] Among them, EI key represents the enhanced key frame binary image, k represents a single Gaussian distribution of the key frame binary image in the mixed Gaussian model, K represents the total number of Gaussian distributions of the key frame binary image in the mixed Gaussian model, ω krepresents the weight value of the kth Gaussian distribution, μ(·) represents the probability density function of the Gaussian distribution, X represents the key frame binary image, M X represents the Gaussian distribution mean of the key frame binary image, V X represents the Gaussian distribution variance of the key frame binary image, ⊙ represents the corrosion operation, represents the expansion operation;

[0084] S3.4: Extract the contour features of the key frame binary image through the edge detection algorithm, perform short-time Fourier transform on the contour features of the key frame binary image, calculate the time domain spectrum features of the contour features, perform logarithmic operation and size reconstruction on the time domain spectrum features, and obtain the discharge contour features of the retired power battery.

[0085] See also Figure 3 , a schematic diagram of the denoising steps of the infrared video differential image according to an embodiment of the present invention, wherein the specific steps of denoising the infrared video differential image are as follows:

[0086] S3.1.1: extract feature information from the infrared video difference image of the current frame and the next frame according to Gaussian difference filtering, and obtain a four-layer two-dimensional Gaussian difference filtering infrared video difference image;

[0087] S3.1.2: Segment the four-layer two-dimensional Gaussian difference filtering infrared video difference image into four directions, filter out high-frequency noise from the segmented four-layer two-dimensional Gaussian difference filtering infrared video difference image through spatial domain filtering, fuse the four-layer two-dimensional Gaussian difference filtering infrared video difference image after denoising in four directions, and obtain the four-layer two-dimensional Gaussian difference filtering infrared video difference image after high-frequency noise removal;

[0088] S3.1.3: Fuse the two-dimensional Gaussian difference filtered infrared video difference images after denoising at each layer to obtain the infrared video difference denoised image.

[0089] The battery health status recognition network includes an input layer, a feature fusion layer and an output layer;

[0090] The input layer is used to perform convolution on the input parameters, expand the plane dimension of the input parameters according to the bilinear difference, and convert the contour features of different levels and scales into a unified scale;

[0091] The feature fusion layer includes a fast branch, a slow branch and a 3DResNet50 fusion network, wherein the fast branch is used to extract temporal features from the input parameters of the feature fusion layer, and the slow branch is used to extract spatial features from the input parameters of the feature fusion layer. The 3DResNet50 fusion network is used to perform feature conversion processing on temporal features and spatial features, and autocorrelation weights are assigned to temporal features and spatial features through an attention mechanism, and temporal features and spatial features are fused according to the assigned weights to obtain fusion features;

[0092] The output layer includes an LSTM model, which processes the temperature change context information of the long time series in the fusion feature through the LSTM model, integrates the output through the full connection Dropout, and outputs the health state SOH of the retired power battery.

[0093] Embodiment 2:

[0094] See also Figure 4 , the present invention provides an embodiment: a power battery state detection system based on deep learning, the system comprising a charge state assessment module and a health state assessment module;

[0095] The state of charge evaluation module is used to perform a remaining energy test on the retired power battery, extract the capacity loss characteristic parameters, and calculate the state of charge SOC of the retired power battery according to the capacity loss characteristic parameters;

[0096] The health status assessment module is used to record the discharge process of the retired power battery through an infrared imager, process the infrared video, extract the discharge profile features of the retired power battery, train the discharge profile features of the retired power battery according to the power battery health status recognition network, and output the health status SOH of the retired power battery;

[0097] The state of charge assessment module comprises:

[0098] The residual energy test unit is used to test the residual energy of electric vehicle batteries and extract characteristic parameters of capacity loss;

[0099] A state of charge calculation unit, used to calculate the state of charge SOC of the electric vehicle battery through the capacity loss characteristic parameter according to the SOC evaluation model;

[0100] The health status assessment module includes:

[0101] An infrared video acquisition unit, used for collecting infrared video of the first five minutes of the battery discharge process using an infrared imager;

[0102] An infrared video processing unit, used to process the acquired infrared video and extract the discharge profile features of the electric vehicle;

[0103] The health state calculation unit is used to construct a power battery health state recognition network, take the discharge profile characteristics of the retired power battery as input parameters of the power battery health state recognition network, train the input parameters through the power battery health state recognition network, and output the health state SOH of the retired power battery.

[0104] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. An electric vehicle battery status detection method based on deep learning, characterized in that: The steps include: S1: Perform a residual energy test on the electric vehicle battery, extract the capacity loss characteristic parameters, and calculate the state of charge SOC of the electric vehicle battery according to the capacity loss characteristic parameters; S2: Discharge the electric vehicle battery at a 2C rate and use an infrared imager to collect infrared video of the first five minutes of the battery discharge process; S3: Process the acquired infrared video to extract the discharge profile features of the electric vehicle battery; S4: construct a battery health state recognition network, use the discharge profile characteristics of the electric vehicle battery as input parameters of the battery health state recognition network, train the input parameters through the battery health state recognition network, and output the health state SOH of the electric vehicle battery.

2. The electric vehicle battery status detection method based on deep learning according to claim 1 is characterized in that: The residual energy test includes: S1.1: At standard room temperature, discharge the electric vehicle battery with a current of 1 / 4C. When the electric vehicle battery voltage is equal to the discharge cut-off voltage, stop discharging, record the data and leave it alone. S1.2: Under standard room temperature conditions, charge the electric vehicle battery with a current of 1 / 4C. When the electric vehicle battery voltage is equal to the charging cut-off voltage, switch to constant voltage charging and detect the charging current. When the charging current is equal to the charging cut-off current, stop charging, record the data and leave it alone. S1.3: Under standard room temperature conditions, discharge the fully charged electric vehicle battery at a current of 1 / 4C. When the electric vehicle battery voltage is equal to the discharge cut-off voltage, stop discharging, record the data and leave it alone; S1.4: Under standard room temperature conditions, charge the discharged electric vehicle battery with a current of 1 / 4C. When the electric vehicle battery voltage is equal to the charging cut-off voltage, stop charging, record the data and leave it alone. S1.5: Perform curve fitting on the data recorded in S1.1 to S1.4 according to the pattern search nonlinear optimization algorithm, optimize the fitting results through the loss function, and output the capacity loss characteristic parameters. The calculation formula of the loss function is: Where L represents the loss function, i represents the measurement time point in the residual energy test, J represents the total measurement time point in the residual energy test, E{·} represents the annealing genetic function, and K b represents the Boltzmann constant, y model [·] represents the pattern search output function, U p represents the positive electrode potential, p o Indicates the initial lithium insertion rate of the positive electrode, l i represents the current at the i-th measurement time point, Q p Indicates the initial capacity of the positive electrode, U n represents the negative electrode potential, n o Indicates the initial lithium insertion rate of the negative electrode, Q n represents the initial capacity of the negative electrode, θ represents the model parameter matrix, u i represents the voltage at the i-th measurement time point.

3. The electric vehicle battery status detection method based on deep learning according to claim 2 is characterized in that: The capacity loss characteristic parameters include negative electrode material activity loss parameters, internal resistance material activity loss parameters, positive electrode material activity loss parameters and lithium loss parameters. The calculation of the state of charge (SOC) of the electric vehicle battery based on the capacity loss characteristic parameters includes data preprocessing, SOC evaluation model parameter optimization and SOC evaluation model application.

4. The electric vehicle battery status detection method based on deep learning according to claim 3 is characterized in that: The data preprocessing is used to preprocess the capacity loss characteristic parameters, divide the preprocessed data into a training set and a test set, and perform normalization processing. The SOC evaluation model parameter optimization optimizes the core parameters and penalty factors of the SOC evaluation model according to the training set through the moth optimization algorithm. The SOC evaluation model application is used to evaluate the test set according to the optimized core parameters and penalty factors, and output the state of charge SOC of the electric vehicle battery.

5. The electric vehicle battery status detection method based on deep learning according to claim 4 is characterized in that: The specific steps of S3 are as follows: S3.1: Collect infrared video of the first five minutes of the battery discharge process using an infrared imager, perform differential processing on the infrared video, perform denoising on the infrared video differential image, and extract the infrared video inter-frame image; S3.2: According to the infrared video inter-frame image, a difference operation is performed on the grayscale values ​​of the corresponding pixels of the continuous infrared video inter-frame images, and the absolute value of the pixel difference is compared with the background threshold. If it is greater than or equal to the background threshold, the pixel is defined as the image foreground; if it is less than the background threshold, the pixel is defined as the image background and deleted, and the pixel of the image foreground is binarized to obtain the key frame binary image; S3.3: Perform background subtraction on the key frame binary image after differential denoising, and enhance the key frame binary image through morphological opening operation. The calculation formula for key frame binary image enhancement is: Among them, EI key represents the enhanced key frame binary image, k represents a single Gaussian distribution of the key frame binary image in the mixed Gaussian model, K represents the total number of Gaussian distributions of the key frame binary image in the mixed Gaussian model, ω k represents the weight value of the kth Gaussian distribution, μ(·) represents the probability density function of the Gaussian distribution, X represents the key frame binary image, M X represents the Gaussian distribution mean of the key frame binary image, V X represents the Gaussian distribution variance of the key frame binary image, ⊙ represents the corrosion operation, represents the expansion operation; S3.4: Extract the contour features of the key frame binary image through the edge detection algorithm, perform short-time Fourier transform on the contour features of the key frame binary image, calculate the time domain spectrum features of the contour features, perform logarithmic operation and size reconstruction on the time domain spectrum features, and obtain the discharge contour features of the electric vehicle battery.

6. The electric vehicle battery status detection method based on deep learning according to claim 5 is characterized in that: The specific steps of performing denoising on the infrared video differential image are as follows: S3.1.1: extract feature information from the infrared video difference image of the current frame and the next frame according to Gaussian difference filtering, and obtain a four-layer two-dimensional Gaussian difference filtering infrared video difference image; S3.1.2: Segment the four-layer two-dimensional Gaussian difference filtering infrared video difference image into four directions, filter out high-frequency noise from the segmented four-layer two-dimensional Gaussian difference filtering infrared video difference image through spatial domain filtering, fuse the four-layer two-dimensional Gaussian difference filtering infrared video difference image after denoising in four directions, and obtain the four-layer two-dimensional Gaussian difference filtering infrared video difference image after high-frequency noise removal; S3.1.3: Fuse the two-dimensional Gaussian difference filtered infrared video difference images after denoising at each layer to obtain the infrared video difference denoised image.

7. The electric vehicle battery status detection method based on deep learning according to claim 6 is characterized in that: The battery health status recognition network includes an input layer, a feature fusion layer and an output layer; The input layer is used to perform convolution on the input parameters, expand the plane dimension of the input parameters according to the bilinear difference, and convert the contour features of different levels and scales into a unified scale; The feature fusion layer includes a fast branch, a slow branch and a 3DResNet50 fusion network, wherein the fast branch is used to extract temporal features from the input parameters of the feature fusion layer, and the slow branch is used to extract spatial features from the input parameters of the feature fusion layer. The 3DResNet50 fusion network is used to perform feature conversion processing on temporal features and spatial features, and autocorrelation weights are assigned to temporal features and spatial features through an attention mechanism, and temporal features and spatial features are fused according to the assigned weights to obtain fusion features; The output layer includes an LSTM model, which processes the temperature change contextual information of the long time series in the fusion feature through the LSTM model, integrates the output through the fully connected Dropout, and outputs the health status SOH of the electric vehicle battery.

8. A power battery status detection system based on deep learning, used to implement the electric vehicle battery status detection method based on deep learning as described in any one of claims 1 to 7, characterized in that: The system includes a state of charge assessment module and a health state assessment module; The state of charge evaluation module is used to perform a remaining energy test on the electric vehicle battery, extract the capacity loss characteristic parameters, and calculate the state of charge SOC of the electric vehicle battery according to the capacity loss characteristic parameters; The health status assessment module is used to record the discharge process of the electric vehicle battery through an infrared imager, process the infrared video, extract the discharge profile features of the electric vehicle battery, train the discharge profile features of the electric vehicle battery according to the battery health status recognition network, and output the health status SOH of the electric vehicle battery.

Images