Electric vehicle battery health state prediction method and system based on data driving

Abstract

The invention provides an electric vehicle battery health state prediction method and system based on data driving. The method comprises the following steps: acquiring electric vehicle driving data and battery circulation data to extract a battery charging and discharging segment to obtain a first charging and discharging segment; calculating the battery capacity, and deleting abnormal charge and discharge fragments in the first charge and discharge fragments to obtain second charge and discharge fragments; constructing a charging feature set related to a battery charging mode and a discharging feature set related to a driving behavior of an electric vehicle driver, and constructing a charging and discharging data set; constructing a dual-channel deep learning model based on a convolutional neural network, a time convolutional network and an efficient channel attention mechanism; and training a dual-channel deep learning model by using the charging and discharging data set, and performing real-time prediction on the health state of the electric vehicle battery by taking the charging characteristics and the discharging characteristics as dual-channel input of the dual-channel deep learning model. The method can accurately predict the health state of the battery of the electric vehicle, and has high real-time performance.

Description

Technical Field

[0001] The present invention belongs to the technical field of power batteries, and in particular relates to a data-driven method and system for predicting the health status of an electric vehicle battery. Background Art

[0002] With the vigorous development of electric vehicles, power batteries are widely used in electric vehicles. Power batteries will continue to age during use, and the battery life will continue to shorten. The degree of aging of power batteries is often expressed by the state of health (SOH). SOH not only determines the performance status and driving range of the entire vehicle, but is also an important basis for formulating the charging and discharging and safety management strategies of the battery management system (BMS).

[0003] However, the change mechanism of SOH, a long-term varying parameter, is complex and is easily affected by multiple factors such as the environment, operating conditions and inconsistency of battery cells, making it difficult to estimate and predict SOH online. It is difficult to make an accurate prediction of the SOH of electric vehicle batteries, which is not conducive to the safe operation of new energy electric vehicles, as well as the classification and cascade utilization of power batteries. Summary of the invention

[0004] The present invention provides a data-driven electric vehicle battery health status prediction method and system to solve the problem in the prior art that SOH is difficult to estimate and predict online due to the combined influence of factors such as environment, working conditions and battery cell inconsistency, and it is difficult to accurately predict the SOH of electric vehicle batteries.

[0005] In a first aspect, the present invention provides a data-driven method for predicting the health status of an electric vehicle battery, comprising:

[0006] Obtain electric vehicle driving data and battery cycle data;

[0007] Extracting battery charge and discharge segments from the electric vehicle driving data and the battery cycle data to obtain a first charge and discharge segment;

[0008] Calculate battery capacity based on charging segment data;

[0009] Deleting abnormal charging and discharging segments in the first charging and discharging segments to obtain second charging and discharging segments;

[0010] Based on the second charging and discharging segment data, a charging feature set related to the battery charging mode and a discharging feature set related to the driving behavior of the electric vehicle driver are respectively constructed;

[0011] Construct a charge and discharge data set based on the charge and discharge feature set and battery capacity;

[0012] Build a dual-channel deep learning model based on convolutional neural network, temporal convolutional network and efficient channel attention mechanism;

[0013] The charging and discharging datasets are used to train a dual-channel deep learning model. The charging characteristics and discharging characteristics of electric vehicles are used as the dual-channel inputs of the dual-channel deep learning model to predict the health status of electric vehicle batteries in real time.

[0014] Optionally, extracting the battery charge and discharge segment from the electric vehicle driving data and the battery cycle data to obtain the first charge and discharge segment includes:

[0015] Determine whether each piece of electric vehicle operation data in the electric vehicle driving data and the battery cycle data belongs to charging data or discharging data;

[0016] Save the continuous charging data or discharging data as a charging and discharging segment;

[0017] Among them, the electric vehicle operation data includes vehicle status, charging status, vehicle speed and total current data; if the current vehicle status is parking charging, the vehicle speed is 0 and the total current is less than 0, then the current electric vehicle operation data is determined to be charging data; if the current vehicle status is vehicle starting and the charging status is not charged, then the current electric vehicle operation data is determined to be discharging data.

[0018] Optionally, calculating the battery capacity according to the charging segment data includes:

[0019] Calculate the battery capacity C according to the following formula:

[0020]

[0021] Among them, t 1 and t 2 are the start and end times of the target charging segment respectively; I(t) is the charging current at time t, t∈[t 1 ,t 2 ]; △t is the fixed sampling time; SOC t2 t 2 The state of charge of the battery at all times; SOC t1 t 1 The battery's state of charge at all times.

[0022] Optionally, deleting the abnormal charging and discharging segments in the first charging and discharging segments to obtain the second charging and discharging segments includes:

[0023] Read the SOC in all charging and discharging segments;

[0024] If the SOC variation interval in the target charge and discharge segment does not cover the first target percentage and the second target percentage, determining that the target charge and discharge segment is shorter and deleting the shorter charge and discharge segment;

[0025] If the changes of two consecutive SOCs in a charge and discharge process are greater than a third target percentage, determining that the target charge and discharge segments are discontinuous and deleting the discontinuous charge and discharge segments;

[0026] Plot a scatter plot of battery capacity versus total mileage of an electric vehicle;

[0027] The interquartile range method is used to determine the outliers in the scatter image, and the outliers are mapped to the charging segments and the mapped charging segments are deleted;

[0028] Get all the charging and discharging segments within the target number of days; if there are only charging segments or only discharging segments in a day, delete all the charging segments or discharging segments in the day.

[0029] Optionally, the method of determining outliers in the scatter image by using an interquartile range method, mapping the outliers to charging segments, and deleting the mapped charging segments includes:

[0030] The scatter plot between battery capacity and total mileage of electric vehicles is segmented into multiple separate intervals using a moving window method;

[0031] Sort the sizes of all battery capacities within the target interval to determine an upper quartile Q1 and a lower quartile Q3 within the target interval;

[0032] The upper bound U and lower bound L are determined according to the following formula:

[0033]

[0034] Outliers beyond the upper and lower bounds are mapped into charging segments and the mapped charging segments are deleted.

[0035] Optionally, the step of constructing a charging feature set related to the battery charging mode and a discharging feature set related to the driving behavior of the electric vehicle driver based on the second charging and discharging segment data respectively includes:

[0036] Constructing a charging feature set related to the battery charging mode based on the charging segment data;

[0037] Constructing a discharge feature set related to the driving behavior of electric vehicle drivers based on the discharge segment data;

[0038] Determining the correlation between the charging characteristic or the discharging characteristic and the battery capacity;

[0039] When the correlation is less than the correlation threshold, the corresponding charging feature or discharging feature is deleted;

[0040] Among them, the charging feature set includes features related to the battery charging mode calculated using the voltage, current and temperature of the battery pack and the single cell in the charging segment; the discharging feature set includes features calculated using the voltage, current and temperature of the battery pack in the discharging segment, and also includes features related to driving behavior calculated using the speed, accelerator pedal travel value, engine speed and engine torque during the driving process of the electric vehicle, and also includes features of the accumulated mileage of the electric vehicle.

[0041] Optionally, constructing a charge and discharge data set according to the charge and discharge feature set and the battery capacity includes:

[0042] Taking the target number of days as a cycle, the average value of the charge and discharge characteristic data and capacity data in a cycle is calculated as the characteristics and capacity corresponding to the electric vehicle battery in the current cycle;

[0043] For each electric vehicle’s capacity in all cycles, the capacity is divided by the capacity in the first cycle to obtain a normalized result to represent the health status of the electric vehicle battery;

[0044] For the characteristics of each electric vehicle in all cycles, each characteristic is processed by the maximum and minimum normalization method;

[0045] The charge and discharge data set is constructed based on the normalized features and capacity.

[0046] Optionally, the construction of a dual-channel deep learning model based on a convolutional neural network, a temporal convolutional network and an efficient channel attention mechanism includes:

[0047] Construct a CNN-TCN network based on three one-dimensional convolutional layers, two temporal convolutional network layers, and two linear fully connected layers;

[0048] A layer of efficient channel attention mechanism is added after the three one-dimensional convolutional layers and the two temporal convolutional network layers of the CNN-TCN network to construct the CNN-TCN-ECA network;

[0049] Two identical CNN-TCN-ECA networks are used as two channels, and then the outputs of the two channels are connected, and two linear fully connected layers are added again to obtain a dual-channel deep learning model.

[0050] Optionally, the dual-channel deep learning model is trained using a charging and discharging data set, and the charging characteristics and discharging characteristics of the electric vehicle are used as dual-channel inputs of the dual-channel deep learning model respectively, so as to make a real-time prediction of the health status of the electric vehicle battery, including:

[0051] The charging features and discharging features in the charging and discharging dataset are used as dual-channel inputs of the model, and the dual-channel deep learning model is iteratively trained to obtain the model parameters.

[0052] The model parameters are called, and the charging characteristics and discharging characteristics under the target cycle are input into the two channels of the model to obtain the health status of the electric vehicle battery under the target cycle, so as to realize real-time prediction of the health status of the electric vehicle battery.

[0053] In a second aspect, the present invention provides a data-driven electric vehicle battery health status prediction system, comprising:

[0054] An acquisition module, used for acquiring electric vehicle driving data and battery cycle data of an electric vehicle battery cycle;

[0055] An extraction module, used for extracting battery charge and discharge segments from the electric vehicle driving data and the battery cycle data to obtain a first charge and discharge segment;

[0056] A calculation module, used for calculating the battery capacity according to the charging segment data;

[0057] A deletion module, used for deleting abnormal charging and discharging segments in the first charging and discharging segments to obtain second charging and discharging segments;

[0058] A first construction module is used to construct a charging feature set related to the battery charging mode and a discharging feature set related to the driving behavior of the electric vehicle driver based on the second charging and discharging segment data;

[0059] The second construction module is used to construct a charge and discharge data set according to the charge and discharge feature set and the battery capacity;

[0060] The third building block is used to build a dual-channel deep learning model based on convolutional neural networks, temporal convolutional networks, and efficient channel attention mechanisms;

[0061] The prediction module is used to train a dual-channel deep learning model using a charging and discharging data set, and use the charging characteristics and discharging characteristics of the electric vehicle as dual-channel inputs of the dual-channel deep learning model to make real-time predictions on the health status of the electric vehicle battery.

[0062] The present invention provides a data-driven method and system for predicting the health status of an electric vehicle battery. The method models the electric vehicle operation data under a real environment, and can cope with the errors in the data collected under a real environment. Before modeling, abnormal data is deleted based on multiple standards, and a large amount of data noise is removed. The method simultaneously takes into account the charging characteristics related to the battery charging mode and the discharge characteristics related to the driver's driving behavior, and the extracted features are relatively comprehensive. The model used in the present invention is a dual-channel deep learning model that integrates a convolutional neural network, a temporal convolutional network, and an efficient channel attention mechanism, which can take into account both the charging characteristics and the discharge characteristics, better capture and fuse information, and can predict the health status of an electric vehicle battery in real time with high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0064] Figure 1 A schematic diagram of a flow chart of a data-driven electric vehicle battery health status prediction method provided by an embodiment of the present invention;

[0065] Figure 2 A histogram corresponding to the mean absolute error results of the models on the test set under different network structures provided in the embodiment of the present invention;

[0066] Figure 3 A histogram corresponding to the training time consumed by different network structures provided by the embodiment of the present invention on the training set;

[0067] Figure 4 A curve chart corresponding to the predicted result and the actual result of the battery health status on an electric vehicle data provided by an embodiment of the present invention;

[0068] Figure 5 A schematic structural diagram of a data-driven electric vehicle battery health status prediction system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0069] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0070] Example 1

[0071] like Figure 1 As shown, this embodiment provides a data-driven electric vehicle battery health status prediction method, including:

[0072] Step 101, obtaining electric vehicle driving data and battery cycle data.

[0073] The collection interval of electric vehicle battery cycle charge and discharge data is ten seconds, which includes vehicle status, charging status, state of charge, current, voltage, temperature, speed, accelerator pedal travel value, engine speed, engine torque and other information.

[0074] Step 102 , extracting battery charge and discharge segments from the electric vehicle driving data and the battery cycle data to obtain a first charge and discharge segment.

[0075] The charge and discharge segment is composed of continuous charge and discharge data, which is judged based on the four data of "vehicle status", "charging status", "total current" and "vehicle speed" in a sample.

[0076] Among them, "vehicle status" includes two states: "vehicle started" and "engine off", and "charging status" includes three states: "parked charging", "charging completed" and "not charged". The process of judging charging data is relatively strict. In addition to using "charging status" to judge, "total current" and "vehicle speed" are also introduced for limitation. Among them, charging data needs to meet the three conditions of "charging status" being "parked charging", "total current" being less than zero, and "vehicle speed" being equal to zero; when judging discharge data, the conditions are relatively loose, and data with "vehicle status" being "vehicle started" and "charging status" being "not charged" are considered to be discharge data.

[0077] Exemplarily, this step includes:

[0078] It is determined whether each piece of electric vehicle operation data in the electric vehicle driving data and the battery cycle data belongs to charging data or discharging data.

[0079] The continuous charging data or discharging data is saved as a charging and discharging segment.

[0080] Among them, the electric vehicle operation data includes vehicle status, charging status, vehicle speed and total current data; if the current vehicle status is parking charging, the vehicle speed is 0 and the total current is less than 0, then the current electric vehicle operation data is determined to be charging data; if the current vehicle status is vehicle starting and the charging status is not charged, then the current electric vehicle operation data is determined to be discharging data.

[0081] Step 103: Calculate the battery capacity according to the charging segment data.

[0082] The tram operation data used contains three data: "SOC", "total current" and "collection time". In a charging segment, the collection time interval between each two data is 10 seconds. The current can be integrated in combination with the SOC change, and the battery capacity C corresponding to the current charging segment can be calculated:

[0083]

[0084] Among them, t 1 and t 2 are the start and end times of the target charging segment respectively; I(t) is the charging current at time t, t∈[t 1 ,t 2 ]; △t is the fixed sampling time; t 2 The battery's state of charge at all times; t 1 The battery's state of charge at all times.

[0085] Step 104 , deleting abnormal charging and discharging segments in the first charging and discharging segments to obtain second charging and discharging segments.

[0086] After obtaining the charge and discharge segments, the SOC change process of each segment is checked, and the shorter or discontinuous charge and discharge segments are deleted.

[0087] The electric vehicle operation data used also includes "cumulative mileage (km)" data, which is the total mileage data of the electric vehicle. After calculating the battery capacity corresponding to each charging segment, a scatter plot between the battery capacity and the total mileage is drawn. Then, the moving window method is used to divide the relationship image between battery capacity and mileage into separate intervals. The interval used in this embodiment is 10,000 km, and the moving window step size is 1,000 km; then all battery capacities in the interval are sorted, and the three quartiles Q1 (first quartile), Q2 (second quartile) and Q3 (third quartile) in the interval are calculated.

[0088] In order to match the charging segment with the discharging segment, since directly mapping each charging segment to the discharging segment will ignore more information, this embodiment takes one day as a unit, averages the battery capacity data corresponding to all the charging segments in one day, and takes the average value as the battery capacity of the electric vehicle on that day. Each discharging segment in that day corresponds to this average capacity. If there is only a charging segment or only a discharging segment in one day, it will be deleted.

[0089] Exemplarily, this step includes reading the SOC in all charging and discharging segments.

[0090] If the SOC variation interval in the target charge and discharge segment does not cover the first target percentage (eg, 70%) and the second target percentage (eg, 90%), it is determined that the target charge and discharge segment is shorter and the shorter charge and discharge segment is deleted.

[0091] If the changes in two consecutive SOCs in a charge and discharge process are greater than a third target percentage (eg, 5%), it is determined that the target charge and discharge segments are discontinuous and the discontinuous charge and discharge segments are deleted.

[0092] Plot a scatter plot of battery capacity versus total electric vehicle range.

[0093] The interquartile range method is used to determine the outliers in the scatter image, and the outliers are mapped to the charged segments and the mapped charged segments are deleted.

[0094] Get all the charging and discharging segments within the target number of days; if there are only charging segments or only discharging segments in a day, delete all the charging segments or discharging segments in the day.

[0095] Among them, the interquartile range method is used to determine the outliers in the scatter image, and the outliers are mapped to the charging segments and the mapped charging segments are deleted, including:

[0096] The scatter plot between battery capacity and total mileage of electric vehicles is segmented into separate intervals using a moving window method.

[0097] The sizes of all battery capacities within the target interval are sorted to determine the first quartile Q1 and the third quartile Q3 within the target interval.

[0098] The upper bound U and lower bound L are determined according to the following formula:

[0099]

[0100] Outliers beyond the upper and lower bounds are mapped into charging segments and the mapped charging segments are deleted.

[0101] When determining the upper and lower bounds, the constant selected in the experiment is 2.5, the purpose is to retain more normal data and only delete a small amount of data with large errors. In this way, outliers in the scatter image can be found, and the charging fragments corresponding to the outliers are considered to be abnormal data and need to be deleted;

[0102] Step 105 : constructing a charging feature set related to the battery charging mode and a discharging feature set related to the driving behavior of the electric vehicle driver based on the second charging and discharging segment data.

[0103] Before step 105, the electric vehicle operation data with insignificant capacity degradation is eliminated. For the screened charge and discharge segments, this embodiment takes three days as a cycle, calculates the average capacity of all battery capacities within three days, and draws the battery capacity degradation curve of each electric vehicle. Since the electric vehicle operation data set used is collected in a real environment, there may be certain problems with both the sensor and the electric vehicle itself, and under ideal conditions, the battery capacity of the electric vehicle should show a degradation trend. Therefore, this embodiment determines based on the battery capacity degradation curve and deletes the operation data of the electric vehicle with insignificant degradation trend or particularly large fluctuations.

[0104] Features related to the battery health status are extracted, and charging features and discharging features are extracted in the charging segment and the discharging segment respectively. In the feature extraction process, this embodiment mainly considers two aspects: the mean value that can directly reflect the data characteristics and the standard deviation value that can reflect the data changes.

[0105] The charging characteristics include charging voltage-related characteristics, charging current-related characteristics and charging temperature-related characteristics, a total of 24 characteristics; among them, the charging voltage-related characteristics include the total battery voltage characteristics and the single cell voltage characteristics; the charging current-related characteristics include two data, the mean and standard deviation of the current in the charging segment; the charging temperature-related characteristics include six data, the maximum temperature of the single cell, the minimum temperature of the single cell, and the mean and standard deviation of the maximum and minimum temperature difference; the total battery voltage characteristics include three data, the mean and standard deviation of the total voltage in the charging segment, and the voltage change rate; the single cell voltage characteristics include the average voltage of the single cell in the charging segment, the maximum temperature of the single cell, and the standard deviation of the maximum temperature difference. The battery voltage characteristics include eight data including the maximum voltage, the minimum voltage of the single cell, the mean and standard deviation of the maximum and minimum voltage differences, and four data including the maximum, minimum, and difference of the voltages of all the single cells in the battery pack when the electric vehicle is charged to 80%, as well as the standard deviation of the voltages of these single cells. In addition, the single cell voltage characteristics also include a feature reflecting the consistency of the single cell voltage changes. The calculation method of this feature is to first calculate the standard deviation of all the single cell voltages in each data in the charging segment, and then calculate the standard deviation of these standard deviation values ​​in a charging segment. The obtained data is the feature reflecting the consistency of the single cell voltage changes corresponding to the charging segment.

[0106] The discharge characteristics include discharge voltage related characteristics, discharge current related characteristics, discharge temperature related characteristics and driving behavior related characteristics, a total of 22 characteristics; among them, the discharge voltage related characteristics include 6 data, namely, the mean and standard deviation of the total voltage, the maximum voltage of the single cell and the minimum voltage of the single cell in the discharge segment; the discharge current related characteristics include 2 data, namely, the mean and standard deviation of the current in the discharge segment; the discharge temperature related characteristics include 4 data, namely, the mean and standard deviation of the maximum temperature of the single cell and the minimum temperature of the single cell in the discharge segment; the driving behavior data include 8 data, namely, the mean and standard deviation of the speed, the accelerator pedal travel value, the engine speed and the engine torque in the discharge segment, and also include 2 data, namely, the total mileage and the mileage of the electric vehicle for every 1% of the power consumed; the mileage of the electric vehicle for every 1% of the power consumed is calculated by calculating the mileage of the electric vehicle in a discharge segment based on the total mileage, and then combining the power change of the discharge segment to obtain the mileage of the electric vehicle for every 1% of the power consumed.

[0107] A charging feature set related to the battery charging mode is constructed based on the charging segment data; a discharge feature set related to the driving behavior of the electric vehicle driver is constructed based on the discharge segment data; the correlation between the charging feature or the discharge feature and the battery capacity is determined (for example, the Pearson correlation coefficient between each feature and the capacity is calculated); when the correlation is less than a correlation threshold (for example, 0.5), the corresponding charging feature or discharge feature is deleted.

[0108] Step 106, constructing a charge and discharge data set according to the charge and discharge feature set and the battery capacity.

[0109] The target number of days (eg, 3 days) is taken as a cycle, and the average values ​​of the charge and discharge characteristic data and the battery capacity data in a cycle are taken as the characteristics and capacity corresponding to the battery in the current cycle.

[0110] The battery capacity of each electric vehicle in all cycles is divided by the battery capacity in the first cycle to obtain a normalized result to represent the health status of the electric vehicle battery.

[0111] The characteristics of each electric vehicle in all cycles are normalized to the maximum and minimum values.

[0112] The charge and discharge data set is constructed based on the normalized features and battery capacity.

[0113] Step 107, construct a dual-channel deep learning model based on convolutional neural network, temporal convolutional network and efficient channel attention mechanism.

[0114] The two channels of the constructed dual-channel deep learning model are mainly composed of two identical CNN-TCN network models, and the ECA (Efficient Channel Attention) attention mechanism is added to the CNN-TCN network model. The structure of the CNN-TCN network model consists of three one-dimensional convolutional layers, two TCN layers, two attention mechanism layers and two fully connected layers.

[0115] The convolution layer consists of three one-dimensional convolution operations, which are used to initially extract information from the feature vector. The three convolution layers all use a convolution kernel of size 3×3, and their output vector dimensions are 64, 128, and 256, respectively, in order to gradually extract deeper feature information. At the same time, the result of each convolution layer is input into the ReLU activation function to increase the nonlinear expression ability of the CNN-TCN network model.

[0116] The attention network used in the attention mechanism layer is ECA, which is a lightweight network that can initially improve the expressiveness of features by adaptively assigning weights to different channels, enhancing the representation of important features, and suppressing unimportant features.

[0117] The TCN layer uses two layers of TCN networks (the fifth and sixth layers of the CNN-TCN network model) to capture long-distance dependencies in features. The two layers of networks reduce the number of feature vector channels to 128 and 64, respectively, and gradually compress the feature dimensions. TCN also selects a convolution kernel of size 3×3 to maintain good computational efficiency. In addition, the Dropout parameter in the network is set to 0.2 to prevent overfitting of the model and enhance the generalization ability of the model.

[0118] The attention mechanism layers (the fourth and seventh layers of the CNN-TCN network model, respectively) also use the ECA attention network to further improve the expressiveness of features.

[0119] The fully connected layer consists of two linear layers. The output of the first linear layer is a 128-dimensional vector, which is processed by the ReLU activation function. The second linear layer receives the result of the first linear layer and then outputs a 64-dimensional vector to provide input for the final prediction task.

[0120] The dual-channel deep learning model consists of two CNN-TCN network models and two linear layers, and also includes a splicing function; the two CNN-TCN networks learn charging features and discharging features respectively, and each outputs a 64-dimensional vector. The two vectors are concatenated by the concatenation function to obtain a 128-dimensional vector; the first layer of the two linear layers receives the concatenated 128-dimensional vector, which is also processed by the activation function, and then processed by the second linear layer to output the final prediction result.

[0121] Step 108, using the charging and discharging data set to train a dual-channel deep learning model, using the electric vehicle charging characteristics and discharging characteristics as dual-channel inputs of the dual-channel deep learning model, respectively, to perform real-time prediction of the health status of the electric vehicle battery.

[0122] The data set is divided by selecting the data of three electric vehicles as the test set and all the data of the remaining electric vehicles as the training set.

[0123] The dual-channel deep learning model is trained in such a way that two channels with the same structure learn charging characteristics and discharging characteristics respectively, use the splicing results of the two channels to predict the final result, iteratively train the dual-channel deep learning model and save the model parameters.

[0124] The model parameters are called and the charging and discharging characteristics of the electric vehicle to be tested are input into the dual-channel deep learning model to predict the battery health status of the electric vehicle in real time.

[0125] In this embodiment, the learning rate is set to 8×10 -5 , select Adam optimizer, and perform 100 iterations, saving the dual-channel deep learning model parameters after 100 iterations. The calculated prediction errors include mean absolute error MAE and root mean square error RMSE, and these two indicators are used as standards to judge the quality of the model.

[0126] To further demonstrate the advantages of the provided dual-channel deep learning model in the task of battery health status prediction, this embodiment considers the common LSTM network and Transformer network, and uses a network composed entirely of convolutional layers. The specific method of constructing the network model is to directly replace the TCN layers in the original dual-channel CNN-TCN network model with convolutional layers, LSTM layers, and Transformer layers with the same number of layers, respectively, to obtain a CNN model, a CNN-LSTM model, and a CNN-Transformer model in turn; the number of neurons in the LSTM hidden layer is 64; the CNN-Transformer model replaces the two-layer TCN network with a two-layer Transformer Encoder, and the model dimension of the Transformer Encoder is 256.

[0127] In addition, in order to further demonstrate the advantages of the ECA attention mechanism, this embodiment replaces the attention network with the SE attention network, the CBAM attention network, and the DANet, and deletes the attention mechanism layer for further comparison. Figure 2 As shown; the bar chart after sorting the training time of different models on the training set is as follows Figure 3 shown.

[0128] Through experiments, it is found that among the attention networks used, only ECA can consume less computing resources while maintaining good prediction results. The performance of the replaced network models on the same data set is not as good as the dual-channel deep learning model provided in this embodiment.

[0129] An electric car in the experimental data is selected as an example, and the dual-channel deep learning model provided in this embodiment is used to execute step 108 to output the battery health status result of the electric car, such as Figure 4 shown.

[0130] In summary, the data-driven electric vehicle battery health status prediction method provided in this embodiment is based on electric vehicle operation data modeling in a real environment, which can cope with the errors in the data collected in the real environment; before modeling, abnormal data is deleted based on multiple standards, and a large amount of data noise is removed; the method takes into account the charging characteristics related to the battery charging mode and the discharge characteristics related to the driver's driving behavior, and the extracted features are relatively comprehensive; the model used in the present invention is a dual-channel deep learning model that integrates convolutional neural networks, temporal convolutional networks and efficient channel attention mechanisms, which can take into account both charging characteristics and discharge characteristics, better capture and fuse information, and can predict the health status of electric vehicle batteries in real time with high accuracy.

[0131] Example 2

[0132] Based on the same inventive concept as Example 1, this embodiment provides a data-driven electric vehicle battery health status prediction system. Since the principle of solving the problem by this system is similar to the aforementioned data-driven electric vehicle battery health status prediction method, the implementation of this system can refer to the implementation of the data-driven electric vehicle battery health status prediction method.

[0133] like Figure 5 As shown, the data-driven electric vehicle battery health status prediction system includes:

[0134] The acquisition module 10 is used to acquire the electric vehicle driving data and the battery cycle data of the electric vehicle battery cycle.

[0135] The extraction module 20 is used to extract the battery charging and discharging segment from the electric vehicle driving data and the battery cycle data to obtain the first charging and discharging segment.

[0136] The calculation module 30 is used to calculate the battery capacity according to the charging segment data.

[0137] The deletion module 40 is used to delete the abnormal charging and discharging segments in the first charging and discharging segments to obtain the second charging and discharging segments.

[0138] The first constructing module 50 is used to construct a charging feature set related to the battery charging mode and a discharging feature set related to the driving behavior of the electric vehicle driver based on the second charging and discharging segment data.

[0139] The second constructing module 60 is used to construct a charge and discharge data set according to the charge and discharge feature set and the battery capacity.

[0140] The third building block 70 is used to build a dual-channel deep learning model based on a convolutional neural network, a temporal convolutional network and an efficient channel attention mechanism.

[0141] The prediction module 80 is used to train a dual-channel deep learning model using a charging and discharging data set, and use the charging characteristics and discharging characteristics of the electric vehicle as dual-channel inputs of the dual-channel deep learning model to perform real-time prediction of the health status of the electric vehicle battery.

[0142] For more specific working processes of the above modules, please refer to the corresponding contents disclosed in Example 1, which will not be repeated here.

[0143] Example 3

[0144] This embodiment provides a computer device, including a processor and a memory; wherein, when the processor executes a computer program stored in the memory, the steps of the data-driven electric vehicle battery health status prediction method described in Example 1 are implemented.

[0145] For more specific details of the above method, please refer to the corresponding contents disclosed in Example 1, which will not be repeated here.

[0146] Example 4

[0147] This embodiment provides a computer-readable storage medium for storing a computer program; when the computer program is executed by a processor, the steps of the data-driven electric vehicle battery health status prediction method described in Example 1 are implemented.

[0148] For more specific details of the above method, please refer to the corresponding contents disclosed in Example 1, which will not be repeated here.

[0149] Example 5

[0150] This embodiment provides a computer program product, including computer executable instructions or a computer program. When the computer executable instructions or the computer program are executed by a processor, the steps of the data-driven electric vehicle battery health status prediction method described in Example 1 are implemented.

[0151] For more specific details of the above method, please refer to the corresponding contents disclosed in Example 1, which will not be repeated here.

[0152] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the systems, devices, storage media, and computer program products disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0153] Those skilled in the art can clearly understand that the technology in the embodiments of the present invention can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution in the embodiments of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention or some parts of the embodiments.

[0154] In some embodiments, computer executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0155] As an example, computer-executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file that stores other programs or data, such as in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subroutines, or code portions).

[0156] As an example, computer executable instructions may be deployed to be executed on one electronic device, or on multiple electronic devices located at one site, or on multiple electronic devices distributed at multiple sites and interconnected by a communication network.

[0157] The present invention has been described in detail above in conjunction with specific implementations and exemplary examples, but these descriptions cannot be understood as limiting the present invention. Those skilled in the art understand that, without departing from the spirit and scope of the present invention, a variety of equivalent substitutions, modifications or improvements may be made to the technical solution of the present invention and its implementation methods, all of which fall within the scope of the present invention. The scope of protection of the present invention shall be subject to the attached claims.

Claims

1. A data-driven electric vehicle battery health status prediction method, characterized in that: include: Obtain electric vehicle driving data and battery cycle data; Extracting battery charge and discharge segments from the electric vehicle driving data and the battery cycle data to obtain a first charge and discharge segment; Calculate battery capacity based on charging segment data; Deleting abnormal charging and discharging segments in the first charging and discharging segments to obtain second charging and discharging segments; Based on the second charging and discharging segment data, a charging feature set related to the battery charging mode and a discharging feature set related to the driving behavior of the electric vehicle driver are respectively constructed; Construct a charge and discharge data set based on the charge and discharge feature set and battery capacity; Build a dual-channel deep learning model based on convolutional neural network, temporal convolutional network and efficient channel attention mechanism; The charging and discharging datasets are used to train a dual-channel deep learning model. The charging characteristics and discharging characteristics of electric vehicles are used as the dual-channel inputs of the dual-channel deep learning model to predict the health status of electric vehicle batteries in real time.

2. The electric vehicle battery health status prediction method according to claim 1, characterized in that: The method of extracting the battery charge and discharge segment from the electric vehicle driving data and the battery cycle data to obtain the first charge and discharge segment includes: Determine whether each piece of electric vehicle operation data in the electric vehicle driving data and the battery cycle data belongs to charging data or discharging data; Save the continuous charging data or discharging data as a charging and discharging segment; Among them, the electric vehicle operation data includes vehicle status, charging status, vehicle speed and total current data; if the current vehicle status is parking charging, the vehicle speed is 0 and the total current is less than 0, then the current electric vehicle operation data is determined to be charging data; if the current vehicle status is vehicle starting and the charging status is not charged, then the current electric vehicle operation data is determined to be discharging data.

3. The electric vehicle battery health status prediction method according to claim 1, characterized in that: The calculating the battery capacity according to the charging segment data comprises: Calculate the battery capacity C according to the following formula: Where t1 and t2 are the start and end times of the target charging segment, respectively; I(t) is the charging current at time t, t∈[t1,t2]; △t is the fixed sampling time; is the state of charge of the battery at time t2; is the state of charge of the battery at time t1.

4. The electric vehicle battery health status prediction method according to claim 1, characterized in that: The step of deleting the abnormal charging and discharging segments in the first charging and discharging segments to obtain the second charging and discharging segments includes: Read the SOC in all charging and discharging segments; If the SOC variation interval in the target charge and discharge segment does not cover the first target percentage and the second target percentage, determining that the target charge and discharge segment is shorter and deleting the shorter charge and discharge segment; If the changes of two consecutive SOCs in a charge and discharge process are greater than a third target percentage, determining that the target charge and discharge segments are discontinuous and deleting the discontinuous charge and discharge segments; Plot a scatter plot of battery capacity versus total mileage of an electric vehicle; The interquartile range method is used to determine the outliers in the scatter image, and the outliers are mapped to the charging segments and the mapped charging segments are deleted; Get all the charging and discharging segments within the target number of days; if there are only charging segments or only discharging segments in a day, delete all the charging segments or discharging segments in the day.

5. The electric vehicle battery health status prediction method according to claim 4, characterized in that: The method of using the interquartile range method to determine the outliers in the scattered image, mapping the outliers to the charging segments and deleting the mapped charging segments includes: The scatter plot between battery capacity and total mileage of electric vehicles is segmented into multiple separate intervals using a moving window method; Sort the sizes of all battery capacities within the target interval to determine an upper quartile Q1 and a lower quartile Q3 within the target interval; The upper bound U and lower bound L are determined according to the following formula: Outliers beyond the upper and lower bounds are mapped into charging segments and the mapped charging segments are deleted.

6. The electric vehicle battery health status prediction method according to claim 1, characterized in that: The method of constructing a charging feature set related to the battery charging mode and a discharging feature set related to the driving behavior of the electric vehicle driver based on the second charging and discharging segment data respectively includes: Constructing a charging feature set related to the battery charging mode based on the charging segment data; Constructing a discharge feature set related to the driving behavior of electric vehicle drivers based on the discharge segment data; Determining the correlation between the charging characteristic or the discharging characteristic and the battery capacity; When the correlation is less than the correlation threshold, the corresponding charging feature or discharging feature is deleted; Among them, the charging feature set includes features related to the battery charging mode calculated using the voltage, current and temperature of the battery pack and the single cell in the charging segment; the discharging feature set includes features calculated using the voltage, current and temperature of the battery pack in the discharging segment, and also includes features related to driving behavior calculated using the speed, accelerator pedal travel value, engine speed and engine torque during the driving process of the electric vehicle, and also includes features of the accumulated mileage of the electric vehicle.

7. The electric vehicle battery health status prediction method according to claim 1, characterized in that: The step of constructing a charge and discharge data set according to the charge and discharge feature set and the battery capacity includes: Taking the target number of days as a cycle, the average value of the charge and discharge characteristic data and capacity data in a cycle is calculated as the characteristics and capacity corresponding to the electric vehicle battery in the current cycle; For each electric vehicle’s capacity in all cycles, the capacity is divided by the capacity in the first cycle to obtain a normalized result to represent the health status of the electric vehicle battery; For the characteristics of each electric vehicle in all cycles, each characteristic is processed by the maximum and minimum normalization method; The charge and discharge data set is constructed based on the normalized features and capacity.

8. The electric vehicle battery health status prediction method according to claim 1, characterized in that: The dual-channel deep learning model based on convolutional neural network, temporal convolutional network and efficient channel attention mechanism is constructed, including: Construct a CNN-TCN network based on three one-dimensional convolutional layers, two temporal convolutional network layers, and two linear fully connected layers; A layer of efficient channel attention mechanism is added after the three one-dimensional convolutional layers and the two temporal convolutional network layers of the CNN-TCN network to construct the CNN-TCN-ECA network; Two identical CNN-TCN-ECA networks are used as two channels, and then the outputs of the two channels are connected, and two linear fully connected layers are added again to obtain a dual-channel deep learning model.

9. The electric vehicle battery health status prediction method according to claim 1, characterized in that: The dual-channel deep learning model is trained by using the charging and discharging data set, and the charging characteristics and discharging characteristics of the electric vehicle are used as the dual-channel inputs of the dual-channel deep learning model respectively, so as to make a real-time prediction of the health status of the electric vehicle battery, including: The charging features and discharging features in the charging and discharging dataset are used as dual-channel inputs of the model, and the dual-channel deep learning model is iteratively trained to obtain the model parameters. The model parameters are called, and the charging characteristics and discharging characteristics under the target cycle are respectively input into the two channels of the model to obtain the health status of the electric vehicle battery under the target cycle, so as to realize the real-time prediction of the health status of the electric vehicle battery.

10. A data-driven electric vehicle battery health status prediction system, characterized in that: include: An acquisition module is used to acquire electric vehicle driving data and battery cycle data; An extraction module, used for extracting battery charge and discharge segments from the electric vehicle driving data and the battery cycle data to obtain a first charge and discharge segment; A calculation module, used for calculating the battery capacity according to the charging segment data; A deletion module, used for deleting abnormal charging and discharging segments in the first charging and discharging segments to obtain second charging and discharging segments; A first construction module is used to construct a charging feature set related to the battery charging mode and a discharging feature set related to the driving behavior of the electric vehicle driver based on the second charging and discharging segment data; The second construction module is used to construct a charge and discharge data set according to the charge and discharge feature set and the battery capacity; The third building block is used to build a dual-channel deep learning model based on convolutional neural networks, temporal convolutional networks, and efficient channel attention mechanisms; The prediction module is used to train a dual-channel deep learning model using a charging and discharging data set, and use the charging characteristics and discharging characteristics of the electric vehicle as dual-channel inputs of the dual-channel deep learning model to make real-time predictions on the health status of the electric vehicle battery.

Images