A method, device, equipment and medium for predicting the state of health of a battery

Abstract

The present invention discloses a method, device, equipment and medium for predicting the state of health of a battery, relating to the field of new energy technologies. The method includes: obtaining charge and discharge cycle data of a lithium battery to be detected; inputting the charge and discharge cycle data into a trained battery state of health prediction model to obtain a battery life result output by the battery state of health prediction model; the battery state of health prediction model is trained by means of supervised deep learning based on the sample fusion features of a sample lithium battery and the state of health of the battery in each sample charge and discharge cycle, the sample fusion features are the input data used for training, and the state of health of the battery in each sample charge and discharge cycle is the label used for training. The present invention effectively improves the prediction accuracy of the model, and while effectively solving the problem of unbalanced data distribution, more accurately completes the task of predicting the state of health of the battery.

Description

Technical Field

[0001] The present invention relates to the field of new energy technologies, and in particular to a battery health status prediction method, device, equipment and medium. Background Art

[0002] After repeated charge and discharge cycles, the internal resistance of lithium batteries will gradually increase. The increase in internal resistance will cause the battery to generate significant heat during the charge and discharge process, which will cause irreversible aging and performance degradation of the lithium battery. The state of health (SOH) of the battery can be used to indicate the remaining life of the lithium battery. As the number of charge and discharge cycles of the lithium battery increases, the SOH will decrease, that is, the remaining life of the battery will decrease and the available capacity will gradually decrease. When the SOH drops to a certain threshold, such as below 80%, it is necessary to stop using the lithium battery or limit its use. Therefore, accurately predicting the SOH of the lithium battery is extremely important for its use. At present, machine learning-based methods such as neural network models are widely used to predict the SOH of lithium batteries.

[0003] During the lithium battery charge and discharge cycle, there is a specific time node in the discharge process. Before and after this specific time node, there is a significant difference in the rate of voltage drop. This specific time node is called the inflection point of the discharge voltage-time characteristic curve. The inflection point voltage will gradually decrease with the aging and performance degradation of the lithium battery. It is precisely because of the existence of this inflection point that the data collected during the lithium battery charge and discharge cycle are unevenly distributed. This imbalance may cause the neural network model to over-focus on the data features of dense areas during training, while ignoring important information in sparse areas, resulting in the neural network model failing to fully learn the data features of sparse areas based on these data, and thus limiting the generalization performance of new data.

[0004] Therefore, how to improve the accuracy of neural network models in predicting lithium battery SOH is an important issue that needs to be urgently addressed in the industry. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides a battery health status prediction method, device, equipment and medium to solve the problem.

[0006] According to a first aspect, an embodiment of the present invention provides a method for predicting a battery health state, the method comprising:

[0007] Obtain the charge and discharge cycle data of the lithium battery to be tested;

[0008] Inputting the charge and discharge cycle data into a trained battery health state prediction model to obtain a battery life result output by the battery health state prediction model;

[0009] The battery health state prediction model is trained by using a supervised deep learning method based on the sample fusion features of the sample lithium battery and the battery health state of each sample charge-discharge cycle. The sample fusion features are the input data used for training, and the battery health state of each sample charge-discharge cycle is the label used for training.

[0010] The sample fusion features are extracted from the sample charge-discharge cycle data of the sample lithium battery by identifying the inflection points in the charge-discharge cycle process based on the time-voltage decay relationship and mining the implicit differential features in the data. The battery health state of each sample charge-discharge cycle is determined according to the ratio of the battery capacity of the sample charge-discharge cycle to the battery capacity of the initial sample charge-discharge cycle.

[0011] Combined with the first aspect, in the first embodiment of the first aspect, the battery health state prediction model is trained through the following steps:

[0012] Obtain the sample charge-discharge cycle data of the sample lithium battery, and preprocess the sample charge-discharge cycle data to screen out the sample features of each sample charge cycle; the sample features include: temperature, current, voltage, power, time, and battery capacity.

[0013] Determine the battery health state of each sample charge-discharge cycle according to the ratio of the battery capacity of each sample charge-discharge cycle to the battery capacity of the initial sample charge-discharge cycle.

[0014] Construct a voltage-time decay feature curve for each sample charge-discharge cycle according to the sample features of each sample charge-discharge cycle.

[0015] Merge the sample features with the same voltage value in the voltage-time decay feature curves of each sample charge-discharge cycle period to obtain voltage grouped features, and obtain the derivative features of the voltage grouped features, and perform feature fusion of the voltage grouped features and the derivative features to obtain sample fusion features.

[0016] Use the sample fusion features as the input data for training, and use the battery health state of each sample charge-discharge cycle as the label, and train by using a supervised deep learning method to obtain a battery health state prediction model for predicting the battery life result of the lithium battery.

[0017] Combined with the first embodiment of the first aspect, in the second embodiment of the first aspect, the merging of the sample features with the same voltage value in the voltage-time decay feature curves of each sample charge-discharge cycle period to obtain voltage grouped features, and obtaining the derivative features of the voltage grouped features, and performing feature fusion of the voltage grouped features and the derivative features to obtain sample fusion features specifically includes:

[0018] For each voltage-time decay characteristic curve, perform fitting through a preset experience curve, identify the turning points on the preset experience curve, and obtain the inflection points and the starting points of the inflection points of each voltage-time decay characteristic curve;

[0019] Merge the sample characteristics with the same voltage value in the voltage-time decay characteristic curves of each sample charge-discharge cycle period to obtain voltage grouping characteristics;

[0020] Determine the difference between each voltage value in the voltage grouping characteristics and the initial voltage of the voltage-time decay characteristic curve to obtain a voltage difference, and determine the time offsets of each voltage value in the voltage grouping characteristics relative to the inflection points and the starting points of the inflection points of the voltage-time decay characteristic curve respectively to obtain a first time difference and a second time difference; The derivative features include the voltage difference, the first time difference, and the second time difference;

[0021] Perform feature fusion of the voltage grouping characteristics and the derivative features to obtain sample fusion features.

[0022] Combined with the first aspect, in the third embodiment of the first aspect, the battery health state prediction model is a composite neural network model combining a convolutional neural network and a long short-term memory network;

[0023] The battery health state prediction model includes a convolutional neural network layer and a long short-term memory network layer. The convolutional neural network layer is used to extract local features of the charge-discharge cycle data, and the long short-term memory network layer is used to capture the dependencies of the local features in the time series.

[0024] Combined with the third embodiment of the first aspect, in the fourth embodiment of the first aspect, the convolutional neural network layer includes a convolutional layer, a max pooling layer, a data flattening layer, and a first fully connected layer connected in sequence;

[0025] The long short-term memory network layer includes a first relationship capture layer, a second relationship capture layer, and a third relationship capture layer connected in sequence.

[0026] Combined with the fourth embodiment of the first aspect, in the fifth embodiment of the first aspect, the battery health state prediction model further includes an identity mapping layer, a feature fusion layer, and a rectified linear layer with a residual network structure;

[0027] After the charge-discharge cycle data are processed by the identity mapping layer and then processed by the convolutional neural network layer and the long short-term memory network layer in sequence, they are processed by the feature fusion layer and the rectified linear layer in sequence to obtain the battery life result.

[0028] Combined with the fifth embodiment of the first aspect, in the sixth embodiment of the first aspect, inputting the charge and discharge cycle data into the trained battery health state prediction model to obtain the battery life result output by the battery health state prediction model specifically includes:

[0029] Input the charge and discharge cycle data into the identity mapping layer to obtain the identity mapping feature output after the identity processing by the identity mapping layer;

[0030] Input the charge and discharge cycle data into the convolutional neural network layer to obtain the first output feature output by the convolutional neural network layer;

[0031] Input the first output feature into the long short-term memory network layer to obtain the third abstract feature output by the long short-term memory network layer;

[0032] Input the identity mapping feature and the third abstract feature into the feature fusion layer for feature fusion, and obtain the battery life result after being processed by the rectified linear layer.

[0033] According to the third aspect, an embodiment of the present invention further provides a battery health state prediction device, and the device includes:

[0034] A data acquisition module, configured to acquire the charge and discharge cycle data of the lithium battery to be detected;

[0035] A battery prediction module, configured to input the charge and discharge cycle data into the trained battery health state prediction model to obtain the battery life result output by the battery health state prediction model;

[0036] The battery health state prediction model is trained by using a supervised deep learning method based on the sample fusion features of the sample lithium battery and the battery health state of each sample charge and discharge cycle. The sample fusion features are the input data used for training, and the battery health state of each sample charge and discharge cycle is the label used for training;

[0037] The sample fusion features are extracted from the sample charge and discharge cycle data of the sample lithium battery by identifying the inflection points in the charge and discharge cycle process based on the time-voltage decay relationship and mining the implicit differential features in the data. The battery health state of each sample charge and discharge cycle is determined according to the ratio of the battery capacity of the sample charge and discharge cycle to the battery capacity of the initial sample charge and discharge cycle.

[0038] According to the third aspect, an embodiment of the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the battery health state prediction method as described in any one of the above are implemented.

[0039] According to a fourth aspect, an embodiment of the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of any of the above-described battery health status prediction methods are implemented.

[0040] According to a fifth aspect, an embodiment of the present invention further provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the steps of the battery health status prediction method as described in any one of the above items are implemented.

[0041] The battery health status prediction method, device, equipment and medium of the present invention predict the battery life result of the lithium battery to be tested through a trained battery health status prediction model, wherein the battery health status prediction model locates the inflection point by accurately fitting and calculating the voltage-time decay characteristic relationship, which not only provides key time information for the data, but also can amplify the difference between repeated information before the voltage occurs, which is helpful to complete the subsequent feature derivation and feature fusion process, and secondly, divides the data according to the inflection point information to alleviate the uneven distribution of data, and simultaneously performs feature derivation and feature fusion, amplifies the difference of window data, and mines important information in sparse areas, so that the model can fully learn the data features of sparse areas, effectively improving the prediction accuracy of the model, while effectively solving the uneven distribution of data, and more accurately completing the task of battery health status prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The features and advantages of the present invention will be more clearly understood by referring to the accompanying drawings, which are schematic and should not be construed as limiting the present invention in any way. In the accompanying drawings:

[0043] Figure 1 A schematic diagram showing a flow chart of a battery health status prediction method provided by the present invention is shown;

[0044] Figure 2 A schematic diagram showing a voltage-time decay characteristic curve in a battery health status prediction method provided by the present invention;

[0045] Figure 3 A schematic diagram showing the structure of a battery health state prediction model in a battery health state prediction method provided by the present invention;

[0046] Figure 4 A schematic diagram showing the structure of a battery health status prediction device provided by the present invention is shown;

[0047] Figure 5 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0049] Due to its advantages such as high energy density, environmental friendliness, and long lifespan, lithium batteries are important energy storage devices currently widely used in new energy fields such as mobile devices, electric vehicles, and energy storage systems.

[0050] However, after lithium batteries experience repeated charge-discharge cycles, their internal resistance will gradually increase. The increase in internal resistance will cause significant heat generation during the charge-discharge process of the battery, which in turn leads to irreversible aging and performance degradation of the lithium battery. SOH can be used to represent the remaining life of a lithium battery. As the number of charge-discharge cycles of the lithium battery increases, the SOH will decrease, that is, the remaining life of the battery and the available capacity will gradually decrease. When the SOH drops below a certain threshold, such as 80%, it is necessary to stop using or limit the use of the lithium battery. Therefore, accurately predicting the SOH of a lithium battery is extremely important for its use. It can help users understand the remaining life and capacity attenuation of the lithium battery. At the same time, SOH is also of great help for accurately predicting other related battery health indicators, such as the state of charge (SOC) of a lithium battery. Currently, machine learning-based methods, such as neural network models, are widely used to predict the SOH of lithium batteries.

[0051] The process of predicting the SOH of a lithium battery using a common neural network model is as follows:

[0052] Feature selection: Select reasonable features to describe the state of the lithium battery and select features that have a key impact on the prediction of the SOH of the lithium battery. The above features include the voltage, current, temperature, number of charge-discharge cycles, battery capacity, etc. of the lithium battery;

[0053] Data collection: Collect and record the feature data of the lithium battery under different working states;

[0054] Data preprocessing: Preprocess the collected data. The above preprocessing includes data cleaning, denoising, missing value processing, and feature extraction, etc.;

[0055] Model selection and training: Select a suitable neural network model for training, and use evaluation indicators such as root mean square error (RSME) and mean absolute error (MAE) to evaluate the performance of the neural network model and adjust the parameters of the model;

[0056] SOH prediction: Use the trained neural network model to predict the lithium battery to be evaluated and obtain the SOH prediction result of the lithium battery.

[0057] During the charge and discharge cycle of lithium batteries, there is a specific time node in the discharge process. Before and after this specific time node, there is a significant difference in the rate of voltage drop. This specific time node is called the inflection point of the discharge voltage-time characteristic curve. The inflection point voltage will gradually decrease as the lithium battery ages and its performance declines.

[0058] It is precisely because of the existence of this inflection point that the data distribution of the collected lithium battery charge and discharge cycle process is uneven, which is specifically manifested as follows: the voltage at the time nodes before the inflection point drops slowly and the data distribution is relatively dense, while the voltage at the time nodes after the inflection point drops rapidly and the data distribution is relatively sparse. The above-mentioned voltage change characteristics are not conducive to the application of neural network models in the field of battery health status prediction. This imbalance may cause the neural network model to pay too much attention to the data features of dense areas during training, while ignoring the important information of sparse areas, resulting in the neural network model failing to fully learn the data features of sparse areas based on these data, and thus limiting the generalization performance of new data.

[0059] Therefore, how to improve the accuracy of neural network models in predicting lithium battery SOH is an important issue that needs to be urgently addressed in the industry.

[0060] In order to solve the above problems, a battery health state prediction method is provided in this specification, which aims to solve the above-mentioned data imbalance problem and effectively complete the task of battery health state prediction based on battery data. The battery health state prediction method provided in this specification can be applied to electronic devices capable of predicting the SOH of lithium batteries. The electronic devices may include notebooks, desktop computers, smart phones, smart wearable devices (virtual reality glasses, smart watches, etc.), tablet computers, etc. Of course, the battery health state prediction method provided in this specification can also be applied to applications running in the above-mentioned electronic devices. For example, the battery health state prediction method can be applied to a browser capable of predicting the SOH of lithium batteries, and can also be applied to software capable of predicting the SOH of lithium batteries. Figure 1 FIG. 1 is a flow chart of a method for predicting a battery health state according to an embodiment of the present invention. Figure 1As shown, the method may include the following steps:

[0061] S10. Obtain the charge-discharge cycle data of the lithium battery to be detected.

[0062] In this embodiment, the charge-discharge cycle data of the lithium battery to be detected may be stored in the electronic device in advance, or obtained by the electronic device from the outside world. For example, the electronic device obtains it from a collection device outside. Here, there is no limitation on the specific acquisition form of the charge-discharge cycle data, as long as it is ensured that the electronic device can obtain the charge-discharge cycle data.

[0063] S20. Input the charge-discharge cycle data into the trained battery health state prediction model to obtain the battery life result output by the battery health state prediction model.

[0064] Among them, the battery health state prediction model is trained by using a supervised deep learning method based on the sample fusion features of the sample lithium battery and the battery health state of each sample charge-discharge cycle. The sample fusion features are the input data used for training, and the battery health state of each sample charge-discharge cycle is the label used for training. The sample fusion features are extracted from the sample charge-discharge cycle data of the sample lithium battery by identifying the inflection points in the charge-discharge cycle process based on the time-voltage decay relationship and mining the differential features hidden in the data. The battery health state of each sample charge-discharge cycle is determined according to the ratio of the battery capacity of the sample charge-discharge cycle to the battery capacity of the initial sample charge-discharge cycle.

[0065] The battery health state prediction model uses the battery health state represented by the lithium battery cycle data as the label, and this index is correlated with the capacity of the battery in the current cycle. Through learning the sample data, the battery health state prediction model can predict the battery health state of the current charge-discharge cycle of the lithium battery.

[0066] In this embodiment, the battery health state prediction model is trained through the following steps:

[0067] A10. Obtain the sample charge-discharge cycle data of the sample lithium battery, preprocess the sample charge-discharge cycle data, and screen out the sample features of each sample charge cycle from the sample charge-discharge cycle data.

[0068] In this embodiment, during preprocessing, the sample features will be screened out from the sample charge-discharge cycle data first. The sample features include: temperature, current, voltage, power, time, battery capacity, etc.

[0069] Similarly, in this embodiment, the sample charge-discharge cycle data of the sample lithium battery can also be stored in the electronic device in advance, or can be obtained by the electronic device from the outside world. For example, the electronic device obtains it from an external acquisition device. There is no limitation on the specific acquisition form of the sample charge-discharge cycle data here, as long as it is ensured that the electronic device can obtain the sample charge-discharge cycle data. By fully collecting the sample charge-discharge cycle data of the offline sample lithium battery as sample data, it provides a large amount of data support for subsequent training.

[0070] A20. Determine the battery health state of each sample charge-discharge cycle according to the ratio of the battery capacity of each sample charge-discharge cycle to the battery capacity of the initial sample charge-discharge cycle. Specifically:

[0071]

[0072] Among them, represents the battery health state of the th sample charge-discharge cycle; represents the battery capacity of the initial sample charge-discharge cycle, that is, the first sample charge-discharge cycle; represents the battery capacity of the current charge-discharge cycle.

[0073] The battery health state prediction model will use the battery health state of the th sample charge-discharge cycle as the label of the sample data, and is obtained based on the initial discharge capacity of the selected sample lithium battery sample charge-discharge cycle data, this key indicator directly reflects the health state of the lithium battery. By obtaining the ratio between the discharge capacity in each charge-discharge cycle and the discharge capacity of the starting cycle, it provides label information for the battery health state prediction model, and thus a supervised learning model is established.

[0074] A30. Construct a voltage-time decay characteristic curve for each sample charge-discharge cycle according to the sample characteristics of each sample charge-discharge cycle.

[0075] After data preprocessing, the sample charge-discharge cycle data only contains sample characteristics, which meets the data analysis requirements. Subsequently, calculate and determine the voltage-time decay characteristic curve within each sample charge-discharge cycle period, and provide time information for the lithium battery data within a single sample charge-discharge cycle.

[0076] A40. Combine the sample characteristics with the same voltage value in the voltage-time decay characteristic curves of each sample charge-discharge cycle period to obtain voltage grouped characteristics, and obtain the derivative characteristics of the voltage grouped characteristics, and perform feature fusion of the voltage grouped characteristics and the derivative characteristics to obtain sample fusion characteristics.

[0077] Since the sample characteristics in each sample charge-discharge cycle include characteristics such as temperature, current, voltage, power, time, and battery capacity, and according to the voltage-time decay characteristic curve corresponding to each sample charge-discharge cycle, the time corresponding to the starting point of the inflection point and the time corresponding to the inflection point can be calculated.

[0078] Please refer to Figure 2 , taking the preset empirical curve obtained from expert experience as a benchmark. Next, use the partial least squares regression method to fit the actually measured voltage-time decay characteristic curve with the initialized preset empirical curve, and identify the turning points on the preset empirical curve. These turning points on the preset empirical curve often correspond to the inflection points or the starting points of the inflection points of the voltage-time decay characteristic curve.

[0079]

[0080] Among them, represents the initialized preset empirical curve; represents the voltage-time decay characteristic curve within a complete sample charge-discharge cycle; represents a random variable that follows a normal distribution and is centered around zero representing the residual; represents the inflection point of the voltage-time decay characteristic curve; represents the starting point of the inflection point of the voltage-time decay characteristic curve; represents the intercept at represents the first slope for adjusting the intersecting line, represents the second slope for adjusting the intersecting line, represents the third slope for adjusting the intersecting line; represents a constant for controlling the suddenness of the transition, which is initialized to a relatively low value; represents the hyperbolic tangent function. The two turning points of the fitted curve model are considered to be the starting point of the inflection point and the inflection point of the voltage-time decay characteristic curve.

[0081] For the voltage data before the inflection point, the data distribution is relatively dense, and multiple data points are recorded repeatedly at the same voltage value. In this embodiment, part of the voltage-time decay characteristic curve shows that the voltage drops very gently before the starting point of the inflection point, approaching a straight line, and there are too many identical values at the same time, which results in duplicate data in a relatively small data window. In order to improve the data processing efficiency and solve the data density situation, the sample characteristics at the same voltage value are merged into a group, and the average values of characteristics such as temperature, current, voltage, and power at this voltage value are calculated to obtain the voltage grouped characteristics, and the voltage grouped characteristics are used as the representative data at this voltage level.

[0082] Meanwhile, to comprehensively extract the hidden differential features in the data, each voltage data point is analyzed in depth. The voltage change amount, which is the difference between each voltage value and the initial voltage, and the time offsets of these voltage data points relative to the starting point and the inflection point of the inflection point are calculated. These derived features amplify the differences between the data and make the data set more compact. Further, feature fusion is performed with the original voltage grouping features. The fusion process can integrate information from multiple dimensions and jointly constitute a rich and comprehensive fused feature data set, which is used as input data to provide a data basis for the subsequent prediction of battery life.

[0083] Feature fusion is performed on the voltage grouping features and the derived features of potential differences such as the time difference from the starting point of the inflection point, the time difference from the inflection point, and the voltage difference of the starting voltage, to obtain the input data for training.

[0084] A50. Using the sample fusion features as the input data for training, and the battery health state of each sample charge-discharge cycle as the label, a supervised deep learning method is used for training to obtain a battery health state prediction model for predicting the battery life result of a lithium battery.

[0085] In this embodiment, the battery health state prediction model is a composite neural network model that combines a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM). The input data passes through the CNN and LSTM structures in sequence, and finally returns the predicted battery life result through a fully connected layer. That is, the battery health state prediction model includes a convolutional neural network layer and a long short-term memory network layer. The convolutional neural network layer is used to extract high-level local features of the charge-discharge cycle data, and the long short-term memory network layer is used to capture the long-term dependence relationship of the local features in the time series. The composite neural network model of CNN and LSTM can not only capture spatial local features but also mine the time information in the sequence features, and at the same time effectively prevent the model from overfitting, and the model has good battery health state prediction ability.

[0086] Please refer to Figure 3 , this battery health state prediction model is a residual network structure. In addition to the above-mentioned convolutional neural network layer and long short-term memory network layer, it also includes an identity mapping layer for skip connection in the residual network, a feature fusion layer, a second fully connected layer for learning the non-linear combination between features, and a rectified linear layer (ReLU activation function layer) for enhancing the non-linear ability of the features.

[0087] Specifically, the convolutional neural network layer (CNN layer) includes a convolutional layer, a max pooling layer, a flatten layer, and a first fully connected layer connected in sequence. The convolutional layer is used to perform convolutional processing on the fused features of the input samples. The convolutional layer is a one-dimensional network structure, including three convolutional kernels, and the size of each convolutional kernel is 3, and the stride is 1. The max pooling layer is used to perform pooling processing on the maximum value in each region of the convolutional features input by the convolutional layer. The window size of the max pooling layer is 3, and the stride is 1. A smaller stride can maximize the retention of sequence information. The max pooling layer reduces the spatial dimension of the data by selecting the maximum value within the window, helping the model focus on important features and reducing the computational amount. The flatten layer is used to flatten the pooled features output by the max pooling layer, realizing the conversion of multi-dimensional features into one-dimensional features for transition to the subsequent first fully connected layer. The first fully connected layer is used to integrate the flattened features input by the flatten layer to obtain the first output features. The CNN layer aims to capture local features such as trends and patterns from the fused features of the samples.

[0088] The long short-term memory network layer (LSTM layer) includes three relation capture layers connected in sequence, namely the first, second, and third relation capture layers. The input of the first relation capture layer of the LSTM layer comes from the output of the convolutional neural network layer, and the input of the subsequent two layers comes from the output of the previous layer. The number of neurons in the hidden layer of the LSTM layer is 200. The output of the LSTM layer learns the non-linear combination in the features through two fully connected layers and finally outputs the prediction result. In this way, the LSTM layer learns and remembers the information in the neurons through the gating mechanism (including the forget gate, input gate, output gate, etc.), and can effectively learn the complex features and long-term dependence relationships in the sequence data through stacking multiple layers of LSTM networks.

[0089] Specifically, the first relation capture layer is used to perform the first feature extraction and abstraction processing on the first output features input by the second connection layer to obtain the first abstract features. The second relation capture layer is used to perform the second feature extraction and abstraction processing on the first abstract features input by the first relation capture layer to obtain the second abstract features. The third relation capture layer is used to perform the third feature extraction and abstraction processing on the second abstract features input by the second relation capture layer to obtain the third abstract features. The second and third relation capture layers both use the output of the previous layer to further extract and abstract features, and finally generate a fixed-length vector containing the entire sequence information.

[0090] For the entire battery health state prediction model, its processing process is as follows:

[0091] The battery health state prediction model has two branches: The first branch is that the charge and discharge cycle data as input data is input into the identity mapping layer of the battery health state prediction model, and the identity mapping layer performs identity mapping processing to output identity mapping features. The second branch is that the charge and discharge cycle data as input data is input into the CNN layer of the battery health state prediction model. The CNN layer extracts local features to output the first output features, and then the first output features are input into the LSTM layer. The LSTM layer captures the dependency relationship to output the third abstract features. It should be noted that a second fully connected layer is also set after the LSTM layer in the second branch, and the purpose of the second fully connected layer is also to perform integration processing. After that, the identity mapping features output by the first branch and the integrated third abstract features output by the second branch are fused at the feature fusion layer, and then processed by the rectified linear layer that enhances the nonlinear ability of the features, and finally the battery life result of the lithium battery to be detected is output.

[0092] In order to evaluate the performance of the battery health state prediction model and optimize it, in this embodiment, the battery health state prediction model uses the Mean-Square Error (MSE) as the loss function to intuitively reflect the difference degree between the predicted value and the actual value. Specifically:

[0093]

[0094] Among them, represents the total number of sample data of the true value and the predicted value; represents the true battery health state reflected by the charge and discharge cycle data of the lithium battery, that is, the sample data, and is used as the label of the lithium battery cycle data to measure the actual state of the battery performance; represents the predicted battery health state after being processed by the fused features and input into the trained battery health state prediction model with a CNN-LSTM composite network structure.

[0095] In this embodiment, during the training process of the battery health state prediction model, the Adaptive Moment Estimation (Adam) algorithm in the gradient descent method with an adaptive learning rate is used for model training. In this way, the optimization problem of large-scale data can be effectively processed. In order to fully train the model and make it achieve a better convergence effect, the training cycle (epoch) is set to 15 to obtain better model performance.

[0096] By using data mining feature fusion strategy and combining innovative CNN and LSTM-based battery health status prediction models, detailed charge and discharge cycle data verification has been carried out in a laboratory environment for the selected A and B batteries. In this embodiment, battery cycle data is selected from the discharge cycle of 750 complete charge and discharge cycles of each battery of the above A and B models as the research data set to ensure the comprehensiveness and representativeness of the data.

[0097] More specifically, when training the model, 80% of the historical discharge cycle data collected from the selected lithium battery is allocated as a training set to train and optimize the battery health status prediction model, while the remaining 20% ​​of the data is used as an independent test set to evaluate the prediction and generalization performance of the battery health status prediction model, thereby verifying the effectiveness of the model in complex battery charge and discharge cycles.

[0098] During verification, mean square error (MSE), root mean square error (RMSE) and mean absolute error (MAE) were selected as evaluation indicators of the model. At the same time, the constructed CNN-LSTM battery health status prediction model was compared and evaluated with the standard LSMT model, where:

[0099]

[0100]

[0101]

[0102] The battery health status prediction method of the present invention predicts the battery life result of the lithium battery to be tested through a trained battery health status prediction model, wherein the battery health status prediction model locates the inflection point by accurately fitting and calculating the voltage-time decay characteristic relationship, which not only provides key time information for the data, but also can amplify the difference between repeated information before the voltage occurs, which is helpful to complete the subsequent feature derivation and feature fusion process, and secondly, divides the data according to the inflection point information to alleviate the uneven distribution of data, and simultaneously performs feature derivation and feature fusion, amplifies the difference of window data, and mines important information in sparse areas, so that the model can fully learn the data features of sparse areas, effectively improves the prediction accuracy of the model, and more accurately completes the task of battery health status prediction while effectively solving the uneven distribution of data.

[0103] The battery health state prediction device provided by an embodiment of the present invention is described below. The battery health state prediction device described below and the battery health state prediction method described above can be referenced to each other.

[0104] In order to solve the above problems, a battery health status prediction device is provided in this specification, aiming to. Figure 4Schematic structural diagram of a battery health state prediction device according to an embodiment of the present invention, as Figure 4 shown, the device may include:

[0105] A data acquisition module 10, configured to acquire charge and discharge cycle data of a lithium battery to be detected.

[0106] In this embodiment, the charge and discharge cycle data of the lithium battery to be detected may be stored in the electronic device in advance, or may be obtained by the electronic device from the outside. For example, the electronic device obtains it from an external acquisition device. Here, there is no restriction on the specific acquisition form of the charge and discharge cycle data, as long as it is ensured that the electronic device can obtain the charge and discharge cycle data.

[0107] A battery prediction module 20, configured to input the charge and discharge cycle data into a trained battery health state prediction model, and obtain a battery life result output by the battery health state prediction model.

[0108] Among them, the battery health state prediction model is trained by using a supervised deep learning method based on the sample fusion features of the sample lithium battery and the battery health state of each sample charge and discharge cycle. The sample fusion features are the input data used for training, and the battery health state of each sample charge and discharge cycle is the label used for training. The sample fusion features are extracted from the sample charge and discharge cycle data of the sample lithium battery by identifying the inflection points in the charge and discharge cycle process based on the time-voltage decay relationship and mining the implicit differential features in the data. The battery health state of each sample charge and discharge cycle is determined according to the ratio of the battery capacity of the sample charge and discharge cycle to the battery capacity of the initial sample charge and discharge cycle.

[0109] The battery health state prediction model uses the battery health state represented by the lithium battery cycle data as the label, and this index is correlated with the capacity of the battery under the current cycle. By learning the sample data, the battery health state prediction model can predict the battery health state of the current charge and discharge cycle of the lithium battery.

[0110] The battery health status prediction device of the present invention predicts the battery life result of the lithium battery to be tested through a trained battery health status prediction model, wherein the battery health status prediction model locates the inflection point by accurately fitting and calculating the voltage-time decay characteristic relationship, which not only provides key time information for the data, but also can amplify the difference between repeated information before the voltage occurs, which is helpful to complete the subsequent feature derivation and feature fusion process, and secondly, divides the data according to the inflection point information to alleviate the uneven data distribution, and simultaneously performs feature derivation and feature fusion to amplify the difference of window data and mine important information in sparse areas, so that the model can fully learn the data features of sparse areas, effectively improving the prediction accuracy of the model, while effectively solving the uneven data distribution, and more accurately completing the task of battery health status prediction.

[0111] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor 510 (processor), a communication interface 520 (Communications Interface), a memory 530 (memory) and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic command in the memory 530 to execute the battery health state prediction method, which includes:

[0112] Obtain the charge and discharge cycle data of the lithium battery to be tested;

[0113] Inputting the charge and discharge cycle data into a trained battery health state prediction model to obtain a battery life result output by the battery health state prediction model;

[0114] The battery health status prediction model is based on the sample fusion features of the sample lithium battery and the battery health status of each sample charge and discharge cycle, and is trained using a supervised deep learning method. The sample fusion features are the input data used for training, and the battery health status of each sample charge and discharge cycle is the label used for training;

[0115] The sample fusion feature is extracted from the sample charge and discharge cycle data of the sample lithium battery based on the time-voltage decay relationship to identify the inflection point in the charge and discharge cycle process and to mine the differential features implicit in the data. The battery health status of each sample charge and discharge cycle is determined based on the ratio of the battery capacity of the sample charge and discharge cycle to the battery capacity of the initial sample charge and discharge cycle.

[0116] In addition, when the logical instructions in the above-mentioned memory 530 can be implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0117] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the battery health state prediction method provided by the above-mentioned various methods. The method includes:

[0118] Obtain the charge and discharge cycle data of the lithium battery to be detected;

[0119] Input the charge and discharge cycle data into the trained battery health state prediction model to obtain the battery life result output by the battery health state prediction model;

[0120] The battery health state prediction model is trained by using a supervised deep learning method based on the sample fusion features of the sample lithium battery and the battery health state of each sample charge and discharge cycle. The sample fusion features are the input data used for training, and the battery health state of each sample charge and discharge cycle is the label used for training;

[0121] The sample fusion features are extracted from the sample charge and discharge cycle data of the sample lithium battery by identifying the inflection points in the charge and discharge cycle process based on the time-voltage decay relationship and mining the implicit differential features in the data. The battery health state of each sample charge and discharge cycle is determined according to the ratio of the battery capacity of the sample charge and discharge cycle to the battery capacity of the initial sample charge and discharge cycle.

[0122] On another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is used to execute the battery health state prediction method provided by the above-mentioned various methods. The method includes:

[0123] Obtain the charge and discharge cycle data of the lithium battery to be detected;

[0124] Input the charge and discharge cycle data into the trained battery health state prediction model to obtain the battery life result output by the battery health state prediction model;

[0125] The battery health state prediction model is trained by using a supervised deep learning method based on the sample fusion features of the sample lithium battery and the battery health state of each sample charge and discharge cycle. The sample fusion features are the input data used for training, and the battery health state of each sample charge and discharge cycle is the label used for training;

[0126] The sample fusion features are extracted from the sample charge and discharge cycle data of the sample lithium battery by identifying the inflection points in the charge and discharge cycle process based on the time-voltage decay relationship and mining the implicit differential features in the data. The battery health state of each sample charge and discharge cycle is determined according to the ratio of the battery capacity of the sample charge and discharge cycle to the battery capacity of the initial sample charge and discharge cycle.

[0127] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0128] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0129] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting battery health status, characterized in that: The method comprises: Obtain the charge and discharge cycle data of the lithium battery to be tested; Inputting the charge and discharge cycle data into a trained battery health state prediction model to obtain a battery life result output by the battery health state prediction model; The battery health status prediction model is based on the sample fusion features of the sample lithium battery and the battery health status of each sample charge and discharge cycle, and is trained using a supervised deep learning method. The sample fusion features are the input data used for training, and the battery health status of each sample charge and discharge cycle is the label used for training; The sample fusion feature is extracted from the sample charge and discharge cycle data of the sample lithium battery based on the time-voltage decay relationship to identify the inflection point in the charge and discharge cycle process and to mine the difference characteristics implicit in the data. The battery health status of each sample charge and discharge cycle is determined based on the ratio of the battery capacity of the sample charge and discharge cycle to the battery capacity of the initial sample charge and discharge cycle. The battery health status prediction model is trained by the following steps: Obtaining sample charge and discharge cycle data of a sample lithium battery, preprocessing the sample charge and discharge cycle data, and filtering out sample features of each sample charge cycle from the sample charge and discharge cycle data; the sample features include: temperature, current, voltage, power, time, and battery capacity; Determine the battery health status of each sample charge and discharge cycle according to the ratio of the battery capacity of each sample charge and discharge cycle to the battery capacity of the initial sample charge and discharge cycle; According to the sample characteristics of each sample charge and discharge cycle, a voltage-time decay characteristic curve of each sample charge and discharge cycle is constructed; Merge sample features of the same voltage value in the voltage-time decay characteristic curve of each sample charge-discharge cycle to obtain voltage grouping features, obtain derived features of the voltage grouping features, perform feature fusion of the voltage grouping features and the derived features, and obtain sample fusion features; The sample fusion features are used as input data for training, and the battery health status of each sample charge and discharge cycle is used as a label. A supervised deep learning method is used for training to obtain a battery health status prediction model for predicting the battery life results of lithium batteries. The method of combining sample features of the same voltage value in the voltage-time decay characteristic curve of each sample charge and discharge cycle to obtain voltage grouping features, obtaining derived features of the voltage grouping features, and performing feature fusion of the voltage grouping features and the derived features to obtain sample fusion features specifically includes: Fitting each voltage-time decay characteristic curve respectively through a preset experience curve, identifying the turning point on the preset experience curve, and obtaining the inflection point and the inflection starting point of each voltage-time decay characteristic curve; Merge the sample features with the same voltage value in the voltage-time decay characteristic curve of each sample charge-discharge cycle to obtain the voltage grouping feature; Determine the difference between each voltage value in the voltage grouping feature and the initial voltage of the voltage-time decay characteristic curve to obtain a voltage difference, and determine the time offset of each voltage value in the voltage grouping feature relative to the inflection point and the starting point of the inflection point of the voltage-time decay characteristic curve to obtain a first time difference and a second time difference; the derived feature includes the voltage difference, the first time difference and the second time difference; The voltage grouping feature and the derived feature are fused to obtain the sample fusion feature.

2. The battery health status prediction method according to claim 1, characterized in that: The battery health status prediction model is a composite neural network model that combines a convolutional neural network and a long short-term memory network; The battery health status prediction model includes a convolutional neural network layer and a long short-term memory network layer. The convolutional neural network layer is used to extract local features of the charge and discharge cycle data, and the long short-term memory network layer is used to capture the dependency of the local features on the time series.

3. The battery health status prediction method according to claim 2, characterized in that: The convolutional neural network layer includes a sequentially connected convolutional layer, a maximum pooling layer, a data flattening layer and a first fully connected layer; The long short-term memory network layer includes a first relationship capture layer, a second relationship capture layer and a third relationship capture layer which are connected in sequence.

4. The battery health status prediction method according to claim 3, characterized in that: The battery health status prediction model also includes an identity mapping layer, a feature fusion layer and a corrected linear layer of a residual network structure; The charge and discharge cycle data are processed by the identity mapping layer, the convolutional neural network layer and the long short-term memory network layer in sequence, and then processed by the feature fusion layer and the corrected linear layer in sequence to obtain the battery life result.

5. The battery health status prediction method according to claim 4, characterized in that: The step of inputting the charge and discharge cycle data into a trained battery health state prediction model to obtain a battery life result output by the battery health state prediction model specifically includes: Inputting the charge-discharge cycle data into the identity mapping layer to obtain an identity mapping feature outputted by the identity mapping layer after identity processing; Inputting the charge-discharge cycle data into the convolutional neural network layer to obtain a first output feature output by the convolutional neural network layer; Inputting the first output feature into a long short-term memory network layer to obtain a third abstract feature output by the long short-term memory network layer; The identity mapping feature and the third abstract feature are input into the feature fusion layer for feature fusion, and the battery life result is obtained after being processed by the corrected linear layer.

6. A battery health status prediction device, characterized in that: The device comprises: A data acquisition module, used to acquire the charge and discharge cycle data of the lithium battery to be tested; A battery prediction module, used to input the charge and discharge cycle data into a trained battery health state prediction model to obtain a battery life result output by the battery health state prediction model; The battery health status prediction model is based on the sample fusion features of the sample lithium battery and the battery health status of each sample charge and discharge cycle, and is trained using a supervised deep learning method. The sample fusion features are the input data used for training, and the battery health status of each sample charge and discharge cycle is the label used for training; The sample fusion feature is extracted from the sample charge and discharge cycle data of the sample lithium battery based on the time-voltage decay relationship to identify the inflection point in the charge and discharge cycle process and to mine the difference characteristics implicit in the data. The battery health status of each sample charge and discharge cycle is determined based on the ratio of the battery capacity of the sample charge and discharge cycle to the battery capacity of the initial sample charge and discharge cycle. The battery health status prediction model is trained by the following steps: Obtaining sample charge and discharge cycle data of a sample lithium battery, preprocessing the sample charge and discharge cycle data, and filtering out sample features of each sample charge cycle from the sample charge and discharge cycle data; the sample features include: temperature, current, voltage, power, time, and battery capacity; Determine the battery health status of each sample charge and discharge cycle according to the ratio of the battery capacity of each sample charge and discharge cycle to the battery capacity of the initial sample charge and discharge cycle; According to the sample characteristics of each sample charge and discharge cycle, a voltage-time decay characteristic curve of each sample charge and discharge cycle is constructed; Merge sample features of the same voltage value in the voltage-time decay characteristic curve of each sample charge-discharge cycle to obtain voltage grouping features, obtain derived features of the voltage grouping features, perform feature fusion of the voltage grouping features and the derived features, and obtain sample fusion features; The sample fusion features are used as input data for training, and the battery health status of each sample charge and discharge cycle is used as a label. A supervised deep learning method is used for training to obtain a battery health status prediction model for predicting the battery life results of lithium batteries. The method of combining sample features of the same voltage value in the voltage-time decay characteristic curve of each sample charge and discharge cycle to obtain voltage grouping features, obtaining derived features of the voltage grouping features, and performing feature fusion of the voltage grouping features and the derived features to obtain sample fusion features specifically includes: Fitting each voltage-time decay characteristic curve respectively through a preset experience curve, identifying the turning point on the preset experience curve, and obtaining the inflection point and the inflection starting point of each voltage-time decay characteristic curve; Merge the sample features with the same voltage value in the voltage-time decay characteristic curve of each sample charge-discharge cycle to obtain the voltage grouping feature; Determine the difference between each voltage value in the voltage grouping feature and the initial voltage of the voltage-time decay characteristic curve to obtain a voltage difference, and determine the time offset of each voltage value in the voltage grouping feature relative to the inflection point and the starting point of the inflection point of the voltage-time decay characteristic curve to obtain a first time difference and a second time difference; the derived feature includes the voltage difference, the first time difference and the second time difference; The voltage grouping feature and the derived feature are fused to obtain the sample fusion feature.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the battery health status prediction method according to any one of claims 1 to 5 are implemented.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the battery health status prediction method as claimed in any one of claims 1 to 5 are implemented.

Images