Model training method, prediction method and device for predicting SOH and RUL of battery

Abstract

The invention provides a model training method, a prediction method and equipment for predicting SOH and RUL of a battery, and the training method comprises the steps: obtaining performance monitoring data and corresponding label data in the historical charging and discharging process of the battery, the performance monitoring data comprises current data, voltage data and temperature data, and the corresponding label data comprises the current data, the voltage data and the temperature data; the label data comprises a battery capacity truth value and a residual cycle index truth value; outputting a battery capacity prediction value and a residual cycle index prediction value through a multi-target prediction model based on the performance monitoring data; and on the basis of the battery capacity prediction value, the residual cycle index prediction value and the label data, generating a loss value through a loss function of the multi-target prediction model, updating the multi-target prediction model until convergence so as to obtain a trained multi-target prediction model, and predicting the battery capacity and the residual cycle index. According to the method, multi-target joint training is adopted, the potential relevance between the two predicted targets can be captured, and the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of battery technology, and in particular to a model training method, a prediction method and a device for predicting battery SOH and RUL. Background Art

[0002] The battery state of health (SOH) reflects the health and remaining service life of the battery. Battery SOH is an important indicator of the battery, and its importance is reflected in the following aspects: reflecting the degree of battery aging, indicating the battery replacement time, evaluating the remaining capacity, optimizing battery management, analyzing battery failures, evaluating battery safety and predicting the remaining service life.

[0003] The remaining useful life (RUL) of a battery is an important indicator for evaluating the health status of a battery. Common battery RULs include the number of remaining battery cycles. Battery RUL is also an important indicator of a battery, which is mainly reflected in several aspects: evaluating the health status of a battery, guiding battery use and maintenance, warning of battery failures, evaluating battery capacity degradation, optimizing battery management strategies, and evaluating the secondary utilization value of a battery. Therefore, predicting battery SOH and RUL has a high application value for judging the performance of a battery.

[0004] The current, voltage and temperature of the battery are closely related to SOH and RUL. Using the current, voltage and temperature of the battery to predict SOH and RUL is a research hotspot. Usually, machine learning models such as neural networks and support vector machines are trained using current, voltage and temperature data to learn the changing patterns of battery SOH and RUL and make predictions. However, a separate prediction model is trained for each prediction target, which is time-consuming and labor-intensive, and has high maintenance costs. Summary of the invention

[0005] The present invention provides a model training method, prediction method and device for predicting battery SOH and RUL. The battery capacity and remaining cycle number of the battery are predicted simultaneously based on a multi-objective prediction model. By adopting multi-objective joint training, the model can capture the potential correlation between the two prediction targets, improve the accuracy of the prediction, and utilize the feature representation shared by the battery current, voltage and temperature data to improve the prediction performance of the two targets and reduce the risk of model overfitting.

[0006] The purpose of the present invention is achieved by the following technical solutions: In a first aspect, the present invention provides a model training method for predicting battery SOH and RUL, comprising: Acquire the performance monitoring data and corresponding label data of the battery in the historical charging and discharging process, wherein the performance monitoring data includes: current data, voltage data and temperature data, and the label data includes: the true value of the battery capacity and the true value of the number of remaining cycles; Based on the performance monitoring data, outputting a battery capacity prediction value and a remaining cycle number prediction value via a multi-objective prediction model; Based on the battery capacity prediction value, the remaining cycle number prediction value and the label data, a loss value is generated via a loss function of the multi-objective prediction model, and the multi-objective prediction model is updated until convergence to obtain the trained multi-objective prediction model; The multi-objective prediction model is used to predict the battery capacity and remaining cycle number of the battery in the current state, so as to serve as data indicators for evaluating the battery SOH and RUL.

[0007] According to a model training method for predicting battery SOH and RUL provided by the present invention, the multi-objective prediction model includes: a convolutional neural network, a bidirectional long short-term memory network, a Transformer model and a prediction network; The step of outputting a battery capacity prediction value and a remaining cycle number prediction value based on the performance monitoring data via a multi-objective prediction model comprises: Inputting the performance monitoring data into the convolutional neural network, and outputting spatial feature data of the performance monitoring data; Inputting the spatial feature data into the bidirectional long short-term memory network, and outputting the temporal feature data of the performance monitoring data; Inputting the time series feature data into the Transformer model, and outputting global feature data represented by different subspace features of the performance monitoring data; The global feature data is input into the prediction network, and a battery capacity prediction value and a remaining cycle number prediction value are output.

[0008] According to a model training method for predicting battery SOH and RUL provided by the present invention, the convolutional neural network includes: at least two convolutional layers and one pooling layer; The step of inputting the performance monitoring data into the convolutional neural network and outputting the spatial feature data of the performance monitoring data comprises: Inputting the performance monitoring data into the convolution layer, learning and identifying key features of the performance monitoring data through convolution operation, and generating convolution calculation results; The convolution calculation result is input into the pooling layer, a maximum pooling operation is performed on the convolution calculation result, and the maximum pooling calculation result is flattened to obtain the spatial feature data.

[0009] According to a model training method for predicting battery SOH and RUL provided by the present invention, the bidirectional long short-term memory network includes: a forward LSTM network and a backward LSTM network; The step of inputting the spatial feature data into the bidirectional long short-term memory network and outputting the temporal feature data of the performance monitoring data comprises: At the same time, the forward LSTM network obtains the forward hidden state of the spatial feature data according to the forward time sequence, and the backward LSTM network obtains the backward hidden state of the spatial feature data according to the reverse time sequence; The forward hidden state and the backward hidden state are concatenated or added as the output state at the moment, thereby obtaining the context information of the performance monitoring data as the time series feature data.

[0010] According to a model training method for predicting battery SOH and RUL provided by the present invention, the Transformer model includes: a position encoding layer, a multi-head attention mechanism network and a feedforward fully connected network; The step of inputting the time series feature data into the Transformer model and outputting the global feature data represented by different subspace features of the performance monitoring data comprises: The position encoding layer sequentially adds position vectors to the time series feature data in time steps, and adds the position vectors to the feature vectors of the time series feature data to provide sequence information for the time series feature data; The multi-head attention mechanism network learns the correlation between the time series feature data at different positions to capture the influence relationship between the current data, the voltage data and the temperature data at different time steps to obtain globally dependent feature information; The feedforward fully connected network extracts and fuses feature information at different positions, obtains the nonlinear relationship between the current data, the voltage data and the temperature data through nonlinear transformation, and obtains a higher-level feature representation as the feature representation of the performance monitoring data in different subspaces.

[0011] According to a model training method for predicting battery SOH and RUL provided by the present invention, the prediction network includes: a fully connected layer and an output layer; The step of inputting the global feature data into the prediction network and outputting a battery capacity prediction value and a remaining cycle number prediction value comprises: Inputting the global feature data into the fully connected layer for feature integration and transformation to generate a feature representation suitable for outputting a prediction value; The feature representation is input into the output layer, and the target value is predicted through two linear regression layers set in the output layer, so as to simultaneously output the battery capacity prediction value and the remaining cycle number prediction value.

[0012] In a second aspect, the present invention provides a method for predicting battery SOH and RUL, comprising: Obtain current data, voltage data and temperature data of the battery to be tested in the current state; The current data, the voltage data and the temperature data are input into a pre-trained multi-objective prediction model to obtain a battery capacity prediction value and a remaining cycle number prediction value output by the multi-objective prediction model, which are used as data indicators for evaluating the battery SOH and RUL. The pre-trained multi-objective prediction model is trained by any of the model training methods for predicting the battery SOH and RUL described above.

[0013] In a third aspect, the present invention provides an electronic device, the electronic device comprising: One or more processors; and a memory storing computer program instructions, wherein the computer program instructions are executed by the processor to implement the steps of the model training method for predicting the battery SOH and RUL as described above, or implement the steps of the battery SOH and RUL prediction method as described above.

[0014] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program / instruction stored thereon, which, when executed by a processor, implements the steps of a model training method for predicting battery SOH and RUL as described in any one of the above, or implements the steps of a method for predicting battery SOH and RUL as described above.

[0015] In a fifth aspect, the present invention provides a computer program product, comprising a computer program / instruction, characterized in that when the computer program / instruction is executed by a processor, it implements the steps of the model training method for predicting the battery SOH and RUL as described in any one of the above, or implements the steps of the battery SOH and RUL prediction method as described above.

[0016] The present invention provides a model training method, prediction method and device for predicting battery SOH and RUL, by obtaining performance monitoring data and corresponding label data in the historical charging and discharging process of the battery, wherein the performance monitoring data includes: current data, voltage data and temperature data, and the label data includes: battery capacity true value and remaining cycle number true value; based on the performance monitoring data, output the battery capacity prediction value and the remaining cycle number prediction value through a multi-objective prediction model; based on the battery capacity prediction value, the remaining cycle number prediction value and the label data, generate a loss value through the loss function of the multi-objective prediction model, and update the multi-objective prediction model until convergence to obtain a trained multi-objective prediction model; the multi-objective prediction model is used to predict the battery capacity and the remaining cycle number of the battery in the current state, as a data indicator for evaluating the battery SOH and RUL. The present invention adopts multi-objective joint training, can extract the spatial characteristics and time-related characteristics of current, voltage and temperature data, and simultaneously learn the time series variation law of current, voltage and temperature data, and model the time-related pattern during the long-term use of the battery to simultaneously predict the battery capacity and the remaining cycle number of the battery, so that the feature extraction ability and robustness of the model are stronger. At the same time, the multi-objective prediction model can capture the potential correlation between the two prediction targets, improve the accuracy of the prediction, and use the shared feature representations between the battery current, voltage and temperature data to improve the prediction performance of the two targets, reducing the risk of model overfitting. Therefore, the model takes into account the comprehensive advantages of multi-objective, feature sharing, accuracy, efficiency and robustness, and is a high-performance and efficient SOH and RUL indicator prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other implementation methods can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 It is a flow chart of a model training method for predicting battery SOH and RUL provided in an embodiment of the present invention; Figure 2 is a model structure diagram of a multi-objective prediction model provided by an embodiment of the present invention; Figure 3 It is a schematic diagram of a process of training a multi-objective prediction model to predict battery SOH and RUL provided by an embodiment of the present invention; Figure 4 Schematic diagram of convolution calculation of a convolutional neural network provided by an embodiment of the present invention; Figure 5is a model structure diagram of a bidirectional long short-term memory network provided by an embodiment of the present invention; Figure 6 is a model structure diagram of the Transformer model provided by an embodiment of the present invention; Figure 7 It is a flow chart of a method for predicting battery SOH and RUL provided by an embodiment of the present invention; Figure 8 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0019] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following specifically cites a preferred embodiment and describes it in detail with the accompanying drawings as follows.

[0020] It should be noted that it is explicitly and implicitly understood by those of ordinary skill in the art that the embodiments described in the present invention can be combined with other embodiments without conflict. Unless otherwise defined, the technical terms or scientific terms involved in the present invention should be the usual meanings understood by people with ordinary skills in the technical field to which the present invention belongs. The words "one", "a", "a", "the" and the like involved in the present invention do not indicate a quantitative limitation and may represent the singular or plural. The terms "including", "comprising", "having" and any of their variations involved in the present invention are intended to cover non-exclusive inclusions; the terms "first", "second", "third", etc. involved in the present invention are merely to distinguish similar objects and do not represent a specific ordering of objects.

[0021] When it comes to method steps, the sequence illustrated in this document represents an exemplary solution, but does not mean a limitation on the sequence.

[0022] Convolution neural network (CNN), CNN network can be used to extract spatial features of current, voltage and temperature data.

[0023] Bidirectional Long Short-Term Memory (BiLSTM) network can simultaneously learn the time series change rules of current, voltage and temperature from the forward sequence and the direction sequence.

[0024] Transformer model,The Transformer model can use a multi-head attention mechanism and a feedforward network to learn the feature representation of input data.

[0025] The applicant has found that in the prior art, a separate prediction model is trained for each prediction target for monitoring battery performance, and each prediction model is used to predict each performance indicator separately. The prediction process is time-consuming and labor-intensive, and the maintenance cost is also high. Therefore, the present invention provides a model training method, prediction method and device for predicting battery SOH and RUL, which adopts multi-objective joint training to simultaneously predict the SOH and RUL performance indicators of the battery in one model, thereby greatly reducing the model training and maintenance costs and improving the prediction accuracy of the model.

[0026] Embodiment 1 Reference Figure 1 As shown, this embodiment provides a model training method for predicting battery SOH and RUL, including: Step S1: Acquire the performance monitoring data and corresponding label data of the battery during the historical charge and discharge process, wherein the performance monitoring data includes: current data, voltage data and temperature data, and the label data includes: the true value of the battery capacity and the true value of the remaining number of cycles.

[0027] Exemplarily, a charge and discharge experiment may be performed on a sample battery pack in advance, and performance monitoring data corresponding to multiple charge and discharge processes of the sample battery may be recorded, and the performance test data may be used as a training data set.

[0028] Specifically, the battery management system (BMS) can collect the current data, voltage data, temperature data, battery capacity and remaining cycle number of the battery pack. Each performance data can be collected in the following ways: Current data collection: BMS detects the total current at the battery pack connection end using a Hall sensor or a high-precision sampling resistor.

[0029] Voltage data collection: BMS detects the voltage of each battery cell through the battery isolation circuit, thereby obtaining the voltage information of the entire battery pack.

[0030] Temperature data collection: BMS sets a temperature sensor at the battery pack connection end to detect the operating temperature of the battery. A temperature sensor is also set in the battery cell.

[0031] Battery capacity collection: BMS will count the ampere-hours of battery charging and discharging, and combine it with the battery voltage change curve to calculate the battery's charge and discharge capacity.

[0032] Remaining cycle number collection: BMS can calculate the number of charge and discharge cycles completed by the battery by monitoring battery parameters such as voltage, current and time, and then calculate the remaining number of cycles.

[0033] Step S2: Based on the performance monitoring data, the battery capacity prediction value and the remaining cycle number prediction value are output through a multi-objective prediction model.

[0034] Exemplarily, the multi-objective prediction model includes: a convolutional neural network, a bidirectional long short-term memory network, a Transformer model and a prediction network.

[0035] like Figure 2 As shown in the figure, the input performance monitoring data is the battery current data, voltage data and temperature data, specifically the measured voltage, measured current, measured temperature, load current and load voltage. The input performance monitoring data passes through a 2-layer convolution layer (Convolution) structure, whose activation function is relu. It is necessary to perform a maximum pooling operation on the convolution calculation result, and then perform a flatten operation on the maximum pooling calculation result, that is, flatten it; then, the feature vector of the flattening layer is input into the bidirectional long short-term memory network (BiLSTM structure), whose activation function is relu, and the output result of the BiLSTM network is input into the Transformer model; then, the output result of the Transformer model is connected to a fully connected layer; finally, through the linear regression layer, the battery capacity prediction value and the remaining cycle number prediction value are output simultaneously.

[0036] The present invention combines a convolutional neural network (CNN), a bidirectional long short-term memory network (BiLSTM) and a Transformer model to construct a multi-objective prediction model, which simultaneously predicts the battery capacity and the remaining number of cycles based on the battery's current data, voltage data and temperature data.

[0037] like Figure 3 As shown in the figure, the CNN network extracts the spatial features of current data, voltage data, and temperature data; the BiLSTM network simultaneously learns the time series change rules of current, voltage, and temperature from the forward sequence and the reverse sequence; the Transformer model can model the time-related patterns of the battery during long-term use. Therefore, the multi-target prediction model composed of CNN-BiLSTM-Transformer has stronger feature extraction capabilities and robustness.

[0038] In this embodiment, step S2 specifically includes: Step S201: input the performance monitoring data into a convolutional neural network, and output the spatial feature data of the performance monitoring data.

[0039] Exemplarily, the convolutional neural network includes: at least two convolutional layers and one pooling layer; step S201 specifically includes: Step S2011: input the performance monitoring data into the convolution layer, learn and identify the key features of the performance monitoring data through the convolution operation, and generate the convolution calculation results; Step S2012: input the convolution calculation result into the pooling layer, perform a maximum pooling operation on the convolution calculation result, and flatten the maximum pooling calculation result to obtain spatial feature data.

[0040] Specifically, the convolutional neural network (CNN structure) contains two convolutional layers and one pooling layer. The activation functions of the convolutional layers are all relu functions, the number of filters is set to 32, the size of the convolution kernel is 1, and the pooling size of the pooling layer is also 1.

[0041] The convolutional neural network model is used to extract the spatial feature data of the battery's current data, voltage data, and temperature data. Specifically, the current data, voltage data, and temperature data are used as the input data of the convolutional neural network, and feature extraction is performed through the convolution layer. The convolution kernel in the convolution layer slides the input data, calculates the convolution operation, and obtains the feature map. The convolution kernels of different convolution layers can extract different features.

[0042] Among them, the convolution calculation of the convolution layer is as follows: Figure 4 As shown in the figure, there is a 1*6*6 rectangle as the input feature map, which is convolved through a 3*3 filter (also called a convolution kernel), that is, the tensor product is calculated starting from the 3*3 size field area in the upper left corner of the matrix, and moves to the right each time the calculation is completed, with a stride of 1. After completing this calculation, a 1*4*4 output feature map will be obtained.

[0043] Step S202: Input the spatial feature data into a bidirectional long short-term memory network, and output the temporal feature data of the performance monitoring data.

[0044] The bidirectional long short-term memory network is a special RNN network that can learn long-term dependencies and capture the dynamic changes of time series data. The bidirectional long short-term memory network consists of the following components: (1) Input gate: determines which parts of the current input information are allowed to enter the state unit.

[0045] (2) Forget gate: determines which information in the state unit needs to be forgotten or removed.

[0046] (3) Status unit: a core component used to store historical status information.

[0047] (4) Output gate: determines the information output to the next hidden layer.

[0048] (5) Activation function (Sigmoid, Tanh): used for nonlinear transformation.

[0049] (6) Output layer: LSTM output data corresponding to the input data.

[0050] The BiLSTM network controls the flow of information entry, retention and output by opening and closing the input gate, forget gate and output gate, and realizes the memory and processing of time series information. Compared with the simple RNN network, the BiLSTM network can better capture the long-term dependencies in time series data, process longer time series, and has more powerful modeling capabilities. It has the following advantages: it can access more contextual information and have a more comprehensive understanding of sequence patterns; it can theoretically process long-distance dependencies of any length; the training effect is generally better than that of unidirectional LSTM, solving the gradient disappearance problem; it has strong feature extraction capabilities for sequence data; it can effectively learn the time context of data; and it has better fitting performance for heterogeneous time series.

[0051] Exemplarily, the bidirectional long short-term memory network includes: a forward LSTM network and a backward LSTM network; step S202 specifically includes: Step S2021: At the same time, the forward LSTM network obtains the forward hidden state of the spatial feature data according to the forward time series, and the backward LSTM network obtains the backward hidden state of the spatial feature data according to the reverse time series; Step S2022: concatenate or add the forward hidden state and the backward hidden state as the output state at that moment, thereby obtaining the context information of the performance monitoring data as the time series feature data.

[0052] like Figure 5 As shown in the figure, the BiLSTM network is composed of a forward LSTM network and a backward LSTM network. In the figure, X represents the input amount, and each time step i outputs hi=[h f i ·h b n-i ], [·] represents a concatenation operation. For example, if the hidden layer dimension of the forward LSTM network and the backward LSTM network is 50, then the dimension of the BiLSTM network is 100.

[0053] The spatial feature data of current, voltage and temperature extracted by the CNN network are used as the input data of the BiLSTM network to further extract the time variation characteristics of current, voltage and temperature. Its forward LSTM network and backward LSTM contain forward hidden states and backward hidden states respectively, which can capture past and future contextual information. At the same time, the forward LSTM network processes the forward sequence, and the backward LSTM network processes the reverse sequence. Finally, the hidden states of the two are concatenated or added as the output state at that moment. The input gate, forget gate and output gate in its network structure can control the flow of information and selectively save long-term information. The memory unit can save the historical state without being disturbed by short-term data. Therefore, the BiLSTM network extracts the time series variation law of current, voltage and temperature through bidirectional structure, gated unit, sequence learning and other methods, and is an effective model for dealing with timing problems.

[0054] Step S203: input the time series feature data into the Transformer model, and output the global feature data of the performance monitoring data represented by different subspace features.

[0055] like Figure 6 As shown in the figure, the main structure and functions of the Transformer model are as follows: (1) Positional Encoding: Provides order information for the input sequence to help Attention learn sequence order features.

[0056] (2) Multi-Head Attention Mechanism: This module is used to learn the correlation between different positions of the input sequence and obtain the feature representation of the input.

[0057] (3) Feed Forward Network: This module performs nonlinear transformation on the output of the attention mechanism module to learn more complex feature representations.

[0058] (4) Residual connection and normalization layer (Add & Norm): Residual connection Add is used to avoid the gradient vanishing problem in deep networks; Layer Norm is used to regularize the model, speed up training, and increase model stability.

[0059] The features learned by the Transformer model have the following advantages: Global feature learning: The Transformer model can learn the dependencies between features of any distance through the self-attention mechanism and construct a global feature representation instead of just a local feature. Separation of content and position: The feature representation in Transformer separates the input content from its position. The content feature is not restricted by position, making it easier to model long sequences. Feature richness: The multi-head self-attention mechanism network enables the Transformer model to learn feature representations of different subspaces and encode richer information. In general, the features learned by the Transformer model are more global and richer.

[0060] Exemplarily, the Transformer model includes: a position encoding layer, a multi-head attention mechanism network and a feedforward fully connected network; step S203 specifically includes: Step S2031: the position encoding layer adds a position vector to the time series feature data in time step order, and adds the position vector to the feature vector of the time series feature data to provide sequence information for the time series feature data; Step S2032: The multi-head attention mechanism network learns the correlation between the time series feature data at different positions to capture the influence relationship between the current data, voltage data and temperature data at different time steps, and obtains the globally dependent feature information; Step S2033: The feedforward fully connected network extracts and fuses feature information at different positions, obtains the nonlinear relationship between current data, voltage data and temperature data through nonlinear transformation, and obtains a higher-level feature representation as the feature representation of performance monitoring data in different subspaces.

[0061] In the Transformer model, the relationship and collaboration mechanism between the position encoding layer, the multi-head attention mechanism network, the feedforward fully connected network, the residual connection and the normalization layer are described in detail as follows: (1) For the battery’s current data, voltage data, and temperature data, which are time series data, the position encoding layer adds a position vector to the time series feature data of each time step. The position vector is added to the feature vector of the data itself to represent the relative position of the input in the sequence, providing order information for the time series feature data.

[0062] (2) Multi-head attention mechanism network In the context of battery performance monitoring data, the multi-head attention mechanism can help the model understand the mutual influence between current, voltage, and temperature at different time steps. By splitting the input vector into multiple "heads", each head can independently learn the attention weights of different parts, thereby enhancing the model's ability to pay attention to different features in the input sequence. This helps the model capture the complex relationship between current, voltage, and temperature, as well as the changing trends of these parameters over time.

[0063] (3) Applying a feedforward fully connected network separately and identically at each position can help the model learn the nonlinear relationship between current, voltage, and temperature, as well as the complex mapping relationship between these parameters and the battery SOH and RUL, and further extract and fuse the feature information of different positions to enhance the nonlinear ability and expression ability of the model.

[0064] (4) Residual connections pass the output features of the multi-head attention and feed-forward fully connected networks to subsequent layers, which is beneficial for gradient propagation and feature learning.

[0065] (5) The Layer Norm layer regularizes the output of the residual connection, accelerates network convergence, and stabilizes the feature extraction process.

[0066] In summary, these modules work together to gradually learn feature representations of the input and support sequence transformation tasks. Position encoding introduces position information, multi-head attention learns global dependencies, fully connected networks enhance features, and residuals and regularization accelerate and stabilize feature learning.

[0067] Exemplarily, this embodiment adopts a 1-layer Transformer structure, in which the parameter d_model (indicating the dimension of feature embedding embedding) in the structure is 32, num_heads (indicating the number of attention heads of the multi-head attention mechanism) is 4, and dff (indicating the step size of the residual connection) is 1.

[0068] Step S204: input the global feature data into the prediction network, and output the battery capacity prediction value and the remaining cycle number prediction value.

[0069] Exemplarily, the prediction network includes: a fully connected layer and an output layer; step S204 specifically includes: Step S2041: inputting the global feature data into the fully connected layer for feature integration and transformation to generate a feature representation suitable for outputting the predicted value; Step S2042: input the feature representation into the output layer, and perform target value prediction through two linear regression layers set in the output layer to simultaneously output the battery capacity prediction value and the remaining cycle number prediction value.

[0070] Specifically, after the Transformer model is the prediction network, which includes a fully connected layer and an output layer. The number of neurons in the fully connected layer is 16, and its activation function is the relu function. After the fully connected layer is the output layer of the model, the output layer is two linear regression layers, and the predicted values ​​are the battery capacity (SOH indicator) and the number of remaining cycles (RUL indicator), and their activation functions are all linear functions. Therefore, the present invention uses a variety of deep learning models to construct a multi-objective prediction model, which improves the prediction accuracy of the model.

[0071] Step S3: Based on the battery capacity prediction value, the remaining cycle number prediction value and the label data, a loss value is generated through the loss function of the multi-objective prediction model, and the multi-objective prediction model is updated until convergence to obtain a trained multi-objective prediction model; the multi-objective prediction model is used to predict the battery capacity and the remaining cycle number of the battery in the current state as data indicators for evaluating the battery SOH and RUL.

[0072] Since the battery capacity and the number of remaining cycles are continuous values, the loss function of each prediction target is the mean square error (MSE). The total loss function of the model is the sum of the two loss functions. The loss value is generated according to the loss function, and the multi-objective prediction model is updated until convergence to obtain a trained multi-objective prediction model. Among them, the optimizer is preferably RMSprop; the evaluation metric is the mean absolute error (MAE), which is the sum of the absolute values ​​of the difference between the target value and the predicted value.

[0073] The present invention uses the performance monitoring data of the B0018 battery in the charge and discharge stage in the public battery data set as a data set to test the model prediction effect, and uses current data, voltage data and temperature data to predict the battery capacity and the remaining number of cycles. The data set is divided into a training set and a test set in a ratio of 8:2, and the multi-objective prediction model is trained on the training set, and the effect is tested on the test set. The test data has a total of 5 features (measured voltage, measured current, measured temperature, load current and load voltage) and 2 prediction targets (battery capacity and remaining number of cycles discharged to 2.7V). The error between the predicted value and the true value is compared by the mean and standard deviation. The data distribution of the predicted result is basically consistent with the data distribution of the true data, indicating that the prediction effect of the model is good.

[0074] In summary, the model training method for predicting battery SOH and RUL provided by the present invention adopts multi-objective joint training, which can extract the spatial characteristics and time-related characteristics of current, voltage, and temperature data, and simultaneously learn the time series change law of current, voltage, and temperature data, and model the time-related patterns during long-term use of the battery, so as to simultaneously predict the battery capacity and the number of remaining cycles of the battery, so that the feature extraction ability and robustness of the model are stronger. At the same time, the multi-objective prediction model can capture the potential correlation between the two prediction targets, improve the accuracy of the prediction, and use the mutually shared feature representations between the battery's current, voltage, and temperature data to improve the prediction performance of the two targets, reducing the risk of model overfitting. Simultaneously predicting battery SOH and RUL in one model greatly reduces the cost of model training and maintenance. Therefore, the model takes into account the comprehensive advantages of multi-objective, feature sharing, accuracy, efficiency, and robustness, and is a high-performance and efficient SOH and RUL indicator prediction model.

[0075] Embodiment 2 Reference Figure 7 As shown, an embodiment of the present invention provides a method for predicting battery SOH and RUL, including: Obtain current data, voltage data and temperature data of the battery to be tested in the current state; The current data, voltage data and temperature data are input into a pre-trained multi-objective prediction model to obtain the battery capacity prediction value and the remaining cycle number prediction value output by the multi-objective prediction model, which are used as data indicators for evaluating the battery SOH and RUL. The pre-trained multi-objective prediction model is trained by the model training method for predicting the battery SOH and RUL provided in Example 1.

[0076] In this embodiment, the current data, voltage data and temperature data of the battery to be tested in the current state are collected, and the battery capacity and remaining cycle times in the current state are predicted through the trained multi-objective prediction model. The performance of the battery is evaluated through the predicted value, which reflects the energy storage capacity and durability of the battery, and provides a basis for the subsequent formulation of reasonable use and maintenance strategies.

[0077] The present invention can also extend the multi-objective prediction model to other prediction scenarios, such as energy storage systems, industrial equipment maintenance, digital electronic products, aerospace and other related fields.

[0078] In addition, some embodiments of the present application also provide an electronic device. The electronic device may be a digital computer in various forms, such as a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, etc. The electronic device may also be a mobile device in various forms, such as a personal digital processing, a cellular phone, a smart phone, a wearable device, and other similar computing devices.

[0079] The electronic device includes: one or more processors; and a memory storing computer program instructions, wherein when the computer program instructions are executed, the processor executes the steps of the model training method for predicting battery SOH and RUL as provided in the above-mentioned embodiment 1, or implements the steps of the battery SOH and RUL prediction method as provided in the above-mentioned embodiment 2. Figure 8 An exemplary structural diagram of the electronic device is disclosed. Figure 8 As shown, the electronic device includes: one or more processors 1101, a memory 1102, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. The various components are connected to each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed in the electronic device, including instructions stored in or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some other embodiments, if necessary, multiple processors and / or multiple buses can be used with multiple memories and multiple memories. Similarly, multiple electronic devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Among them, the components shown in this article, their connections and relationships, and their functions are only examples, and are not intended to limit the implementation of the present application described and / or required herein.

[0080] The electronic device may further include: an input device 1103 and an output device 1104. The processor 1101, the memory 1102, the input device 1103 and the output device 1104 may be connected via a bus or other means. Figure 8 The example of connecting through bus is taken in the following.

[0081] The input device 1103 can receive input digital or character information, and generate key signal input related to the user settings and function control of the electronic device, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator rod, one or more mouse buttons, a trackball, a joystick and other input devices. The output device 1104 may include a display device, an auxiliary lighting device (e.g., an LED) and a tactile feedback device (e.g., a vibration motor), etc. The display device may include, but is not limited to, a liquid crystal display (LCD), a light emitting diode (LED) display and a plasma display. In some embodiments, the display device may be a touch screen.

[0082] To provide interaction with a user, the electronic device may be a computer. The computer has: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball), through which the user can provide input to the computer. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0083] In an embodiment of the present application, a computer program / instruction is stored on a computer-readable medium, and when the computer program / instruction is executed by a processor, the steps of the model training method for predicting battery SOH and RUL as provided in the above-mentioned embodiment 1 are implemented, or the steps of the method for predicting battery SOH and RUL as provided in the above-mentioned embodiment 2 are implemented. The computer-readable medium may be included in the electronic device described in the above-mentioned embodiment; or it may exist independently without being assembled into the device. The above-mentioned computer-readable medium carries one or more computer-readable instructions.

[0084] The memory 1102 can be used as a non-transient computer-readable storage medium, which can be used to store non-transient software programs, non-transient computer executable programs and modules. The processor 1101 executes various functional applications and data processing of the server by running the non-transient software programs, instructions and modules stored in the memory 1102, so as to implement the program instructions / modules corresponding to the method provided by any one or more embodiments in the embodiments of the present application.

[0085] The memory 1102 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 1102 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory 1102 may optionally include a memory remotely arranged relative to the processor 1101, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0086] It should be noted that more specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0087] Computer-readable storage media include permanent and non-permanent, removable and non-removable media, and can be implemented by any method or technology to store information. Information can be computer-readable instructions, data structures, modules of programs or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

[0088] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0089] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. For example, an application specific integrated circuit (ASIC), a general-purpose computer or any other similar hardware device may be used to implement the embodiments. In some embodiments, the software program of the present application may be executed by a processor to implement the above steps or functions. Similarly, the software program of the present application (including related data structures) may be stored in a computer-readable recording medium, such as a RAM memory, a magnetic or optical drive or a floppy disk and the like. In addition, some steps or functions of the present application may be implemented by hardware, for example, as a circuit that cooperates with a processor to perform various steps or functions.

[0090] The computer program product provided in the embodiment of the present application includes one or more computer programs / instructions, and when the computer program / instructions are executed by the processor, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website site, a computer, a server, or a data center to another website site, a computer, a server, or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integrations. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid state disk (SSD)), etc.

[0091] The flow chart or block diagram in the accompanying drawings shows the possible architecture, function and operation of the equipment, method and computer program product according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented with a dedicated system for hardware that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0092] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily mention changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims, and the above embodiments should be regarded as exemplary and non-restrictive.

Claims

1. A model training method for predicting battery SOH and RUL, characterized in that: include: Acquire the performance monitoring data and corresponding label data of the battery in the historical charging and discharging process, wherein the performance monitoring data includes: current data, voltage data and temperature data, and the label data includes: the true value of the battery capacity and the true value of the number of remaining cycles; Based on the performance monitoring data, a battery capacity prediction value and a remaining cycle number prediction value are simultaneously output via a multi-objective prediction model; Based on the battery capacity prediction value, the remaining cycle number prediction value and the label data, a loss value is generated via a loss function of the multi-objective prediction model, and the multi-objective prediction model is updated until convergence to obtain the trained multi-objective prediction model; The multi-objective prediction model is used to simultaneously predict the battery capacity and the number of remaining cycles of the battery in the current state, so as to serve as data indicators for evaluating the battery SOH and RUL; The multi-objective prediction model includes: a convolutional neural network, a bidirectional long short-term memory network, a Transformer model and a prediction network; The step of outputting a battery capacity prediction value and a remaining cycle number prediction value based on the performance monitoring data via a multi-objective prediction model comprises: Inputting the performance monitoring data into the convolutional neural network, and outputting spatial feature data of the performance monitoring data; Inputting the spatial feature data into the bidirectional long short-term memory network, and outputting the temporal feature data of the performance monitoring data; Inputting the time series feature data into the Transformer model, and outputting global feature data represented by different subspace features of the performance monitoring data; The global characteristic data is input into the prediction network, and a battery capacity prediction value and a remaining cycle number prediction value are output simultaneously.

2. The model training method for predicting battery SOH and RUL according to claim 1, characterized in that: The convolutional neural network includes: at least two convolutional layers and one pooling layer; The step of inputting the performance monitoring data into the convolutional neural network and outputting the spatial feature data of the performance monitoring data comprises: Inputting the performance monitoring data into the convolution layer, learning and identifying key features of the performance monitoring data through convolution operation, and generating convolution calculation results; The convolution calculation result is input into the pooling layer, a maximum pooling operation is performed on the convolution calculation result, and the maximum pooling calculation result is flattened to obtain the spatial feature data.

3. The model training method for predicting battery SOH and RUL according to claim 1, characterized in that: The bidirectional long short-term memory network includes: a forward LSTM network and a backward LSTM network; The step of inputting the spatial feature data into the bidirectional long short-term memory network and outputting the temporal feature data of the performance monitoring data comprises: At the same time, the forward LSTM network obtains the forward hidden state of the spatial feature data according to the forward time sequence, and the backward LSTM network obtains the backward hidden state of the spatial feature data according to the reverse time sequence; The forward hidden state and the backward hidden state are concatenated or added as the output state at the moment, thereby obtaining the context information of the performance monitoring data as the time series feature data.

4. The model training method for predicting battery SOH and RUL according to claim 1, characterized in that: The Transformer model includes: a position encoding layer, a multi-head attention mechanism network and a feedforward fully connected network; The step of inputting the time series feature data into the Transformer model and outputting the global feature data represented by different subspace features of the performance monitoring data comprises: The position encoding layer sequentially adds position vectors to the time series feature data in time steps, and adds the position vectors to the feature vectors of the time series feature data to provide sequence information for the time series feature data; The multi-head attention mechanism network learns the correlation between the time series feature data at different positions to capture the influence relationship between the current data, the voltage data and the temperature data at different time steps to obtain globally dependent feature information; The feedforward fully connected network extracts and fuses feature information at different positions, obtains the nonlinear relationship between the current data, the voltage data and the temperature data through nonlinear transformation, and obtains a higher-level feature representation as the feature representation of the performance monitoring data in different subspaces.

5. The model training method for predicting battery SOH and RUL according to claim 1, characterized in that: The prediction network includes: a fully connected layer and an output layer; The step of inputting the global feature data into the prediction network and outputting a battery capacity prediction value and a remaining cycle number prediction value comprises: Inputting the global feature data into the fully connected layer for feature integration and transformation to generate a feature representation suitable for outputting a prediction value; The feature representation is input into the output layer, and the target value is predicted through two linear regression layers set in the output layer, so as to simultaneously output the battery capacity prediction value and the remaining cycle number prediction value.

6. A method for predicting battery SOH and RUL, characterized in that: include: Obtain current data, voltage data and temperature data of the battery to be tested in the current state; The current data, the voltage data and the temperature data are input into a pre-trained multi-objective prediction model to obtain a battery capacity prediction value and a remaining cycle number prediction value output by the multi-objective prediction model, which are used as data indicators for evaluating the battery SOH and RUL. The pre-trained multi-objective prediction model is trained by the model training method for predicting the battery SOH and RUL as described in any one of claims 1 to 5.

7. An electronic device, characterized in that: The electronic device comprises: One or more processors; and a memory storing computer program instructions, wherein the computer program instructions, when executed by the processor, implement the steps of the model training method for predicting battery SOH and RUL as described in any one of claims 1 to 5, or implement the steps of the method for predicting battery SOH and RUL as described in claim 6.

8. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by the processor, the steps of the model training method for predicting the battery SOH and RUL as described in any one of claims 1 to 5 are implemented, or the steps of the method for predicting the battery SOH and RUL as described in claim 6 are implemented.

9. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the steps of the model training method for predicting battery SOH and RUL as described in any one of claims 1 to 5 are implemented, or the steps of the method for predicting battery SOH and RUL as described in claim 6 are implemented.

Images