A battery status prediction method, system, device and storage medium

Abstract

The present invention provides a battery status prediction method, system, device and storage medium, which relate to the field of battery management technology. By obtaining multi-dimensional operating feature information corresponding to the target battery; using a multi-layer perceptron network based on channel attention to extract and weightedly fuse the multi-dimensional operating features to generate a first extraction result; obtaining the historical data of the target battery, extracting the temporal relationship of the historical data according to a bidirectional long short-term memory neural network to generate a second extraction result; fusing the first extraction result and the second extraction result to generate a fusion result; predicting the health status of the target battery based on the fusion result and generating a prediction result. The historical health status is integrated with the operating feature change information to achieve the technical effect of accurately predicting the health status of the target battery.

Description

Technical Field

[0001] The present application relates to the field of battery management technology, and in particular to a battery status prediction method, system, device and storage medium. Background Art

[0002] As an efficient and lightweight energy storage solution, lithium batteries are highly favored due to their high energy density, long life and low self-discharge rate, and play an important role in portable electronic devices, electric vehicles, grid energy storage and other fields. However, with the increase of usage time and the number of charge and discharge cycles, the performance of lithium batteries gradually decreases, affecting the device usage experience and system reliability.

[0003] Existing health status prediction methods are mainly divided into three categories: direct measurement method, model-based method and data-driven method. The direct measurement method measures the impedance, voltage, current and other parameters of the battery to infer its health status, but the application scenario of the direct measurement method is limited by the measurement accuracy and real-time performance of the instrument. The model-based method estimates the health status by establishing an equivalent circuit model or physical model of the battery, but it is often difficult to fully consider the various electrical characteristics and electrochemical processes of the battery, resulting in low model accuracy or overly complex model establishment, resulting in high computational costs, and it is difficult to achieve online prediction in practical applications.

[0004] Therefore, how to accurately predict the health status of lithium batteries has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] In order to accurately predict the health status of a lithium battery, the present application provides a battery status prediction method, system, device and storage medium.

[0006] In a first aspect, a battery status prediction method provided by the present application adopts the following technical solution:

[0007] A battery state prediction method, comprising:

[0008] When it is determined that the target battery meets the detection conditions, multi-dimensional operation characteristic information corresponding to the target battery is obtained, wherein the multi-dimensional operation characteristic information includes: internal resistance information, constant current charging time information, constant voltage charging time information, and peak characteristic information of the capacity increment curve;

[0009] Using a multi-layer perceptron network based on channel attention to perform feature extraction and weighted fusion on the multi-dimensional operation features to generate a first extraction result;

[0010] Acquire historical data of the target battery, and extract the temporal relationship of the historical data according to a bidirectional long short-term memory neural network to generate a second extraction result;

[0011] fusing the first extraction result and the second extraction result to generate a fusion result;

[0012] The health status of the target battery is predicted based on the fusion result and a prediction result is generated.

[0013] Optionally, when determining that the target battery meets the detection condition, the step of obtaining multi-dimensional operation characteristic information corresponding to the target battery includes:

[0014] Real-time monitoring of the charging status of the target battery;

[0015] When it is determined that the target battery meets the detection condition, obtaining internal resistance information of the target battery;

[0016] Obtaining constant current charging time information of the target battery in the constant current charging stage;

[0017] Acquiring constant voltage charging time information of the target battery in the constant voltage charging stage;

[0018] Acquire peak characteristic information of the capacity increment curve corresponding to the target battery by means of incremental capacity analysis;

[0019] Multi-dimensional operation characteristic information is generated according to the internal resistance information, the constant current charging time information, the constant voltage charging time information and the peak characteristic information of the capacity increment curve.

[0020] Optionally, before the step of extracting and weighted fusion the multi-dimensional operation features using a channel attention-based multi-layer perceptron network to generate a first extraction result, the step further includes:

[0021] Performing average pooling processing on the multi-dimensional operation features to determine overall feature information;

[0022] A maximum pooling operation is performed on the multi-dimensional running features to determine salient feature information.

[0023] Optionally, the step of extracting and weighted fusing the multi-dimensional operation features using a channel attention-based multi-layer perceptron network to generate a first extraction result includes:

[0024] Determine the hidden layer neuron information and output value of the multilayer perceptron network to generate multilayer perceptron network information;

[0025] Generating a channel attention weight by combining the overall feature information, the prominent feature information, and the multi-layer perceptron network information;

[0026] Multiplying the channel attention weight by the corresponding salient feature information to determine corresponding feature data;

[0027] A first extraction result is generated according to the feature data.

[0028] Optionally, the step of acquiring the historical data of the target battery and extracting the temporal relationship of the historical data according to a bidirectional long short-term memory neural network to generate a second extraction result includes:

[0029] Acquiring historical data of the target battery;

[0030] Extracting the historical data through a bidirectional long short-term memory neural network to obtain a historical data extraction result;

[0031] A health status feature sequence is determined according to the historical data extraction result as a second extraction result.

[0032] Optionally, after the step of determining the health status feature sequence as the second extraction result according to the historical data extraction result, the step further includes:

[0033] Introducing a time attention mechanism, and focusing on the fluctuation results corresponding to the health status feature sequence according to the time attention mechanism;

[0034] According to the fluctuation result combined with the health status feature sequence To determine the attention weight corresponding to each element in the health status feature sequence :

[0035]

[0036]

[0037] in, , For the The attention score corresponding to each element, represents the exponential function, represents the weight, represents bias;

[0038] The attention weight corresponding to each element in the health status feature sequence is multiplied by the corresponding value to generate a second extraction result.

[0039] Optionally, after the step of predicting the health state of the target battery based on the fusion result and generating a prediction result, the method further includes:

[0040] The prediction result is updated by using a preset loss function

[0041]

[0042] in, , and are adjustable parameters that control the gradient descent speed and positive and negative gradients of the loss function, respectively, and .

[0043] In a second aspect, the present application provides a battery state prediction system, the battery state prediction system comprising:

[0044] An operation characteristic acquisition module, used to acquire multi-dimensional operation characteristic information corresponding to the target battery when determining that the target battery meets the detection conditions, wherein the multi-dimensional operation characteristic information includes: internal resistance information, constant current charging time information, constant voltage charging time information, and peak characteristic information of the capacity increment curve;

[0045] A first extraction result acquisition module, used for performing feature extraction and weighted fusion on the multi-dimensional operation features by using a channel attention-based multi-layer perceptron network to generate a first extraction result;

[0046] A second extraction result acquisition module, used to acquire the historical data of the target battery, and extract the temporal relationship of the historical data according to the bidirectional long short-term memory neural network to generate a second extraction result;

[0047] A result fusion module, used for fusing the first extraction result and the second extraction result to generate a fusion result;

[0048] A result prediction module is used to predict the health state of the target battery based on the fusion result and generate a prediction result.

[0049] In a third aspect, the present application provides a computer device, comprising: a memory and a processor, wherein the processor executes the method described above when running computer instructions stored in the memory.

[0050] In a fourth aspect, the present application provides a computer-readable storage medium, comprising instructions, which, when executed on a computer, enable the computer to execute the method described above.

[0051] In summary, the present application includes the following beneficial technical effects:

[0052] The present application obtains multi-dimensional operation feature information corresponding to the target battery; uses a multi-layer perceptron network based on channel attention to extract and weightedly fuse the multi-dimensional operation features to generate a first extraction result; obtains historical data of the target battery, and extracts the temporal relationship of the historical data according to a bidirectional long short-term memory neural network to generate a second extraction result;

[0053] The first extraction result and the second extraction result are fused to generate a fusion result; based on the fusion result, the health state of the target battery is predicted and a prediction result is generated. The historical health state is fused with the operation characteristic change information to achieve the technical effect of accurately predicting the health state of the target battery. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a schematic diagram of the computer device structure of the hardware operating environment involved in the embodiment of the present application;

[0055] Figure 2 It is a flowchart of the first embodiment of the battery state prediction method of the present application;

[0056] Figure 3 It is a structural block diagram of the first embodiment of the battery status prediction system of the present application. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below through the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0058] Reference Figure 1 , Figure 1 A schematic diagram of the computer device structure of the hardware operating environment involved in the embodiment of the present application.

[0059] like Figure 1 As shown, the computer device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity (Wireless-Fidelity, Wi-Fi) interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM), or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk storage. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0060] Those skilled in the art will understand that Figure 1The structure shown in the figure does not constitute a limitation on the computer device, and may include more or less components than shown in the figure, or combine some components, or arrange the components differently.

[0061] like Figure 1 As shown, the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module, and a battery status prediction program.

[0062] exist Figure 1 In the computer device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the present application can be set in the computer device, and the computer device calls the battery status prediction program stored in the memory 1005 through the processor 1001, and executes the battery status prediction method provided in the embodiment of the present application.

[0063] The present application embodiment provides a battery status prediction method, referring to Figure 2 , Figure 2 This is a flow chart of the first embodiment of the battery status prediction method of the present application.

[0064] In this embodiment, the battery status prediction method includes the following steps:

[0065] Step S10: When it is determined that the target battery meets the detection conditions, multi-dimensional operation characteristic information corresponding to the target battery is obtained, and the multi-dimensional operation characteristic information includes: internal resistance information, constant current charging time information, constant voltage charging time information and peak characteristic information of the capacity increment curve.

[0066] It should be noted that the detection conditions will be obtained before the target battery performs health status prediction, and it will be determined whether the target battery currently meets the detection conditions. By setting the detection conditions, the detection efficiency can be improved, and the situations that do not meet the corresponding states of the detection conditions can be eliminated.

[0067] It should be noted that the charging process of lithium-ion batteries is divided into two stages: constant current charging (CCC) and constant voltage charging (CCV). In the CCC stage, a large amount of electrical energy is quickly charged into the lithium battery through a constant current. When the lithium battery voltage reaches a certain threshold, the CCV method is used for charging, and the charging rate is controlled by the gradually decreasing voltage difference between the charging voltage and the lithium battery voltage. The constant current charging time (CCCT) and constant voltage charging time (CVCT) will change with the aging of the lithium battery and are related to the historical health state (SOH). From the perspective of the battery's operating mechanism, the capacity decay process of the lithium battery is related to the loss of lithium ions during the charging and discharging process, the formation of the solid electrolyte interface layer, and the degradation of the electrolyte. The above processes will lead to the obstruction of ion transmission and the reduction of electronic conduction efficiency, which is manifested as the increase of the internal resistance of the lithium battery on a macro scale. Therefore, the impedance of the lithium battery is also related to the SOH.

[0068] This embodiment extracts multi-dimensional operation characteristics related to SOH from the battery operation process, and combines the change information of the operation characteristics with the SOH historical information to provide more comprehensive information for the model and effectively improve the prediction accuracy of the model.

[0069] It should be noted that this embodiment proposes a battery status prediction method based on multi-operation feature fusion, extracting multi-dimensional operation features that have a strong correlation with the health status from the battery operation process, integrating feature information from both operation features and historical health status, and realizing accurate health status prediction. The proposed method includes four steps:

[0070] (1) Extract four operating characteristics from the battery operation process, which are strongly correlated with SOH: internal resistance, constant current charging time, constant voltage charging time, and peak characteristics of the capacity increment curve.

[0071] (2) Channel attention-based MLP is used to realize data feature extraction and dynamic weighted fusion of running features.

[0072] (3) The BiLSTM neural network is used to extract the temporal relationship in the SOH historical data, and the temporal attention mechanism is introduced to enhance the ability to extract temporal features to cope with the SOH fluctuations caused by capacity regeneration.

[0073] (4) The operating feature change information extracted by the MLP network based on channel attention is fused with the health status history information extracted by the BiLSTM network based on time attention to predict the current SOH. In particular, in the SOH prediction process, overestimation of SOH may cause the lithium battery to operate beyond its limit, causing problems such as overcharging and over-discharging, and posing safety risks. To this end, a parameter-adjustable asymmetric loss function is proposed to give a higher weight to the positive error in the model training process, guiding the model training to focus more on reducing SOH overestimation.

[0074] Furthermore, in order to reasonably obtain multi-dimensional operating characteristic information and accurately predict the battery health status, when it is determined that the target battery meets the detection conditions, the step of obtaining the multi-dimensional operating characteristic information corresponding to the target battery includes: real-time monitoring of the charging status of the target battery; when it is determined that the target battery meets the detection conditions, obtaining the internal resistance information of the target battery; obtaining the constant current charging time information of the target battery in the constant current charging stage; obtaining the constant voltage charging time information of the target battery in the constant voltage charging stage; obtaining the peak characteristic information of the capacity increment curve corresponding to the target battery by means of incremental capacity analysis; and generating multi-dimensional operating characteristic information according to the internal resistance information, constant current charging time information, constant voltage charging time information and the peak characteristic information of the capacity increment curve.

[0075] In the specific implementation, this embodiment uses the incremental capacity analysis (ICA) method to extract the operating characteristics related to the health status of the lithium battery. During the charging process of the lithium battery, the voltage rises slowly in the middle of charging, and it is difficult to observe the charging process of the lithium battery through voltage changes. However, most of the capacity of the lithium battery is charged in the middle of charging, and the characteristics of the lithium battery in the middle of charging are closely related to the health status. Therefore, this embodiment uses the ICA method to analyze the voltage charging process, and converts the lithium battery voltage change that is not easy to observe into an IC curve by processing the capacity and voltage differential of the lithium battery charging process, so as to judge the performance differences of lithium batteries in different health states and extract the peak characteristics of the IC curve.

[0076] In summary, the operating characteristics of the four dimensions of CCCT, CVCT, resistance and IC curve peak can well reflect the aging process of lithium batteries. This embodiment extracts the above operating characteristics from the battery operation process for SOH estimation. In order to verify the correlation, this embodiment uses the Pearson correlation coefficient to judge the four-dimensional operating characteristics. With SOH sequence Related relationships:

[0077] The calculation formula of the Pearson correlation coefficient between sequences is:

[0078]

[0079] in is the covariance, is the standard deviation.

[0080] The results show that the correlation coefficients between each operating characteristic and SOH are greater than 0.8, indicating a strong correlation.

[0081] Step S20: Using a channel attention-based multi-layer perceptron network to perform feature extraction and weighted fusion on the multi-dimensional operation features to generate a first extraction result.

[0082] It can be understood that the first extraction result and the second extraction result in this embodiment are only used to distinguish different results, and do not limit their importance or order. This embodiment does not limit the naming of the extraction results.

[0083] Furthermore, in order to improve the rationality of the first extraction result, before the step of using a multi-layer perceptron network based on channel attention to perform feature extraction and weighted fusion on the multi-dimensional operation features to generate the first extraction result, it also includes: performing average pooling processing on the multi-dimensional operation features to determine the overall feature information; performing maximum pooling operation on the multi-dimensional operation features to determine the prominent feature information.

[0084] It should be noted that the step of using a multi-layer perceptron network based on channel attention to perform feature extraction and weighted fusion on multi-dimensional operating features to generate a first extraction result includes: determining the hidden layer neuron information and output values ​​of the multi-layer perceptron network to generate multi-layer perceptron network information; combining the overall feature information, prominent feature information and multi-layer perceptron network information to generate channel attention weights; multiplying the channel attention weights with the corresponding prominent feature information to determine the corresponding feature data; and generating a first extraction result based on the feature data.

[0085] It should be noted that different operating characteristics have different importance for health status prediction. In order to measure the difference in the contribution of different operating characteristics to SOH prediction, the channel attention mechanism is used to weight these four-dimensional operating characteristics, and the multi-layer perceptron is used to extract data features from the weighted operating characteristic data. The specific operation characteristic data feature extraction process is as follows:

[0086] After selecting the operating features, in order to measure the importance of different operating features to SOH prediction and focus on the operating features that have a greater impact on SOH, this embodiment uses a channel attention mechanism to assign weights to the four-dimensional operating features to measure the contribution of different operating features. The channel attention mechanism helps the model automatically learn and emphasize the most important channels in the input data. Dynamically adjust the weight of each channel, assign a larger weight to the key channel, and realize the fusion of multi-channel features, so that the model can capture key operating feature information. First, perform average pooling and maximum pooling operations on the input four-dimensional operating feature data to extract overall features and salient features:

[0087]

[0088] in, is the input operation characteristic data, is the average pooling function, which obtains the overall trend of the running feature changes through the average pooling operation; It is the maximum pooling function, which focuses on prominent data features through the maximum pooling operation.

[0089] Then, the extracted features are passed through the MLP layer with shared weights of a single hidden layer and the channel attention weights are obtained through the sigmoid activation function. and the output of the MLP The calculation is as follows:

[0090]

[0091]

[0092] in, The input layer The value of a neuron, and is related to the hidden layer The input layer of neurons connected to The weights and biases of the neurons, is the nonlinear activation function of the hidden layer, and is the hidden layer connected to the output layer The weights and biases corresponding to each neuron, is the non-linear activation function of the output layer.

[0093] The calculation formula of the attention weight of the shared multi-layer perceptron channel is as follows:

[0094]

[0095] in, is the sigmoid activation function, represents the MLP layer, , is the shared weight of the multilayer perceptron shared at the input.

[0096] Multiply the attention weight vector by the input running feature data of the corresponding channel to obtain the running feature data that assigns the channel attention weight. Finally, the weighted data is subjected to feature fusion through a multi-layer perceptron to obtain the features extracted from the running feature data:

[0097]

[0098] in, Represents the attention weights for feature data running on different channels.

[0099] Based on this, feature extraction and weighted fusion of four-dimensional operating characteristics that are strongly related to the aging process of lithium batteries are achieved, which introduces more feature information for health status prediction and improves the performance of the prediction model.

[0100] Step S30: Acquire historical data of the target battery, and extract the temporal relationship of the historical data according to the bidirectional long short-term memory neural network to generate a second extraction result.

[0101] It should be noted that it can be observed in the actual data that the health state decline is not a monotonic decay process. Affected by capacity regeneration, the health state sequence of lithium batteries fluctuates, affecting the prediction accuracy. In order to solve the problem of being unable to accurately mine the data features contained in the SOH sequence whose data fluctuates due to capacity regeneration, this embodiment starts from two aspects: On the one hand, in order to mine the time dependency in SOH historical data, this embodiment uses a bidirectional long short-term memory neural network to extract time features and long-term dependencies of data. On the other hand, in order to deal with SOH fluctuations caused by capacity regeneration, this embodiment introduces a time attention mechanism to pay attention to the fluctuations of the SOH sequence, giving greater weight to important information in the features. First, the historical health state data is passed through the BiLSTM layer to extract data features.

[0102] Furthermore, in order to improve the accuracy of the second extraction result, the historical data of the target battery is obtained, and the temporal relationship of the historical data is extracted according to the bidirectional long short-term memory neural network to generate the second extraction result, including: obtaining the historical data of the target battery; extracting the historical data through the bidirectional long short-term memory neural network to obtain the historical data extraction result; determining the health status feature sequence as the second extraction result according to the historical data extraction result.

[0103] It should be noted that after the step of determining the health status feature sequence as the second extraction result based on the historical data extraction result, it also includes: introducing a time attention mechanism, and paying attention to the fluctuation results corresponding to the health status feature sequence according to the time attention mechanism; combining the health status feature sequence with the fluctuation results To determine the attention weight corresponding to each element in the health status feature sequence :

[0104]

[0105]

[0106] in, , For the The attention score corresponding to each element, represents the exponential function, represents the weight, represents bias;

[0107] The second extraction result is generated by multiplying the attention weight corresponding to each element in the health status feature sequence by the corresponding value.

[0108] In the specific implementation, the BiLSTM unit is composed of two LSTM units with the same structure but opposite directions. LSTM is an improved recursive network model proposed to overcome the gradient vanishing and gradient exploding problems of traditional RNN. It has good performance in capturing long-term dependencies. The LSTM unit has three gates: input gate , Forget Gate , output gate , and two status information: temporary unit status and cell status The calculation process of LSTM can be summarized as forgetting useless information in the unit state and memorizing new information. Selectively forget information about the previous cell state and decide what information needs to be retained in the cell state:

[0109]

[0110] Input Gate Get the output of the sigmoid activation function and the output of the tanh activation function And update the cell state:

[0111]

[0112]

[0113]

[0114] The process of obtaining the current neuron output value through the output gate is defined as:

[0115]

[0116]

[0117] in, yes The output value at time, is the vector concatenation operator, and are the weight matrix and bias vector of each unit respectively. yes The input value at the time, yes The output value of the forget gate at all times, yes The output value of the sigmoid activation function in the input gate at any given moment, The output gate is The state of the moment, The input gate is The output value of the tanh activation function at time instant, The output gate is The state of the moment. The output gate is The output value at time, The output gate is Output value at time. Represents the tanh activation function.

[0118] In order to facilitate the representation of this embodiment, the hidden layer outputs of the forward and backward LSTM units of the BiLSTM unit are represented by and Indicates that the final calculated output of the BiLSTM unit is:

[0119]

[0120] in , Represent the weights of the forward and backward LSTM units respectively, is the bias of the BiLSTM unit.

[0121] Then, in order to cope with the health status fluctuation caused by capacity regeneration, this embodiment introduces a temporal attention mechanism to focus on the fluctuation of the health status sequence. Calculate the attention weights corresponding to each element :

[0122]

[0123]

[0124] in, , For the The attention score corresponding to each element, represents the exponential function, and Represent the corresponding weights and biases respectively.

[0125] Multiply the calculated attention weight of each element by the corresponding value of the point to obtain the attention-weighted health status feature sequence.

[0126]

[0127] Step S40: Fusing the first extraction result and the second extraction result to generate a fusion result.

[0128] In the specific implementation, the first extraction result is the lithium battery operation characteristics, and the second extraction result is the historical health status. After obtaining the two parts of data features of the lithium battery operation characteristics and the historical health status, this embodiment splices the feature information of the two aspects and obtains the final prediction result through the full connection of the Dense layer.

[0129]

[0130] in, represents the fully connected layer, Represents a data concatenation operation.

[0131] Step S50: predicting the health status of the target battery based on the fusion result and generating a prediction result.

[0132] It should be noted that in the actual use of lithium batteries, the predicted value of SOH will be used to guide the operation of the battery. Among them, overestimation of SOH will cause the lithium battery to operate beyond its limit, resulting in overcharging, over-discharging, etc., which increases the management and maintenance costs of lithium batteries, and may even cause safety risks. To this end, the SOH prediction model needs to minimize overestimation while ensuring the accuracy of the model. In response to this practical need, this embodiment innovatively proposes an asymmetric loss function with adjustable parameters. The designed loss function can adjust the parameters of the loss function according to the data characteristics of SOH, and assign different weights to overestimation and underestimation of the health status in the process of guiding model training, so that the training process pays more attention to the forward prediction error, effectively reducing SOH overestimation while ensuring the accuracy of model prediction. At the same time, the loss function also satisfies the continuous differentiability of the gradient, thereby effectively ensuring model convergence.

[0133] Furthermore, in order to effectively ensure the convergence of the model, after the step of predicting the health status of the target battery based on the fusion result and generating the prediction result, it also includes:

[0134] Update the prediction results by using the preset loss function

[0135]

[0136] in, , and are adjustable parameters that control the gradient descent speed and positive and negative gradients of the loss function, respectively, and By adjusting the three parameters of the loss function, controlling the overall gradient descent speed and the descent gradients of the forward error and reverse error, we can focus on the overestimation of the model in the training process and effectively reduce the overestimation of SOH.

[0137] This embodiment obtains multi-dimensional operation feature information corresponding to the target battery; uses a multi-layer perceptron network based on channel attention to extract and weightedly fuse the multi-dimensional operation features to generate a first extraction result; obtains historical data of the target battery, extracts the temporal relationship of the historical data according to a bidirectional long short-term memory neural network to generate a second extraction result; fuses the first extraction result and the second extraction result to generate a fusion result; predicts the health status of the target battery based on the fusion result and generates a prediction result. The historical health status is integrated with the operation feature change information to achieve the technical effect of accurately predicting the health status of the target battery.

[0138] In addition, an embodiment of the present application further proposes a computer-readable storage medium, on which a program for battery status prediction is stored. When the program for battery status prediction is executed by a processor, the steps of the method for battery status prediction as described above are implemented.

[0139] Reference Figure 3 , Figure 3 This is a structural block diagram of the first embodiment of the battery status prediction system of the present application.

[0140] like Figure 3 As shown, the battery status prediction system proposed in the embodiment of the present application includes:

[0141] The operation characteristic acquisition module 10 is used to obtain multi-dimensional operation characteristic information corresponding to the target battery when it is determined that the target battery meets the detection conditions, and the multi-dimensional operation characteristic information includes: internal resistance information, constant current charging time information, constant voltage charging time information and peak characteristic information of the capacity increment curve;

[0142] A first extraction result acquisition module 20, configured to extract and weightedly fuse multi-dimensional operation features using a channel attention-based multi-layer perceptron network to generate a first extraction result;

[0143] A second extraction result acquisition module 30 is used to acquire historical data of the target battery, and extract the temporal relationship of the historical data according to the bidirectional long short-term memory neural network to generate a second extraction result;

[0144] A result fusion module 40, used to fuse the first extraction result and the second extraction result to generate a fusion result;

[0145] The result prediction module 50 is used to predict the health state of the target battery based on the fusion result and generate a prediction result.

[0146] It should be understood that the above is only an example and does not constitute any limitation on the technical solution of the present application. In specific applications, technicians in this field can make settings as needed, and the present application does not impose any limitation on this.

[0147] This embodiment obtains multi-dimensional operation feature information corresponding to the target battery; uses a multi-layer perceptron network based on channel attention to extract and weightedly fuse the multi-dimensional operation features to generate a first extraction result; obtains historical data of the target battery, extracts the temporal relationship of the historical data according to a bidirectional long short-term memory neural network to generate a second extraction result; fuses the first extraction result and the second extraction result to generate a fusion result; predicts the health status of the target battery based on the fusion result and generates a prediction result. The historical health status is integrated with the operation feature change information to achieve the technical effect of accurately predicting the health status of the target battery.

[0148] In one embodiment, the operation characteristic acquisition module 10 is also used to monitor the charging state of the target battery in real time; when it is determined that the target battery meets the detection conditions, the internal resistance information of the target battery is obtained; the constant current charging time information of the target battery in the constant current charging stage is obtained; the constant voltage charging time information of the target battery in the constant voltage charging stage is obtained; the peak characteristic information of the capacity increment curve corresponding to the target battery is obtained by incremental capacity analysis; and multi-dimensional operation characteristic information is generated according to the internal resistance information, the constant current charging time information, the constant voltage charging time information and the peak characteristic information of the capacity increment curve.

[0149] In one embodiment, the first extraction result acquisition module 20 further performs average pooling processing on the multi-dimensional operation features to determine the overall feature information; and performs maximum pooling operation on the multi-dimensional operation features to determine the prominent feature information.

[0150] In one embodiment, the first extraction result acquisition module 20 is also used to determine the hidden layer neuron information and output values ​​of the multi-layer perceptron network to generate multi-layer perceptron network information; combine the overall feature information, prominent feature information and multi-layer perceptron network information to generate channel attention weights; multiply the channel attention weights with the corresponding prominent feature information to determine the corresponding feature data; and generate the first extraction result based on the feature data.

[0151] In one embodiment, the second extraction result acquisition module 30 is also used to obtain historical data of the target battery; extract the historical data through a bidirectional long short-term memory neural network to obtain historical data extraction results; determine the health status feature sequence as the second extraction result based on the historical data extraction results.

[0152] In one embodiment, the second extraction result acquisition module 30 is further used to introduce a time attention mechanism, and pay attention to the fluctuation results corresponding to the health status feature sequence according to the time attention mechanism; To determine the attention weight corresponding to each element in the health status feature sequence :

[0153]

[0154]

[0155] in, , For the The attention score corresponding to each element, represents the exponential function, represents the weight, Represents a bias; multiplying the attention weight corresponding to each element in the health status feature sequence by the corresponding value to generate a second extraction result.

[0156] In one embodiment, the result prediction module 50 is further used to update the prediction result by using a preset loss function.

[0157]

[0158] in, , and are adjustable parameters that control the gradient descent speed and positive and negative gradients of the loss function, respectively, and .

[0159] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of the present application. In practical applications, technicians in this field can select part or all of it according to actual needs to achieve the purpose of the present embodiment, and no limitation is made here.

[0160] In addition, for technical details not fully described in this embodiment, reference can be made to the battery status prediction method provided in any embodiment of the present application, which will not be repeated here.

[0161] In addition, it should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0162] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0163] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory (ROM) / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods of each embodiment of the present application.

[0164] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A battery status prediction method, characterized in that: include: When it is determined that the target battery meets the detection conditions, multi-dimensional operation characteristic information corresponding to the target battery is obtained, wherein the multi-dimensional operation characteristic information includes: internal resistance information, constant current charging time information, constant voltage charging time information, and peak characteristic information of the capacity increment curve; Using a multi-layer perceptron network based on channel attention to perform feature extraction and weighted fusion on the multi-dimensional operation features to generate a first extraction result; Acquire historical data of the target battery, and extract the temporal relationship of the historical data according to a bidirectional long short-term memory neural network to generate a second extraction result; fusing the first extraction result and the second extraction result to generate a fusion result; Predicting the health status of the target battery based on the fusion result and generating a prediction result; Wherein, when determining that the target battery meets the detection condition, the step of obtaining the multi-dimensional operation characteristic information corresponding to the target battery includes: Real-time monitoring of the charging status of the target battery; When it is determined that the target battery meets the detection condition, obtaining internal resistance information of the target battery; Obtaining constant current charging time information of the target battery in the constant current charging stage; Acquiring constant voltage charging time information of the target battery in the constant voltage charging stage; Acquire peak characteristic information of the capacity increment curve corresponding to the target battery by means of incremental capacity analysis; Generate multi-dimensional operation characteristic information according to the internal resistance information, the constant current charging time information, the constant voltage charging time information and the peak characteristic information of the capacity increment curve; Before the step of extracting and weighted fusing the multi-dimensional operation features using a channel attention-based multi-layer perceptron network to generate a first extraction result, the method further includes: Performing average pooling processing on the multi-dimensional operation features to determine overall feature information; Performing a maximum pooling operation on the multi-dimensional running features to determine salient feature information; The step of extracting and weighted fusing the multi-dimensional operation features using a channel attention-based multi-layer perceptron network to generate a first extraction result includes: Determine the hidden layer neuron information and output value of the multilayer perceptron network to generate multilayer perceptron network information; Generating a channel attention weight by combining the overall feature information, the prominent feature information, and the multi-layer perceptron network information; Multiplying the channel attention weight by the corresponding salient feature information to determine corresponding feature data; generating a first extraction result according to the feature data; Wherein, after the step of predicting the health state of the target battery based on the fusion result and generating the prediction result, the method further includes: The prediction result is updated by using a preset loss function in, , and are adjustable parameters that control the gradient descent speed and positive and negative gradients of the loss function, respectively, and .

2. The battery status prediction method according to claim 1, characterized in that: The step of acquiring the historical data of the target battery and extracting the temporal relationship of the historical data according to the bidirectional long short-term memory neural network to generate a second extraction result includes: Acquiring historical data of the target battery; Extracting the historical data through a bidirectional long short-term memory neural network to obtain a historical data extraction result; A health status feature sequence is determined according to the historical data extraction result as a second extraction result.

3. The battery status prediction method according to claim 2, characterized in that: After the step of determining the health status feature sequence as the second extraction result according to the historical data extraction result, the method further includes: Introducing a time attention mechanism, and focusing on the fluctuation results corresponding to the health status feature sequence according to the time attention mechanism; According to the fluctuation result combined with the health status feature sequence To determine the attention weight corresponding to each element in the health status feature sequence : in, , For the The attention score corresponding to each element, represents the exponential function, represents the weight, represents bias; The attention weight corresponding to each element in the health status feature sequence is multiplied by the corresponding value to generate a second extraction result.

4. A battery status prediction system, characterized in that: The battery state prediction system comprises: An operation characteristic acquisition module, used to acquire multi-dimensional operation characteristic information corresponding to the target battery when determining that the target battery meets the detection conditions, wherein the multi-dimensional operation characteristic information includes: internal resistance information, constant current charging time information, constant voltage charging time information, and peak characteristic information of the capacity increment curve; A first extraction result acquisition module, used for performing feature extraction and weighted fusion on the multi-dimensional operation features by using a multi-layer perceptron network based on channel attention to generate a first extraction result; A second extraction result acquisition module, used to acquire the historical data of the target battery, and extract the temporal relationship of the historical data according to the bidirectional long short-term memory neural network to generate a second extraction result; A result fusion module, used for fusing the first extraction result and the second extraction result to generate a fusion result; A result prediction module, used to predict the health state of the target battery based on the fusion result and generate a prediction result; The running feature acquisition module is also used to monitor the charging state of the target battery in real time; When it is determined that the target battery meets the detection condition, obtaining internal resistance information of the target battery; Obtaining constant current charging time information of the target battery in the constant current charging stage; Acquiring constant voltage charging time information of the target battery in the constant voltage charging stage; Acquire peak characteristic information of the capacity increment curve corresponding to the target battery by means of incremental capacity analysis; Generate multi-dimensional operation characteristic information according to the internal resistance information, the constant current charging time information, the constant voltage charging time information and the peak characteristic information of the capacity increment curve; The first extraction result acquisition module is further used to perform average pooling processing on the multi-dimensional operation features to determine overall feature information; Performing a maximum pooling operation on the multi-dimensional running features to determine salient feature information; The first extraction result acquisition module is further used to determine the hidden layer neuron information and output values ​​of the multi-layer perceptron network to generate multi-layer perceptron network information; Generating a channel attention weight by combining the overall feature information, the prominent feature information, and the multi-layer perceptron network information; Multiplying the channel attention weight by the corresponding salient feature information to determine corresponding feature data; generating a first extraction result according to the feature data; The result prediction module is also used to update the prediction result through a preset loss function in, , and are adjustable parameters that control the gradient descent speed and positive and negative gradients of the loss function, respectively, and .

5. A computer device, characterized in that: The device comprises: a memory and a processor, wherein the processor executes the method according to any one of claims 1 to 3 when running computer instructions stored in the memory.

6. A computer-readable storage medium, characterized in that: The method comprises instructions, which, when executed on a computer, cause the computer to execute the method according to any one of claims 1 to 3.

Images