CN-122017610-A - Method and system for estimating battery state of health based on segment charging data

CN122017610ACN 122017610 ACN122017610 ACN 122017610ACN-122017610-A

Abstract

The invention belongs to the technical field of battery evaluation, and provides a method and a system for estimating the state of health of a battery based on fragment charging data, wherein charging data of a target fragment acquired in real time are calculated to obtain corresponding IC and IE fragments, and peak/valley information corresponding to each fragment is extracted; if the acquired data of the target segment contains a plurality of groups of combined features corresponding to the IC curves and the IE curves, respectively inputting the groups of combined features into corresponding pre-trained sub-models for processing, and carrying out self-adaptive weighted fusion on the output results of the sub-models by using an attention mechanism so as to acquire a final battery health state estimated value; if the data of the target segment only comprises a single group of combined features, the group of features are input into the corresponding submodel, and the estimated value of the battery health state is directly obtained. The invention improves the robustness and generalization capability of SOH estimation.

Inventors

ZHANG CHENGHUI
YU MIAO
SHANG YUNLONG
ZHU YUHAO

Assignees

山东大学

Dates

Publication Date: 20260512
Application Date: 20260416

Claims (10)

1. A method for estimating battery state of health based on segment charge data, comprising the steps of: Acquiring battery parameter data in the charge-discharge test cycle process of the battery, and preprocessing the acquired data; Calculating to obtain an IC curve and an IE curve according to the preprocessed data, matching the corresponding IC curve with peak/valley characteristics in the IE curve in the same voltage interval, combining the corresponding IC curve with the peak/valley characteristics as a group of combined characteristics, and taking the combined characteristics as input and battery health state as output; Building sub-models, and respectively training the single sub-model by utilizing each group of extracted combined characteristics and corresponding battery health state values to obtain each sub-model after training; charging data of a target segment obtained in real time are calculated to obtain corresponding IC and IE segments, and peak/valley information corresponding to each segment is extracted; If the acquired data of the target segment contains a plurality of groups of combined features corresponding to the IC curves and the IE curves, respectively inputting the groups of combined features into corresponding pre-trained sub-models for processing, and carrying out self-adaptive weighted fusion on the output results of the sub-models by using an attention mechanism so as to acquire a final battery health state estimated value; if the data of the target segment only comprises a single group of combined features, the group of features are input into the corresponding submodel, and the estimated value of the battery health state is directly obtained.
2. The method of estimating a state of health of a battery based on segment charge data of claim 1, wherein said battery parameter data comprises battery stress, current, voltage and capacity data; The process of calculating the IC curve and the IE curve according to the preprocessed data comprises the following specific calculation formulas of the IC curve and the IE curve: ; ; wherein Q represents battery capacity, V represents voltage, E represents battery expansion stress, deltaX represents variation of characteristic X of adjacent sampling points, and X is Q, V or E.
3. The method for estimating the battery health state based on the segment charging data according to claim 1, wherein the structure of the sub-model is based on a BP neural network and comprises an input layer, a hidden layer and an output layer which are sequentially connected, the training process of the sub-model comprises the steps of initializing the neural network, determining the number of neurons of the output layer, the hidden layer and the output layer, initializing the connection weight between the layers by using a random initialization method, determining an activation function, correcting a threshold value by using error back propagation of the sub-model, correcting the threshold value by using a gradient descent method according to the output error, judging whether the training of the sub-model reaches the expected state, if not, returning the hidden layer, continuing to iterate the training until the expected state is met, and outputting an SOH estimated value.
4. The method for estimating a state of health of a battery based on segment charge data as recited in claim 1, wherein the hidden layer output calculation formula is as follows: ; Wherein w ij is a weight coefficient between the input layer and the hidden layer, f is an activation function, wherein a Tanh activation function is selected, x i is an output vector, a j is a hidden layer neuron threshold value, and n represents the number of input layer nodes; the output layer calculation formula is as follows: ; wherein b k is a threshold value, w jk is a weight coefficient h j between the hidden layer and the output layer, and l represents the number of neurons in the hidden layer; the neural network error is a mean square error, and the calculation formula is as follows: ; Wherein, the Representing an estimated value of SOH, wherein SOH is a true value expected to be output by the sub-model; The process for correcting the threshold value by adopting the gradient descent method according to the output error comprises the following steps: ; ; ; ; Where η is the learning rate, x i is the input feature, h j is the hidden layer output, e k is the network error, and m is the number of output layer nodes.
5. The method of estimating battery health based on segment charge data of claim 1 wherein matching corresponding IC curves with peak/valley features in IE curves and combining them as a set of joint features in the same voltage interval comprises regarding detectable combinations of peaks/valleys of IE and IC curves in different voltage intervals as joint features and constructing a sub-model for them that is identical in structure but independent of parameters.
6. The method for estimating battery health status based on segment charging data according to claim 1, wherein the process of adaptively weighting and fusing the output results of each sub-model by using an attention mechanism comprises the steps of introducing the attention mechanism to automatically learn hidden features output by all sub-models, dynamically adjusting contribution of different peaks through normalization weights to adapt to the situation of peak quantity change and incomplete data, inputting the weighted and fused global health features to a regression layer to obtain a battery health status estimation result, calculating an estimation error, outputting a battery health status result if an error setting expectation is reached, and updating each neuron weight of the corresponding sub-model if the error setting expectation is not reached until the expectation is reached.
7. The method for estimating battery health based on segment charge data of claim 6 wherein the process of automatically learning hidden features of all submodel outputs by the attention introducing mechanism comprises linearly transforming each hidden feature, projecting BP submodel outputs into a unified attention space, and compressing the mapped features into scalar scores: ; Wherein e i represents the attention score of the ith sub-model, W a represents the attention weight matrix, b a represents the attention bias, v is a learnable attention vector, and tanh (·) represents the hyperbolic tangent activation function for non-linear mapping of the linear transformation result; The normalized weights are calculated as: ; Wherein, in order to adapt to the variation of the peak-valley number, the mask variable m i is introduced, if the peak i exists, m i =1, if the peak i does not exist, m i =0, n represents the peak/valley feature number contained in the segment curve, and α i is the attention weight of the ith sub-model; and carrying out weighted summation on hidden features output by the BP network by using the attention weight, and inputting the weighted summation into a regression layer to obtain an SOH estimated value: ; Wherein, the Representing the SOH estimate, h i representing the submodel output, g (·) representing the fully connected network.
8. The method for estimating battery health based on segment charging data according to claim 1, wherein in the training process of the sub-model, a physical loss function is constructed based on prior physical knowledge in the aging process of the battery, and a model total loss function is constructed in combination with data loss, and when the loss function has not converged or does not meet a preset precision requirement, an iterative optimization process of forward propagation and backward propagation is continuously performed until convergence criteria are met, so that the training is completed.
9. The method of claim 8, wherein a mean square error is used as a data driving loss term for measuring a deviation between a model predicted value and a true value, and the expression is as follows: ; Wherein Y k represents the true SOH value, Representing an SOH estimated value, and m represents the number of samples; Based on the monotonic evolution relation presented between the corresponding voltage values of the IC and IE curve characteristics and the SOH, constructing a loss item driven by physical information: ; Wherein DeltaV k represents the variation of the peak/valley characteristic voltage value corresponding to the data of each curve adjacent segment, Representing the variation of the SOH estimation value in the adjacent fragment data, wherein delta is a tolerance coefficient for emphasizing a monotonous trend of increasing the voltage value corresponding to each peak/valley characteristic as the SOH of the battery is reduced, and simultaneously allowing local non-strict monotonous variation; The total loss function of the model is formed by weighting the data-driven loss term and the physical information constraint term.
10. A system for estimating battery state of health based on segment charge data, comprising: the data acquisition module is configured to acquire battery parameter data in the charge-discharge test cycle process of the battery and preprocess the acquired data; The curve characteristic extraction module is configured to calculate an IC curve and an IE curve according to the preprocessed data, match the corresponding peak/valley characteristics in the IC curve and the IE curve in the same voltage interval, and combine the corresponding IC curve and the corresponding IE curve as a group of combined characteristics, wherein the combined characteristics are taken as input, and the state of health of the battery is taken as output; The sub-model construction module is configured to construct sub-models, and respectively train the single sub-model by utilizing each group of extracted combined characteristics and corresponding battery health state values to obtain each trained sub-model; The operation feature extraction module is configured to calculate the charging data of the target fragment acquired in real time to obtain corresponding IC and IE fragments, and extract peak/valley information corresponding to each fragment; The battery state of health estimation module is configured to input each group of joint characteristics into a corresponding pre-trained sub-model for processing if the acquired data of the target segment contains the joint characteristics corresponding to the plurality of groups of IC curves and IE curves, and adaptively weighting and fusing the output results of each sub-model by using an attention mechanism to acquire a final battery state of health estimation value, and input the group of characteristics into the corresponding sub-model if the data of the target segment contains only a single group of joint characteristics, so as to directly acquire the battery state of health estimation value.

Description

Method and system for estimating battery state of health based on segment charging data Technical Field The invention belongs to the technical field of battery evaluation, and particularly relates to a method and a system for estimating the state of health of a battery based on fragment charging data. Background The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art. With the rapid development of new energy electric vehicles, energy storage systems and portable electronic devices, lithium ion batteries have become the current mainstream energy storage devices due to the advantages of high energy density, long cycle life, low self-discharge rate and the like. During the long-term use of the battery, the capacity and the performance of the battery are inevitably degraded under the influence of electrochemical reaction, accumulation of side reactions, structural change and other factors. Therefore, accurate assessment of the State of Health (SOH) of a lithium ion battery is an important precondition for ensuring safe operation of a battery system, optimizing an energy management strategy, and prolonging the service life of the battery. Among existing SOH estimation methods, an analysis method based on an incremental capacity (INCREMENTAL CAPACITY, IC) curve is receiving attention because it can reflect the characteristics of the electrochemical reaction inside the battery. By differentiating the voltage-capacity curve, characteristic peak information in the IC curve can be obtained, thereby being used for characterizing the battery aging degree. Meanwhile, with the development of battery mechanical property testing means, the volume expansion behavior of the battery in the charge and discharge process is gradually used for representing the internal structure change of the battery, and an incremental expansion (INCREMENTAL EXPANSION, IE) curve constructed based on expansion characteristics also provides a new information dimension for battery health state evaluation. However, existing SOH estimation methods based on IC or IE curves mostly rely on complete charge-discharge cycle data, generally requiring continuous sampling over a wide voltage interval. The premise is difficult to meet in practical application, particularly in the online running process of an electric automobile and an energy storage system, the battery is often in an incomplete charge-discharge or random working condition state, and the acquired data are mostly fragmented and discontinuous partial charge data. In addition, some existing methods only use a single electrochemical or mechanical feature, and complementary information between different physical features cannot be fully mined, so that SOH estimation accuracy and robustness are limited under complex working conditions. Disclosure of Invention In order to solve the problems, the invention provides a method and a system for estimating the state of health of a battery based on fragment charging data. According to some embodiments, the present invention employs the following technical solutions: A method of estimating battery state of health based on segment charge data, comprising the steps of: Acquiring battery parameter data in the charge-discharge test cycle process of the battery, and preprocessing the acquired data; Calculating to obtain an IC curve and an IE curve according to the preprocessed data, matching the corresponding IC curve with peak/valley characteristics in the IE curve in the same voltage interval, combining the corresponding IC curve with the peak/valley characteristics as a group of combined characteristics, and taking the combined characteristics as input and battery health state as output; Building sub-models, and respectively training the single sub-model by utilizing each group of extracted combined characteristics and corresponding battery health state values to obtain each sub-model after training; charging data of a target segment obtained in real time are calculated to obtain corresponding IC and IE segments, and peak/valley information corresponding to each segment is extracted; If the acquired data of the target segment contains a plurality of groups of combined features corresponding to the IC curves and the IE curves, respectively inputting the groups of combined features into corresponding pre-trained sub-models for processing, and carrying out self-adaptive weighted fusion on the output results of the sub-models by using an attention mechanism so as to acquire a final battery health state estimated value; if the data of the target segment only comprises a single group of combined features, the group of features are input into the corresponding submodel, and the estimated value of the battery health state is directly obtained. Alternatively, the battery parameter data includes battery stress, current, voltage, and capacity data. In an alternative embodiment, the