Lithium battery SOC estimation method based on multi-modal data fusion

Abstract

The invention provides a lithium battery SOC estimation method based on multi-modal data fusion, and relates to the field of battery SOC estimation, and the method comprises the steps: S1, in the charging and discharging circulation of a lithium battery, acoustic data and electrical data of the lithium battery are obtained respectively, the acoustic data comprise an ultrasonic waveform, and the electrical data comprise voltage and current; s2, the acoustic data and the electrical data are preprocessed, and acoustic features of the preprocessed acoustic data are extracted; s3, processing the preprocessed electrical data and acoustic features by using two multi-scale convolution modules to obtain electrical depth features and acoustic depth features; s4, fusing the electrical depth feature and the acoustic depth feature by using bidirectional cross attention to obtain a fused feature; and S5, inputting the fusion features into a lightweight TimeMixer model for prediction, and realizing SOC estimation of the lithium battery. The method can improve the estimation precision, and has high robustness.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of battery SOC estimation, and in particular to a lithium battery SOC estimation method based on multi-modal data fusion. BACKGROUND

[0002] Lithium-ion batteries have become the core energy carrier for electric vehicles, portable electronic devices and energy storage systems due to their high energy density, long cycle life and low discharge rate. The battery state of charge (SOC) as a key parameter to characterize the remaining available energy of the battery, its accuracy of estimation has a decisive significance for the reliable operation, life optimization and capacity control strategy of the battery management system (BMS).

[0003] Traditional SOC estimation methods can be roughly divided into ampere-hour integration method, open circuit voltage method, model-driven method and data-driven method. The ampere-hour integration method calculates the charge by time integration of current, but this method is easily affected by the uncertainty of the initial SOC of the battery and the cumulative error in the battery. The open circuit voltage method uses the mapping relationship between the battery SOC and the open circuit voltage, which needs to be static for a period of time to achieve electrochemical equilibrium, so it cannot meet the demand of dynamic working conditions. The model-driven method is divided into equivalent circuit model and electrochemical model, which uses electrical elements or equations to describe the electrochemical behavior inside the lithium battery to obtain accurate SOC estimation. But the identification of model parameters needs to rely on a large number of experiments and calculations, and is sensitive to the nonlinear degradation of the battery. The data-driven method usually uses voltage, current and temperature time series data to construct a regression relationship, but it is easily affected by environmental temperature, battery aging and measurement noise, and is difficult to meet the actual engineering demand of high precision and strong robustness, and only relies on external electrical parameters, it is difficult to reflect the deep information of the internal state of the battery. SUMMARY

[0004] The main purpose of the present application is to provide a lithium battery SOC estimation method based on multi-modal data fusion, which can improve the estimation accuracy and has high robustness.

[0005] The present application is realized by the following technical solutions:

[0006] A lithium battery SOC estimation method based on multi-modal data fusion, comprising the following steps:

[0007] Step S1, in the lithium battery charging and discharging cycle, the acoustic data and electrical data of the lithium battery are obtained respectively, the acoustic data includes ultrasonic waveform, and the electrical data includes voltage and current;

[0008] Step S2, the electrical data and acoustic data are preprocessed respectively, and the acoustic features of the preprocessed acoustic data are extracted;

[0009] Step S3, the electrical data and the acoustic features extracted in step S2 are processed by two multi-scale convolution modules respectively to obtain electrical deep features and acoustic deep features;

[0010] Step S4, the electrical deep features and the acoustic deep features obtained in step S3 are fused by a bidirectional cross attention to obtain fusion features.

[0011] Step S5, the fusion features are input into a lightweight TimeMixer model for prediction to realize SOC estimation of the lithium battery.

[0012] Further, in step S1, the acoustic data of the lithium battery is collected by pulse echo method, the ultrasonic probe is placed on one side of the lithium battery, and the acoustic data is obtained by reflection of ultrasonic waves. The acoustic data is an ultrasonic waveform composed of surface waves with larger amplitude and bottom echo with smaller amplitude.

[0013] Further, in step S2, the preprocessing includes time sequence alignment and removal of outliers.

[0014] Further, in step S2, the bottom echo of the preprocessed ultrasonic waveform is intercepted, Hilbert transform and Fourier transform are performed on the bottom echo respectively, and then SG filter is used for smoothing processing to extract time domain features and frequency domain features respectively. The time domain features include time domain mean, time domain variance, peak-to-peak value and energy integral, and the frequency domain features include frequency domain amplitude, frequency domain variance and spectral fluctuation.

[0015] Further, in step S3, the multi-scale convolution module includes three parallel one-dimensional convolution layer branches, a splicing layer connected with the three one-dimensional convolution layer branches, one-dimensional convolution layers, activation function layers, maximum pooling layers and Dropout layers connected in turn. The one-dimensional convolution layer is connected with the splicing layer. The electrical data preprocessed in step S2 and the acoustic features extracted in step S2 are input into the three one-dimensional convolution layer branches of the corresponding multi-scale convolution module respectively. After the corresponding Dropout layer, the electrical deep features and the acoustic deep features are obtained.

[0016] Further, the convolution kernels of the three one-dimensional convolution layer branches are 1, 3 and 5 respectively.

[0017] Further, in step S4, in the first direction, the acoustic deep features are taken as the Query value, the electrical deep features are taken as the Key value and the Value value, in the second direction, the electrical deep features are taken as the Query value, and the acoustic deep features are taken as the Key value and the Value value. The outputs in the first direction and the second direction are added to obtain the fusion features.

[0018] Further, in the step S4, the fusion feature is represented as , is the output in the first direction, is the output in the second direction, wherein, , , , , , , is the acoustic depth feature, is the electrical depth feature, , is the weight matrix representing Query, , is the weight matrix representing Key, , is the weight matrix representing Value.

[0019] Further, the step S5 includes the following steps:

[0020] Step S51, according to the formula extract the trend component of the fusion feature, represents the trend component at time step t, W represents the number of adjacent windows corresponding to the window at time step t, which is an odd number, represents the fusion feature data intercepted at the t+i window;

[0021] Step S52, according to the formula obtain the seasonal component of the fusion feature ;

[0022] Step S53, send the seasonal component into the multi-layer perception machine for linear transformation, the multi-layer perception machine includes two fully connected layers and GELU activation function, and the transformation process is represented as wherein, , are the weight and bias of the first fully connected layer of the MLP respectively, , are the weight and bias of the second fully connected layer of the MLP respectively;

[0023] Step S54, add the result obtained in step S53 to the trend component, and perform residual connection and normalization to realize the SOC estimation of the lithium battery.

[0024] From the above description of the present application, compared with the prior art, the present application has the following beneficial effects:

[0025] This invention first acquires and preprocesses acoustic and electrical data of a lithium battery. Then, it uses a multi-scale convolution module to process the preprocessed electrical data and the extracted acoustic features, obtaining electrical depth features and acoustic depth features, respectively. Next, it uses bidirectional cross-attention to fuse the electrical and acoustic depth features, resulting in fused features. Finally, the fused features are input into a lightweight TimeMixer model for prediction, achieving SOC estimation of the lithium battery. In this process, the multi-scale convolution module captures key information from the data at different scales, deeply and comprehensively extracts complex features from the acoustic data, and uses bidirectional cross-attention for feature fusion. This achieves effective alignment and deep fusion of feature information from different modalities, comprehensively describing the lithium battery state, overcoming the limitations of single-modal data, ensuring the comprehensiveness and robustness of the information used for estimation, and thus effectively improving the accuracy and robustness of SOC estimation. Attached Figure Description

[0026] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] Figure 1 This is a flowchart of the present invention.

[0028] Figure 2 This is a schematic diagram of the ultrasonic waveform of the present invention.

[0029] Figure 3 This is a schematic diagram of the multi-scale convolution structure of the present invention.

[0030] Figure 4 This is a schematic diagram of the bidirectional cross-attention structure of the present invention. Detailed Implementation

[0031] The present invention will be further described below through specific embodiments.

[0032] like Figure 1 As shown, the lithium battery SOC estimation method based on multimodal data fusion includes the following steps:

[0033] Step S1: During the lithium battery charge-discharge cycle, acquire the acoustic data and electrical data of the lithium battery respectively. The acoustic data includes ultrasonic waveforms, and the electrical data includes voltage and current.

[0034] Specifically, acoustic data of the lithium battery is acquired using a pulse-echo method. An ultrasonic probe is placed on one side of the lithium battery, and acoustic data is obtained by utilizing the reflection of ultrasonic waves. Figure 2 As shown, the acoustic data is an ultrasonic waveform composed of a surface wave with a larger amplitude and a bottom echo with a smaller amplitude. The process of acquiring the electrical data is existing technology.

[0035] Step S2, respectively, pre-process the electrical data and the acoustic data, and extract acoustic features of the pre-processed acoustic data;

[0036] The pre-processing specifically includes time alignment and removal of outliers. Since the acoustic data and the electrical data are collected simultaneously using two independent devices, time alignment, i.e. time alignment, is needed when processing the data. The specific process of time alignment and removal of outliers is prior art.

[0037] The bottom surface echo of the pre-processed ultrasonic waveform is intercepted, Hilbert transform and Fourier transform are performed on the bottom surface echo respectively, and SG filter (Savitzky-Golay) is used for smoothing processing to extract time domain features and frequency domain features respectively. The time domain features include time domain mean, time domain variance, peak-to-peak value and energy integral, and the frequency domain features include frequency domain amplitude, frequency domain variance and spectral fluctuation. The process of intercepting the bottom surface echo is prior art.

[0038] Step S3, using two multi-scale convolution modules to process the electrical data pre-processed in step S2 and the acoustic features extracted in step S2, to obtain electrical deep features and acoustic deep features;

[0039] As shown in Figure 3 The multi-scale convolution module includes three parallel one-dimensional convolution layer branches, a splicing layer connected with the three one-dimensional convolution layer branches respectively, one-dimensional convolution layers, activation function layers, maximum pooling layers and Dropout layers connected in turn, and the one-dimensional convolution layers are connected with the splicing layer. The pre-processed electrical data and the extracted acoustic features are respectively input into the three one-dimensional convolution layer branches of the corresponding multi-scale convolution module, the outputs of each one-dimensional convolution layer branch are spliced in the feature dimension to obtain high-dimensional features containing multi-scale information, and the high-dimensional features are sequentially subjected to one-dimensional convolution layers, activation function layers, maximum pooling layers and Dropout layers to obtain electrical deep features and acoustic deep features. The convolution kernels of the three one-dimensional convolution layer branches are 1, 3 and 5 respectively, the convolution kernel of the one-dimensional convolution layer is 1, the pooling size of the maximum pooling layer is 2, the activation function layer adopts Softmax activation function, and the dropout rate of the Dropout layer is 0.1.

[0040] Step S4, using bidirectional cross attention to fuse the electrical deep features and the acoustic deep features obtained in step S3 to obtain fusion features;

[0041] As shown in Figure 4As shown, in the first direction, the acoustic depth feature is taken as the Query value, the electrical depth feature is taken as the Key value and the Value value, in the second direction, the electrical depth feature is taken as the Query value, and the acoustic depth feature is taken as the Key value and the Value value, and addition is performed to obtain the fusion feature, and then the outputs in the first direction and the second direction are obtained to obtain the fusion feature .

[0042] Specifically: in the first direction, first linear mapping is performed, that is, the acoustic depth feature and the electrical depth feature are respectively mapped into the Query, Key, and Value vectors required by the attention mechanism by being multiplied by respective weight matrices: , , Then, the Query, Key, and Value vectors are substituted into the attention formula to obtain the output in the first direction ;

[0043] The process in the second direction is completely symmetrical to the process in the first direction, except that the sources of the Query, Key, and Value vectors are changed. Specifically: first linear mapping is performed, that is, the electrical depth feature and the acoustic depth feature are respectively mapped into the Query, Key, and Value vectors required by the attention mechanism by being multiplied by respective weight matrices: , , Then, the Query, Key, and Value vectors are substituted into the attention formula to obtain the output in the first direction ; wherein, is the acoustic depth feature, is the electrical depth feature, , is the weight matrix representing Query, , is the weight matrix representing Key, , is the weight matrix representing Value;

[0044] The fusion feature is represented as is the output in the first direction, is the output in the second direction.

[0045] Step S5, input the fusion feature into a lightweight TimeMixer model for prediction to realize SOC estimation of the lithium battery;

[0046] Specifically, the following steps are included:

[0047] Step S51, according to the formula extract the trend component of the fusion feature, denotes the trend component at time step t, W denotes the number of adjacent windows corresponding to the window at time step t, which is an odd number, denotes the fusion feature data intercepted in the t+i window;

[0048] Step S52, according to the formula obtain the seasonal component of the fusion feature ;

[0049] Step S53, the seasonal component is sent to the multi-layer perception for linear transformation, the multi-layer perception includes two fully connected layers and GELU activation function, and the transformation process is represented as , wherein, , are the weight and bias of the first fully connected layer of the MLP respectively, , are the weight and bias of the second fully connected layer of the MLP respectively;

[0050] Step S54, the result obtained in step S53 is added to the trend component, and residual connection and normalization are performed to realize the SOC estimation of the lithium battery.

[0051] Experiments have verified that the average absolute error MAE of the present application is 1.07%, which is about 15% higher than the existing method, effectively improving the estimation accuracy.

[0052] In the present application, the terms "first", "second", "third" and the like are only used to distinguish similar objects, and do not have to be used to describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. In the description, the directions or positional relationships indicated by "up", "down", "left", "right", "front" and "back" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present application, and do not indicate or imply that the devices referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the scope of protection of the present application. For ordinary skilled persons in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0053] In addition, in the description of the present application, "a plurality of" means two or more, unless otherwise specified. The association relationship between the associated objects is described as "and / or", which means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone. The character " / " generally represents an "or" relationship between the associated objects before and after it.

[0054] The above merely illustrates the specific embodiments of the present application, but the design concept of the present application is not limited thereto, and any non-essential modification of the present application by using the concept shall be deemed as the infringement of the protection scope of the present application.

Claims

1. A lithium battery SOC estimation method based on multi-modal data fusion, characterized in that: The method comprises the following steps: Step S1, in the lithium battery charge-discharge cycle, the acoustic data and the electrical data of the lithium battery are obtained respectively, the acoustic data includes ultrasonic waveform, and the electrical data includes voltage and current; Step S2, the electrical data and the acoustic data are preprocessed respectively, and the acoustic features of the preprocessed acoustic data are extracted; Step S3, the electrical data preprocessed in step S2 and the acoustic features extracted in step S2 are processed by using two multi-scale convolution modules to obtain electrical depth features and acoustic depth features; Step S4, the electrical depth features and the acoustic depth features obtained in step S3 are fused by using a bidirectional cross attention to obtain fusion features; Step S5, the fusion features are input into a lightweight TimeMixer model for prediction to realize SOC estimation of the lithium battery.

2. The lithium battery SOC estimation method based on multi-modal data fusion according to claim 1, characterized in that: In the step S1, the acoustic data of the lithium battery is collected by using the pulse echo method, the ultrasonic probe is placed on one side of the lithium battery, and the acoustic data is obtained by using the reflection of ultrasonic waves, which is an ultrasonic waveform composed of surface waves with larger amplitude and bottom echo with smaller amplitude.

3. The lithium battery SOC estimation method based on multi-modal data fusion according to claim 2, characterized in that: In the step S2, the preprocessing includes time sequence alignment and removal of abnormal values.

4. The lithium battery SOC estimation method based on multi-modal data fusion according to claim 1 or 2 or 3, characterized in that: In the step S2, the bottom echo of the preprocessed ultrasonic waveform is intercepted, Hilbert transform and Fourier transform are performed on the bottom echo respectively, and then SG filter is used for smoothing processing to extract time domain features and frequency domain features respectively, the time domain features include time domain mean, time domain variance, peak-to-peak value and energy integral, and the frequency domain features include frequency domain amplitude, frequency domain variance and spectral fluctuation.

5. The lithium battery SOC estimation method based on multi-modal data fusion according to claim 1 or 2 or 3, characterized in that: In the step S3, the multi-scale convolution module comprises three parallel one-dimensional convolution layer branches, a splicing layer connected with the three one-dimensional convolution layer branches, one-dimensional convolution layers connected in turn, an activation function layer, a maximum pooling layer and a Dropout layer, the one-dimensional convolution layer is connected with the splicing layer, the electrical data preprocessed in step S2 and the acoustic features extracted in step S2 are input into the three one-dimensional convolution layer branches of the corresponding multi-scale convolution module, and the electrical depth features and the acoustic depth features are obtained after the corresponding Dropout layer.

6. The lithium battery SOC estimation method based on multi-modal data fusion according to claim 5, characterized in that: The convolution kernels of the three one-dimensional convolution layer branches are 1, 3 and 5 respectively.

7. The lithium battery SOC estimation method based on multi-modal data fusion according to claim 1 or 2 or 3, characterized in that: In the step S4, in the first direction, the acoustic depth features are taken as the Query value, and the electrical depth features are taken as the Key value and the Value value, in the second direction, the electrical depth features are taken as the Query value, and the acoustic depth features are taken as the Key value and the Value value, the outputs in the first direction and the second direction are added to obtain the fusion features.

8. The lithium battery SOC estimation method based on multi-modal data fusion according to claim 7, characterized in that: The step S4, the fusion feature is expressed as , is the output in the first direction, is the output in the second direction, wherein, , , , , , , is the acoustic depth feature, is the electrical depth feature, , is the weight matrix representing Query, , is the weight matrix representing Key, , is the weight matrix representing Value.

9. The lithium battery SOC estimation method based on multi-modal data fusion according to claim 7, characterized in that: The step S5 comprises the following steps: Step S51, according to the formula extracting the trend component of the fusion feature, denotes the trend component at time step t, W denotes the number of adjacent windows corresponding to the window at time step t, which is an odd number, denotes the fusion feature data intercepted at the t+i th window; Step S52, obtaining seasonal components of the fused features according to the formula ;​ Step S53, the seasonal component is sent into a multi-layer perception for linear transformation, the multi-layer perception includes two fully connected layers and a GELU activation function, and the transformation process is represented as wherein, , , are the weight and bias of the first fully connected layer of the MLP, respectively, , are the weight and bias of the second fully connected layer of the MLP, respectively. Step S54, the result obtained in step S53 is added to the trend component, residual connection and normalization are performed to realize SOC estimation of the lithium battery.

Images