Battery health prediction method and system based on adaptive feature fusion and double-branch frequency domain decomposition, and storage medium

Abstract

The invention relates to the technical field of new energy and artificial intelligence, in particular to a battery health prediction method and system based on adaptive feature fusion and double-branch frequency domain decomposition, and a storage medium. Fusion key features obtained based on attention mechanism fusion are sorted to form a fusion key feature sequence, the fusion key feature sequence is divided into a low-frequency feature sequence and a high-frequency feature sequence through a threshold value, the low-frequency feature sequence serves as a sampling sequence of a battery degradation trend, and the high-frequency feature sequence serves as a sampling sequence of a capacity regeneration phenomenon. And a battery degradation trend characteristic and a nonlinear fluctuation characteristic are obtained after extraction by a bidirectional gating circulation unit and extraction and capture by a complex valued neural network, so that a battery health result is predicted based on the two characteristics. The invention aims to solve the problem of how to capture the characteristics of the long-term attenuation trend and the capacity regeneration phenomenon of the battery.

Description

Technical Field

[0001] This application relates to the fields of new energy and artificial intelligence technology, and in particular to a battery health prediction method, system and storage medium based on adaptive feature fusion and dual-branch frequency domain decomposition. Background Technology

[0002] With the rapid development of the new energy industry, the health status of batteries, as core energy storage components, directly determines the operational reliability, range, and lifespan of equipment. Battery State of Health (SOH) is typically defined as the ratio of the battery's current capacity to its rated capacity, and is a key indicator for measuring battery performance degradation. However, during long-term cyclic use, batteries are affected by factors such as material aging (e.g., electrode active material shedding, electrolyte decomposition), fluctuations in operating conditions (e.g., dynamic charge / discharge current, temperature changes), and environmental disturbances (e.g., humidity, vibration), causing SOH to decline non-linearly, which poses challenges to traditional prediction methods.

[0003] Related technical solutions often employ time-domain modeling to predict the battery's state of equilibrium (SOH) degradation curve. However, this method of plotting the SOH curve overlooks two phenomena that are difficult to capture simultaneously: the long-term degradation trend and capacity regeneration. The former refers to the irreversible, slow decline in battery capacity with increasing time and cycle count, and this trend increases irregularly with the complexity of battery operating conditions. The latter refers to the short-term, small-scale, temporary recovery or fluctuation in battery capacity under a long-term degradation trend, but it is not a true capacity rebound. Both of these phenomena can lead to a significant error between the predicted and actual SOH values.

[0004] In view of this, this application proposes a new method for predicting the state of battery health, which aims to capture the characteristics of long-term degradation trends and capacity regeneration phenomena, thereby improving the accuracy of battery SOH prediction under complex operating conditions. Summary of the Invention

[0005] The main objective of this application is to provide a battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition, which aims to solve the problem of how to capture the characteristics of the long-term degradation trend and capacity regeneration phenomenon of batteries.

[0006] To achieve the above objectives, this application provides a battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition, the method comprising:

[0007] S10, obtain the fusion key features of the battery to be predicted, wherein the fusion key features are obtained by weighted fusion of the candidate battery health features of the battery to be predicted using an attention mechanism;

[0008] S20: Acquire the fusion key features collected within a preset time period and sort them in time sequence to form a fusion key feature sequence; determine the sequence representation of the fusion key feature sequence in the frequency domain; and divide the sequence representation into low-frequency feature sequences and high-frequency feature sequences according to a preset frequency threshold.

[0009] S30, extract battery degradation trend features from the low-frequency feature sequence based on a bidirectional gated cyclic unit; and extract nonlinear fluctuation features from the high-frequency feature sequence based on a complex-valued neural network;

[0010] S40, Based on the battery degradation trend characteristics and the nonlinear fluctuation characteristics, generate the battery health prediction result of the battery to be predicted.

[0011] Optionally, in step S30, the step of extracting battery degradation trend features includes:

[0012] S31, Define the low-frequency signal sequence:

[0013]

[0014] In the formula, The sequence length is given; for each time step... The low-frequency signal input is used, and the low-frequency branch performs gating state updates through the GRU unit:

[0015] ;

[0016] ;

[0017] ;

[0018] ;

[0019] In the formula, Represents the door update. This represents resetting the door. Represents a hidden state. , and These represent the input weight matrix of the update gate, the input weight matrix of the reset gate, and the input weight matrix of the candidate hidden state, respectively. , and Let represent the hidden state weight matrix of the update gate, the hidden state weight matrix of the reset gate, and the hidden state weight matrix of the candidate hidden state, respectively. , and These represent the bias vectors of the update gate, the reset gate, and the candidate hidden state, respectively. This represents the bias vector of the update gate, the bias vector of the reset gate, and the bias vector of the candidate hidden state.

[0020] S32, introduces a bidirectional structure to extract low-frequency signal input. Historical information And future information :

[0021] ;

[0022] ;

[0023] Battery degradation trend characteristics were obtained. , is represented as:

[0024] ;

[0025] The steps for extracting nonlinear fluctuation characteristics include:

[0026] S33, converts the high-frequency signal of the high-frequency branch. Represented in complex form :

[0027] ;

[0028] In the formula, It is amplitude information. It is phase information, where T represents the length of the time series involved in the modeling;

[0029] S32, obtains the high-frequency signal in complex form through complex linear mapping. :

[0030] l=1,2,…,L;

[0031] In the formula, W represents the complex weight matrix, b represents the complex bias, and L represents the neural network layer;

[0032] definition:

[0033] ;

[0034] Will Expand into real and imaginary parts for calculation:

[0035] ;

[0036] The activation function uses a complex-valued nonlinear mapping form:

[0037] l=1,2,…,L;

[0038] S34, by stacking complex-valued neural network layers, obtains nonlinear fluctuation characteristics. :

[0039]

[0040] In the formula, It is a complex-valued mapping function.

[0041] Optionally, step S40 includes:

[0042] S41, calculate the frequency domain energy distribution of the low-frequency feature sequence and the high-frequency feature sequence respectively:

[0043]

[0044]

[0045] In the formula, Indicates low-frequency feature sequences, Represents high-frequency feature sequences; Represents the frequency domain energy distribution of low-frequency characteristic sequences; Represents the frequency domain energy distribution of high-frequency characteristic sequences;

[0046] S42, Calculate the corresponding attention weights for the two based on their frequency domain energy distributions:

[0047]

[0048]

[0049] In the formula, Attention weights for low-frequency feature sequences; Attention weights for high-frequency feature sequences;

[0050] S43, feature fusion is performed based on the attention weights of the two to obtain fused features. :

[0051]

[0052] In the formula, This indicates the characteristics of battery degradation trends. Indicates nonlinear fluctuation characteristics;

[0053] S44, based on fusion features A prediction is made, generating a battery health prediction result for the battery to be predicted.

[0054] Optionally, the selection step for the candidate battery health features includes:

[0055] Select initial features from the collected preprocessed battery running dataset;

[0056] Determine the correlation coefficient between the initial features and the measured battery health values ​​in the battery health feature dataset;

[0057] Initial features with a correlation coefficient greater than a preset value are used as candidate battery health features.

[0058] Optionally, the initial features include time-domain features, statistical features, and cyclic features;

[0059] The time-domain features include at least one of the following: charging voltage plateau length, discharge voltage drop rate, and charge / discharge energy difference.

[0060] The statistical characteristics include at least one of voltage fluctuation variance, peak current percentage, and mean temperature.

[0061] The cycle characteristics include at least one of unit cycle capacity decay and cycle lifetime percentage.

[0062] Optionally, the preprocessing steps for the preprocessed battery running dataset include:

[0063] The dynamic current data / dynamic voltage data in the initial battery operation dataset are determined, and the peak noise in the dynamic current data / dynamic voltage data is eliminated by a sliding window mid-range filter. The length of the sliding window is adjusted based on the data fluctuation frequency of the dynamic current data / dynamic voltage data.

[0064] Identify abnormal data in the initial battery running dataset and repair the abnormal data using linear interpolation.

[0065] The initial battery running dataset, which has been repaired by sliding window midpoint filtering and linear interpolation, is normalized to obtain the preprocessed battery running dataset.

[0066] Optionally, the step of generating the fusion key features includes:

[0067] Determine the health characteristics of each candidate battery Corresponding feature weights :

[0068]

[0069] In the formula, This is the weighting adjustment coefficient; Let be the correlation coefficient between the j-th candidate battery health feature and the battery health state. Let be the attention function. This represents the correlation coefficient between the j-th candidate battery health characteristic and the battery health state;

[0070] Health characteristics of each candidate battery and their corresponding trophy weights By performing a weighted sum, the key features of the fusion are obtained:

[0071]

[0072] In the formula, m is the number of candidate battery health features.

[0073] Optionally, the battery health prediction result includes a battery health prediction value;

[0074] The battery health level of the battery to be predicted is determined based on the range in which the predicted battery health value falls.

[0075] In addition, to achieve the above objectives, this application also provides a computer system, the computer system comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, it implements the steps of the battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition as described in any of the preceding claims.

[0076] In addition, to achieve the above objectives, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition as described in any of the preceding claims.

[0077] This application has at least the following beneficial effects:

[0078] 1. By fusing key features obtained based on attention mechanisms, a precise mapping between features and SOH is established;

[0079] 2. After sorting the key features within a certain time window to form a key feature sequence, the key features are divided into low-frequency feature sequences and high-frequency feature sequences by a threshold. The low-frequency feature sequence is used as the sampling sequence for battery degradation trend, and the high-frequency feature sequence is used as the sampling sequence for capacity regeneration phenomenon, so as to realize data sampling of battery degradation trend and capacity regeneration phenomenon.

[0080] 3. After extraction by a bidirectional gated recurrent unit and extraction by a complex-valued neural network, the battery degradation trend characteristics and nonlinear fluctuation characteristics are obtained. Based on these characteristics, the battery health outcome is predicted, and dynamic prediction of SOH is achieved. Attached Figure Description

[0081] Figure 1 This is a flowchart illustrating the battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition involved in the embodiments of this application.

[0082] Figure 2 This is a schematic diagram of the hardware operating environment of the computer system involved in the embodiments of this application.

[0083] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0084] To better understand the above technical solutions, exemplary embodiments of this disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of this disclosure to those skilled in the art.

[0085] First Embodiment

[0086] Reference Figure 1 This embodiment provides a battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition, the method comprising the following steps:

[0087] S10, obtain the fusion key features of the battery to be predicted, wherein the fusion key features are obtained by weighted fusion of the health features of the candidate batteries to be predicted using an attention mechanism;

[0088] In this embodiment, the innovative key feature fusion proposed in this embodiment is first obtained. The key feature fusion refers to integrating multiple candidate battery health features that are strongly correlated with the health status of the battery to be predicted into a feature set. It is obtained by weighted fusion of the candidate battery health features of the battery to be predicted using an attention mechanism.

[0089] Optionally, prior to performing this step, the selection of candidate battery health features may include:

[0090] S11, Select the initial features from the collected preprocessed battery running dataset;

[0091] S12, determine the correlation coefficient between the initial features and the measured battery health values ​​of the battery health feature dataset;

[0092] S13, the initial features that are greater than the preset correlation coefficient are used as the candidate battery health features.

[0093] In some optional implementations, this embodiment uses a modified Pearson correlation analysis method to calculate the correlation coefficient. Notably, the improvement lies in introducing a time decay weight, with a higher weight given to more recent battery data, thus enhancing the impact of recent data on the correlation. The specific expression is:

[0094] r

[0095] In the formula, Let i be the eigenvalue of the i-th initial feature; This corresponds to the measured battery health value; , These represent the mean of the eigenvalues ​​and the mean of the measured battery health values, respectively, where n represents the total number of initial eigenvalues. The time decay weight is expressed as follows:

[0096]

[0097] In some optional implementations, the preset correlation coefficient is set to 0.6, and the correlation coefficient is... The characteristics were used to determine the set of candidate battery health characteristics. .

[0098] In some alternative implementations, the characteristic values ​​of the initial features are calculated based on the battery's historical charge and discharge data or relevant measurement data.

[0099] Optionally, considering that different batteries to be predicted may exhibit different characteristics under different operating conditions, this embodiment also provides a method for selecting initial features. In this embodiment, we classify initial features into three types: time-domain features, statistical features, and cyclic features. When constructing initial feature data, the following types of data can be referenced for selection:

[0100] Time-domain characteristics include the length of the charging voltage plateau. Discharge voltage drop rate and the difference in charge and discharge energy At least one of them;

[0101] The statistical characteristics include voltage fluctuation variance. Current peak percentage and temperature average At least one of them;

[0102] The cyclic characteristics include the unit cyclic capacity decay. and cycle life percentage At least one of them.

[0103] For example:

[0104]

[0105]

[0106] In the formula, Let N be the capacity of the Nth iteration; Rated cycle life; N represents the current number of battery cycles; This indicates the battery's rated cycle life.

[0107] Optionally, the initial battery run dataset can be processed using the following steps before selecting initial features:

[0108] S14, determine the dynamic current data / dynamic voltage data in the initial battery operation dataset, and use a sliding window mid-range filter to eliminate the spike noise in the dynamic current data / dynamic voltage data, wherein the length of the sliding window is adjusted based on the data fluctuation frequency of the dynamic current data / dynamic voltage data;

[0109] In some alternative implementations, a sliding window mid-range filter is used to eliminate spike noise, and the window length L is adaptively adjusted based on the data fluctuation frequency.

[0110]

[0111] In the formula, The preset initial sliding window length The standard deviation of the internal data is calculated as follows:

[0112]

[0113] In the formula, k is the window size. Internal data points; Let i be the i-th data point; For window Internal data mean.

[0114] S15, Identify abnormal data in the initial battery operation dataset and repair the abnormal data using linear interpolation;

[0115] In some alternative implementations, based on 3 The principle for identifying outlier data is as follows: if data point x satisfies |x− |>3 If the problem persists, linear interpolation will be used for repair. The interpolation formula is:

[0116]

[0117] In the formula, t is the timestamp corresponding to the abnormal data; , These are the timestamps of the adjacent valid data before and after the abnormal data; , For the corresponding data values, Let x be the standard deviation of the data point.

[0118] S16, the initial battery running dataset that has been repaired by sliding window midpoint filtering and linear interpolation is normalized to obtain the preprocessed battery running dataset.

[0119] In some alternative implementations, Min-Max normalization is used to map the preprocessed data to the [0, 1] interval to eliminate the influence of dimensions:

[0120]

[0121] In the formula, These are the minimum and maximum values ​​of the dataset, respectively.

[0122] Furthermore, and optionally, in this step, the fusion of key features can be generated using the following steps:

[0123] S17, Determine the health characteristics of each candidate battery. Corresponding feature weights :

[0124]

[0125] In the formula, This is the weighting adjustment coefficient; Let be the correlation coefficient between the j-th candidate battery health feature and the battery health state. Let be the attention function. This represents the correlation coefficient between the k-th candidate battery health feature and the battery health status;

[0126] S18, the health characteristics of each candidate battery and their corresponding trophy weights By performing a weighted sum, the key features of the fusion are obtained:

[0127]

[0128] In the formula, m is the number of candidate battery health features.

[0129] It should be noted that the order of the above steps only represents the order in which the steps appear in the embodiment, and does not represent their execution order. For the specific execution order, please refer to the textual limitations above.

[0130] S20: Acquire the fusion key features collected within a preset time period and sort them in time sequence to form a fusion key feature sequence; determine the sequence representation of the fusion key feature sequence in the frequency domain; and divide the sequence representation into low-frequency feature sequences and high-frequency feature sequences according to a preset frequency threshold.

[0131] In this embodiment, after obtaining the key fusion features, the key fusion features are sorted in chronological order to form a key fusion feature sequence:

[0132] For example, the mathematical expression for fusing key feature sequences is:

[0133]

[0134] Then, the sequence representation of the fusion key feature sequence in the frequency domain is determined.

[0135] For example, the sequence representation in the frequency domain is obtained by using the Fast Fourier Transform. :

[0136]

[0137] Next, the sequence representation is divided into low-frequency feature sequences and high-frequency feature sequences according to a preset frequency threshold:

[0138] For example, let the frequency threshold be... Then we have:

[0139]

[0140]

[0141] In the formula, It is a low-frequency characteristic sequence representing the long-term degradation trend of batteries. It is a high-frequency characteristic sequence that characterizes local capacity regeneration and local fluctuation characteristics.

[0142] The frequency threshold Determined by preset parameters:

[0143]

[0144] In the formula, The energy percentage threshold is set to 0.9.

[0145] In this step, we perform modeling in the frequency domain to obtain low-frequency and high-frequency feature sequences. The low-frequency feature sequence reflects the long-term degradation trend of the battery under test, while the high-frequency feature sequence reflects the capacity regeneration and local fluctuation characteristics of the battery under test, thus providing input for subsequent extraction of corresponding battery degradation features and high-frequency features.

[0146] S30, extract battery degradation trend features from the low-frequency feature sequence based on a bidirectional gated cyclic unit; and extract nonlinear fluctuation features from the high-frequency feature sequence based on a complex-valued neural network;

[0147] After obtaining the low-frequency and high-frequency feature sequences, this embodiment innovatively proposes a dual-branch temporal coding layer architecture. Specifically, it includes:

[0148] S31, Define the low-frequency signal sequence:

[0149]

[0150] In the formula, The sequence length is given; for each time step... The low-frequency signal input is used, and the low-frequency branch performs gating state updates through the GRU unit:

[0151]

[0152]

[0153]

[0154]

[0155] In the formula, Represents the door update. This represents resetting the door. Represents a hidden state. , and These represent the input weight matrix of the update gate, the input weight matrix of the reset gate, and the input weight matrix of the candidate hidden state, respectively. , and Let represent the hidden state weight matrix of the update gate, the hidden state weight matrix of the reset gate, and the hidden state weight matrix of the candidate hidden state, respectively. , and These represent the bias vectors of the update gate, the reset gate, and the candidate hidden state, respectively. This represents the bias vector of the update gate, the bias vector of the reset gate, and the bias vector of the candidate hidden state.

[0156] S32, introduces a bidirectional structure to extract low-frequency signal input. Historical information And future information :

[0157]

[0158]

[0159] Battery degradation trend characteristics were obtained. , is represented as:

[0160]

[0161] The steps for extracting nonlinear fluctuation characteristics include:

[0162] S33, converts the high-frequency signal of the high-frequency branch. Represented in complex form :

[0163]

[0164] In the formula, It is amplitude information. It is phase information, where T represents the length of the time series involved in the modeling;

[0165] S32, obtains the high-frequency signal in complex form through complex linear mapping. :

[0166] l=1,2,…,L

[0167] In the formula, W represents the complex weight matrix, b represents the complex bias, and L represents the neural network layer;

[0168] definition:

[0169]

[0170] Will Expand into real and imaginary parts for calculation:

[0171]

[0172] The activation function uses a complex-valued nonlinear mapping form:

[0173] l=1,2,…,L

[0174] S34, by stacking complex-valued neural network layers, obtains nonlinear fluctuation characteristics. :

[0175]

[0176] In the formula, It is a complex-valued mapping function.

[0177] S40, Based on the battery degradation trend characteristics and the nonlinear fluctuation characteristics, generate the battery health prediction result of the battery to be predicted.

[0178] After obtaining the battery degradation trend characteristics and nonlinear fluctuation characteristics, these two are input into the prediction model for prediction, thereby generating the battery health prediction result of the battery to be predicted.

[0179] Furthermore, and optionally, to facilitate better prediction by the prediction model, the two are processed as follows:

[0180] S41, calculate the frequency domain energy distribution of the low-frequency feature sequence and the high-frequency feature sequence respectively:

[0181]

[0182]

[0183] In the formula, Indicates low-frequency feature sequences, Represents high-frequency feature sequences; Represents the frequency domain energy distribution of low-frequency characteristic sequences; Represents the frequency domain energy distribution of high-frequency characteristic sequences;

[0184] S42, Calculate the corresponding attention weights for the two based on their frequency domain energy distributions:

[0185]

[0186]

[0187] In the formula, Attention weights for low-frequency feature sequences; Attention weights for high-frequency feature sequences;

[0188] S43, feature fusion is performed based on the attention weights of the two to obtain fused features. :

[0189]

[0190] In the formula, This indicates the characteristics of battery degradation trends. This indicates nonlinear fluctuation characteristics.

[0191] S44, based on fusion features A prediction is made, generating a battery health prediction result for the battery to be predicted.

[0192] In some alternative implementations, step S44 may involve inputting the fused features into the fully connected output layer of the prediction model for result prediction. The fully connected output layer comprises two hidden layers (32 and 16 dimensions respectively, with ReLU activation function) and one output layer (with Sigmoid activation function).

[0193] Further and optionally, the battery health prediction result includes a battery health prediction value; the steps include:

[0194] The battery health level of the battery to be predicted is determined based on the range in which the predicted battery health value falls.

[0195] For example, the battery health level includes the following three levels:

[0196] Level 1 warning (battery health prediction value) ): Normal operation, it is recommended to update the prediction model every 100 cycles;

[0197] Level 2 warning (battery health prediction value) Performance degradation: It is recommended to reduce the charge / discharge current to 80% of the rated value and increase the frequency of temperature monitoring.

[0198] Level 3 Warning (Battery Health Prediction) ): If the battery needs to be replaced, a maintenance work order will be triggered, and a backup battery replacement plan will be pushed out simultaneously.

[0199] In the technical solution provided in this embodiment, the key features obtained by fusion based on the attention mechanism are sorted to form a key feature sequence and then divided into a low-frequency feature sequence and a high-frequency feature sequence by a threshold. The low-frequency feature sequence is used as the sampling sequence for battery degradation trend and the high-frequency feature sequence is used as the sampling sequence for capacity regeneration phenomenon. After extraction by a bidirectional gated recurrent unit and extraction by a complex value neural network, the battery degradation trend feature and nonlinear fluctuation feature are obtained, and the battery health result is predicted based on the two.

[0200] Second Embodiment

[0201] Based on the first embodiment, in order to ensure the accuracy of battery health prediction results, this embodiment provides a training process for a prediction model, specifically including:

[0202] (1) Loss function

[0203] The loss is weighted by root mean square error (RMSE) and mean absolute percentage error (MAPE):

[0204]

[0205] in,

[0206]

[0207]

[0208] In the formula, This represents the predicted SOH value for the i-th sample. This represents the true SOH value of the i-th sample. This indicates the total number of samples used to evaluate the number of cycles or data points.

[0209] (2) Optimizer

[0210] The Adam optimizer was used with an initial learning rate of lr=0.001 and a learning rate decay strategy of decaying to 0.9 times the previous rate every 100 epochs.

[0211] (3) Training termination conditions

[0212] The validation set loss does not decrease for 20 consecutive epochs, or the total number of epochs reaches 1000.

[0213] (4) Verify the prediction results using a multi-index verification mechanism and output the final SOH assessment report:

[0214] Three core indicators for calculating prediction results:

[0215] (4.1) The normalized root mean square error (NRMSE) is:

[0216]

[0217] (4.2) Coefficient of determination ( ):

[0218]

[0219] (4.3) Maximum Absolute Error (MAE):

[0220]

[0221] If all three metrics meet the threshold requirements, the prediction is considered valid; otherwise, increase the training data by 5% or adjust the attention coefficient. This triggers model retraining.

[0222] (5) Robustness check:

[0223] Add to input features Calculate the rate of change of prediction error under random noise interference. :

[0224]

[0225] In the formula, RMSE after noise interference; This represents the RMSE under noise-free conditions. If If so, optimize the median filter window length or increase the hidden layer dimension of the attention layer.

[0226] (6) Credibility Indicators

[0227] Calculate the overall reliability C based on the verification results:

[0228]

[0229] Among them, the credibility level is as follows:

[0230] Green indicates high reliability; Yellow indicates a reliable color; Red indicates low credibility.

[0231] As one implementation scheme, Figure 2 This is a schematic diagram of the hardware operating environment of the computer system involved in the embodiments of this application.

[0232] like Figure 2 As shown, the computer system may include: a processor 1001, such as a CPU; a memory 1005; a user interface 1003; a network interface 1004; and a communication bus 1002. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as a disk drive. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0233] Those skilled in the art will understand that Figure 2 The computer system architecture shown does not constitute a limitation on the computer system and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0234] like Figure 2 As shown, the memory 1005, as a storage medium, may include an operating system, a network communication module, a user interface module, and computer programs. The operating system is a program that manages and controls the hardware and software resources of the computer system, as well as the operation of the computer programs and other software or programs.

[0235] exist Figure 2 In the computer system shown, the user interface 1003 is mainly used to connect to the terminal and communicate with the terminal; the network interface 1004 is mainly used to communicate with the backend server; and the processor 1001 can be used to call the computer program stored in the memory 1005.

[0236] In this embodiment, the computer system includes: a memory 1005, a processor 1001, and a computer program stored in the memory and executable on the processor, wherein:

[0237] When processor 1001 calls a computer program stored in memory 1005, it performs the following operations:

[0238] S10, obtain the fusion key features of the battery to be predicted, wherein the fusion key features are obtained by weighted fusion of the candidate battery health features of the battery to be predicted using an attention mechanism;

[0239] S20: Acquire the fusion key features collected within a preset time period and sort them in time sequence to form a fusion key feature sequence; determine the sequence representation of the fusion key feature sequence in the frequency domain; and divide the sequence representation into low-frequency feature sequences and high-frequency feature sequences according to a preset frequency threshold.

[0240] S30, extract battery degradation trend features from the low-frequency feature sequence based on a bidirectional gated cyclic unit; and extract nonlinear fluctuation features from the high-frequency feature sequence based on a complex-valued neural network;

[0241] S40, Based on the battery degradation trend characteristics and the nonlinear fluctuation characteristics, generate the battery health prediction result of the battery to be predicted.

[0242] Furthermore, those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in a computer system to implement the process steps of the embodiments of the above methods.

[0243] Therefore, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the various steps of the battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition as described in the above embodiments.

[0244] The computer-readable storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0245] It should be noted that, since the storage medium provided in the embodiments of this application is the storage medium used to implement the methods of the embodiments of this application, those skilled in the art can understand the specific structure and variations of the storage medium based on the methods described in the embodiments of this application, and therefore will not be repeated here. All storage media used in the methods of the embodiments of this application fall within the scope of protection of this application.

[0246] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0247] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0248] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0249] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0250] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0251] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition, characterized in that, The method includes the following steps: S10, obtain the fusion key features of the battery to be predicted, wherein the fusion key features are obtained by weighted fusion of the candidate battery health features of the battery to be predicted using an attention mechanism; S20: Acquire the fusion key features collected within a preset time period and sort them in time sequence to form a fusion key feature sequence; determine the sequence representation of the fusion key feature sequence in the frequency domain; and divide the sequence representation into low-frequency feature sequences and high-frequency feature sequences according to a preset frequency threshold. S30, extract battery degradation trend features from the low-frequency feature sequence based on a bidirectional gated cyclic unit; and extract nonlinear fluctuation features from the high-frequency feature sequence based on a complex-valued neural network; S40, Based on the battery degradation trend characteristics and the nonlinear fluctuation characteristics, generate the battery health prediction result of the battery to be predicted.

2. The battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition as described in claim 1, characterized in that, In step S30, the extraction step of battery degradation trend characteristics includes: S31, Define the low-frequency signal sequence: ； In the formula, The sequence length; for each time step The low-frequency signal input is used, and the low-frequency branch performs gating state updates through the GRU unit: ； ； ； ； In the formula, Represents the door update. This represents resetting the door. Represents a hidden state. , and These represent the input weight matrix of the update gate, the input weight matrix of the reset gate, and the input weight matrix of the candidate hidden state, respectively. , and Let represent the hidden state weight matrix of the update gate, the hidden state weight matrix of the reset gate, and the hidden state weight matrix of the candidate hidden state, respectively. , and These represent the bias vectors of the update gate, the reset gate, and the candidate hidden state, respectively. This represents the bias vector of the update gate, the bias vector of the reset gate, and the bias vector of the candidate hidden state. S32, introduces a bidirectional structure to extract low-frequency signal input. Historical information and future information : ； ； Battery degradation trend characteristics were obtained. , is represented as: ； The steps for extracting nonlinear fluctuation characteristics include: S33, converts the high-frequency signal of the high-frequency branch. Represented in complex form : ； In the formula, It is amplitude information. It is phase information, where T represents the length of the time series involved in the modeling; S32, obtains the high-frequency signal in complex form through complex linear mapping. : ，l=1,2,…,L； In the formula, W represents the complex weight matrix, b represents the complex bias, and L represents the neural network layer; definition: ； Will Expand into real and imaginary parts for calculation: ； The activation function uses a complex-valued nonlinear mapping form: ，l=1,2,…,L； S34, by stacking complex-valued neural network layers, obtains nonlinear fluctuation characteristics. : ； In the formula, It is a complex-valued mapping function.

3. The battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition as described in claim 1, characterized in that, Step S40 includes: S41, calculate the frequency domain energy distribution of the low-frequency feature sequence and the high-frequency feature sequence respectively: ； ； In the formula, Indicates low-frequency feature sequences, Represents high-frequency feature sequences; Represents the frequency domain energy distribution of low-frequency characteristic sequences; Represents the frequency domain energy distribution of high-frequency characteristic sequences; S42, Calculate the corresponding attention weights for the two based on their frequency domain energy distributions: ； ； In the formula, Attention weights for low-frequency feature sequences; Attention weights for high-frequency feature sequences; S43, feature fusion is performed based on the attention weights of the two to obtain fused features. : ； In the formula, This indicates the characteristics of battery degradation trends. Indicates nonlinear fluctuation characteristics; S44, based on fusion features A prediction is made, generating a battery health prediction result for the battery to be predicted.

4. The battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition as described in claim 1, characterized in that, The selection steps for the candidate battery health features include: Select initial features from the collected preprocessed battery running dataset; Determine the correlation coefficient between the initial features and the measured battery health values ​​in the battery health feature dataset; Initial features with a correlation coefficient greater than a preset value are used as candidate battery health features.

5. The battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition as described in claim 4, characterized in that, The initial features include time-domain features, statistical features, and cyclic features; The time-domain features include at least one of the following: charging voltage plateau length, discharge voltage drop rate, and charge / discharge energy difference. The statistical characteristics include at least one of voltage fluctuation variance, peak current percentage, and mean temperature. The cycle characteristics include at least one of unit cycle capacity decay and cycle lifetime percentage.

6. The battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition as described in claim 4, characterized in that, The preprocessing steps for the preprocessed battery runtime dataset include: The dynamic current data / dynamic voltage data in the initial battery operation dataset are determined, and the peak noise in the dynamic current data / dynamic voltage data is eliminated by a sliding window mid-range filter. The length of the sliding window is adjusted based on the data fluctuation frequency of the dynamic current data / dynamic voltage data. Identify abnormal data in the initial battery running dataset and repair the abnormal data using linear interpolation. The initial battery running dataset, which has been repaired by sliding window midpoint filtering and linear interpolation, is normalized to obtain the preprocessed battery running dataset.

7. The battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition as described in any one of claims 4 to 6, characterized in that, The steps for generating the fusion key features include: Determine the health characteristics of each candidate battery Corresponding feature weights : ； In the formula, This is the weighting adjustment coefficient; Let be the correlation coefficient between the j-th candidate battery health feature and the battery health state. Let be the attention function. This represents the correlation coefficient between the k-th candidate battery health feature and the battery health status; Health characteristics of each candidate battery and their corresponding trophy weights By performing a weighted sum, the key features of the fusion are obtained: ； In the formula, m is the number of candidate battery health features.

8. The battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition as described in claim 1, characterized in that, The battery health prediction results include battery health prediction values; The battery health level of the battery to be predicted is determined based on the range in which the predicted battery health value falls.

9. A computer system, characterized in that, The computer system includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the battery health prediction method based on adaptive feature fusion and dual-branch frequency domain decomposition as described in any one of claims 1 to 8.

Images