Parallel lithium-ion battery fault diagnosis method based on complete set empirical mode decomposition and time series convolution network

By combining complete set empirical mode decomposition and temporal convolutional networks, the problem of detecting minute voltage changes in parallel lithium-ion battery packs is solved, achieving efficient and accurate fault diagnosis, and maintaining good robustness, especially during battery aging.

CN121831545BActive Publication Date: 2026-11-10QINGDAO UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202610039619.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-01-13
Publication Date
2026-11-10
Estimated Expiration
2046-01-13

AI Technical Summary

Technical Problem

Existing fault diagnosis methods for lithium-ion battery packs are difficult to achieve rapid and accurate fault detection under minute voltage changes, and the diagnostic accuracy is not high under different operating conditions and aging states. In particular, minute voltage changes in parallel batteries are difficult to detect, and traditional methods have poor robustness, resulting in frequent false alarms and missed alarms.

Method used

The voltage signal is decomposed into multiple scales using the complete set empirical mode decomposition algorithm. The intrinsic mode function components with significant fault characteristics are screened using information entropy for signal reconstruction. The reconstructed signal is then input into a temporal convolutional network with an attention mechanism for training and diagnosis. Fault types are distinguished by comparing the error between the predicted output and the actual signal.

Benefits of technology

It achieves high-accuracy fault diagnosis under minute voltage changes, improves the robustness and accuracy of fault detection, maintains good diagnostic performance during battery aging, and can diagnose internal connection faults of parallel batteries using only voltage signals.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121831545B_ABST
    Figure CN121831545B_ABST
Patent Text Reader

Abstract

The application discloses a parallel lithium ion battery fault diagnosis method based on complete set empirical mode decomposition and time sequence convolution network, and comprises the following steps: step one, collecting voltage signals; step two, decomposing the voltage signals by using a complete set empirical mode decomposition algorithm to obtain multiple intrinsic mode functions and residual errors; step three, selecting intrinsic mode function components with relatively significant entropy changes in low frequency or medium frequency based on information entropy changes of the intrinsic mode function components, reconstructing the original signals to obtain reconstructed fault signals; step four, inputting the reconstructed fault signals into a time convolution network fused with an attention mechanism for training and diagnosis, and performing feature extraction and prediction on the signals; and step five, diagnosing and distinguishing fault types. The application can realize rapid and accurate detection of faults under slight voltage changes, and improve fault diagnosis accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of battery fault diagnosis technology, and in particular relates to a fault diagnosis method for parallel lithium-ion batteries based on complete set empirical mode decomposition and temporal convolutional networks. Background Technology

[0002] With the energy crisis worsening, developing new energy sources is becoming increasingly important. Currently, lithium battery technology is at the core of the new energy field, benefiting from its extremely wide applicability in portable devices, satellites, power storage, and electric vehicles.

[0003] The technology has seen comprehensive development across various fields. In my country's current strong promotion of "secondary power sources," lithium batteries possess excellent performance characteristics.

[0004] The potential and broad prospects of lithium-ion batteries are significant. Lithium-ion batteries are highly complex nonlinear time-varying systems, with early fault characteristics being subtle and insidious. Furthermore, the internal electrochemical reaction mechanism of lithium-ion batteries is extremely complex, making it difficult to accurately capture battery parameter changes caused by independent fault diagnosis (ISC). Existing threshold-based or model-based ISC diagnosis and detection methods have many shortcomings and deficiencies, failing to adequately meet the needs of practical applications. The safety of lithium batteries is crucial, constraining the development of the new energy industry; therefore, the use of lithium battery fault diagnosis technology is indispensable.

[0005] In the fields of electric vehicles and energy storage systems, lithium-ion battery packs are often configured using a combination of parallel and series connections in practical applications. Because parallel batteries exhibit significant self-balancing effects among individual cells when a fault occurs, often manifesting only as weak voltage fluctuations, traditional fault diagnosis methods relying on voltage / current differences struggle to accurately determine the fault type. Currently, mainstream fault diagnosis methods for lithium-ion battery packs mainly fall into two categories: model-based and data-driven. The former relies on accurate equivalent models, but due to the strong nonlinearity and time-varying characteristics of lithium-ion battery packs, it is difficult to establish universal models, leading to poor robustness and frequent false alarms and missed alarms. The latter, while not relying on mechanistic models, is constrained by difficulties in obtaining fault samples and the scarcity of labeled data, resulting in insufficient model training and limited generalization ability. Therefore, both methods struggle to simultaneously achieve both diagnostic accuracy and real-time performance.

[0006] Therefore, there is an urgent need for a diagnostic method that can efficiently extract weak fault signal features and adapt to different operating conditions and aging states, in order to make up for the shortcomings of existing technologies in detecting small voltage changes in parallel lithium-ion batteries and the low accuracy of fault diagnosis. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the technical problem to be solved by this invention is to provide a parallel lithium-ion battery fault diagnosis method based on complete set empirical mode decomposition and temporal convolutional networks, which can achieve rapid and accurate fault detection under minute voltage changes, exhibit good robustness during battery aging, and improve fault diagnosis accuracy.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for fault diagnosis of parallel lithium-ion batteries based on complete set empirical mode decomposition and temporal convolutional networks, comprising the following steps:

[0009] Step 1: Collect voltage signals of parallel batteries under different operating conditions;

[0010] Step 2: Decompose the acquired voltage signal using the complete set empirical mode decomposition algorithm to obtain multiple intrinsic mode functions and residuals;

[0011] Step 3: Based on the information entropy changes of each intrinsic mode function component, select the intrinsic mode function components with more significant entropy changes in low or medium frequency to reconstruct the original signal and obtain the reconstructed fault signal;

[0012] Step 4: Input the reconstructed fault signal into a temporal convolutional network that incorporates an attention mechanism for training and diagnosis, and perform feature extraction and prediction on the signal;

[0013] Step 5: By comparing the error between the predicted output and the actual reconstructed signal, a predetermined error threshold is set to diagnose and distinguish the fault type.

[0014] The above-mentioned fault diagnosis method for parallel lithium-ion batteries based on complete set empirical mode decomposition and temporal convolutional networks, in step two, the decomposition using the complete set empirical mode decomposition algorithm includes the following steps:

[0015] (1) Add a pair of white noise sequences with opposite signs to the original voltage signal to generate multiple noise superposition signals;

[0016] (2) Empirical mode decomposition is performed on each noise superimposed signal to obtain a set of intrinsic mode functions;

[0017] (3) Average the intrinsic mode functions corresponding to each pair of noise signals to eliminate the influence of noise and obtain the final intrinsic mode function components.

[0018] In the above-mentioned parallel lithium-ion battery fault diagnosis method based on complete set empirical mode decomposition and temporal convolutional networks, step two includes:

[0019] The specific process of step (1) is as follows: The acquired raw voltage signal... ,Add to White noise sequences with paired positive and negative signals ,generate Noise signal The calculation formula is as follows:

[0020] ;

[0021] The specific process of step (2) is as follows: perform empirical mode decomposition on each noisy signal to obtain its intrinsic mode function components, and average the results of paired components to cancel out the noise effect, so as to obtain the final intrinsic mode function components. The calculation formula is as follows:

[0022] ;

[0023] Final residual signal The calculation formula is as follows:

[0024] ;

[0025] The specific process of step (3) is as follows: After decomposition, the original signal can be represented as all intrinsic mode function components and residuals. The sum is calculated using the following formula:

[0026] .

[0027] The specific process of step three in the above-mentioned parallel lithium-ion battery fault diagnosis method based on complete set empirical mode decomposition and temporal convolutional networks is as follows:

[0028] (1) Intrinsic mode function screening based on information entropy:

[0029] The information entropy of each intrinsic mode function component is calculated using the following formula:

[0030] ,

[0031] Using the entropy value of each intrinsic mode function of the normal signal as a baseline, the entropy values ​​of each intrinsic mode function of the three fault signals are calculated respectively, and the intrinsic mode function component with the most significant change in entropy value is identified.

[0032] (2) Signal reconstruction:

[0033] Those with significant changes in entropy value were selected. The intrinsic mode function components are added together to form the reconstructed signal. The calculation formula is as follows:

[0034] ,

[0035] The reconstructed signal effectively amplifies the fault characteristics.

[0036] The above-mentioned fault diagnosis method for parallel lithium-ion batteries based on complete set empirical mode decomposition and temporal convolutional networks, in step four, the TCN-Attention model, based on the traditional temporal convolutional network structure, extracts long-term features by extending convolution and residual connections, and introduces an attention mechanism to weightedly fuse features from different time steps to enhance the extraction effect of fault-sensitive features. The specific process is as follows:

[0037] (1) Model building:

[0038] The TCN-Attention model utilizes the dilated causal convolution and residual structure of the TCN to capture long-term dependencies in time series data. By introducing an attention mechanism, the model dynamically assigns weights to features at different time steps, focusing on the critical time periods most relevant to the fault. The calculation formula for the attention mechanism is as follows:

[0039] ;

[0040] (2) Model training:

[0041] The reconstructed voltage signal dataset obtained in step three is randomly divided into a training set and a test set in a 7:3 ratio. The TCN-Attention model is trained using the reconstructed voltage data under normal conditions in the training set, enabling it to predict the voltage signal under normal conditions with high accuracy. The trained model is then used to predict the test set data under all conditions.

[0042] In the above-mentioned parallel lithium-ion battery fault diagnosis method based on complete set empirical mode decomposition and temporal convolutional networks, step five involves calculating the model's predicted values. Compared with the actual reconstructed voltage value Error between The specific formula for diagnosis is as follows:

[0043] ,

[0044] The diagnostic step uses a preset error threshold. Implement this by setting a diagnostic threshold. When the error Exceeding the diagnostic threshold If the condition is not met, it is determined to be a fault state, and different fault types are distinguished based on error characteristics; otherwise, it is a normal state.

[0045] In the above-mentioned parallel lithium-ion battery fault diagnosis method based on complete set empirical mode decomposition and temporal convolutional network, in step (1), the input time series data is divided into multiple time steps, each time step containing multiple features, which will be used as input to the TCN model. When processing voltage data, the TCN first ensures that the model only depends on the current and previous voltage data through causal convolution to ensure that the model conforms to the causality of the time series. Then, the TCN captures the long-term dependency by introducing intervals between the convolution kernel elements. Finally, through a fully connected layer, these extracted voltage features are transformed into specific prediction results. When processing voltage data, the attention mechanism calculates weights based on the features of the voltage signal. The model performs a weighted summation of the features of each time step according to the calculated attention weights.

[0046] The above-mentioned parallel lithium-ion battery fault diagnosis method based on complete set empirical mode decomposition and temporal convolutional network includes normal signals and three types of fault signals in step (2).

[0047] The advantages of the parallel lithium-ion battery fault diagnosis method based on complete set empirical mode decomposition and temporal convolutional networks are as follows: First, the complete set empirical mode decomposition algorithm is used to perform multi-scale decomposition on the acquired voltage signal, and the intrinsic mode function components with significant fault characteristics are screened using the information entropy index for signal reconstruction, thereby enhancing the fault information in the signal; then, a temporal convolutional network model with an attention mechanism is used to extract features from the reconstructed signal, and fault detection and type differentiation are achieved by comparing the error between the predicted output and the actual signal. The method diagnoses internal connection faults in parallel batteries by using the acquired voltage signal, without the need to directly measure the current of each branch, thus overcoming the technical challenge of diagnosing faults due to small voltage changes in parallel batteries; the complete set empirical mode decomposition algorithm can accurately extract fault information hidden in the voltage signal, and then reconstruct the fault signal based on components with significant entropy changes, improving the ability to identify voltage signal faults; and the fault diagnosis model is trained by a temporal convolutional network with an attention mechanism, achieving a high accuracy rate in fault diagnosis. Attached Figure Description

[0048] Figure 1 This is a flowchart of the parallel lithium-ion battery fault diagnosis method of the present invention;

[0049] Figure 2 A resistance change trend curve to simulate a connection fault;

[0050] Figure 3 A schematic diagram of all intrinsic mode function signals and residual signals for normal voltage decomposition;

[0051] Figure 4A graph showing the changes in entropy values ​​of fault signals and normal signals;

[0052] Figure 5 A comparison chart showing the reconstruction of fault signals and normal signals;

[0053] Figure 6 The image shows the prediction results and error distribution of the TCN-Attention model in fault cases and normal cases. Detailed Implementation

[0054] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] In this invention, unless otherwise stated, directional terms such as "upper" and "lower" generally refer to the upper and lower positions of the device in its actual use or operating state, specifically the drawing directions in the accompanying drawings; while "inner" and "outer" refer to the outline of the device. Furthermore, in the description of this application, the term "comprising" means "including but not limited to". The terms first, second, third, etc., are used merely as illustrative purposes and do not impose numerical requirements or establish an order. The term "multiple" means "two or more".

[0056] like Figure 1 As shown, this invention presents a fault diagnosis method for parallel lithium-ion batteries based on complete set empirical mode decomposition and temporal convolutional networks. The method mainly includes: parallel battery data acquisition, complete set empirical mode decomposition and entropy reconstruction of the voltage signal, TCN-Attention model training, and fault diagnosis based on the reconstructed voltage prediction error. First, the acquired voltage signal is decomposed into multiple scales using the complete set empirical mode decomposition algorithm. Then, the intrinsic mode function components with significant fault characteristics are selected using the information entropy index for signal reconstruction, thereby enhancing the fault information in the signal. Subsequently, a temporal convolutional network model with an attention mechanism is used to extract features from the reconstructed signal. Fault detection and type differentiation are achieved by comparing the error between the predicted output and the actual signal. This invention can achieve high-accuracy fault diagnosis even under minute voltage changes.

[0057] Specifically, the steps include the following:

[0058] Step 1: Collect voltage signals of parallel batteries under different operating conditions;

[0059] In actual vehicle operation, vibration or metallic foreign objects can cause changes in the resistance of connection points. This invention collects typical connection fault data: Case: The connection resistance increases linearly continuously, simulating a severe fault caused by poor contact leading to continuous heating and a vicious cycle. The trend of resistance change is as follows. Figure 2As shown, the aforementioned resistance changes are introduced into urban driving cycle conditions, generating a large amount of voltage-time series data containing normal and fault states, which constitutes the original dataset.

[0060] To address the problem of weak voltage fluctuations in parallel battery faults that are difficult to directly identify, this invention employs a complete set empirical mode decomposition algorithm in steps two and three to decompose the original voltage signal, such as... Figure 3 As shown, information entropy is used to filter key components to reconstruct the signal and enhance fault characteristics, such as... Figure 4 As shown.

[0061] Step 2: Decompose the acquired voltage signal using the complete ensemble empirical mode decomposition algorithm to obtain multiple intrinsic mode functions and residuals. The decomposition using the complete ensemble empirical mode decomposition algorithm includes the following steps:

[0062] (1) Add a pair of white noise sequences with opposite signs to the original voltage signal to generate multiple noise superposition signals; specifically as follows: for the acquired original voltage signal ,Add to White noise sequences with paired positive and negative signals ,generate Noise signal The calculation formula is as follows:

[0063]

[0064] (2) Perform empirical mode decomposition on each noise-added signal to obtain a set of intrinsic mode functions; specifically as follows: Perform empirical mode decomposition on each noisy signal to obtain its intrinsic mode function components, and average the results of the paired components to cancel out the noise effect, and obtain the final intrinsic mode function components. The calculation formula is as follows:

[0065] ;

[0066] Final residual signal The calculation formula is as follows:

[0067] ;

[0068] (3) Averaging the intrinsic mode functions corresponding to each pair of noise signals eliminates the influence of noise, yielding the final intrinsic mode function components. Specifically, after decomposition, the original signal can be represented as all intrinsic mode function components and the residual. The sum is calculated using the following formula:

[0069] .

[0070] Step 3: Based on the information entropy changes of each intrinsic mode function component, select the intrinsic mode function components with more significant entropy changes in low or mid-frequency ranges to reconstruct the original signal and obtain the reconstructed fault signal; the specific process is as follows:

[0071] (1) Intrinsic mode function screening based on information entropy:

[0072] The information entropy of each intrinsic mode function component is calculated using the following formula:

[0073] ,

[0074] Using the entropy value of each intrinsic mode function of the normal signal as a baseline, the entropy values ​​of each intrinsic mode function of the three fault signals are calculated respectively, and the intrinsic mode function component with the most significant change in entropy value is identified.

[0075] Using the entropy values ​​of the intrinsic mode functions (IMFs) of the normal signal as a baseline, the entropy values ​​of the IMFs of the three fault signals are calculated respectively, and the IMF component with the most significant entropy change is identified. For example... Figure 4 As shown, the entropy values ​​of some low- and mid-frequency intrinsic mode function components in the case study are significantly increased compared to the normal state, indicating that these components contain rich fault information. High-frequency intrinsic mode function components, however, typically contain noise and their entropy values ​​do not change much, so they are discarded.

[0076] (2) Signal reconstruction:

[0077] Those with significant changes in entropy value were selected. The intrinsic mode function components are added together to form the reconstructed signal. The calculation formula is as follows:

[0078] ,

[0079] The reconstructed signal effectively amplifies the fault characteristics. For example... Figure 5 As shown, the reconstructed signals of the fault case and the normal state were compared. It can be seen that there are obvious differences between the reconstructed fault signal and the normal signal, which lays the foundation for subsequent fault diagnosis.

[0080] Step 4: The reconstructed fault signal is input into a temporal convolutional network with an attention mechanism for training and diagnosis, and features are extracted and predicted from the signal. The TCN-Attention model, based on the traditional temporal convolutional network structure, extracts long-term features by extending convolutions and residual connections, and introduces an attention mechanism to weightedly fuse features from different time steps to enhance the extraction of fault-sensitive features. The specific process is as follows:

[0081] (1) Model building:

[0082] The TCN-Attention model utilizes the dilated causal convolution and residual structure of the TCN to capture long-term dependencies in time series data. By introducing an attention mechanism, the model dynamically assigns weights to features at different time steps, focusing on the critical time periods most relevant to the fault. The calculation formula for the attention mechanism is as follows:

[0083] ,

[0084] First, the input time-series data is segmented into multiple time steps, each containing multiple features. These features serve as input to the TCN model. When processing voltage data, TCN first ensures that the model depends only on current and previous voltage data through causal convolution, thus ensuring the model conforms to the causality of time series data. Next, TCN captures dependencies over long time spans by introducing intervals between convolutional kernel elements.

[0085] TCN prevents the vanishing or exploding gradient problems that occur during the training of deep networks through residual connections. After each convolutional layer, the input and output are added together to form residual learning. Residual connections can effectively preserve input information and help gradients propagate more smoothly through addition, thereby accelerating model training. Finally, through fully connected layers, these extracted voltage features are transformed into specific prediction results, such as whether the battery voltage is abnormal or the future voltage trend, thus achieving accurate voltage anomaly detection or fault prediction.

[0086] When processing voltage data, the attention mechanism first generates a weight for each time step, reflecting the importance of that time step to the final prediction result. Specifically, the attention mechanism calculates weights based on the characteristics of the voltage signal. For battery fault detection tasks, certain key moments in the voltage signal (such as voltage fluctuations during battery connection failures) are assigned higher weights.

[0087] Next, the model performs a weighted summation of the features at each time step based on the calculated attention weights. This allows the model to pay more attention to features at important time steps and less attention to those at less important time steps. Through this process, TCN can focus more on the parts of the voltage data that are crucial for fault diagnosis, thereby improving the model's ability to identify abnormal patterns and the accuracy of its predictions.

[0088] (2) Model training:

[0089] The reconstructed voltage signal dataset obtained in step three is randomly divided into a training set and a test set in a 7:3 ratio. The TCN-Attention model is trained using the reconstructed voltage data under normal conditions in the training set, so that it can predict the voltage signal under normal conditions with high accuracy. The trained model is then used to predict the test set data for all states (normal signal and three types of fault signals).

[0090] Step 5: By comparing the error between the predicted output and the actual reconstructed signal, a predetermined error threshold is set to diagnose and distinguish the fault type. This is done by calculating the model's predicted value. Compared with the actual reconstructed voltage value Error between The specific formula for diagnosis is as follows:

[0091] ,

[0092] The diagnostic steps use a preset error threshold. Implement this by setting a diagnostic threshold. When the error Exceeding the diagnostic threshold If the condition is not met, it is determined to be a fault state, and different fault types are distinguished based on error characteristics; otherwise, it is a normal state.

[0093] By analyzing the error distribution under normal and fault conditions, this invention sets a threshold. Setting it to 0.015 effectively distinguishes different states. The prediction results and error distribution of the TCN-Attention model in four cases are shown below. Figure 6 As shown, it can clearly distinguish between fault and normal states.

[0094] In summary, this invention combines a complete set empirical mode decomposition algorithm with a temporal convolutional network (TCN-Attention) model incorporating an attention mechanism. Its core innovation lies in generating signals rich in fault characteristics through a combination of complete set empirical mode decomposition and entropy; these signals are then input into the TCN-Attention model for feature extraction and fault prediction. Entropy reconstruction via complete set empirical mode decomposition effectively enhances the weak fault voltage characteristics in the parallel battery system. Combined with the excellent time-series prediction and feature focusing capabilities of the TCN-Attention model, it achieves rapid and accurate diagnosis of internal connection faults and exhibits good robustness during battery aging. This invention relies solely on voltage signals to diagnose internal connection faults in parallel batteries, eliminating the need for direct measurement of branch currents, thus overcoming the technical challenge of detecting minute voltage changes in parallel batteries. In experimental verification, using data collected under urban driving cycle conditions for model training and testing, the TCN-Attention model achieved an accuracy of 98.37% in fault diagnosis.

[0095] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should be protected by the present invention.

Claims

1. A method for fault diagnosis of parallel lithium-ion batteries based on complete set empirical mode decomposition and temporal convolutional networks, characterized in that, Includes the following steps: Step 1: Collect voltage signals of parallel batteries under different operating conditions; Step 2: Decompose the acquired voltage signal using the complete set empirical mode decomposition algorithm to obtain multiple intrinsic mode functions and residuals; Step 3: Based on the information entropy changes of each intrinsic mode function component, select the intrinsic mode function components with more significant entropy changes in low or mid-frequency ranges to reconstruct the original signal and obtain the reconstructed fault signal; the specific process is as follows: (1) Intrinsic mode function screening based on information entropy: The information entropy of each intrinsic mode function component is calculated using the following formula: , Using the entropy value of each intrinsic mode function of the normal signal as a baseline, the entropy values ​​of each intrinsic mode function of the three fault signals are calculated respectively, and the intrinsic mode function component with the most significant change in entropy value is identified. (2) Signal reconstruction: The k intrinsic mode function components with significant entropy changes selected from the pool are summed to form the reconstructed signal x. rec (t), the calculation formula is as follows: , The reconstructed signal effectively amplifies the fault characteristics; Step 4: Input the reconstructed fault signal into a temporal convolutional network that incorporates an attention mechanism for training and diagnosis, and perform feature extraction and prediction on the signal; Step 5: By comparing the error between the predicted output and the actual reconstructed signal, a predetermined error threshold is set to diagnose and distinguish the fault type, and the model prediction value U is calculated. pre Compared with the actual reconstructed voltage value U rec The error e between them is diagnosed using the following formula: , The diagnostic steps are implemented by a preset error threshold r. A diagnostic threshold r is set. When the error e exceeds the diagnostic threshold r, it is determined to be a fault state, and different fault types are distinguished according to the error characteristics. Otherwise, it is a normal state.

2. The parallel lithium-ion battery fault diagnosis method based on complete set empirical mode decomposition and temporal convolutional networks as described in claim 1, characterized in that, Step two, which uses the complete set empirical mode decomposition algorithm for decomposition, includes the following steps: (1) Add a pair of white noise sequences with opposite signs to the original voltage signal to generate multiple noise superposition signals; (2) Empirical mode decomposition is performed on each noise superimposed signal to obtain a set of intrinsic mode functions; (3) Average the intrinsic mode functions corresponding to each pair of noise signals to eliminate the influence of noise and obtain the final intrinsic mode function components.

3. The parallel lithium-ion battery fault diagnosis method based on complete set empirical mode decomposition and temporal convolutional networks as described in claim 2, characterized in that, In step two: The specific process of step (1) is as follows: For the acquired original voltage signal x(t), add N pairs of positive and negative white noise sequences η. i (t), generating 2N noisy signals X i (t), the calculation formula is as follows: ; The specific process of step (2) is as follows: perform empirical mode decomposition on each noisy signal to obtain its intrinsic mode function components, and average the results of paired components to cancel out the noise effect, so as to obtain the final intrinsic mode function component C. k (t), the calculation formula is as follows: ; The formula for calculating the final residual signal R(t) is as follows: ; The specific process of step (3) is as follows: After decomposition, the original signal can be expressed as the sum of all intrinsic mode function components and the residual R(t), and the calculation formula is as follows: 。 4. The parallel lithium-ion battery fault diagnosis method based on complete set empirical mode decomposition and temporal convolutional network according to claim 1, characterized in that, In step four, the TCN-Attention model, based on the traditional temporal convolutional network structure, extracts long-term features by expanding convolutions and residual connections, and introduces an attention mechanism to weightedly fuse features from different time steps to enhance the extraction effect of fault-sensitive features. The specific process is as follows: (1) Model building: The TCN-Attention model utilizes the dilated causal convolution and residual structure of the TCN to capture long-term dependencies in time series data. By introducing an attention mechanism, the model dynamically assigns weights to features at different time steps, focusing on the critical time periods most relevant to the fault. The calculation formula for the attention mechanism is as follows: ; (2) Model training: The reconstructed voltage signal dataset obtained in step three is randomly divided into a training set and a test set in a 7:3 ratio. The TCN-Attention model is trained using the reconstructed voltage data under normal conditions in the training set, enabling it to predict the voltage signal under normal conditions with high accuracy. The trained model is then used to predict the test set data under all conditions.

5. The parallel lithium-ion battery fault diagnosis method based on complete set empirical mode decomposition and temporal convolutional network according to claim 4, characterized in that: In step (1), the input time series data is divided into multiple time steps, each containing multiple features. These data will be used as input to the TCN model. When processing voltage data, the TCN first ensures that the model depends only on the current and previous voltage data through causal convolution to ensure that the model conforms to the causality of the time series. Then, the TCN captures the long-term dependencies by introducing intervals between the convolution kernel elements. Finally, through a fully connected layer, these extracted voltage features are transformed into specific prediction results. When processing voltage data, the attention mechanism calculates weights based on the features of the voltage signal. The model performs a weighted summation of the features of each time step according to the calculated attention weights.

6. The parallel lithium-ion battery fault diagnosis method based on complete set empirical mode decomposition and temporal convolutional network according to claim 4, characterized in that: All states in step (2) include normal signals and three types of fault signals.

Citation Information

Patent Citations

  • Low-voltage series fault arc identification method, device, equipment and medium

    CN117031199A

  • Transformer fault diagnosis method based on double attention mechanism

    CN117272102A