Battery health status monitoring and prediction method and system based on multimodal feature fusion

Abstract

The present invention provides a battery health status monitoring and prediction method and system using multimodal feature fusion, which relates to the field of battery monitoring technology. The method obtains multimodal data by periodically collecting static data of the battery and real-time collecting dynamic data of the battery. The method is then implemented through steps such as data monitoring, shared underlying feature extraction, SOH prediction, RUL prediction, and multi-objective joint optimization. The present invention can perform online monitoring and analysis on the multimodal feature data generated by the battery, introduces a multimodal feature fusion and multi-objective joint prediction and optimization mechanism, and uses independent branches based on the SOH prediction model and the RUL prediction model to predict the SOH and RUL respectively, thereby predicting the battery health status. This method can not only realize automatic analysis of the battery's multimodal massive data, reduce workload, and improve analysis efficiency, but also improve the accuracy of the prediction results. At the same time, the constructed feature sharing mechanism can reduce computational complexity and improve prediction efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of battery monitoring, and in particular to a battery health status monitoring and prediction method and system based on multimodal feature fusion. Background Art

[0002] The safe and stable operation of battery packs is one of the important guarantees for energy supply and economic development. In recent years, with the continuous advancement of new energy large-scale model technology, the battery pack monitoring method that relied on manual operations in the past has emerged with many shortcomings. For example, the massive amount of data generated by important indicators such as battery temperature, voltage, and internal resistance is mainly generated by the terminal equipment continuously sending massive multi-dimensional battery data to the system through the protocol, comparing it with the pre-set battery parameter critical values, and then predicting the battery health status mainly through manual analysis, without using algorithms to predict the battery health status. This traditional method of relying solely on manual analysis of massive data has the following main defects:

[0003] 1. The massive amount of data generated by batteries is inefficient if analyzed solely through manual labor.

[0004] 2. Operation of battery packs, such as capacity verification, requires numerous parameters and a complex system interface, which places high demands on the workers' knowledge of electricity and their ability to operate the system proficiently.

[0005] 3. Manual prediction of the remaining life of the battery is inaccurate, which greatly wastes battery energy.

[0006] In short, the traditional method of relying solely on manual analysis of massive data is not only inefficient and labor-intensive, but also unable to accurately predict the remaining life of the battery, resulting in a huge waste of energy.

[0007] Based on this, the present invention is proposed. Summary of the Invention

[0008] In order to solve the problems of low efficiency and large errors in traditional manual analysis methods, the purpose of the present invention is to provide a battery health status monitoring and prediction method and system with multimodal feature fusion, which can not only realize automatic analysis of massive multimodal data of batteries, reduce workload and improve analysis efficiency, but also improve the accuracy of prediction results.

[0009] To achieve the above object, the present invention provides the following technical solutions:

[0010] In a first aspect, the present invention provides a battery health status monitoring and prediction method with multimodal feature fusion, including: data acquisition, periodically acquiring static data of the battery and real-time acquiring dynamic data of the battery to obtain multimodal data; data monitoring, aggregating and monitoring the acquired multimodal data; shared underlying feature extraction, characterizing the multimodal data based on a shared underlying feature extraction model to obtain shared underlying features; SOH prediction, extracting and fusing the obtained shared underlying features based on the SOH prediction model, and outputting the SOH prediction result; RUL prediction, extracting and fusing the obtained shared underlying features based on the RUL prediction model, and outputting the RUL prediction result; multi-objective joint optimization, respectively calculating the SOH loss and RUL loss and designing a weighted multi-objective loss function to obtain the total loss and optimize the prediction error.

[0011] In a first aspect, the present invention provides a preferred solution, wherein the static data includes electrolyte concentration and self-discharge rate; the dynamic data includes battery cell voltage, charge and discharge current, ambient temperature, internal resistance parameters and charge and discharge cycle number.

[0012] In a first aspect, the present invention provides a preferred solution, wherein the shared underlying feature extraction model adopts LSTM, SAE or attention network.

[0013] In the first aspect, the present invention provides a preferred solution, wherein the shared underlying features include shared underlying features derived from static data and shared underlying features derived from dynamic data; in the SOH prediction step, the shared underlying features derived from static data are subjected to cross-period feature encoding to extract trend features including long-term aging features, and the shared underlying features derived from dynamic data are subjected to intra-period encoding to extract time series dynamic features including abnormal features.

[0014] In a first aspect, the present invention provides a preferred solution, wherein the SOH prediction model includes a multi-layer perceptron encoder, a transposed convolutional encoder, a splicing layer, and a fully connected layer; the shared underlying features derived from static data are encoded across cycles by the multi-layer perceptron encoder to obtain cross-cycle feature vectors; the shared underlying features derived from dynamic data are encoded within cycles by the transposed convolutional encoder to obtain intra-cycle feature vectors; the cross-cycle and intra-cycle feature vectors are spliced ​​by the splicing layer, and a dynamic weight distribution mechanism is introduced to adjust the contribution of each feature vector according to the battery type and usage scenario, and then output through the fully connected layer.

[0015] In the first aspect, the present invention provides a preferred solution, wherein the fully connected layer outputs an estimated value of the capacity attenuation rate or an estimated value of the internal resistance change rate, and calculates an estimated SOH value based on the estimated value of the capacity attenuation rate or the estimated value of the internal resistance change rate as the SOH prediction result.

[0016] In the first aspect, the present invention provides a preferred solution, which is based on the RUL prediction model and includes: a feature extraction module, which is based on a multi-layer feedforward network and a residual connection structure, shares underlying features for feature extraction, and extracts high-dimensional representations of degraded features; a multi-head attention module, which captures the nonlinear coupling relationship between multi-source features through parallel subspace attention calculation; a self-supervised learning module, which designs auxiliary tasks based on time series prediction and jointly optimizes feature extraction and RUL prediction main tasks; a RUL prediction output module, which outputs RUL prediction values ​​through linear layer and matrix fusion operations.

[0017] In a first aspect, the present invention provides a preferred solution, wherein in the multi-objective joint optimization step, the weighted multi-objective loss function is expressed as:

[0018]

[0019] in, is the total loss, is the SOH loss, is the RUL loss, is the weight.

[0020] In the first aspect, the present invention provides a preferred solution, wherein the weighted multi-objective loss function adopts a dynamic weighted multi-objective loss function. It is a dynamic weight, which is adjusted according to the task priority.

[0021] In a second aspect, the present invention provides a battery health status monitoring and prediction system with multimodal feature fusion, which is used to execute the above method, including: a data acquisition unit, which is used to periodically collect static data of the battery and real-time dynamic data of the battery to obtain multimodal data; a data monitoring unit, which is used to collect and monitor the acquired multimodal data; a shared underlying feature extraction unit, which is used to characterize the multimodal data based on a shared underlying feature extraction model to obtain shared underlying features; an SOH prediction unit, which is used to extract and fuse the obtained shared underlying features based on the SOH prediction model, and output the SOH prediction result; a RUL prediction unit, which is used to extract and fuse the obtained shared underlying features based on the RUL prediction model, and output the RUL prediction result; a multi-objective joint optimization unit, which is used to calculate the SOH loss and RUL loss respectively and design a weighted multi-objective loss function to obtain the total loss and optimize the prediction error.

[0022] Compared with the prior art, the present invention has the following beneficial technical effects:

[0023] This invention enables online monitoring and analysis of multimodal feature data generated by batteries. It also incorporates a multimodal feature fusion and multi-objective joint prediction and optimization mechanism. Based on the SOH prediction model and the RUL prediction model, independent branches are used to predict the state of health (SOH) and the remaining useful life (RUL), respectively, thereby predicting the battery's state of health. This not only enables automated analysis of massive amounts of multimodal battery data, reducing workload and improving analysis efficiency, but also enhances the accuracy of prediction results. Furthermore, by characterizing multimodal data based on a shared underlying feature extraction model, shared underlying features are obtained, and a feature sharing mechanism is established, reducing redundant computations, lowering computational complexity, and improving prediction efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0025] Figure 1 This is a flowchart of the battery health status monitoring and prediction method using multimodal feature fusion provided in Example 1 of the present invention;

[0026] Figure 2 This is a module block diagram of the battery health status monitoring and prediction system with multimodal feature fusion provided in Example 1 of the present invention;

[0027] Figure 3 This is a diagram showing the overall architecture of the battery health status monitoring and prediction system using multimodal feature fusion provided in Example 2 of the present invention;

[0028] Figure 4 This is a diagram showing the architecture of the battery health status monitoring and prediction method and the SOH prediction model in the system using multimodal feature fusion provided in Example 2 of the present invention;

[0029] Figure 5 This is a diagram of the RUL prediction model architecture in the battery health status monitoring and prediction method and system using multimodal feature fusion provided in Example 2 of the present invention.

[0030] Figure numerals: data acquisition unit 1, acquisition module 11, charge and discharge module 12, data monitoring unit 2, battery monitoring host 21, shared underlying feature extraction unit 3, SOH prediction unit 4, RUL prediction unit 5, multi-objective joint optimization unit 6, first battery group 71, second battery group 72. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention. Example 1

[0032] Please refer to Figure 1 This embodiment provides a battery health status monitoring and prediction method based on multimodal feature fusion, including:

[0033] S1. Data acquisition: periodically collect static data of the battery and collect dynamic data of the battery in real time to obtain multimodal data;

[0034] S2. Data monitoring: collect and monitor the acquired multimodal data;

[0035] S3. Shared underlying feature extraction: characterize the multimodal data based on the shared underlying feature extraction model to obtain shared underlying features;

[0036] S4.SOH prediction: After extracting and fusing the obtained shared underlying features based on the SOH prediction model, the SOH prediction result is output;

[0037] S5. RUL prediction: After extracting and fusing the shared underlying features obtained based on the RUL prediction model, the RUL prediction result is output;

[0038] S6. Multi-objective joint optimization, calculate the SOH loss and RUL loss separately and design a weighted multi-objective loss function to obtain the total loss and optimize the prediction error.

[0039] Please refer to Figure 2Accordingly, this embodiment provides a battery health status monitoring and prediction system with multimodal feature fusion, which is used to execute the above method and is mainly composed of the following units: a data acquisition unit 1, which is used to periodically collect static data of the battery and real-time dynamic data of the battery to obtain multimodal data; a data monitoring unit 2, which is used to aggregate and monitor the acquired multimodal data; a shared underlying feature extraction unit 3, which is used to perform feature processing on the multimodal data based on the shared underlying feature extraction model to obtain shared underlying features; an SOH prediction unit 4, which is used to extract and fuse the obtained shared underlying features based on the SOH prediction model, and output the SOH prediction result; a RUL prediction unit 5, which is used to extract and fuse the obtained shared underlying features based on the RUL prediction model, and output the RUL prediction result; a multi-objective joint optimization unit 6, which is used to calculate the SOH loss and RUL loss respectively and design a weighted multi-objective loss function to obtain the total loss and optimize the prediction error.

[0040] The battery health status monitoring and prediction method and system using multimodal feature fusion in this embodiment can perform online monitoring and analysis of multimodal feature data generated by batteries. Simultaneously, it introduces a multimodal feature fusion and multi-objective joint prediction and optimization mechanism. Based on the SOH prediction model and the RUL prediction model, independent branches are used to predict the battery health status, respectively. This not only enables automatic analysis of massive multimodal battery data, reducing workload and improving analysis efficiency, but also enhances the accuracy of prediction results. Furthermore, by characterizing multimodal data based on a shared underlying feature extraction model, shared underlying features are obtained, and a feature sharing mechanism is established, which reduces redundant computations, reduces computational complexity, and improves prediction efficiency. Example 2

[0041] Please refer to Figure 3 Based on Example 1, this embodiment provides a more preferred multi-modal feature fusion battery health status monitoring and prediction system, which will be described in detail below:

[0042] This embodiment takes two battery groups as an example, namely the first battery group 71 and the second battery group 72. Each battery group includes multiple single batteries. The data acquisition unit 1 used in this embodiment includes multiple acquisition modules 11 and charge and discharge modules 12. The acquisition modules acquire battery status data (including dynamic parameters and dynamic parameters). Specifically, the first battery group 71 periodically acquires static data of the first battery group 71 and real-time dynamic data of the first battery group 71 through the corresponding acquisition module 11 to obtain multi-modal data; the second battery group 72 also periodically acquires static data of the second battery group 72 and real-time dynamic data of the second battery group 72 through the corresponding acquisition module 11 to obtain multi-modal data. Figure 3 As shown, in one embodiment, 10 individual batteries share one data acquisition module. For a battery pack consisting of 108 individual batteries, 11 data acquisition modules are required. Of course, in other implementations, depending on the size of the battery storage cabinet, a certain number of individual batteries may be stored and managed by one data acquisition module. For example, 20 batteries may be stored in the same cabinet and served by one data acquisition module.

[0043] Each acquisition module 11 is connected to the charge and discharge module 12, and the charge and discharge module 12 is used to control the charge and discharge of the single battery group in each battery group according to the acquisition instructions of the acquisition module 11, so as to obtain the battery status data (including dynamic parameters and dynamic parameters) of each single battery, including static data and dynamic data.

[0044] In this embodiment, the static data mainly include: electrolyte concentration and self-discharge rate. The dynamic data mainly include: battery cell voltage, charge and discharge current, ambient temperature, internal resistance parameters and charge and discharge cycle times.

[0045] The data monitoring unit 2 in this embodiment uses a battery pack monitoring host 21 to aggregate and monitor the multimodal data acquired by each acquisition module 11. Furthermore, the energy storage cabinet / battery cabinet housing the battery packs is monitored online 24 / 7 via bus-powered sensors to prevent battery loss.

[0046] The shared underlying feature extraction model adopted by the shared underlying feature extraction unit 3 of this embodiment preferably adopts LSTM, SAE or attention network. It is preferred to use a multi-layer LSTM network to extract the time series degradation features in the battery charging and discharging data. The time series degradation features are part of the shared underlying features and are manifested as the changes in data within a certain period of time. The shared underlying features are mainly composed of shared underlying features derived from static data and shared underlying features derived from dynamic data. The shared underlying features are manifested as charging and discharging cycle features. Through the shared underlying feature mechanism, it is convenient to compare the actual data with the predicted data and evaluate the accuracy of the model.

[0047] Please refer to Figure 4In the SOH prediction unit 4, cross-cycle feature encoding is performed on the shared underlying features from the static data to extract trend features, including long-term aging features. Intra-cycle encoding is performed on the shared underlying features from the dynamic data to extract time-series dynamic features, including anomaly features. Cross-cycle feature encoding primarily captures trend features, such as continuous decreases or increases in data. This can occur across multiple battery charge and discharge cycles, primarily analyzing data variation patterns from a macro perspective. Intra-cycle feature encoding primarily captures anomaly features, such as selecting the highest voltage or temperature to determine if anomalies exist. This involves one battery charge and discharge cycle (full charge and complete consumption = one charge and discharge cycle), primarily analyzing data from a micro perspective. More specifically, the SOH prediction model underlying this implementation consists of the following modules: a multi-layer perceptron encoder, a transposed convolutional encoder, a concatenation layer, and a fully connected layer. For feature processing, this embodiment uses a multi-layer perceptron encoder (MLP encoder) to perform cross-cycle feature encoding on the shared underlying features from the static data to obtain a cross-cycle feature vector. The shared underlying features from the dynamic data are intra-cycle encoded through the transposed convolution encoder (TransConv Encoder), also known as the intra-cycle feature encoder, to obtain the intra-cycle feature vector.

[0048] The Multilayer Perceptron (MLP) encoder uses a multilayer perceptron (MLP) to perform nonlinear mapping on static data / parameters (such as battery structure parameters, cycle count, and ambient temperature) to extract long-term aging characteristics. Fuzzy inference rules are integrated within the MLP encoder to convert expert experience into weight parameters, enhancing model interpretability.

[0049] The TransConv Encoder primarily consists of a Transformer layer, a convolutional layer, and a generalization layer. The Transformer layer (collectively referred to as the Transformer model) incorporates two deep learning models. It uses a self-attention mechanism to capture long-term temporal dependencies during the charging and discharging process, such as voltage fluctuations and current spikes. The convolutional and generalization layers (also known as the Transformer layer) extract local temporal features, such as short-term internal resistance changes and temperature gradients, suppressing noise interference and enhancing the model's versatility.

[0050] To facilitate classification, the cross-cycle and intra-cycle feature vectors output by the multi-layer perceptron encoder and the transposed convolutional encoder are flattened into a column, converting the multi-dimensional time series signal into a one-dimensional vector to adapt to downstream feature fusion.

[0051] The stitching layer concatenates the flattened cross-cycle and intra-cycle feature vectors. A dynamic weighting mechanism is introduced to adjust the contribution of each feature vector based on the battery type and usage scenario. Different batteries require different feature processing, especially in extreme environments such as fast charging and high and low temperatures. In high-temperature environments, the weight of internal resistance is increased. Different battery types also have different weighting coefficients for each feature. For example, ternary batteries may degrade faster at high temperatures, while iron-lithium batteries are more sensitive to overdischarge.

[0052] Fully connected layer: Feature dimensionality reduction and nonlinear combination are achieved through the fully connected layer.

[0053] In an optional embodiment, the fully connected layer outputs an estimated capacity decay rate or an estimated internal resistance change rate, and calculates an estimated SOH value based on the estimated capacity decay rate or the estimated internal resistance change rate as the SOH prediction result. There are two methods for calculating the SOH estimate: one is based on the estimated capacity decay rate, and the other is based on the internal resistance change rate.

[0054] (1) Based on the capacity decay rate, the capacity decay rate formula is as follows:

[0055]

[0056] Then use the capacity fade rate to calculate the estimated SOH value, the formula is as follows:

[0057]

[0058] (2) Based on the internal resistance change rate, the internal resistance change rate formula is as follows:

[0059]

[0060] Then use the internal resistance change rate to calculate the estimated SOH value. The formula is as follows:

[0061] ‌

[0062] The SOH prediction model adopted in this embodiment first performs nonlinear transformation on cross-cycle static parameters through a multi-layer perceptron encoder to extract long-term aging features, which can realize deep mapping of static features, improve the static feature representation capability, and greatly reduce the SOH estimation error; secondly, through the transposed convolutional encoder, it captures local mutations in the short-term charging and discharging process, suppresses high-frequency noise, and realizes time series dynamic feature extraction. At the same time, combined with the Transformer self-attention mechanism, it captures long-term degradation patterns such as voltage fluctuations, realizes long-term dependency modeling, and enhances the dynamic feature modeling capability, which can identify early degradation more in advance; finally, through the splicing layer and the fully connected layer, the static and dynamic features are spliced, and the long-term aging and short-term dynamic information are combined to realize the fusion of multimodal features, improve the comprehensive representation capability of the fused features, and reduce the collaborative error of SOH and RUL prediction.

[0063] Please refer to Figure 5 In the RUL prediction unit 5, the RUL prediction model based on this embodiment mainly consists of the following modules:

[0064] Feature extraction module (left)

[0065] Composition: Residual Connection, LayerNorm and FeedForward Neural Network.

[0066] Input: Shared low-level features including parameters such as voltage, current, and temperature.

[0067] Processing flow:

[0068] ① Input features → layer normalization → feedforward neural network;

[0069] ② Original input + feedforward output → residual connection;

[0070] ③ Repeat the above steps (double-layer structure in the figure).

[0071] Output: High-dimensional representation of degenerate features , to characterize the global trend of battery performance degradation, that is, to preserve long-term degradation trends, such as linear capacity decay, and prevent the attention mechanism from over-focusing on short-term fluctuations. Mathematical expression:

[0072]

[0073] in, is the output feature (high-dimensional representation of the degenerate feature), is the input feature (shared underlying features, including voltage, current, temperature and other data), FeedForward is the fully connected layer, and LayerNorm is the layer normalization operation.

[0074] Multi-head attention module (middle)

[0075] Composition: Parallel attention heads ("multi-head attention mechanism" in the figure), matrix multiplication operation ("matrix multiplication").

[0076] Input: Output of feature extraction module .

[0077] Processing flow:

[0078] ①H out Split into Q / K / V vectors;

[0079] ② Parallel calculation of multiple attention heads, that is, capturing the nonlinear coupling relationship between multi-source features (voltage / current / temperature, etc.) through parallel subspace attention calculation to obtain the dynamic correlation between them.

[0080] Output: Capture the features of nonlinear coupling relationships as multi-head attention output .

[0081] Key functions: Identify key degradation factors, such as abnormal internal resistance, and enhance the model's sensitivity to sudden operating conditions such as current spikes and temperature changes, avoiding ignoring local anomalies.

[0082] Self-supervised learning module (lower middle)

[0083] Composition: auxiliary prediction head ("self-supervision" in the figure); time series prediction task branch. Design an auxiliary task based on time series prediction to jointly optimize feature extraction and RUL prediction main tasks. By constructing an auxiliary learning objective related to the main task (RUL prediction), the model is forced to mine more robust time series degradation features from the original data. Use unlabeled data (or the same batch of data) for self-supervised training to reduce dependence on labeled data. The auxiliary task is to predict the trend of battery dynamic parameter changes (voltage, current, temperature, etc.) in future time steps. The auxiliary task is essentially a self-supervision strategy. By designing simple time series prediction objectives, such as predicting future battery parameters, the model is guided to learn more discriminative feature expressions for the main task (RUL prediction). This design reduces the labeling cost while improving the robustness of the model in noisy environments and small sample scenarios.

[0084] Input: Intermediate output of the feature extraction module.

[0085] Processing flow:

[0086] ① Extract intermediate features from the feature extraction layer;

[0087] ② Predict future time step characteristics, such as voltage change trends;

[0088] ③ Calculate the auxiliary loss by calculating the mean square error (MSE) loss. The specific loss function is as follows:

[0089]

[0090] in, is the loss value, t is the time step index, is the true value of RUL, is the RUL predicted value, Indicates that only the time steps with valid mask labels are summed to avoid invalid data interfering with model training.

[0091] Output: prediction results of auxiliary tasks.

[0092] RUL prediction output module (right side),

[0093] Composition: feature fusion layer ("connection"), linear regression layer ("linear regression"), matrix multiplication fusion ("matrix multiplication").

[0094] Input: The output of feature extraction and the output of multi-head attention.

[0095] Processing flow: ① Matrix fusion; ② Linear regression.

[0096] Output: Remaining useful life prediction value, also known as RUL prediction value. The RUL prediction value is output through linear layer and matrix fusion operation, and the mathematical expression is as follows:

[0097]

[0098] Among them, RUL is the RUL prediction value output by the RUL prediction model, that is, the predicted value of the remaining service life. and are the linear regression layer parameters, representing the weight matrix and bias term, which are used to realize the linear mapping from feature vector to RUL value and compensate the reference offset of the prediction result respectively. is the degradation feature fusion matrix, including the global degradation feature (residual network output ) and local correlation features (attention output ), To flatten, multi-dimensional features are compressed into a one-dimensional vector.

[0099] In a preferred embodiment, the remaining useful life (RUL) and the consistency index of the single battery are predicted simultaneously. The consistency index calculation formula is as follows:

[0100]

[0101] in, σ is the standard deviation, and mean is the average value. Standard deviation (σ): refers to the standard deviation of the voltage values ​​of all cells in the battery pack measured at the same time. Mean: refers to the arithmetic mean of the voltage values ​​of all cells at the same time.

[0102] The difference in cell voltage within a battery pack directly reflects the imbalance in capacity, internal resistance, and aging between cells. Selecting cell voltage directly reflects inconsistency. By calculating consistency indicators, we can further achieve the following goals:

[0103] 1. Early warning of abnormal cells: Identify cells with high / low voltage through voltage standard deviation and locate potential battery failures, such as micro-short circuits and accelerated aging.

[0104] 2. Assess the health of the battery pack: Poor voltage consistency will accelerate overall degradation. Deterioration of indicators indicates a decrease in battery pack life, that is, a shortened RUL.

[0105] 3. Optimize charging and discharging strategies: Dynamically adjust balancing strategies, such as supplementing charging of low-voltage cells to delay consistency degradation.

[0106] 4. Support RUL joint prediction: Provide key input of the "barrel effect" for the RUL prediction model, that is, the weakest unit determines the overall lifespan.

[0107] The RUL prediction model adopted in this embodiment first adopts a feature extraction module of a multi-layer feedforward network + residual connection, which alleviates the gradient vanishing problem through residual connection, enhances the network depth, and realizes the extraction of complex high-dimensional degradation features in the battery degradation process, thereby improving the degradation feature characterization capability and reducing the RUL prediction error; secondly, a multi-head attention module is used to capture the dynamic correlation between multi-source features through parallel subspace attention calculation and identify key degradation driving factors. On the one hand, it can realize the capture of nonlinear coupling relationships, and on the other hand, it suppresses irrelevant noise interference and improves the anti-noise capability through the adaptive allocation of attention weights, thereby making the present invention more effective in prediction under complex working conditions. The measurement stability is improved and the false alarm rate is reduced; then, the self-supervised learning module is used to design auxiliary tasks based on time series prediction, jointly optimize the feature extraction and RUL prediction main tasks, enhance the model's generalization ability of time series degradation laws, use unlabeled data for pre-training, reduce dependence on labeled data, improve data utilization, and improve the prediction accuracy of the model in small sample scenarios; finally, the RUL prediction output module is combined with the linear layer + matrix fusion, and the feature extraction and attention output are integrated through matrix fusion to retain global and local degradation information. The linear layer also realizes efficient regression to meet real-time requirements, greatly shortening the prediction response time and realizing online real-time monitoring.

[0108] In general, this embodiment uses a multi-layer LSTM network to extract the timing degradation features in the battery status data (including dynamic parameters and dynamic parameters). Then, based on the shared features, the SOH branch outputs the capacity decay rate through the fully connected layer, and the RUL branch predicts the remaining number of cycles based on the degradation trend.

[0109] In the multi-objective joint optimization unit 6, the weighted multi-objective loss function of this embodiment is expressed as:

[0110]

[0111] in, is the total loss, is the SOH loss, is the RUL loss, is the weight.

[0112] The weighted multi-objective loss function adopts a dynamically weighted multi-objective loss function. Dynamic weights are adjusted based on task priorities. This embodiment designs a weighted loss function during the multi-objective joint optimization process, dynamically adjusts weight parameters based on prediction errors, prioritizes key tasks, and increases their weights when SOH is abnormal. This adapts to task priority management, avoids multi-task conflicts, and enhances prediction stability and accuracy.

[0113] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0114] The technical features of the above-described embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. The above-described embodiments only express several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they cannot be understood as limiting the scope of the present invention. For those of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all fall within the scope of protection of the present invention.

Claims

1. A battery health status monitoring and prediction method based on multimodal feature fusion, characterized in that: include: Data acquisition: periodically collect static data of the battery and collect dynamic data of the battery in real time to obtain multimodal data; Data monitoring: collect and monitor the acquired multimodal data; Shared underlying feature extraction: Based on the shared underlying feature extraction model, multimodal data is characterized to obtain shared underlying features; SOH prediction: After extracting and fusing the shared underlying features obtained based on the SOH prediction model, the SOH prediction result is output; RUL prediction: After extracting and fusing the shared underlying features obtained based on the RUL prediction model, the RUL prediction results are output; Multi-objective joint optimization: calculate the SOH loss and RUL loss separately and design a weighted multi-objective loss function to obtain the total loss and optimize the prediction error; The shared underlying features include shared underlying features derived from static data and shared underlying features derived from dynamic data; in the SOH prediction step, cross-period feature encoding is performed on the shared underlying features derived from static data to extract trend features including long-term aging features, and intra-period encoding is performed on the shared underlying features derived from dynamic data to extract time series dynamic features including abnormal features; The SOH prediction model includes a multi-layer perceptron encoder, a transposed convolutional encoder, a splicing layer, and a fully connected layer. The multi-layer perceptron encoder performs cross-cycle feature encoding on the shared underlying features derived from static data to obtain a cross-cycle feature vector. The transposed convolutional encoder performs intra-cycle encoding on the shared underlying features derived from dynamic data to obtain an intra-cycle feature vector. The splicing layer splices the cross-cycle and intra-cycle feature vectors, and introduces a dynamic weight allocation mechanism to adjust the contribution of each feature vector according to the battery type and usage scenario, and then outputs them through the fully connected layer.

2. The battery health status monitoring and prediction method based on multimodal feature fusion according to claim 1 is characterized in that: The static data includes electrolyte concentration and self-discharge rate; the dynamic data includes battery cell voltage, charge and discharge current, ambient temperature, internal resistance parameters and charge and discharge cycle times.

3. The battery health status monitoring and prediction method based on multimodal feature fusion according to claim 1 is characterized in that: The shared underlying feature extraction model uses LSTM, SAE or attention network.

4. The battery health status monitoring and prediction method based on multimodal feature fusion according to claim 1 is characterized in that: The fully connected layer outputs an estimated value of the capacity attenuation rate or an estimated value of the internal resistance change rate, and calculates an estimated SOH value based on the estimated value of the capacity attenuation rate or the estimated value of the internal resistance change rate as the SOH prediction result.

5. The battery health status monitoring and prediction method based on multimodal feature fusion according to claim 1 is characterized in that: The RUL prediction model includes: The feature extraction module, based on a multi-layer feedforward network and residual connection structure, shares the underlying features for feature extraction and extracts high-dimensional representations of degraded features; The multi-head attention module captures the nonlinear coupling relationship between multi-source features through parallel subspace attention calculation; The self-supervised learning module designs auxiliary tasks based on time series prediction and jointly optimizes the main tasks of feature extraction and RUL prediction; The RUL prediction output module outputs the RUL prediction value through linear layer and matrix fusion operation.

6. The battery health status monitoring and prediction method based on multimodal feature fusion according to claim 1 is characterized in that: In the multi-objective joint optimization step, the weighted multi-objective loss function is expressed as: ； in, is the total loss, is the SOH loss, is the RUL loss, is the weight.

7. The battery health status monitoring and prediction method based on multimodal feature fusion according to claim 6 is characterized in that: The weighted multi-objective loss function adopts a dynamic weighted multi-objective loss function. It is a dynamic weight, which is adjusted according to the task priority.

Images