Lithium battery health state prediction method and system based on multi-parameter fusion

Abstract

The invention discloses a lithium battery health state prediction method and system based on multi-parameter fusion, and relates to the technical field of lithium batteries, and the method comprises the following steps: collecting multi-dimensional parameter data in a lithium battery operation process, preprocessing the multi-dimensional parameter data, and constructing a time sequence data set; extracting multi-source degradation characteristics from the time sequence data set, and generating degradation influence factors; building a hybrid model based on an LSTM network and a Transform encoder, inputting the degradation impact factors into the hybrid model for model training until the model converges, and obtaining a lithium battery health state prediction model; and performing health detection on a to-be-detected lithium battery based on the lithium battery health state prediction model, and outputting an SOH prediction value. Multi-source battery parameters can be fused to predict the health state of the lithium battery, and the prediction precision and the real-time performance are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium batteries, and more specifically, to a method and system for predicting the state of health of lithium batteries based on multi-parameter fusion. Background Art

[0002] In today's technology-driven era, lithium batteries are widely used in multiple fields, including transportation power sources, power energy storage systems, consumer electronics, medical equipment and security, as well as aerospace and military, due to their advantages such as high energy density, long life, light weight, and environmental protection characteristics. However, during long-term use, the performance of lithium batteries will degrade. The temperature, charge-discharge rate, and depth of discharge during battery operation are all factors affecting battery aging and life. The state of health (SOH) of lithium batteries has become an important indicator for measuring the remaining life and performance of batteries, but it cannot be obtained through direct measurement. Therefore, the prediction of the state of health of lithium batteries has become one of the core research technologies in battery management systems.

[0003] In traditional technologies, the state of health of lithium batteries is often predicted by establishing physical and chemical models inside the battery. However, such methods rely on a deep understanding of the internal mechanism of the battery, are computationally complex, and require a large amount of experimental data to calibrate parameters; or empirical formulas are constructed based on battery aging data, with poor generality and difficulty in adapting to complex working conditions. In addition, in terms of parameter selection, SOH prediction often indirectly estimates through capacity attenuation or internal resistance change, which is easily affected by noise interference and has insufficient accuracy; when using machine learning models for prediction, multi-dimensional data throughout the battery life cycle is not fully integrated, making it difficult to adapt to dynamic and complex working conditions.

[0004] Therefore, how to fuse multi-source battery parameters for predicting the state of health of lithium batteries and improve the prediction accuracy and real-time performance is a technical problem that those skilled in the art urgently need to solve. Summary of the Invention

[0005] In view of this, the present invention provides a method and system for predicting the state of health of lithium batteries based on multi-parameter fusion, which solves the problems existing in the background art.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for predicting the state of health of lithium batteries based on multi-parameter fusion includes the following steps:

[0008] Collect multi-dimensional parameter data during the operation of lithium batteries, preprocess the multi-dimensional parameter data, and construct a time series data set;

[0009] Extract multi-source degradation features from the time series data set to generate degradation influence factors;

[0010] Build a hybrid model based on the LSTM network and the Transformer encoder, input the degradation impact factors into the hybrid model for model training until the model converges, and obtain the lithium battery health state prediction model;

[0011] Based on the lithium battery health state prediction model, perform health detection on the lithium battery to be detected and output the SOH prediction value.

[0012] Optionally, the multi-dimensional parameter data is divided into three types: electrochemical parameters, environmental parameters, and usage parameters; among them, the electrochemical parameters include voltage, current, capacity, and internal resistance, the environmental parameters include temperature and humidity, and the usage parameters include the number of cycles, charge-discharge depth, and charging rate.

[0013] Optionally, preprocess the multi-dimensional parameter data, which specifically includes the following steps:

[0014] Based on the physical characteristics of the battery, eliminate the abnormal data exceeding the preset reasonable range from the multi-dimensional parameter data, fill in the missing values using the interpolation method, and adopt the data standardization and normalization methods to eliminate the dimension difference, obtaining the first data set;

[0015] Select a suitable wavelet basis to process the first data set and remove the high-frequency noise components to obtain the second data set;

[0016] Divide the second data set into independent charge-discharge cycles, each cycle corresponding to a number of cycles, and add SOH labels to the data of each charge-discharge cycle through standard charge-discharge tests.

[0017] Optionally, generate the degradation impact factors, which specifically includes the following steps:

[0018] Calculate the capacity attenuation rate of each cycle, extract the slope of the change curve of the internal resistance over time, and obtain the time-domain characteristics;

[0019] Through Fourier transform, calculate the low-frequency energy ratio and the root mean square value of the high-frequency components of the signal to obtain the frequency-domain characteristics;

[0020] Based on the distribution characteristic analysis and correlation analysis, calculate the voltage distribution skewness and the correlation coefficient between the capacity and the internal resistance to obtain the statistical characteristics;

[0021] Adopt the attention mechanism to perform weighted fusion on the time-domain characteristics, frequency-domain characteristics, and statistical characteristics to generate the degradation impact factors.

[0022] Optionally, the hybrid model includes an input layer, a time series feature extraction layer, a global feature interaction layer, a fully connected layer, and an output layer connected in sequence;

[0023] The input layer is used to receive the degradation impact factors and organize the degradation impact factors in time series to form a tensor form suitable for model input;

[0024] The time series feature extraction layer uses a double-layer LSTM unit to capture the slow change trend of battery parameters over time during multiple charge and discharge cycles of lithium batteries;

[0025] The global feature interaction layer is used to globally model different positions in the input sequence and explore potential correlations between data;

[0026] The fully connected layer is used to transform the dimension and integrate the information of the feature vector output by the global feature interaction layer, and add a bias term;

[0027] The output layer is used to output the current health status prediction results and future health trends of the lithium battery.

[0028] Optionally, the global feature interaction layer includes: an attention mechanism module and a feedforward network;

[0029] The attention mechanism module adopts a multi-head self-attention mechanism to project the hidden state sequence output by the temporal feature extraction layer onto different linear transformation matrices and concatenate and fuse the calculation results to obtain the multi-head self-attention output;

[0030] Feedforward network, used to transform and enhance the features of multi-head self-attention outputs and explore the potential connections between battery parameters;

[0031] Residual connections and layer normalization are introduced in both the attention mechanism module and the feedforward network. The residual connection is used to add the input to the result transformed by the attention mechanism module or the feedforward network, and the layer normalization is used to normalize all feature dimensions of each input data.

[0032] Optionally, the specific steps of model training are:

[0033] The degradation influencing factors after feature extraction are divided into a training set and a validation set according to a preset ratio. The training set is used to update the model parameters, and the validation set is used to adjust the model hyperparameters.

[0034] Select mean square error as the loss function and Adam optimizer to update the model parameters. The weights and biases of each layer in the model are continuously adjusted by minimizing the loss function.

[0035] The training set is input into the hybrid model in batches, and passes through the time series feature extraction layer, global feature interaction layer, fully connected layer and output layer in turn to calculate the loss value between the prediction result and the true label; the gradient of each parameter is calculated through the back propagation algorithm, and the Adam optimizer updates the parameters according to the gradient; this process is repeated until the performance of the model on the validation set no longer improves or reaches the preset number of training rounds, and the optimal parameters are retained as the lithium battery health status prediction model.

[0036] Optionally, it further includes:

[0037] When it is determined that the health state of the lithium battery is abnormal according to the predicted SOH value of the output and the preset abnormal threshold, trigger a light or sound for reminder, and transmit the predicted SOH value of the current state of the lithium battery to the user interface to notify the user or operator of the possible safety hazards of the lithium battery.

[0038] A lithium battery health state prediction system based on multi-parameter fusion, applying the method for predicting the health state of a lithium battery based on multi-parameter fusion described in any one of the above, includes a data acquisition module, a feature extraction module, a model construction and training module, and a prediction module connected in sequence;

[0039] The data acquisition module is used to collect multi-dimensional parameter data during the operation of the lithium battery, preprocess the multi-dimensional parameter data, and construct a time series data set;

[0040] The feature extraction module is used to extract multi-source degradation features from the time series data set and generate degradation influence factors;

[0041] The model construction and training module is used to construct a hybrid model based on the LSTM network and the Transformer encoder, input the degradation influence factors into the hybrid model for model training until the model converges, and obtain a lithium battery health state prediction model;

[0042] The prediction module is used to perform health detection on the lithium battery to be detected through the lithium battery health state prediction model and output the predicted SOH value.

[0043] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method and system for predicting the health state of a lithium battery based on multi-parameter fusion, which fuses multi-source battery parameters, constructs a hybrid model based on the LSTM network and the Transformer encoder, significantly improves the prediction accuracy and real-time performance of SOH, and timely reminds the user or operator of possible safety hazards, solving the problems of single parameter, poor dynamic adaptability, and insufficient real-time performance existing in the existing prediction technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0045] Figure 1 It is a flowchart of the method for predicting the health state of a lithium battery based on multi-parameter fusion provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] Next, in combination with the accompanying drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0047] Accurately predicting the health state of lithium batteries can timely detect potential battery failures and safety hazards, take measures in advance to avoid safety accidents caused by battery failures, such as fires, explosions, etc., and ensure the safety of personnel's lives and equipment and property. By understanding the health state of the battery, a more reasonable charge and discharge strategy can be formulated to avoid situations that damage the battery such as overcharging and over-discharging, extend the battery life, and reduce the use cost. In addition, with the rapid development of the new energy industry, accurate health state prediction technology helps to improve the quality and market competitiveness of battery products, promote the healthy and sustainable development of the lithium battery industry, and at the same time provide strong support for the development of related industries.

[0048] The health state of lithium batteries is affected by various factors. For example, the magnitude of the current will affect the chemical reaction rate inside the battery, the temperature change will affect the internal resistance and chemical activity of the battery, and the number of cycles is directly related to the degree of battery aging, etc. If only relying on a single piece of information for judgment, the extracted features are difficult to comprehensively reflect the complex physical and chemical processes inside the battery, and the prediction model is difficult to capture the complex non-linear relationship between the battery health state and various influencing factors, resulting in low stability and reliability of the prediction results.

[0049] To solve the above problems, the embodiments of the present invention disclose a method for predicting the health state of lithium batteries based on multi-parameter fusion, as Figure 1 shown, including the following steps:

[0050] Collect multi-dimensional parameter data during the operation of the lithium battery, preprocess the multi-dimensional parameter data, and construct a time series dataset;

[0051] Extract multi-source degradation features from the time series dataset to generate degradation impact factors;

[0052] Build a hybrid model based on the LSTM network and the Transformer encoder, input the degradation impact factors into the hybrid model for model training until the model converges to obtain a lithium battery health state prediction model;

[0053] Perform health detection on the lithium battery to be detected based on the lithium battery health state prediction model, and output the SOH prediction value.

[0054] Based on Figure 1According to the shown process, in this embodiment, multiple battery parameters are collected and key influencing factors related to the degradation of lithium batteries are extracted from them. A hybrid model is built by combining the LSTM network and the Transformer encoder, giving full play to the advantages of both, which can more comprehensively and accurately mine the time series features and global correlation information in the multi-parameter data of lithium batteries, thereby further improving the accuracy of lithium battery state of health prediction.

[0055] Further, the multi-dimensional parameter data is divided into three types: electrochemical parameters, environmental parameters, and usage parameters; among them, the electrochemical parameters include voltage, current, capacity, and internal resistance, the environmental parameters include temperature and humidity, and the usage parameters include cycle number, charge-discharge depth, and charging rate.

[0056] Among them, the battery voltage can intuitively reflect the current charging state and working condition. By monitoring the voltage, it is possible to initially judge whether the battery is working properly and estimate the remaining power; the current determines the charge-discharge rate of the battery. Understanding the magnitude of the current helps to analyze the performance of the battery under different working conditions and calculate the charge-discharge capacity of the battery; lithium batteries are extremely sensitive to temperature. Too high or too low temperature will change the chemical reaction rate inside the battery, affecting the capacity and internal resistance; the cycle number is directly related to the aging degree of the battery, the internal resistance can reflect the degree of obstruction to the current inside the battery, and the capacity represents the amount of electric charge that the battery can store. In this embodiment, selecting the above-mentioned several battery parameters for monitoring and comprehensive analysis can more accurately evaluate the battery state of health, and further provide reliable data for the battery management system to achieve reasonable use and effective maintenance of the battery.

[0057] Further, to ensure data quality and improve model performance, this embodiment preprocesses the multi-dimensional parameter data, which specifically includes the following steps:

[0058] Based on the physical characteristics of the battery, abnormal data exceeding the preset reasonable range is removed from the multi-dimensional parameter data, the missing values are filled using the interpolation method, and the data standardization and normalization methods are used to eliminate the dimension difference, obtaining the first data set;

[0059] Select a suitable wavelet basis to process the first data set to remove the high-frequency noise components, obtaining the second data set;

[0060] The second data set is divided into independent charge-discharge cycles, each cycle corresponding to a cycle number, and SOH labels are added to the data of each charge-discharge cycle through standard charge-discharge tests.

[0061] Based on the above preprocessing operations, in this embodiment, through data cleaning and outlier handling, noise data can be removed to ensure data validity; missing value filling processing can solve the problem of data loss caused by sensor failures or communication interruptions; data standardization and normalization can eliminate the dimensional differences of different parameters and accelerate model convergence; wavelet basis decomposition can improve signal quality and highlight key features related to battery degradation. Therefore, this embodiment can reduce the impact of noise on the model, significantly improve the accuracy of subsequent feature extraction and model training, and provide a high-quality data basis for lithium battery state of health prediction.

[0062] Furthermore, to generate degradation impact factors, this embodiment proposes time-domain analysis, frequency-domain analysis, and statistical analysis methods for time-series data sets, which specifically include the following steps:

[0063] Calculate the capacity attenuation rate of each cycle, extract the slope of the curve of internal resistance changing with time, and obtain time-domain features; where, Cn is the capacity of the nth cycle, and C0 is the initial capacity;

[0064] Through Fourier transform, calculate the low-frequency energy proportion and the root mean square value of the high-frequency component of the signal to obtain frequency-domain features; where, an increase in the low-frequency energy proportion may reflect the slow accumulation of side reactions inside the battery, and a sudden increase in the high-frequency noise energy may indicate local micro-short circuit or lithium plating;

[0065] Based on distribution characteristic analysis and correlation analysis, calculate the skewness of voltage distribution and the correlation coefficient between capacity and internal resistance to obtain statistical features; where, an increase in the skewness of voltage distribution may reflect the intensification of inhomogeneous reactions inside the battery,

[0066] Adopt the attention mechanism to perform weighted fusion on time-domain features, frequency-domain features, and statistical features to generate degradation impact factors.

[0067] Based on the above feature extraction operations, the frequency-domain features extracted in this embodiment can capture microscopic degradation signs early, and the statistical features provide a physical basis for the model and enhance the credibility of the results; at the same time, time-domain analysis, frequency-domain analysis, and statistical analysis methods are adopted to avoid the limitations of a single perspective, provide high-quality input for the subsequent hybrid model, and achieve high-precision SOH prediction.

[0068] LSTM is good at processing time series data and capturing long-term dependencies, and is suitable for extracting time series features during battery degradation; while the self-attention mechanism of Transformer can capture the association between global features and parameters and solve the problem of long-distance dependencies. Therefore, in order to combine the time series modeling capability of LSTM with the global feature interaction of Transformer to improve the prediction accuracy, this embodiment proposes a hybrid model, including an input layer, a time series feature extraction layer (LSTM), a global feature interaction layer (Transformer encoder), a fully connected layer, and an output layer connected in sequence;

[0069] The input layer is used to receive the degradation influencing factors and organize them in time series to form a tensor form suitable for model input;

[0070] The time series feature extraction layer uses a double-layer LSTM unit to capture the slow change trend of battery parameters over time during multiple charge and discharge cycles of lithium batteries. The LSTM unit receives the hidden state of the previous moment and the input data of the current moment, and selectively memorizes and updates information through the collaborative work of the forget gate, input gate, and output gate. The hidden state of its output will carry the dynamic change characteristics of the battery parameters in the time dimension, providing a basis for subsequent processing.

[0071] The global feature interaction layer is used to globally model different positions in the input sequence and explore the potential correlation between data. The hidden state of the LSTM output is further input into the Transformer encoder, which can analyze the relationship between these features in the entire time series and overcome the limitations that LSTM may have when processing long sequences.

[0072] The fully connected layer is used to transform the dimension and integrate the information of the feature vector output by the global feature interaction layer, and add a bias term. Based on this, the extracted complex features can be mapped to specific dimensions related to the health status of the lithium battery, providing a usable feature representation for the final prediction;

[0073] The output layer is used to output the current health status prediction results and future health trends of the lithium battery.

[0074] Furthermore, from the perspective of enhancing the feature extraction and analysis capabilities of multi-parameter time series data of lithium batteries, this embodiment designs the following Transformer encoder layer. Specifically, the global feature interaction layer includes: an attention mechanism module, a feedforward network;

[0075] The attention mechanism module adopts a multi-head self-attention mechanism to project the hidden state sequence output by the temporal feature extraction layer onto different linear transformation matrices and concatenate and fuse the calculation results to obtain multi-head self-attention outputs. By calculating attention in parallel in different representation subspaces, it can capture richer features and relationships in the data.

[0076] The feedforward network is used to perform feature conversion and enhancement on the output of multi-head self-attention and explore the potential connections between battery parameters. The first layer of the feedforward network is a linear transformation layer, which projects the input data into a higher-dimensional space and uses the ReLU activation function to increase the nonlinear expression ability of the model. The second layer projects the data in the high-dimensional space back to the original dimension. Based on this method, the feedforward network can learn complex patterns in the data and further explore the potential connections between lithium battery parameters.

[0077] Residual connections and layer normalization are introduced in both the attention mechanism module and the feedforward network. Residual connections are used to add the input to the result after transformation by the attention mechanism module or the feedforward network, and layer normalization is used to normalize all feature dimensions of each input data. Residual connections help retain important information in the original input, so that the model can still effectively learn long-term dependencies in the deep structure and avoid information loss during transmission; layer normalization can accelerate the convergence of the model, improve the stability of training, and ensure that the model learns data features faster and more accurately.

[0078] Based on the design of the above Transformer layer, this embodiment can lay a solid foundation for accurately analyzing the global correlation information in the multi-parameter data of lithium batteries, and effectively improve the performance of the lithium battery health status prediction model.

[0079] Furthermore, the specific steps of model training are:

[0080] The degradation influencing factors after feature extraction are divided into training set and validation set according to the preset ratio. The training set is used to update the model parameters, and the validation set is used to adjust the model hyperparameters to prevent overfitting.

[0081] Select mean square error as the loss function and Adam optimizer to update the model parameters. The weights and biases of each layer in the model are continuously adjusted by minimizing the loss function.

[0082] The training set is input into the hybrid model in batches, and passes through the time series feature extraction layer, global feature interaction layer, fully connected layer and output layer in turn to calculate the loss value between the prediction result and the true label; the gradient of each parameter is calculated through the back propagation algorithm, and the Adam optimizer updates the parameters according to the gradient; this process is repeated until the performance of the model on the validation set no longer improves or reaches the preset number of training rounds, and the optimal parameters are retained as the lithium battery health status prediction model.

[0083] Further, the method for predicting the state of health of the lithium battery in this embodiment further includes the following steps:

[0084] When it is determined that the state of health of the lithium battery is abnormal according to the predicted SOH value output and the preset abnormal threshold, trigger a light or sound for reminder, and transmit the predicted SOH value of the current state of the lithium battery to the user interface to notify the user or operator of the possible safety hazards of the lithium battery.

[0085] And Figure 1 Corresponding to the Figure 1 method described above, the embodiment of the present invention further provides a system for predicting the state of health of a lithium battery based on multi-parameter fusion, which is used for

[0086] The data acquisition module is used to collect multi-dimensional parameter data during the operation of the lithium battery, preprocess the multi-dimensional parameter data, and construct a time series data set;

[0087] The feature extraction module is used to extract multi-source degradation features from the time series data set and generate degradation impact factors;

[0088] The model construction and training module is used to construct a hybrid model according to the LSTM network and the Transformer encoder, input the degradation impact factor into the hybrid model for model training until the model converges, and obtain a prediction model for the state of health of the lithium battery;

[0089] The prediction module is used to perform health detection on the lithium battery to be detected through the prediction model of the state of health of the lithium battery and output the predicted SOH value.

[0090] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same and similar parts among the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0091] The foregoing description of the disclosed embodiments enables those skilled in the art to practice or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting the health state of a lithium battery based on multi-parameter fusion, characterized in that, The following steps are involved: Collect multi-dimensional parameter data during the operation of lithium batteries, pre-process the multi-dimensional parameter data, and construct a time series data set; Extract multi-source degradation features from time series data sets and generate degradation impact factors; A hybrid model is built based on the LSTM network and the Transformer encoder, and the degradation influencing factors are input into the hybrid model for model training until the model converges to obtain a lithium battery health status prediction model; Based on the lithium battery health status prediction model, the health of the lithium battery to be tested is tested, and the SOH prediction value is output.

2. The method for predicting the state of health of a lithium battery based on multi-parameter fusion according to claim 1, wherein, Multidimensional parameter data is divided into three types: electrochemical parameters, environmental parameters and usage parameters; among them, electrochemical parameters include voltage, current, capacity, internal resistance, environmental parameters include temperature and humidity, and usage parameters include number of cycles, charge and discharge depth, and charging rate.

3. A method for predicting the health state of a lithium battery based on multi-parameter fusion according to claim 1, characterized in that, Preprocessing the multidimensional parameter data includes the following steps: Based on the physical characteristics of the battery, the abnormal data exceeding the preset reasonable range is eliminated from the multi-dimensional parameter data, the missing values ​​are filled by interpolation method, and the dimension differences are eliminated by data standardization and normalization method to obtain the first data set; Selecting a suitable wavelet basis to process the first data set, removing high-frequency noise components, and obtaining a second data set; The second data set is divided into independent charge and discharge cycles, each cycle corresponds to one cycle number, and a SOH label is added to the data of each charge and discharge cycle through a standard charge and discharge test.

4. A method for predicting the health state of a lithium battery based on multi-parameter fusion according to claim 1, characterized in that, Generate the degradation impact factor, specifically including the following steps: Calculate the capacity decay rate of each cycle, extract the slope of the internal resistance change curve over time, and obtain the time domain characteristics; Through Fourier transform, the low-frequency energy ratio and the root mean square value of the high-frequency component of the signal are calculated to obtain the frequency domain characteristics; Based on distribution characteristic analysis and correlation analysis, the voltage distribution skewness and the correlation coefficient between capacity and internal resistance are calculated to obtain statistical characteristics; The attention mechanism is used to perform weighted fusion of time domain features, frequency domain features and statistical features to generate degradation impact factors.

5. A method for predicting the health state of a lithium battery based on multi-parameter fusion according to claim 1, characterized in that, The hybrid model includes a sequentially connected input layer, a temporal feature extraction layer, a global feature interaction layer, a fully connected layer, and an output layer; The input layer is used to receive the degradation influencing factors and organize them in time series to form a tensor form suitable for model input; The time series feature extraction layer uses a double-layer LSTM unit to capture the slow change trend of battery parameters over time during multiple charge and discharge cycles of lithium batteries; The global feature interaction layer is used to globally model different positions in the input sequence and explore potential correlations between data; The fully connected layer is used to transform the dimension and integrate the information of the feature vector output by the global feature interaction layer, and add a bias term; The output layer is used to output the current health status prediction results and future health trends of the lithium battery.

6. The method for predicting the health state of a lithium battery based on multi-parameter fusion according to claim 5, wherein The global feature interaction layer includes: attention mechanism module and feedforward network; The attention mechanism module adopts a multi-head self-attention mechanism to project the hidden state sequence output by the temporal feature extraction layer onto different linear transformation matrices and concatenate and fuse the calculation results to obtain the multi-head self-attention output; Feedforward network, used to transform and enhance the features of multi-head self-attention outputs and explore the potential connections between battery parameters; Residual connections and layer normalization are introduced in both the attention mechanism module and the feedforward network. The residual connection is used to add the input to the result transformed by the attention mechanism module or the feedforward network, and the layer normalization is used to normalize all feature dimensions of each input data.

7. A method for predicting the health state of a lithium battery based on multi-parameter fusion according to claim 5, characterized in that The specific steps of model training are: The degradation influencing factors after feature extraction are divided into a training set and a validation set according to a preset ratio. The training set is used to update the model parameters, and the validation set is used to adjust the model hyperparameters. Select mean square error as the loss function and Adam optimizer to update the model parameters. The weights and biases of each layer in the model are continuously adjusted by minimizing the loss function. The training set is input into the hybrid model in batches, and passes through the time series feature extraction layer, global feature interaction layer, fully connected layer and output layer in turn to calculate the loss value between the prediction result and the true label; the gradient of each parameter is calculated through the back propagation algorithm, and the Adam optimizer updates the parameters according to the gradient; this process is repeated until the performance of the model on the validation set no longer improves or reaches the preset number of training rounds, and the optimal parameters are retained as the lithium battery health status prediction model.

8. A method for predicting the health state of a lithium battery based on multi-parameter fusion according to claim 1, characterized in that Also includes: When the health status of the lithium battery is judged to be abnormal based on the output SOH prediction value and the preset abnormal threshold, the light or sound is triggered to remind, and the SOH prediction value of the current status of the lithium battery is transmitted to the user interface to inform the user or operator of the possible safety hazards of the lithium battery.

9. A lithium battery health state prediction system based on multi-parameter fusion, characterized in that, A lithium battery health status prediction method based on multi-parameter fusion as described in any one of claims 1 to 8 is applied, comprising a data acquisition module, a feature extraction module, a model building and training module, and a prediction module connected in sequence; The data acquisition module is used to collect multi-dimensional parameter data during the operation of the lithium battery, pre-process the multi-dimensional parameter data, and construct a time series data set; The feature extraction module is used to extract multi-source degradation features from the time series data set and generate degradation impact factors; The model building and training module is used to build a hybrid model based on the LSTM network and the Transformer encoder, input the degradation influencing factors into the hybrid model for model training until the model converges, and obtain a lithium battery health status prediction model; The prediction module is used to perform health detection on the lithium battery to be detected through the lithium battery health status prediction model and output the SOH prediction value.

Images