Sodium-ion battery health state evaluation method and system based on multi-source data fusion

Abstract

The application discloses a kind of based on multi-source data fusion's sodium ion battery health state evaluation method and system, method includes: continuously collecting the operating state parameter and environmental parameter of sodium ion battery as multi-source data and carries out data preprocessing;Correlation between the health state SOH of multi-source data is calculated, and several characteristic parameters with correlation greater than first preset threshold are screened out;Information entropy calculation is carried out to the characteristic parameters after correlation screening, and the characteristic parameters with information entropy value greater than second preset threshold are retained, and information enhancement is executed to voltage fluctuation class and capacity attenuation class characteristic parameters with information entropy in preset entropy value range;The deep learning model based on BERT network is constructed, and the deep learning model is trained with the processed characteristic parameters as training set data, to obtain battery health state evaluation model.The present application can realize the high-precision dynamic prediction of sodium ion battery health state.

Description

Technical Field

[0001] This invention relates to the field of sodium-ion battery health status monitoring technology, specifically to a method and system for assessing the health status of sodium-ion batteries based on multi-source data fusion. Background Technology

[0002] With the rapid development of renewable energy and electric vehicles, sodium-ion batteries have gradually become an important alternative to lithium-ion batteries due to their abundant resources, low cost, and environmental friendliness. However, the State of Health (SOH) assessment of sodium-ion batteries still faces many technical challenges. Existing SOH assessment methods are mainly divided into the following categories, but all have significant drawbacks: 1) Methods based on a single data source. Traditional methods usually rely on a single signal (such as voltage, current, or temperature) for SOH assessment. For example, Kalman filtering and equivalent circuit models estimate capacity decay through voltage-current curves. Although computationally simple, these methods show a significant decrease in accuracy under complex operating conditions (such as rapid charging and discharging, and temperature fluctuations). For example, in low-temperature (-20℃) environments, voltage signals are easily affected by polarization effects, leading to prediction errors exceeding 10%. In addition, a single signal cannot comprehensively reflect the combined effects of internal chemical reactions and mechanical aging in the battery; 2) Methods based on physical models. These methods predict SOH by constructing physical models such as sodium-ion diffusion equations and electrode reaction kinetic models. For example, pseudo-two-dimensional (P2D) models simulate battery behavior by coupling electrochemical equations. However, physical models are highly complex and require precise battery parameters (such as electrode porosity and diffusion coefficient), which are difficult to calibrate in real time in practical applications. Furthermore, model computation is time-consuming (a single simulation can take several hours), failing to meet online monitoring requirements; 3) Data-driven methods. In recent years, machine learning (such as support vector machines and random forests) and deep learning (such as LSTM and CNN) have been widely used in SOH prediction, but overfitting occurs due to feature redundancy when fusing multi-source data. The bottleneck of data-driven methods lies in their dependence on high-quality data, and the models' generalization ability under unknown conditions (such as extreme temperatures and vibrations) is insufficient; 4) Statistical methods. Principal component analysis (PCA) and factor analysis process multivariate data through dimensionality reduction, but lose nonlinear features. In addition, statistical methods struggle to handle the coupled effects of dynamic environmental factors (such as humidity and mechanical stress). Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide a method and system for assessing the health status of sodium-ion batteries based on multi-source data fusion, which addresses the problems of existing technologies such as reliance on a single data source, insufficient nonlinear processing capabilities, and poor environmental adaptability, thereby achieving high-precision dynamic prediction of the health status of sodium-ion batteries.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for assessing the health status of sodium-ion batteries based on multi-source data fusion includes the following steps: S1, continuously collect the operating status parameters and environmental parameters of the sodium-ion battery as multi-source data and perform data preprocessing; S2, calculate the correlation between the multi-source data and the health status SOH based on the Pearson correlation coefficient, and filter out several feature parameters in the multi-source data whose correlation is greater than a first preset threshold; S3, calculate the information entropy of the feature parameters after correlation screening, retain the feature parameters whose information entropy value is greater than the second preset threshold, and perform information enhancement on the voltage fluctuation type and capacity decay type feature parameters whose information entropy is within the preset entropy value range; S4. Construct a deep learning model based on the BERT network, and train the deep learning model using the feature parameters processed in step S3 as the training set data to obtain a battery health status assessment model. S5 uses a trained battery health status assessment model to estimate the health status of sodium-ion batteries.

[0005] Further, in step S2, the correlation between the multi-source data and the health status SOH is calculated based on the Pearson correlation coefficient, and the calculation expression is:

[0006] In the above formula, x i and y i These are variables from multiple data sources. x and y The i-th observation, and They are variables x and y The mean of the values, where n is the number of observations.

[0007] Furthermore, in step S3, the expression for calculating the information entropy of the feature parameters after relevance filtering is as follows:

[0008] In the above formula, H( x ) is the information entropy of the random variable X, p( x i ) is a random variable x Values x i The probability, x i and y i These are variables from multiple data sources. x andy The i-th observation, where n is the number of observations.

[0009] Furthermore, information enhancement is performed on characteristic parameters of voltage fluctuation and capacity attenuation types whose information entropy is within a preset entropy range. The calculation expression is as follows:

[0010] In the above formula, These are the original values ​​of the feature parameters. The mean of the feature parameters, The standard deviation of the characteristic parameter, For the enhanced feature parameter values, This is the enhancement coefficient.

[0011] Further, in step S4, when training the deep learning model, the training objective is to minimize the loss function constrained by the mixed mean square error and information entropy. The deep learning model's parameters are updated when adjacent temperature changes or voltage fluctuations in the feature parameters exceed the trigger condition. The expression for the loss function is:

[0012]

[0013]

[0014] In the above formula, It is the actual value. These are predicted values, where n is the sample size and L is the number of samples. entropy For information regular expressions, L MSE Let λ be the mixed mean square error, and λ be the weighting parameter. For summation, j is the category index. To belong to category The probability of; The expression for parameter update is:

[0015] In the above formula, For model parameters, The learning rate; The steps for updating parameters are as follows: When the trigger condition is met or When calculating the loss function gradient with respect to parameter θ The parameters are then adjusted according to the learning rate η to obtain new parameters. , where Δ Δ represents the temperature change between adjacent samples. This represents the change in voltage between adjacent samples. , These are the corresponding temperature and voltage thresholds.

[0016] Furthermore, step S1, data preprocessing, specifically includes: Outlier data points exceeding a preset range in multi-source data are removed. The preset range is a range extending upwards and downwards from the mean by a certain number of standard deviations. Normalization and Kalman filtering are performed on the multi-source data after removing outlier data points.

[0017] Furthermore, in step S1, data preprocessing also includes using the voltage fluctuation range of the low-voltage plateau region as a reference for data removal, based on the operating characteristics of sodium-ion batteries, and using a threshold of the cycle capacity decay rate, based on the actual degradation law of sodium-ion batteries, to assist in identifying abnormal data. The expression for the voltage fluctuation amplitude in the low-voltage plateau region is:

[0018] In the above formula, This represents the maximum voltage change over time within the low-voltage plateau region. This represents the minimum voltage change over time within the low-voltage plateau region. The expression for the cycle capacity decay rate is:

[0019] In the above formula, C k Let ΔN be the capacity of the k-th iteration, and ΔN be the number of iteration intervals. This represents the capacity decay.

[0020] Furthermore, the deep learning model includes an input layer, multiple Transformer encoder layers, and an output layer. The Transformer encoder layers employ a multi-head self-attention mechanism, expressed as:

[0021] In the above formula, Q, K, and V are the query matrix, key matrix, and value matrix, respectively, and d k It is the dimension of the key matrix K. This is the dot product similarity matrix.

[0022] A sodium-ion battery health status assessment system based on multi-source data fusion includes an interconnected microprocessor and a memory, wherein the microprocessor is programmed or configured to execute the steps of a sodium-ion battery health status assessment method based on multi-source data fusion.

[0023] A computer-readable storage medium storing a computer program / instructions programmed or configured to execute, via a processor, steps of a sodium-ion battery health status assessment method based on multi-source data fusion.

[0024] Compared with the prior art, the advantages of the present invention are as follows: This invention achieves high-precision assessment of the health status of sodium-ion batteries through multi-source data fusion and intelligent feature selection. By employing dual screening using Pearson correlation coefficient and information entropy, redundant and low-contribution features are effectively eliminated, while key influencing factors are retained. Information enhancement is applied to voltage fluctuation and capacity decay features, making these feature parameters more consistent with the performance degradation patterns of sodium-ion batteries. The deep learning model based on the BERT network possesses powerful feature representation and temporal relationship modeling capabilities, accurately capturing battery aging patterns and significantly improving the prediction accuracy, robustness, and real-time performance of sodium-ion battery SOH estimation. This makes it suitable for scenarios such as energy storage power stations and electric vehicles. Attached Figure Description

[0025] Figure 1 This is a flowchart of a sodium-ion battery health status assessment method based on multi-source data fusion, according to an embodiment of the present invention.

[0026] Figure 2 This is a schematic diagram of the BERT network structure in a specific application embodiment.

[0027] Figure 3 This is a flowchart illustrating the updating of weight parameters using the Adam optimizer in a specific application example. Detailed Implementation

[0028] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0029] like Figure 1 As shown, the sodium-ion battery health status assessment method based on multi-source data fusion in this embodiment includes the following steps: S1, continuously collect the operating status parameters and environmental parameters of the sodium-ion battery as multi-source data and perform data preprocessing; S2, calculate the correlation between the multi-source data and the health status SOH based on the Pearson correlation coefficient, and filter out several feature parameters in the multi-source data whose correlation is greater than a first preset threshold; S3, calculate the information entropy of the feature parameters after correlation screening, retain the feature parameters whose information entropy value is greater than the second preset threshold, and perform information enhancement on the voltage fluctuation type and capacity decay type feature parameters whose information entropy is within the preset entropy value range; S4. Construct a deep learning model based on the BERT network, and train the deep learning model using the feature parameters processed in step S3 as the training set data to obtain a battery health status assessment model. S5 uses a trained battery health status assessment model to estimate the health status of sodium-ion batteries.

[0030] In this embodiment, step S1, data preprocessing, specifically includes: Outlier data points exceeding a preset range in multi-source data are removed. The preset range is a range extending upwards and downwards from the mean by a certain number of standard deviations. Normalization and Kalman filtering are performed on the multi-source data after removing outlier data points.

[0031] Outlier data points exceeding a preset range in the multi-source data are removed. The preset range is a range extending upwards and downwards from the mean by several standard deviations. The multi-source data after removing outlier data points is then normalized and processed using Kalman filtering.

[0032] In this embodiment, when performing data preprocessing in step S1, it further includes using the voltage fluctuation range of the low-voltage plateau region as a reference for data rejection based on the operating characteristics of sodium-ion batteries, and using the threshold of the cycle capacity decay rate based on the actual degradation law of sodium-ion batteries to assist in identifying abnormal data. The expression for the voltage fluctuation amplitude of the low-voltage plateau region is as follows:

[0033] In the above formula, This represents the maximum voltage change over time within the low-voltage plateau region. This represents the minimum voltage change over time within the low-voltage plateau range; specifically, based on the voltage stability characteristics of the low-voltage plateau (2.0V~3.0V) of sodium-ion batteries, The threshold range for judging abnormal data needs to be limited to a reasonable range of voltage fluctuations within this range to avoid normal platform fluctuations being mistakenly judged as abnormal data (in practical applications, it needs to be calibrated through a large amount of experimental data, and the specific range needs to be determined in combination with battery model, charge and discharge rate and other operating conditions). The expression for the cycle capacity decay rate is:

[0034] In the above formula, C k Let ΔN be the capacity of the k-th iteration, and ΔN be the number of iteration intervals. This refers to the capacity decay rate. The threshold range for judging abnormal data of cycle capacity decay rate needs to conform to the actual decay law of sodium-ion batteries. The threshold needs to distinguish between normal capacity decay and abnormal decay (for example, the common range may be 0.01%~0.5% / cycle, and the specific value needs to be calibrated according to factors such as battery design life and usage environment).

[0035] It is understandable that this embodiment, in its data screening for sodium-ion battery characteristics, specifically adds a voltage fluctuation range (2.0–3.0 V) as a reference for data removal during the outlier removal process, taking into account the operating characteristics of sodium-ion batteries. This avoids misjudging fluctuations as outliers during this critical plateau phase. In capacity-related data processing, the threshold of the cycle decay rate ΔC / ΔN is used to assist in identifying anomalies, ensuring that the removal results more closely match the actual decay patterns of sodium-ion batteries.

[0036] Specifically, step S1 involves data acquisition and preprocessing, simultaneously acquiring the voltage, current, temperature, and environmental parameters of the sodium-ion battery at a sampling frequency of 10Hz~1kHz, and performing data cleaning, noise reduction, and Min-Max normalization processing.

[0037] During data collection, the following parameters are collected: Signal types: voltage (sampling frequency 1kHz), current (1kHz), temperature (10Hz), humidity (1Hz), vibration acceleration (100Hz). Equipment: High-precision voltage sensor (error ±0.1mV), Hall current sensor (error ±0.5%), integrated temperature and humidity module (accuracy ±0.5℃).

[0038] The steps of data preprocessing are as follows: Outlier removal: Based on the 3σ principle, outliers exceeding the limit are removed. μ ±3 σ The range of data points; Normalization: Min-Max standardization is used, and the formula is:

[0039] Where X is the original input value, X min X is the minimum value of the corresponding feature in the entire dataset. max X represents the maximum value of the corresponding feature across the entire dataset. norm This is the normalized value.

[0040] Kalman filtering: The state equation and observation equation are as follows:

[0041] in, Let A be the state vectors at time k and time k-1, respectively; let A be the state transition matrix; and let B be the control matrix. This is the control input at time k-1. This represents the process noise at time k-1.

[0042] In this embodiment, in step S2, the correlation between the multi-source data and the health status SOH is calculated based on the Pearson correlation coefficient, specifically by calculating the input feature x. i The correlation coefficient with the SOH label y is calculated using the following expression:

[0043] In the above formula, x i and y i These are variables from multiple data sources. x and y The i-th observation, and They are variables x and y The mean of the values, where n is the number of observations.

[0044] It is understandable that this embodiment proposes a two-level screening mechanism of "hierarchical weighted Pearson + information entropy" in combination with the characteristics of sodium-ion batteries: firstly, the correlation is calculated and weighted in different temperature ranges, rate and low voltage plateaus, and then the information entropy is combined to measure the information content of the features, thereby avoiding the omission of nonlinear or degenerate correlation features by the single correlation method, and in particular, it can better capture key indicators such as low voltage plateau fluctuations and capacity decay rate.

[0045] Based on Pearson correlation screening, information entropy determination and information enhancement are further introduced. This mechanism can identify and retain features that have high information content but no significant linear correlation in health status estimation.

[0046] In this embodiment, the expression for calculating the information entropy of the feature parameters after relevance filtering in step S3 is as follows:

[0047] In the above formula, H( x ) is the information entropy of the random variable X, p( x i ) is a random variable x Values x i The probability; specifically, entropy preservation processing is performed on features with a correlation ≥ 0.7 (such as voltage plateau slope, temperature gradient).

[0048] In this embodiment, information enhancement is performed on characteristic parameters of voltage fluctuation and capacity attenuation types whose information entropy is within a preset entropy value range. The calculation expression is as follows:

[0049] In the above formula, These are the original values ​​of the feature parameters. The mean of the feature parameters, The standard deviation of the characteristic parameter, For the enhanced feature parameter values, The enhancement factor has a value range of 0.1-0.3.

[0050] It is understandable that the above formula first standardizes the features to eliminate the influence of different units; then it utilizes... Scaling and shifting the standardized results amplifies the range of feature fluctuations while introducing slight perturbations, thus increasing sample diversity while ensuring data rationality. By selectively enhancing key features of sodium-ion batteries, such as voltage fluctuations and capacity decay rates in the low-voltage plateau region, these parameters reflect early degradation and temperature sensitivity issues, better aligning with the performance degradation patterns of sodium-ion batteries. Entropy value determination of capacity decay rate and low-voltage plateau voltage fluctuation features compensates for the inadequacy of single correlation screening in reflecting differences in feature information, enabling more refined feature selection in sodium-ion battery applications. The enhancement coefficient α serves as both a scaling factor and a shift factor in the calculation, achieving a dual enhancement effect. Limiting α to the range of 0.1-0.3 effectively avoids model training shift problems caused by excessive perturbations, ensuring that the enhancement results are both differentiated and reasonable.

[0051] In this embodiment, when training the deep learning model in step S4, the training objective is to minimize the loss function constrained by the mixed mean square error and information entropy. The deep learning model's parameters are updated when adjacent temperature changes or voltage fluctuations in the feature parameters exceed the trigger condition. The expression for the loss function L is:

[0052]

[0053]

[0054] In the above formula, It is the actual value. These are predicted values, where n is the sample size and L is the number of samples. entropy For information regular expressions, L MSE The mean squared error is the mixed mean squared error, and λ is the weighting parameter (preferably 0.5). For summation, j is the category index. To belong to category The probability of; The expression for parameter update is:

[0055] In the above formula, For model parameters, The learning rate; The steps for updating parameters are as follows: When the trigger condition is met or When calculating the loss function gradient with respect to parameter θ The parameters are then adjusted according to the learning rate η to obtain new parameters. ,in This represents the temperature change between adjacent samples. This represents the change in voltage between adjacent samples. These are the corresponding temperature and voltage thresholds.

[0056] Specifically, Figure 2 This is a schematic diagram of the BERT network structure. The training data for the BERT network includes at least 10,000 sets of charge-discharge cycle data, covering a temperature range of -20℃ to 60℃. The dynamic update trigger condition is that the average ambient temperature change within the sliding window is ≥5℃, or the instantaneous voltage fluctuation is ≥10% for 10 seconds. . Figure 3 The flowchart illustrates the process of updating weight parameters using the Adam optimizer. Model parameters are updated in real-time via a sliding window strategy (window length T=100, step size s=10, real-time updates of input data), with the Adam optimizer (learning rate η=0.001). Furthermore, the model can be integrated into a battery management system (BMS) for millisecond-level online evaluation. Edge deployment is also possible, pruning the BERT network to 6 layers, reducing the number of parameters by 60%. The deployment platform is an FPGA (Xilinx Zynq UltraScale+), with inference latency ≤20ms.

[0057] In this embodiment, the deep learning model includes an input layer, multiple Transformer encoder layers, and an output layer. The Transformer encoder layers employ a multi-head self-attention mechanism, expressed as:

[0058] In the above formula, Q, K, and V are the query matrix, key matrix, and value matrix, respectively, and d k It is the dimension of the key matrix K. This is the dot product similarity matrix.

[0059] Specifically, the BERT network model's input layer consists of a multi-source data sequence with 100 time steps (dimension: 10×100); the Transformer encoder has 12 layers, each containing an 8-head self-attention mechanism; and the output layer is a fully connected layer (activation function: Sigmoid) used to output SOH prediction values ​​(0%~100%). It is understandable that in the construction of the deep learning model, this embodiment introduces a multi-channel embedding at the input end to separately encode the unique degradation features of sodium-ion batteries and give them higher weight in the attention mechanism. Simultaneously, a capacity decay trend prediction head is added at the output end, achieving joint learning of numerical estimation and trend prediction of the state of health (SOH). By introducing a multi-source feature embedding module at the input layer, voltage, current, temperature, and the unique low-voltage plateau fluctuations and capacity decay rate of sodium-ion batteries are encoded separately and then fused, thus avoiding the weakening of key features. In the attention mechanism, higher weights are assigned to degradation-related features, enabling the model to have stronger representation capabilities under critical operating conditions. Furthermore, a dual-channel structure is designed at the output layer, which not only outputs the SOH value but also predicts the capacity decay trend, achieving joint learning of point estimation and trend estimation. These improvements optimize both feature selection and model modeling for the characteristics of sodium-ion batteries, allowing the model to better capture the degradation characteristics of sodium-ion batteries, thereby improving the accuracy and robustness of SOH estimation.

[0060] The following two specific application examples illustrate the effectiveness of the method in this embodiment.

[0061] Example 1: State of Health (SOH) Assessment of Electric Vehicle Power Batteries 1. Experimental setup Battery pack: 100 sodium-ion batteries (model SN-21700, nominal capacity 3.5Ah).

[0062] Environment: Temperature cycling test (-10℃→25℃→50℃, 2 hours per cycle).

[0063] Data acquisition: voltage (1kHz), current (1kHz), temperature (10Hz).

[0064] 2. Implementation Steps Step 1: Collect 10,000 sets of data and remove outliers (voltage <2.0V or >3.8V).

[0065] Step 2: Calculate the Pearson correlation coefficient, retaining the voltage plateau slope (PCC=0.85) and temperature gradient (PCC=0.72).

[0066] Step 3: Train the BERT model (epoch=200, batch_size=32), and the loss function converges to 0.023.

[0067] Step 4: Deploy to the vehicle-mounted BMS to predict SOH in real time. The update is triggered when the temperature change is ≥5℃.

[0068] 3. Results Prediction accuracy: MAE=2.5% (MAE=5.6% for traditional LSTM model).

[0069] Latency: Inference time ≤ 15ms, meeting real-time requirements.

[0070] Example 2: SOH monitoring of power grid energy storage system 1. Experimental setup System size: 1MW / 4MWh sodium-ion battery pack (4000 cells).

[0071] Sampling frequency: voltage / current (1kHz), temperature (10Hz).

[0072] 2. Implementation Steps Step 1: Distributed data acquisition, with independent preprocessing for each battery module.

[0073] Step 2: Global feature fusion, with the entropy preservation threshold set to the information entropy value. H ( X ≥2.8 (unit: bits) Only retain feature parameters whose information entropy meets the threshold (such as low voltage platform fluctuation amplitude, cycle capacity decay rate, temperature gradient, etc.), and remove redundant features with low information content.

[0074] Step 3: Lightweight the model (compress the number of parameters to 50MB) and deploy it to the FPGA.

[0075] Step 4: Dynamically update the frequency to 10Hz, supporting second-level frequency adjustment of the power grid.

[0076] 3. Results Extreme operating conditions: MAE=3.8% at -20℃, which is better than the traditional model (MAE=7.2%).

[0077] Economic benefits: Battery replacement cycle extended by 25%, maintenance costs reduced by 18%.

[0078] Table 1

[0079] Table 1 shows the mean absolute error (MAE) of the SOH prediction accuracy of the method of the present invention and the traditional LSTM method at room temperature and low temperature. It can be seen that the method of the present invention has smaller error and higher prediction accuracy.

[0080] The present invention further provides a sodium-ion battery health status assessment system based on multi-source data fusion, including a microprocessor and a memory interconnected thereto, wherein the microprocessor is programmed or configured to execute the steps of the sodium-ion battery health status assessment method based on multi-source data fusion.

[0081] The present invention further provides a computer-readable storage medium storing a computer program / instructions, the computer program / instructions being programmed or configured to execute, via a processor, the steps of a sodium-ion battery health status assessment method based on multi-source data fusion.

[0082] The system and medium of the present invention, corresponding to the methods described above, also have the advantages described above.

[0083] The present invention can implement all or part of the processes in the methods of the above embodiments, or it can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the above method embodiments. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. Computer-readable media include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. The memory is used to store computer programs and / or modules. The processor implements various functions by running or executing the computer programs and / or modules stored in the memory, and by calling data stored in the memory. The memory may include high-speed random access memory, as well as non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital (SD) cards, flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0084] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for assessing the health status of sodium-ion batteries based on multi-source data fusion, characterized in that, Including the following steps: S1, continuously collect the operating status parameters and environmental parameters of the sodium-ion battery as multi-source data and perform data preprocessing; S2, calculate the correlation between the multi-source data and the health status SOH based on the Pearson correlation coefficient, and filter out several feature parameters in the multi-source data whose correlation is greater than a first preset threshold; S3, calculate the information entropy of the feature parameters after correlation screening, retain the feature parameters whose information entropy value is greater than the second preset threshold, and perform information enhancement on the voltage fluctuation type and capacity decay type feature parameters whose information entropy is within the preset entropy value range; S4. Construct a deep learning model based on the BERT network, and train the deep learning model using the feature parameters processed in step S3 as the training set data to obtain a battery health status assessment model. S5 uses a trained battery health status assessment model to estimate the health status of sodium-ion batteries.

2. The method for assessing the health status of sodium-ion batteries based on multi-source data fusion according to claim 1, characterized in that, In step S2, the correlation between the multi-source data and the health status SOH is calculated based on the Pearson correlation coefficient. The calculation expression is as follows: In the above formula, x i and y i These are variables from multiple data sources. x and y The i-th observation, and They are variables x and y The mean of the values, where n is the number of observations.

3. The method for assessing the health status of sodium-ion batteries based on multi-source data fusion according to claim 1, characterized in that, In step S3, the expression for calculating the information entropy of the feature parameters after relevance filtering is as follows: In the above formula, H( x ) is the information entropy of the random variable X, p( x i ) is a random variable x Values x i The probability, x i and y i These are variables from multiple data sources. x and y The i-th observation, where n is the number of observations.

4. The method for assessing the health status of sodium-ion batteries based on multi-source data fusion according to claim 3, characterized in that, Information enhancement is performed on voltage fluctuation and capacity decay feature parameters whose information entropy is within a preset entropy range. The calculation expression is as follows: In the above formula, These are the original values ​​of the feature parameters. The mean of the feature parameters, The standard deviation of the characteristic parameter, For the enhanced feature parameter values, This is the enhancement coefficient.

5. The method for assessing the health status of sodium-ion batteries based on multi-source data fusion according to claim 3, characterized in that, Step S4: When training the deep learning model, the training objective is to minimize the loss function constrained by the mixed mean square error and information entropy. The deep learning model's parameters are updated when adjacent temperature changes or voltage fluctuations in the feature parameters exceed the trigger condition. The expression for the loss function L is: In the above formula, It is the actual value. These are predicted values, where n is the sample size and L is the number of samples. entropy For information regular expressions, L MSE Let λ be the mixed mean square error, and λ be the weighting parameter. For summation, j is the category index. To belong to category The probability of; The expression for parameter update is: In the above formula, For model parameters, The learning rate; The steps for updating parameters are as follows: When the trigger condition is met or When calculating the loss function gradient with respect to parameter θ The parameters are then adjusted according to the learning rate η to obtain new parameters. , where Δ Δ represents the temperature change between adjacent samples. This represents the change in voltage between adjacent samples. , These are the corresponding temperature and voltage thresholds.

6. The method for assessing the health status of sodium-ion batteries based on multi-source data fusion according to any one of claims 1-5, characterized in that, Step S1, data preprocessing, specifically includes: Outlier data points exceeding a preset range in multi-source data are removed. The preset range is a range extending upwards and downwards from the mean by a certain number of standard deviations. Normalization and Kalman filtering are performed on the multi-source data after removing outlier data points.

7. The method for assessing the health status of sodium-ion batteries based on multi-source data fusion according to claim 6, characterized in that, Step S1, during data preprocessing, also includes using the voltage fluctuation range of the low-voltage plateau region as a reference for data removal, based on the operating characteristics of sodium-ion batteries, and using a threshold for the cycle capacity decay rate, based on the actual degradation pattern of sodium-ion batteries, to assist in identifying abnormal data. The expression for the voltage fluctuation amplitude in the low-voltage plateau region is: In the above formula, This represents the maximum voltage change over time within the low-voltage plateau region. This represents the minimum voltage change over time within the low-voltage plateau region. The expression for the cycle capacity decay rate is: In the above formula, C k Let ΔN be the capacity of the k-th iteration, and ΔN be the number of iteration intervals. This represents the capacity decay.

8. The method for assessing the health status of sodium-ion batteries based on multi-source data fusion according to any one of claims 1-5, characterized in that, The deep learning model includes an input layer, multiple Transformer encoder layers, and an output layer. The Transformer encoder layers employ a multi-head self-attention mechanism, expressed as follows: In the above formula, Q, K, and V are the query matrix, key matrix, and value matrix, respectively, and d k It is the dimension of the key matrix K. This is the dot product similarity matrix.

9. A sodium-ion battery health status assessment system based on multi-source data fusion, comprising a microprocessor and a memory interconnected, characterized in that, The microprocessor is programmed or configured to perform the steps of the sodium-ion battery health status assessment method based on multi-source data fusion as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program / instructions, characterized in that, The computer program / instructions are programmed or configured to execute the steps of the sodium-ion battery health status assessment method based on multi-source data fusion as described in any one of claims 1 to 8 via a processor.

Images