Sodium-ion battery Pack health state online evaluation method and system

Abstract

The invention discloses a sodium-ion battery pack health state online evaluation method and system, and the method comprises the following steps: S1, constructing a sodium-ion battery aging model, collecting and preprocessing pack operation data, and generating a data sequence; s2, calculating a feature entropy value, adjusting the state of the liquid neural network, and extracting a fusion feature sequence; s3, performing path signature transformation on the fusion feature and the simulation sequence, calculating a structural deviation, and generating a deviation sequence; s4, updating model parameters by using particle filtering, and reasoning a capacity estimation value to form a capacity sequence; s5, performing Loess fitting and smooth filtering on the capacity sequence to generate a capacity estimation curve; and S6, evaluating the Pack health state according to the capacity estimation curve, and outputting an online result. According to the method, digital twin modeling, a liquid neural network and the like are fused, and the method has the advantages of high evaluation precision, high response speed and high adaptability.

Description

Technical Field

[0001] This invention relates to the field of battery management technology, and in particular to an online health status assessment method and system for sodium-ion battery packs. Background Technology

[0002] Sodium-ion batteries, as a novel energy storage battery system that has attracted much attention in recent years, are gradually becoming one of the potential alternatives to lithium batteries due to their abundant resources, low raw material costs, and strong adaptability to high temperatures. In fields such as power storage and low-speed electric vehicles, the practical application scenarios of sodium-ion battery pack systems are constantly expanding, and the requirements for their operational reliability and safety are also increasing. Against this backdrop, accurately assessing the state of health (SOH) of sodium-ion battery packs has become a key factor affecting system operation and maintenance and lifespan management.

[0003] In existing technologies, battery health status assessment methods are mostly based on empirical models or electrochemical modeling structures of lithium-ion batteries. Common methods include static internal resistance inference, capacity curve comparison, and feature fitting regression. These methods have certain effects under experimental conditions, but they have the following shortcomings in practical applications: First, most traditional models are based on individual cells, making it difficult to effectively reflect the impact of inconsistencies between cells at the pack level on the overall health status. Second, lithium-ion batteries and sodium-ion batteries have significant differences in aging mechanisms, capacity decay characteristics, and electrochemical responses. Directly transferring lithium battery models to sodium battery scenarios results in low assessment accuracy. Third, most existing methods use static models or offline assessment methods, making it difficult to achieve real-time dynamic monitoring during operation and lacking a model update mechanism that synchronizes with actual operating conditions.

[0004] In addition, a few studies have attempted to introduce data-driven models such as neural networks and LSTM for SOH estimation, but there are still problems such as weak generalization ability, lagging feature response and difficulty in expressing high-dimensional data structures. Especially in the scenario where the operating characteristics of sodium-ion batteries exhibit nonlinear fluctuations and long-period changes, the existing network structure is difficult to adapt to the dual challenges of temporal dynamics and structural perturbations.

[0005] Therefore, how to provide an online health status assessment method and system for sodium-ion battery packs is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an online health status assessment method and system for sodium-ion battery packs. This invention integrates digital twin modeling, liquid neural networks, path signature transformation, and particle filtering algorithms. It constructs a dedicated dynamic model for the aging characteristics of sodium batteries, extracts multi-dimensional sensor features in real time during operation, and adaptively corrects the SOH estimation results. The invention details the entire process of achieving accurate pack-level health status assessment under non-stationary time-series operating conditions, and possesses the advantages of high assessment accuracy, fast response speed, and strong adaptability.

[0007] An online health status assessment method for a sodium-ion battery pack according to an embodiment of the present invention includes the following steps: S1. Construct a battery aging model based on a preset sodium battery aging mechanism, collect multidimensional operating data of the sodium-ion battery pack at a preset frequency and preprocess it to generate an operating data sequence. S2. Extract the feature entropy values ​​of each time step of the running data sequence. In the liquid neural network, dynamically adjust the neural state of the liquid neural network according to the feature entropy values, and perform feature extraction and fusion operations on the running data sequence to obtain the fused feature sequence. S3. Extract the simulated fusion sequence generated in the battery aging model, perform path signature transformation operation on the simulated fusion sequence and the fusion feature sequence respectively, and perform structural error analysis operation on the two transformation results to calculate the structural deviation value and construct the deviation sequence. S4. Using the particle filter algorithm, the parameters of the battery aging model are adaptively updated according to the deviation sequence. The updated battery aging model is then used to perform inference calculations at each time step to obtain the corresponding capacity estimates, forming a capacity estimation sequence. S5. Use the Loess algorithm to perform curve fitting on the capacity estimation sequence, and perform smoothing filtering on the fitting results to generate the capacity estimation curve. S6. Perform a health status assessment on the sodium-ion battery pack based on the capacity estimation curve and generate online health status assessment results.

[0008] Optionally, the sodium battery aging model is a digital twin model used to simulate the operating state of a sodium-ion battery pack. The multidimensional operating data includes the voltage, current, temperature, and environmental parameters of each cell in the sodium-ion battery pack. The preprocessing includes missing data filling, time alignment, and anomaly removal operations. The neural state represents the internal response vector of the liquid neural network that evolves in each time step based on the operating data sequence and the corresponding feature entropy value.

[0009] Optionally, the internal calculation process of the battery aging model specifically includes: The cumulative usage time, cumulative charge-discharge cycles, historical highest operating temperature, and average discharge rate of the sodium-ion battery pack were collected and normalized to linearly map the original values ​​to the [0,1] interval. A battery feature vector was constructed, which included usage time factor, temperature factor, cycle factor, and rate factor. The usage time factor represents the ratio of cumulative usage time to rated service life. The temperature factor represents the ratio of historical highest operating temperature to the rated safe operating temperature set by the manufacturer. The cycle factor represents the ratio of cumulative charge-discharge cycles to the battery's designed cycle life. The rate factor represents the ratio of average discharge rate to standard rated rate. The total simulation cycle length is set according to the duration factor in the battery feature vector, and the capacity decay coefficient for each simulation time step is set according to the cycle factor. In the first time step, the sodium-ion battery pack is set to have the battery capacity as the simulated capacity value of the previous time step. At each time step, the simulated capacity value from the previous time step is used as the starting capacity, and capacity decay calculation is performed, specifically including: The temperature adjustment amount is calculated by multiplying the current temperature factor by the set temperature sensitivity coefficient. The product of the current multiplier factor and the cycle factor is multiplied by a preset multiplier sensitivity coefficient to calculate the multiplier effect. According to the preset weight ratio, the temperature adjustment amount and the multiplier effect amount are weighted and superimposed. The superimposed result is multiplied by the preset capacity decay coefficient to obtain the capacity decay value at the current time step. The capacity decay value is then subtracted from the initial capacity to obtain the current simulated capacity value. Arrange all simulated capacity values ​​in chronological order to construct a simulated capacity sequence; Within each time step, based on the battery feature vector and the current time step index, the simulated voltage, simulated current and simulated temperature values ​​for the current time step are deduced and then spliced ​​with the real-time acquired environmental parameters to form a simulated running sequence. By performing feature extraction and fusion operations on the simulated running sequence using a liquid neural network, a simulated fused sequence that is consistent with the fused feature sequence in terms of time dimension, feature dimension, and arrangement order is obtained.

[0010] Optionally, S2 specifically includes: S21. Expand the running data sequence according to time steps. For the voltage, current, temperature and environmental parameters of each cell in each time step, calculate the maximum value, minimum value, mean value and rate of change of the four types of data in the sliding window, and form the feature statistics set of the current time step. S22. Based on the feature statistics set for each time step, calculate the feature entropy value for that time step, arrange all feature entropy values ​​to form a feature entropy sequence, and align it with the running data sequence in time. The calculation process of the feature entropy value specifically includes: The feature statistics set of the current time step is divided into voltage subset, current subset, temperature subset and environmental parameter subset according to data category. Min-Max normalization operation is performed on the maximum value, minimum value, mean value and rate of change within each subset. Within each subset, based on the normalized feature values ​​of the corresponding feature in the current sliding window, the feature values ​​are divided into multiple continuous and non-overlapping value range intervals according to the preset interval division rules. The frequency of each normalized feature value falling into each value range interval within the current sliding window is counted as the interval frequency. The interval frequency is then divided by the number of normalized feature values ​​within the sliding window to obtain the normalized probability value corresponding to each value range interval. Combine the normalized probability values ​​corresponding to all value ranges into a probability distribution vector of the current feature within the sliding window; Based on the probability distribution vector of each normalized feature, calculate the local entropy value corresponding to each subset, and then sum the local entropy values ​​of the four subsets according to the preset weight coefficient set to obtain the feature entropy value of the current time step. S23. Input the time alignment result into the liquid neural network. In each time step, dynamically adjust the neural state of the liquid neural network according to the feature entropy value and update the internal response vector of that time step. S24. Based on the internal response vector and running data at each time step, perform feature enhancement operations in the liquid neural network, extract the fused feature vector, and generate the fused feature sequence in chronological order.

[0011] Optionally, S23 specifically includes: S231. In each time step, extract the feature entropy value of the current time step, and match the feature entropy value with the preset adjustment function to calculate the adjustment weight; S232. Based on the adjustment weights, the time constants of each neuron in the liquid neural network are dynamically scaled, and the product of the original time constant and the adjustment weights is used as the updated time constant for the current time step. S233. Based on the updated time constant, the running data of the current time step and the internal response vector of the previous time step are weighted and fused to generate the internal response vector of the current time step. S234. Arrange the internal response vectors at all time steps in chronological order to construct an internal response sequence that is consistent with the time dimension of the running data sequence, while maintaining consistency in the feature dimensions.

[0012] Optionally, S3 specifically includes: S31. Based on the time dimension of the fusion feature sequence, read the simulated fusion sequence of the corresponding time step in the battery aging model; S32. Expand the fused feature sequence and the simulated fused sequence according to time steps, and construct the corresponding path sequences according to a fixed arrangement order; S33. Take the path vector difference between adjacent time steps in each path sequence as the path increment, and gradually accumulate the path increment according to the time order to generate the path representation. S34. At each time step, perform a path signature transformation operation on the path representation, extract the path integral values ​​at different orders, and form a fused signature sequence and a simulated signature sequence. S35. Calculate the difference between the two path integral values ​​at the corresponding time steps in the fused signature sequence and the simulated signature sequence, sum the absolute values ​​of the differences in each dimension, and perform Z-Score standardization to obtain the structural deviation value at the current time step. S36. Arrange all structural deviation values ​​in chronological order to construct a deviation sequence.

[0013] Optionally, S34 specifically includes: S341. In each time step, according to the order of path increments in the fused path representation and the simulated path representation, the current time step is taken as the cutoff point, and the path increment sequence between the first time step and the current time step is extracted as the path representation segment of the current time step. S342. Perform multi-order path integral calculation operation on the path representation segment, calculate the first-order, second-order to the preset highest-order path integral values ​​in sequence, and construct the path integral set corresponding to the current time step. The path integral value of each order is calculated by accumulating the multi-dimensional feature quantities in the path increment sequence in a fixed order. S343. Perform feature mapping operation on the path integral set, and arrange the path integral values ​​of different orders according to the feature type by using dimensional expansion to generate a path signature vector that matches the path representation. S344. Arrange the path signature vectors generated at all time steps in chronological order, and form a fused signature sequence and a simulated signature sequence according to their source.

[0014] Optionally, S4 specifically includes: S41. Set the capacity decay coefficient, temperature sensitivity coefficient and rate sensitivity coefficient in the battery aging model as the set of parameters to be estimated by the particle filter algorithm, and use the fusion feature vector of the corresponding time step in the fusion feature sequence as the state driver to initialize the particle set containing the weights of each particle. S42. In each time step, based on the particle set of the previous time step, the state prediction update operation is performed on the parameter set to be estimated. The prediction result is substituted into the battery aging model to calculate the predicted capacity value of the current time step. The predicted capacity value is matched with the structural deviation value of the corresponding time step in the deviation sequence to construct the observation error of the current time step. S43. Based on the observation error at the current time step, perform a weight update operation on each particle in the particle set. The smaller the observation error, the higher the weight of the particle. Then, perform Min-Max normalization on the updated weights of all particles to generate an updated particle set, specifically including: Z-Score normalization is performed on the observation error of each particle to calculate the error score. The lower the error score, the closer the particle is to the center of the overall error distribution. A mapping operation is performed based on the error score value. A monotonically decreasing preset weight function is used to map the error score value to non-negative weights, and an initial weight set is constructed for the current time step. The initial weight set is subjected to Min-Max normalization, which linearly compresses the weights of all particles to a set range, forming a normalized weight set. Map the normalized weight set back to the particle set and update the weights of all particles at the current time step; S44. Perform a resampling operation on the updated particle set, remove particles with weights lower than a preset threshold, sort each particle according to its weight, select the particles with the highest weights by a set percentage and perform a resampling copy operation to form a resampled particle set for the current time step. S45. Use the mean value of the particle parameters in the resampled particle set as the update parameter of the battery aging model at the current time step, and substitute the update parameter into the battery aging model to perform capacity inference calculation at the current time step to obtain the corresponding capacity estimate. S46. Arrange the capacity estimates obtained at each time step in chronological order to construct a capacity estimation sequence.

[0015] Optionally, S5 specifically includes: S51. Set a fixed-length sliding window for the capacity estimation sequence according to the time step, and use the weighted multinomial regression method to locally fit the capacity estimation value in each sliding window to generate the local fitting value for the corresponding time step. S52. Arrange the local fitted values ​​at all time steps in the order of time steps to construct the initial capacity curve, and perform a continuity detection operation on the initial capacity curve to identify abrupt change points. S53. Perform curve splicing and adjustment operations on the local fitted values ​​before and after the mutation point, and perform Savitzky-Golay filtering on the adjustment results to generate intermediate capacity curves, specifically including: Symmetrical splicing intervals are set on both sides of the mutation point, and local fitting values ​​within the corresponding time period are extracted to construct fitting curve segments; Smoothing weights are set based on the trend of slope change and numerical difference at the endpoints of the fitted curve segments, and a weighted superposition operation is performed on the local fitted values ​​within the splicing interval to generate a smooth transition curve. The smooth transition curve and the fitted curve segment are spliced ​​together in chronological order to construct the corrected initial capacity curve as the adjustment result; Based on the adjustment results, the sliding window length and polynomial order are set, and the Savitzky-Golay filtering algorithm is used to smooth the adjustment results and extract the filter value at each time step. Arrange all filter values ​​in chronological order to generate an intermediate capacity curve; S54. Perform monotonicity discrimination processing on the intermediate capacity curve. If an abnormal interval where the local fluctuation amplitude exceeds the set threshold is found, perform curve segment replacement operation to form a capacity estimation curve.

[0016] An online health status assessment system for a sodium-ion battery pack according to an embodiment of the present invention includes: The data acquisition module is used to construct a battery aging model based on a preset sodium battery aging mechanism, collect multi-dimensional operating data of the sodium-ion battery pack at a preset frequency and perform preprocessing to generate an operating data sequence. The feature fusion module is used to extract the feature entropy values ​​of each time step of the running data sequence. In the liquid neural network, the neural state of the liquid neural network is dynamically adjusted according to the feature entropy values, and feature extraction and fusion operations are performed on the running data sequence to obtain the fused feature sequence. The deviation analysis module is used to extract the simulated fusion sequence generated in the battery aging model, perform path signature transformation operations on the simulated fusion sequence and the fusion feature sequence respectively, and perform structural error analysis operations on the two transformation results to calculate the structural deviation value and construct the deviation sequence. The capacity estimation module is used to perform adaptive parameter update operations on the battery aging model based on the deviation sequence using the particle filter algorithm, and to perform inference calculation operations on each time step using the updated battery aging model to obtain the corresponding capacity estimate value and form a capacity estimation sequence. The curve fitting module is used to perform curve fitting on the capacity estimation sequence using the Loess algorithm, and to perform smoothing filtering on the fitting results to generate the capacity estimation curve. The status assessment module is used to perform a health status assessment operation on the sodium-ion battery pack based on the capacity estimation curve and generate online health status assessment results.

[0017] The beneficial effects of this invention are: First, this invention avoids the problem of insufficient evaluation accuracy of traditional lithium battery empirical models in sodium battery scenarios by constructing a digital twin model for the aging characteristics of sodium-ion batteries. This model integrates factors unique to sodium batteries, such as cycle degradation, rate response, and temperature sensitivity, and can more realistically simulate the health evolution process of the pack under different operating conditions, effectively improving the physical consistency and engineering applicability of health status estimation.

[0018] Secondly, this invention introduces a dynamic feature extraction mechanism that combines liquid neural networks with path signature transformation. This mechanism can extract the nonlinear correlation structure between multidimensional sensor features under continuously changing operating data conditions. It also improves the modeling ability for inconsistencies within the Pack level by adjusting the neural state through feature entropy. At the same time, this invention uses path signature transformation to extract temporal structural differences and combines it with particle filtering algorithm to achieve online adaptive updating of the aging model, ensuring that the SOH estimation results still have high stability and real-time performance in long-term operation.

[0019] Finally, this invention integrates Loess fitting and Savitzky-Golay filtering algorithms in the capacity estimation stage to achieve high-precision smoothing of the SOH curve, effectively suppressing the impact of sensor noise and abnormal fluctuations. The overall method possesses advantages such as online availability, model adaptability, and high evaluation accuracy, making it suitable for health status monitoring and maintenance scheduling of sodium-ion battery packs under actual operating conditions, providing reliable data support for subsequent battery life management and safety control. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of an online health status assessment method for a sodium-ion battery pack proposed in this invention; Figure 2 This is a flowchart of the path signature transformation and particle filter update process for an online health status assessment method for a sodium-ion battery pack proposed in this invention. Figure 3 This is a module structure diagram of an online health status assessment system for a sodium-ion battery pack proposed in this invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0022] refer to Figures 1-2An online health status assessment method for a sodium-ion battery pack includes the following steps: S1. Construct a battery aging model based on a preset sodium battery aging mechanism, collect multidimensional operating data of the sodium-ion battery pack at a preset frequency and preprocess it to generate an operating data sequence. S2. Extract the feature entropy values ​​of each time step of the running data sequence. In the liquid neural network, dynamically adjust the neural state of the liquid neural network according to the feature entropy values, and perform feature extraction and fusion operations on the running data sequence to obtain the fused feature sequence. S3. Extract the simulated fusion sequence generated in the battery aging model, perform path signature transformation operation on the simulated fusion sequence and the fusion feature sequence respectively, and perform structural error analysis operation on the two transformation results to calculate the structural deviation value and construct the deviation sequence. S4. Using the particle filter algorithm, the parameters of the battery aging model are adaptively updated according to the deviation sequence. The updated battery aging model is then used to perform inference calculations at each time step to obtain the corresponding capacity estimates, forming a capacity estimation sequence. S5. Use the Loess algorithm to perform curve fitting on the capacity estimation sequence, and perform smoothing filtering on the fitting results to generate the capacity estimation curve. S6. Perform a health status assessment on the sodium-ion battery pack based on the capacity estimation curve and generate online health status assessment results.

[0023] In this embodiment, the sodium battery aging model is a digital twin model used to simulate the operating state of a sodium-ion battery pack. The multidimensional operating data includes the voltage, current, temperature and environmental parameters of each cell in the sodium-ion battery pack. The preprocessing includes missing data filling, time alignment and anomaly removal operations. The neural state representation is the internal response vector of the liquid neural network that evolves according to the operating data sequence and the corresponding feature entropy value at each time step.

[0024] In this embodiment, the internal calculation process of the battery aging model specifically includes: The cumulative usage time, cumulative charge-discharge cycles, historical highest operating temperature, and average discharge rate of the sodium-ion battery pack were collected and normalized to linearly map the original values ​​to the [0,1] interval. A battery feature vector containing usage time factor, temperature factor, cycle factor, and rate factor was constructed. The usage time factor represents the ratio of cumulative usage time to rated lifespan, the temperature factor represents the ratio of historical highest operating temperature to the manufacturer's rated safe operating temperature, the cycle factor represents the ratio of cumulative charge-discharge cycles to the battery's designed cycle life, and the rate factor represents the ratio of average discharge rate to standard rated rate. The total simulation cycle length is set according to the duration factor in the battery feature vector, and the capacity decay coefficient for each simulation time step is set according to the cycle factor. In the first time step, the sodium-ion battery pack is set to have the battery capacity as the simulated capacity value of the previous time step. At each time step, the simulated capacity value from the previous time step is used as the starting capacity, and capacity decay calculation is performed, specifically including: The temperature adjustment amount is calculated by multiplying the current temperature factor by the set temperature sensitivity coefficient. The product of the current multiplier factor and the cycle factor is multiplied by a preset multiplier sensitivity coefficient to calculate the multiplier effect. According to the preset weight ratio, the temperature adjustment amount and the multiplier effect amount are weighted and superimposed. The superimposed result is multiplied by the preset capacity decay coefficient to obtain the capacity decay value at the current time step. The capacity decay value is then subtracted from the initial capacity to obtain the current simulated capacity value. Arrange all simulated capacity values ​​in chronological order to construct a simulated capacity sequence; Within each time step, based on the battery feature vector and the current time step index, the simulated voltage, simulated current and simulated temperature values ​​for the current time step are deduced and then spliced ​​with the real-time acquired environmental parameters to form a simulated running sequence. By performing feature extraction and fusion operations on the simulated running sequence using a liquid neural network, a simulated fused sequence that is consistent with the fused feature sequence in terms of time dimension, feature dimension, and arrangement order is obtained.

[0025] In this embodiment, S2 specifically includes: S21. Expand the running data sequence according to time steps. For the voltage, current, temperature and environmental parameters of each cell in each time step, calculate the maximum value, minimum value, mean value and rate of change of the four types of data in the sliding window, and form the feature statistics set of the current time step. S22. Based on the feature statistics set at each time step, calculate the feature entropy value for that time step, arrange all feature entropy values ​​to form a feature entropy sequence, and align it with the running data sequence in time. The specific calculation process of the feature entropy value includes: The feature statistics set of the current time step is divided into voltage subset, current subset, temperature subset and environmental parameter subset according to data category. Min-Max normalization operation is performed on the maximum value, minimum value, mean value and rate of change within each subset. Within each subset, based on the normalized feature values ​​of the corresponding feature in the current sliding window, the feature values ​​are divided into multiple continuous and non-overlapping value range intervals according to the preset interval division rules. The frequency of each normalized feature value falling into each value range interval within the current sliding window is counted as the interval frequency. The interval frequency is then divided by the number of normalized feature values ​​within the sliding window to obtain the normalized probability value corresponding to each value range interval. Combine the normalized probability values ​​corresponding to all value ranges into a probability distribution vector of the current feature within the sliding window; Based on the probability distribution vector of each normalized feature, calculate the local entropy value corresponding to each subset, and then sum the local entropy values ​​of the four subsets according to the preset weight coefficient set to obtain the feature entropy value of the current time step. S23. Input the time alignment result into the liquid neural network. In each time step, dynamically adjust the neural state of the liquid neural network according to the feature entropy value and update the internal response vector of that time step. S24. Based on the internal response vector and running data at each time step, perform feature enhancement operations in the liquid neural network, extract the fused feature vector, and generate the fused feature sequence in chronological order.

[0026] In this embodiment, S23 specifically includes: S231. In each time step, extract the feature entropy value of the current time step, and match the feature entropy value with the preset adjustment function to calculate the adjustment weight; S232. Based on the adjustment weights, the time constants of each neuron in the liquid neural network are dynamically scaled, and the product of the original time constant and the adjustment weights is used as the updated time constant for the current time step. S233. Based on the updated time constant, the running data of the current time step and the internal response vector of the previous time step are weighted and fused to generate the internal response vector of the current time step. S234. Arrange the internal response vectors at all time steps in chronological order to construct an internal response sequence that is consistent with the time dimension of the running data sequence, while maintaining consistency in the feature dimensions.

[0027] In this embodiment, S3 specifically includes: S31. Based on the time dimension of the fusion feature sequence, read the simulated fusion sequence of the corresponding time step in the battery aging model; S32. Expand the fused feature sequence and the simulated fused sequence according to time steps, and construct the corresponding path sequences according to a fixed arrangement order; S33. Take the path vector difference between adjacent time steps in each path sequence as the path increment, and gradually accumulate the path increment according to the time order to generate the path representation. S34. At each time step, perform a path signature transformation operation on the path representation, extract the path integral values ​​at different orders, and form a fused signature sequence and a simulated signature sequence. S35. Calculate the difference between the two path integral values ​​at the corresponding time steps in the fused signature sequence and the simulated signature sequence, sum the absolute values ​​of the differences in each dimension, and perform Z-Score standardization to obtain the structural deviation value at the current time step. S36. Arrange all structural deviation values ​​in chronological order to construct a deviation sequence.

[0028] In this embodiment, S34 specifically includes: S341. In each time step, according to the order of path increments in the fused path representation and the simulated path representation, the current time step is taken as the cutoff point, and the path increment sequence between the first time step and the current time step is extracted as the path representation segment of the current time step. S342. Perform multi-order path integral calculation operation on the path representation segment, calculate the first-order, second-order to the preset highest-order path integral values ​​in sequence, and construct the path integral set corresponding to the current time step. The path integral value of each order is calculated by accumulating the multi-dimensional feature quantities in the path increment sequence in a fixed order. S343. Perform feature mapping operation on the path integral set, and arrange the path integral values ​​of different orders according to the feature type by using dimensional expansion to generate a path signature vector that matches the path representation. S344. Arrange the path signature vectors generated at all time steps in chronological order, and form a fused signature sequence and a simulated signature sequence according to their source.

[0029] In this embodiment, S4 specifically includes: S41. Set the capacity decay coefficient, temperature sensitivity coefficient and rate sensitivity coefficient in the battery aging model as the set of parameters to be estimated by the particle filter algorithm, and use the fusion feature vector of the corresponding time step in the fusion feature sequence as the state driver to initialize the particle set containing the weights of each particle. S42. In each time step, based on the particle set of the previous time step, the state prediction update operation is performed on the parameter set to be estimated. The prediction result is substituted into the battery aging model to calculate the predicted capacity value of the current time step. The predicted capacity value is matched with the structural deviation value of the corresponding time step in the deviation sequence to construct the observation error of the current time step. S43. Based on the observation error at the current time step, perform a weight update operation on each particle in the particle set. The smaller the observation error, the higher the weight of the particle. Then, perform Min-Max normalization on the updated weights of all particles to generate an updated particle set, specifically including: For each particle's observation error, Z-Score normalization is performed to calculate the error score. The lower the error score, the closer the particle is to the center of the overall error distribution. A mapping operation is performed based on the error score value. A monotonically decreasing preset weight function is used to map the error score value to non-negative weights, and an initial weight set is constructed for the current time step. The initial weight set is subjected to Min-Max normalization, which linearly compresses the weights of all particles to a set range, forming a normalized weight set. Map the normalized weight set back to the particle set and update the weights of all particles at the current time step; S44. Perform a resampling operation on the updated particle set, remove particles with weights lower than a preset threshold, sort each particle according to its weight, select the particles with the highest weights by a set percentage and perform a resampling copy operation to form a resampled particle set for the current time step. S45. Use the mean value of the particle parameters in the resampled particle set as the update parameter of the battery aging model at the current time step, and substitute the update parameter into the battery aging model to perform capacity inference calculation at the current time step to obtain the corresponding capacity estimate. S46. Arrange the capacity estimates obtained at each time step in chronological order to construct a capacity estimation sequence.

[0030] In this embodiment, the state prediction update operation specifically includes: In each time step, a state transition operation is performed on the parameters of each particle in the particle set of the previous time step, and the parameter prediction value of the current time step is generated according to the set state evolution function. A pseudo-random perturbation term is introduced into the parameter prediction values ​​to construct a set of prediction parameters after the perturbation is added. The perturbation term follows a set distribution and is used to simulate the impact of environmental fluctuations on model parameters. Each prediction parameter vector in the prediction parameter set is input into the battery aging model, and the capacity prediction calculation is performed by combining the corresponding feature values ​​in the fused feature vector to generate a set of predicted capacity values ​​for all particles at the current time step.

[0031] In this embodiment, S5 specifically includes: S51. Set a fixed-length sliding window for the capacity estimation sequence according to the time step, and use a weighted multinomial regression method to locally fit the capacity estimate within each sliding window to generate the local fitted value for the corresponding time step, specifically including: Set the length of the sliding window and divide the sliding window into time steps in the capacity estimation sequence; Within each sliding window, a set of data pairs is constructed with the center time step as the fitting time step, using the time step index as the independent variable and the capacity estimate as the dependent variable. For each data set, a weighted polynomial regression fitting operation is performed. The weighting coefficients for each capacity estimate are set according to the distance of the time step position from the center of the window. The weighted least squares method is used to fit a polynomial function of a fixed order. Use the function value of the polynomial function at the current time step as the local fitted value; S52. Arrange the local fitted values ​​at all time steps in the order of time steps to construct the initial capacity curve, and perform a continuity detection operation on the initial capacity curve to identify abrupt change points. S53. Perform curve splicing and adjustment operations on the local fitted values ​​before and after the mutation point, and perform Savitzky-Golay filtering on the adjustment results to generate intermediate capacity curves, specifically including: Symmetrical splicing intervals are set on both sides of the mutation point, and local fitting values ​​within the corresponding time period are extracted to construct fitting curve segments; Smoothing weights are set based on the trend of slope change and numerical difference at the endpoints of the fitted curve segments, and a weighted superposition operation is performed on the local fitted values ​​within the splicing interval to generate a smooth transition curve. The smooth transition curve and the fitted curve segment are spliced ​​together in chronological order to construct the corrected initial capacity curve as the adjustment result; Based on the adjustment results, the sliding window length and polynomial order are set, and the Savitzky-Golay filtering algorithm is used to smooth the adjustment results and extract the filter value at each time step. Arrange all filter values ​​in chronological order to generate an intermediate capacity curve; S54. Perform monotonicity discrimination processing on the intermediate capacity curve. If an abnormal interval where the local fluctuation amplitude exceeds the set threshold is found, perform curve segment replacement operation to form a capacity estimation curve.

[0032] In this embodiment, the process of generating online health status assessment results specifically includes: Based on the capacity estimation curve, extract the capacity value, capacity decline rate and local fluctuation amplitude at the current time step, and construct the evaluation feature set for the corresponding time step; The set of evaluation features is matched with preset health status threshold rules to determine whether the current time step is in the normal range, the mild decay range, or the severe degradation range. Based on the evaluation feature set and matching results, assign a health level label to the current time step and mark the status type of the current time step, including normal, mild decline and severe degradation; Arrange all health level labels in chronological order to construct a health status sequence; The health status sequence and capacity estimation curve are jointly visualized and encoded to generate online health status assessment results.

[0033] refer to Figure 3 An online health status assessment system for a sodium-ion battery pack includes: The data acquisition module is used to construct a battery aging model based on a preset sodium battery aging mechanism, collect multi-dimensional operating data of the sodium-ion battery pack at a preset frequency and perform preprocessing to generate an operating data sequence. The feature fusion module is used to extract the feature entropy values ​​of each time step of the running data sequence. In the liquid neural network, the neural state of the liquid neural network is dynamically adjusted according to the feature entropy values, and feature extraction and fusion operations are performed on the running data sequence to obtain the fused feature sequence. The deviation analysis module is used to extract the simulated fusion sequence generated in the battery aging model, perform path signature transformation operations on the simulated fusion sequence and the fusion feature sequence respectively, and perform structural error analysis operations on the two transformation results to calculate the structural deviation value and construct the deviation sequence. The capacity estimation module is used to perform adaptive parameter update operations on the battery aging model based on the deviation sequence using the particle filter algorithm, and to perform inference calculation operations on each time step using the updated battery aging model to obtain the corresponding capacity estimate value and form a capacity estimation sequence. The curve fitting module is used to perform curve fitting on the capacity estimation sequence using the Loess algorithm, and to perform smoothing filtering on the fitting results to generate the capacity estimation curve. The status assessment module is used to perform a health status assessment operation on the sodium-ion battery pack based on the capacity estimation curve and generate online health status assessment results.

[0034] Example 1: To verify the feasibility of this invention in practice, it was applied to the scenario of monitoring the operating status and health assessment of a sodium-ion battery pack in an energy storage system. This system uses 48 sodium-ion cells to construct the pack structure. Each cell has a nominal capacity of 120 Ah and a nominal voltage of 3.2V. The overall system is capable of long-term continuous charge and discharge operation and is used under conditions of significant day-night temperature differences and frequent changes in current load. Because sodium-ion batteries are more prone to capacity decay, increased internal resistance, and increased inconsistency between cells under high-rate discharge conditions, traditional static model-based SOH assessment methods are difficult to adapt to the dynamic fluctuations during pack-level operation. Especially when slight drift in cell parameters or localized overheating occurs, the existing methods cannot capture subtle degradation trends in a timely manner, resulting in delayed or even misjudged assessments.

[0035] To address the aforementioned issues, this invention deploys an online health status assessment model within the Battery Management Unit (BMS) and integrates a digital twin modeling component and a high-frequency data acquisition interface into the operation control system. In actual operation, the system collects real-time data on voltage, current, temperature, and environmental information for each battery cell at a fixed sampling period of 5 seconds, and synchronously sends the data to the edge analysis module for preprocessing. The preprocessing includes outlier removal, sliding time alignment, and missing data imputation to ensure that the data for subsequent analysis is complete, clean, and time-consistent.

[0036] During the feature extraction stage, the system constructs a set of four types of feature statistical indicators for each time step and calculates the feature entropy value corresponding to that time step. The entropy value not only reflects the degree of change of multidimensional features under the current operating state, but also serves as a driving parameter to dynamically adjust the neural state of the liquid neural network, enabling the network to automatically adapt to operating environments of different complexities and achieve more robust feature extraction and fusion.

[0037] In the digital twin modeling section, the system constructs a simulated aging model based on battery manufacturing data and current cumulative cycle information, incorporating duration, temperature, rate, and cycle factors. By inputting normalized operational characteristics, the model can generate simulated capacity sequences and simulated operational data streams online, which are then used for structural difference analysis with actual data. The quantification of structural differences employs a path signature transformation algorithm, mapping simulated and actual data into path signature vectors respectively, and performing difference analysis in a high-dimensional space to construct a structural deviation sequence.

[0038] Based on structural bias, the system employs a particle filter algorithm to adaptively update the aging model parameters in real time, enabling the model to continuously adjust its fitting ability to the capacity evolution trend as the cell state changes. At each time step, the system outputs a capacity estimate based on the latest model inference, ultimately forming a capacity estimation sequence that evolves over time. This sequence is then subjected to local curve fitting using the Loess algorithm, combined with Savitzky-Golay filtering and a monotonicity discrimination mechanism, to achieve a smooth and stable output of the capacity estimation curve, providing an accurate basis for the final SOH assessment.

[0039] The entire process is performed continuously online during normal battery operation, without system interruption or specific operational intervention, greatly improving the automation and real-time performance of health assessment. To verify the effectiveness of the method of this invention, it is compared with the traditional SOH estimation method based on capacity regression, and the SOH value calculated from the measured capacity under standard operating conditions of periodic charge and discharge is used as a real reference, as shown in the table below: Table 1. Comparative Data of Online Health Status Assessment for Sodium-ion Battery Packs

[0040] As shown in Table 1, the method of this invention consistently maintains a SOH estimation error below 0.3% throughout the entire evaluation cycle, far superior to the performance of traditional methods, which reach errors exceeding 6% in later cycles. Furthermore, the method demonstrates significant fluctuation control, effectively mitigating the problem of evaluation curve jumps caused by drastic changes in operating conditions in traditional methods. Particularly after 120 cycles, the error of traditional methods rapidly expands, becoming unreliable; while this invention, through a continuous model update mechanism, maintains a high degree of consistency between the estimated value and the measured SOH, significantly improving the accuracy and stability of the evaluation.

[0041] Therefore, this invention not only solves the problems of poor generalization ability and weak inconsistency expression of existing methods in sodium battery scenarios, but also realizes the online, dynamic and refined evaluation of SOH throughout the entire process, providing key support for battery operation and maintenance strategies, and has good engineering application value and promotion potential.

[0042] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for online assessment of the health status of a sodium-ion battery pack, characterized in that, Includes the following steps: S1. Construct a battery aging model based on a preset sodium battery aging mechanism, collect multidimensional operating data of the sodium-ion battery pack at a preset frequency and preprocess it to generate an operating data sequence. S2. Extract the feature entropy values ​​of each time step of the running data sequence. In the liquid neural network, dynamically adjust the neural state of the liquid neural network according to the feature entropy values, and perform feature extraction and fusion operations on the running data sequence to obtain the fused feature sequence. S3. Extract the simulated fusion sequence generated in the battery aging model, perform path signature transformation operation on the simulated fusion sequence and the fusion feature sequence respectively, and perform structural error analysis operation on the two transformation results to calculate the structural deviation value and construct the deviation sequence. S4. Using the particle filter algorithm, the parameters of the battery aging model are adaptively updated according to the deviation sequence. The updated battery aging model is then used to perform inference calculations at each time step to obtain the corresponding capacity estimates, forming a capacity estimation sequence. S5. Use the Loess algorithm to perform curve fitting on the capacity estimation sequence, and perform smoothing filtering on the fitting results to generate the capacity estimation curve. S6. Perform a health status assessment on the sodium-ion battery pack based on the capacity estimation curve and generate online health status assessment results.

2. The online health status assessment method for a sodium-ion battery pack according to claim 1, characterized in that, The sodium battery aging model is a digital twin model used to simulate the operating state of a sodium-ion battery pack. The multidimensional operating data includes the voltage, current, temperature, and environmental parameters of each cell in the sodium-ion battery pack. The preprocessing includes missing data filling, time alignment, and anomaly removal operations. The neural state represents the internal response vector of the liquid neural network that evolves in each time step based on the operating data sequence and the corresponding feature entropy value.

3. The online health status assessment method for a sodium-ion battery pack according to claim 1, characterized in that, The internal calculation process of the battery aging model specifically includes: The cumulative usage time, cumulative charge-discharge cycles, historical highest operating temperature, and average discharge rate of the sodium-ion battery pack were collected and normalized to linearly map the original values ​​to the [0,1] interval. A battery feature vector was constructed, which included usage time factor, temperature factor, cycle factor, and rate factor. The usage time factor represents the ratio of cumulative usage time to rated service life. The temperature factor represents the ratio of historical highest operating temperature to the rated safe operating temperature set by the manufacturer. The cycle factor represents the ratio of cumulative charge-discharge cycles to the battery's designed cycle life. The rate factor represents the ratio of average discharge rate to standard rated rate. The total simulation cycle length is set according to the duration factor in the battery feature vector, and the capacity decay coefficient for each simulation time step is set according to the cycle factor. In the first time step, the sodium-ion battery pack is set to have the battery capacity as the simulated capacity value of the previous time step. At each time step, the simulated capacity value from the previous time step is used as the starting capacity, and capacity decay calculation is performed, specifically including: The temperature adjustment amount is calculated by multiplying the current temperature factor by the set temperature sensitivity coefficient. The product of the current multiplier factor and the cycle factor is multiplied by a preset multiplier sensitivity coefficient to calculate the multiplier effect. According to the preset weight ratio, the temperature adjustment amount and the multiplier effect amount are weighted and superimposed. The superimposed result is multiplied by the preset capacity decay coefficient to obtain the capacity decay value at the current time step. The capacity decay value is then subtracted from the initial capacity to obtain the current simulated capacity value. Arrange all simulated capacity values ​​in chronological order to construct a simulated capacity sequence; Within each time step, based on the battery feature vector and the current time step index, the simulated voltage, simulated current and simulated temperature values ​​for the current time step are deduced and then spliced ​​with the real-time acquired environmental parameters to form a simulated running sequence. By performing feature extraction and fusion operations on the simulated running sequence using a liquid neural network, a simulated fused sequence that is consistent with the fused feature sequence in terms of time dimension, feature dimension, and arrangement order is obtained.

4. The online health status assessment method for a sodium-ion battery pack according to claim 1, characterized in that, S2 specifically includes: S21. Expand the running data sequence according to time steps. For the voltage, current, temperature and environmental parameters of each cell in each time step, calculate the maximum value, minimum value, mean value and rate of change of the four types of data in the sliding window, and form the feature statistics set of the current time step. S22. Based on the feature statistics set for each time step, calculate the feature entropy value for that time step, arrange all feature entropy values ​​to form a feature entropy sequence, and align it with the running data sequence in time. The calculation process of the feature entropy value specifically includes: The feature statistics set of the current time step is divided into voltage subset, current subset, temperature subset and environmental parameter subset according to data category. Min-Max normalization operation is performed on the maximum value, minimum value, mean value and rate of change within each subset. Within each subset, based on the normalized feature values ​​of the corresponding feature in the current sliding window, the feature values ​​are divided into multiple continuous and non-overlapping value range intervals according to the preset interval division rules. The frequency of each normalized feature value falling into each value range interval within the current sliding window is counted as the interval frequency. The interval frequency is then divided by the number of normalized feature values ​​within the sliding window to obtain the normalized probability value corresponding to each value range interval. Combine the normalized probability values ​​corresponding to all value ranges into a probability distribution vector of the current feature within the sliding window; Based on the probability distribution vector of each normalized feature, calculate the local entropy value corresponding to each subset, and then sum the local entropy values ​​of the four subsets according to the preset weight coefficient set to obtain the feature entropy value of the current time step. S23. Input the time alignment result into the liquid neural network. In each time step, dynamically adjust the neural state of the liquid neural network according to the feature entropy value and update the internal response vector of that time step. S24. Based on the internal response vector and running data at each time step, perform feature enhancement operations in the liquid neural network, extract the fused feature vector, and generate the fused feature sequence in chronological order.

5. The online health status assessment method for a sodium-ion battery pack according to claim 4, characterized in that, S23 specifically includes: S231. In each time step, extract the feature entropy value of the current time step, and match the feature entropy value with the preset adjustment function to calculate the adjustment weight; S232. Based on the adjustment weights, the time constants of each neuron in the liquid neural network are dynamically scaled, and the product of the original time constant and the adjustment weights is used as the updated time constant for the current time step. S233. Based on the updated time constant, the running data of the current time step and the internal response vector of the previous time step are weighted and fused to generate the internal response vector of the current time step. S234. Arrange the internal response vectors at all time steps in chronological order to construct an internal response sequence that is consistent with the time dimension of the running data sequence, while maintaining consistency in the feature dimensions.

6. The online health status assessment method for a sodium-ion battery pack according to claim 1, characterized in that, S3 specifically includes: S31. Based on the time dimension of the fusion feature sequence, read the simulated fusion sequence of the corresponding time step in the battery aging model; S32. Expand the fused feature sequence and the simulated fused sequence according to time steps, and construct the corresponding path sequences according to a fixed arrangement order; S33. Take the path vector difference between adjacent time steps in each path sequence as the path increment, and gradually accumulate the path increment according to the time order to generate the path representation. S34. At each time step, perform a path signature transformation operation on the path representation, extract the path integral values ​​at different orders, and form a fused signature sequence and a simulated signature sequence. S35. Calculate the difference between the two path integral values ​​at the corresponding time steps in the fused signature sequence and the simulated signature sequence, sum the absolute values ​​of the differences in each dimension, and perform Z-Score standardization to obtain the structural deviation value at the current time step. S36. Arrange all structural deviation values ​​in chronological order to construct a deviation sequence.

7. The online health status assessment method for a sodium-ion battery pack according to claim 6, characterized in that, S34 specifically includes: S341. In each time step, according to the order of path increments in the fused path representation and the simulated path representation, the current time step is taken as the cutoff point, and the path increment sequence between the first time step and the current time step is extracted as the path representation segment of the current time step. S342. Perform multi-order path integral calculation operation on the path representation segment, calculate the first-order, second-order to the preset highest-order path integral values ​​in sequence, and construct the path integral set corresponding to the current time step. The path integral value of each order is calculated by accumulating the multi-dimensional feature quantities in the path increment sequence in a fixed order. S343. Perform feature mapping operation on the path integral set, and arrange the path integral values ​​of different orders according to the feature type by using dimensional expansion to generate a path signature vector that matches the path representation. S344. Arrange the path signature vectors generated at all time steps in chronological order, and form a fused signature sequence and a simulated signature sequence according to their source.

8. The online health status assessment method for a sodium-ion battery pack according to claim 1, characterized in that, S4 specifically includes: S41. Set the capacity decay coefficient, temperature sensitivity coefficient and rate sensitivity coefficient in the battery aging model as the set of parameters to be estimated by the particle filter algorithm, and use the fusion feature vector of the corresponding time step in the fusion feature sequence as the state driver to initialize the particle set containing the weights of each particle. S42. In each time step, based on the particle set of the previous time step, the state prediction update operation is performed on the parameter set to be estimated. The prediction result is substituted into the battery aging model to calculate the predicted capacity value of the current time step. The predicted capacity value is matched with the structural deviation value of the corresponding time step in the deviation sequence to construct the observation error of the current time step. S43. Based on the observation error at the current time step, perform a weight update operation on each particle in the particle set. The smaller the observation error, the higher the weight of the particle. Then, perform Min-Max normalization on the updated weights of all particles to generate an updated particle set, specifically including: Z-Score normalization is performed on the observation error of each particle to calculate the error score. The lower the error score, the closer the particle is to the center of the overall error distribution. A mapping operation is performed based on the error score value. A monotonically decreasing preset weight function is used to map the error score value to non-negative weights, and an initial weight set is constructed for the current time step. The initial weight set is subjected to Min-Max normalization, which linearly compresses the weights of all particles to a set range, forming a normalized weight set. Map the normalized weight set back to the particle set and update the weights of all particles at the current time step; S44. Perform a resampling operation on the updated particle set, remove particles with weights lower than a preset threshold, sort each particle according to its weight, select the particles with the highest weights by a set percentage and perform a resampling copy operation to form a resampled particle set for the current time step. S45. Use the mean value of the particle parameters in the resampled particle set as the update parameter of the battery aging model at the current time step, and substitute the update parameter into the battery aging model to perform capacity inference calculation at the current time step to obtain the corresponding capacity estimate. S46. Arrange the capacity estimates obtained at each time step in chronological order to construct a capacity estimation sequence.

9. The online health status assessment method for a sodium-ion battery pack according to claim 1, characterized in that, S5 specifically includes: S51. Set a fixed-length sliding window for the capacity estimation sequence according to the time step, and use the weighted multinomial regression method to locally fit the capacity estimation value in each sliding window to generate the local fitting value for the corresponding time step. S52. Arrange the local fitted values ​​at all time steps in the order of time steps to construct the initial capacity curve, and perform a continuity detection operation on the initial capacity curve to identify abrupt change points. S53. Perform curve splicing and adjustment operations on the local fitted values ​​before and after the mutation point, and perform Savitzky-Golay filtering on the adjustment results to generate intermediate capacity curves, specifically including: Symmetrical splicing intervals are set on both sides of the mutation point, and local fitting values ​​within the corresponding time period are extracted to construct fitting curve segments; Smoothing weights are set based on the trend of slope change and numerical difference at the endpoints of the fitted curve segments, and a weighted superposition operation is performed on the local fitted values ​​within the splicing interval to generate a smooth transition curve. The smooth transition curve and the fitted curve segment are spliced ​​together in chronological order to construct the corrected initial capacity curve as the adjustment result; Based on the adjustment results, the sliding window length and polynomial order are set, and the Savitzky-Golay filtering algorithm is used to smooth the adjustment results and extract the filter value at each time step. Arrange all filter values ​​in chronological order to generate an intermediate capacity curve; S54. Perform monotonicity discrimination processing on the intermediate capacity curve. If an abnormal interval where the local fluctuation amplitude exceeds the set threshold is found, perform curve segment replacement operation to form a capacity estimation curve.

10. An online health status assessment system for a sodium-ion battery pack, comprising executing the online health status assessment method for a sodium-ion battery pack according to any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to construct a battery aging model based on a preset sodium battery aging mechanism, collect multi-dimensional operating data of the sodium-ion battery pack at a preset frequency and perform preprocessing to generate an operating data sequence. The feature fusion module is used to extract the feature entropy values ​​of each time step of the running data sequence. In the liquid neural network, the neural state of the liquid neural network is dynamically adjusted according to the feature entropy values, and feature extraction and fusion operations are performed on the running data sequence to obtain the fused feature sequence. The deviation analysis module is used to extract the simulated fusion sequence generated in the battery aging model, perform path signature transformation operations on the simulated fusion sequence and the fusion feature sequence respectively, and perform structural error analysis operations on the two transformation results to calculate the structural deviation value and construct the deviation sequence. The capacity estimation module is used to perform adaptive parameter update operations on the battery aging model based on the deviation sequence using the particle filter algorithm, and to perform inference calculation operations on each time step using the updated battery aging model to obtain the corresponding capacity estimate value and form a capacity estimation sequence. The curve fitting module is used to perform curve fitting on the capacity estimation sequence using the Loess algorithm, and to perform smoothing filtering on the fitting results to generate the capacity estimation curve. The status assessment module is used to perform a health status assessment operation on the sodium-ion battery pack based on the capacity estimation curve and generate online health status assessment results.

Images