Evaluation Methods and Systems for Battery Reuse in Battery Swapping Cabinets

Abstract

This invention relates to the field of battery technology for battery swapping cabinets, and discloses a method and system for evaluating the reuse of batteries in battery swapping cabinets. The method includes: deploying a distributed detection device network in a battery cluster within the battery swapping cabinet, and collecting raw battery datasets from the battery cluster; performing feature classification and vector encoding on the raw battery datasets to obtain multiple battery performance feature vectors; inputting the multiple battery performance feature vectors into a set of battery reuse potential prediction models for prediction, and outputting a comprehensive battery reuse potential prediction result; setting evaluation node parameters based on the comprehensive battery reuse potential prediction result, and performing adaptive sampling and dynamic optimization calculations on the evaluation node parameters to obtain a multi-stage battery evaluation and control scheme. This invention achieves a comprehensive evaluation of battery reuse potential, improves prediction accuracy, and has strong generalization ability.

Description

Technical Field

[0001] This invention relates to the field of battery swapping cabinets, and more particularly to a method and system for evaluating the reuse of batteries in battery swapping cabinets. Background Technology

[0002] During the service life of batteries in battery swapping stations, their performance gradually degrades due to factors such as increased charge-discharge cycles and changes in ambient temperature. When key indicators such as capacity or power can no longer meet the needs of battery swapping, these batteries need to be retired. To maximize the value of these batteries, it is particularly important to conduct a reasonable reuse assessment of retired batteries.

[0003] Currently, conventional battery reuse assessment methods mainly rely on simple capacity testing and internal resistance detection, lacking systematic analysis and in-depth mining of battery lifecycle data. This assessment method struggles to accurately reflect the battery's true health status and cannot precisely assess its remaining value. Furthermore, due to the large number of battery packs in battery swapping cabinets, traditional centralized assessment schemes face problems in practical applications such as low data acquisition efficiency, heavy computational burden, and insufficient assessment accuracy. Summary of the Invention

[0004] This invention provides a method and system for evaluating the reuse potential of batteries in battery swapping cabinets. This invention achieves a comprehensive evaluation of the reuse potential of batteries, improves the prediction accuracy, and has a strong generalization ability.

[0005] In a first aspect, the present invention provides a method for evaluating the reuse of batteries in a battery swapping cabinet, the method comprising:

[0006] A distributed detection device network is deployed in the battery cluster of the battery swapping cabinet, and the raw battery dataset of the battery cluster is collected.

[0007] The original battery dataset is subjected to feature classification and vector encoding to obtain multiple battery performance feature vectors;

[0008] The multiple battery performance feature vectors are input into the battery reuse potential prediction model set for prediction, and the comprehensive battery reuse potential prediction result is output.

[0009] Based on the comprehensive battery reuse potential prediction results, evaluation node parameters are set, and adaptive sampling and dynamic optimization calculations are performed on the evaluation node parameters to obtain a multi-stage battery evaluation and control scheme.

[0010] Secondly, the present invention provides a battery reuse evaluation system for battery swapping cabinets, the battery reuse evaluation system comprising:

[0011] The data acquisition module is used to deploy a distributed detection device network in the battery cluster of the battery swapping cabinet and to collect the raw battery dataset of the battery cluster.

[0012] The encoding module is used to perform feature classification and vector encoding on the original battery dataset to obtain multiple battery performance feature vectors.

[0013] The prediction module is used to input the multiple battery performance feature vectors into the battery reuse potential prediction model set for prediction, and output the comprehensive battery reuse potential prediction result.

[0014] The calculation module is used to set evaluation node parameters based on the comprehensive battery reuse potential prediction results, and to perform adaptive sampling and dynamic optimization calculation on the evaluation node parameters to obtain a multi-stage battery evaluation and control scheme.

[0015] A third aspect of the present invention provides a battery recycling evaluation device for a battery swapping cabinet, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the battery recycling evaluation device for the battery swapping cabinet to perform the above-described battery recycling evaluation method for the battery swapping cabinet.

[0016] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described battery reuse evaluation method for battery swapping cabinets.

[0017] The technical solution provided by this invention, by deploying a distributed detection device network and combining a multi-level data acquisition and processing mechanism, achieves efficient acquisition and analysis of battery lifecycle data in battery swapping cabinets, significantly improving the completeness and accuracy of data acquisition. An evaluation method combining deep learning and variational Bayesian inference, through multi-level feature extraction and dynamic optimization strategies, improves the accuracy of battery performance evaluation. A tiered utilization decision mechanism based on multi-objective optimization, combined with scene feature matching and resource optimization allocation strategies, improves the matching accuracy of battery reuse. Through adaptive sampling and dynamic optimization control, this invention can automatically adjust the evaluation strategy according to changes in battery performance. The introduction of a multi-layer variational Bayesian inference network, combined with analysis of battery performance degradation patterns, enhances the reliability of the evaluation results. By integrating multiple prediction sub-models and a dynamic weight allocation mechanism, this invention achieves a comprehensive evaluation of battery reuse potential, improving prediction accuracy and exhibiting strong generalization ability.

[0018] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of an embodiment of the battery reuse evaluation method for battery swapping cabinets in this invention.

[0021] Figure 2 This is a schematic diagram of one embodiment of the battery reuse evaluation system for battery swapping cabinets in this invention.

[0022] Figure 3 This is a schematic diagram of one embodiment of the battery reuse evaluation device for battery swapping cabinets in this invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] The terms "comprising" and "having," and any variations thereof, used in the embodiments of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0025] To facilitate understanding of this embodiment, a detailed description of the battery reuse evaluation method for a battery swapping cabinet disclosed in this embodiment of the invention will be provided first. For example... Figure 1 As shown, this method includes the following steps:

[0026] 101. Deploy a distributed detection device network in the battery cluster of the battery swapping cabinet and collect the raw battery dataset of the battery cluster.

[0027] It is understood that the executing entity of this invention can be a battery recycling evaluation system for battery swapping cabinets, or it can be a terminal or a server; the specific implementation is not limited here. This embodiment of the invention will be described using a server as an example.

[0028] Specifically, a distributed testing device network is deployed within the battery clusters of the battery swapping cabinets. The battery clusters are grouped according to geographical location information and usage environment parameters, and distributed nodes are optimized based on this grouping information to form a distributed testing device network capable of efficiently covering all batteries. During the construction of this network, the detection range and priority distribution of each node are determined by comprehensively analyzing the distribution characteristics of geographical locations, environmental conditions (such as temperature, humidity, and altitude), battery usage status (such as charging and discharging frequency), and data acquisition requirements, thereby optimizing node layout and achieving efficient operation of the distributed network. In the deployed distributed testing device network, the operating parameters of the network nodes are synchronously calibrated. Through data sampling cycle coordination and timestamp alignment technology, unified synchronous sampling control parameters are set to ensure that all distributed nodes collect data under the same time reference. Based on the synchronous sampling control parameters, key physical parameters of the batteries, including voltage, current, and temperature, are sampled at high frequency to record the battery's operating conditions. The collected battery operating condition data is input into the feature analysis module. Time-domain feature analysis extracts time-series characteristics such as voltage fluctuation amplitude and temperature change rate. Simultaneously, frequency-domain feature analysis uncovers hidden battery performance variation patterns, such as frequency response characteristics and harmonic distribution. Combining the results of time-domain and frequency-domain feature analysis yields a multi-dimensional description of the battery's operating state, forming dynamic characteristic data. Based on this dynamic characteristic data, a voltage-current response curve is constructed. This curve visually reflects the battery's intrinsic electrochemical performance by showing the voltage and current changes under different operating conditions. By analyzing the voltage-current response curve, the battery's internal resistance change value is extracted using relevant calculation models. A temperature compensation coefficient is introduced to correct the results during the calculation of the internal resistance change value, improving the accuracy of the internal resistance parameter. The battery's performance degradation is assessed using the corrected internal resistance parameter. Furthermore, the charging and discharging processes of the batteries in the battery swapping cabinet cluster are recorded at different usage stages. The charge-discharge capacity of the batteries is accurately calculated using the charge integral method, and the number of battery cycles is statistically analyzed using the coulomb counting method. By combining charge / discharge capacity and cycle count calculation methods, a more accurate assessment of battery life and remaining performance is achieved, generating comprehensive battery usage record data. To achieve a holistic evaluation of battery operating status, battery condition data, battery dynamic characteristic data, battery internal resistance parameters, and battery usage record data are merged in a time series. By aligning these data onto a unified time axis, a consistent time-series dataset is formed. Simultaneously, to eliminate dimensional differences and noise in the data, data standardization techniques are employed to normalize all data, enhancing comparability and consistency, ultimately resulting in a raw battery dataset that comprehensively reflects the battery's operating status.

[0029] 102. Perform feature classification and vector encoding on the original battery dataset to obtain multiple battery performance feature vectors;

[0030] Specifically, based on the voltage, current, and temperature parameters in the original battery dataset, data dimensionality reduction is performed. Dimensionality reduction techniques such as principal component analysis or linear discriminant analysis are used to transform high-dimensional multivariate data into a low-dimensional representation, resulting in a battery parameter dimensionality reduction matrix. Time-series correlation analysis is performed on the battery parameter dimensionality reduction matrix to quantify the time-series correlation between different parameters, calculating time-series feature weight coefficients to reflect the relative importance of each parameter to battery performance at different time points. The time-series feature weight coefficients are then weighted and fused with the battery parameter dimensionality reduction matrix, extracting multiple primary battery performance characteristics by combining the dynamic changes between parameters. Simultaneously, to comprehensively evaluate the battery's performance degradation characteristics, time-series decomposition is performed on the capacity-related data in the original battery dataset. A double exponential function is used to fit the capacity degradation curve, thus constructing a capacity degradation model that describes the capacity change pattern. To improve the fitting accuracy, the capacity degradation curve is corrected by incorporating a temperature stress coefficient, enabling the model to adapt to different usage environments. Through this process, capacity degradation model parameters reflecting the battery capacity change pattern are obtained. Based on these parameters, a battery capacity prediction function is constructed. By approximating the capacity prediction function using Taylor expansion, the capacity decay rate under different operating conditions is calculated, yielding the predicted capacity decay value. Simultaneously, to analyze the battery's internal resistance variation characteristics, the internal resistance data in the original dataset is segmented, and an internal resistance variation function is established based on piecewise linear regression. Based on this, parameters are optimized using historical data to obtain internal resistance variation model coefficients that reflect the internal resistance variation pattern. These coefficients are input into the prediction function, and the predicted internal resistance variation values ​​at different time points are calculated recursively. Due to the significant impact of temperature on battery internal resistance variation, a temperature correction coefficient is used to correct the predicted internal resistance value, improving the accuracy and reliability of the prediction, resulting in a corrected predicted internal resistance variation value. The predicted capacity decay value and the predicted internal resistance variation value are respectively processed into feature vectors, transforming these two types of characteristics into an operable multi-dimensional feature space, generating multiple second battery performance features. The first and second battery performance features are combined, and all extracted features are classified using feature classification algorithms (such as K-means clustering or support vector classification). The classification results are then vector-encoded to generate multiple battery performance feature vectors.

[0031] Multiple performance characteristics of the first and second batteries are normalized to eliminate dimensional differences and data distribution biases between different characteristics. Normalization methods such as maximum-minimum standardization or Z-score standardization are used to map all feature values ​​to a unified numerical range, improving the stability of feature calculation and cluster analysis. After normalization, the feature space is analyzed using a density peak detection algorithm. By calculating the density distribution of data points and their distances to high-density regions, the initial values ​​of the cluster centers in the feature space are identified, yielding the first cluster centroids. Using the first cluster centroids as input, the K-means algorithm is applied for iterative optimization. The K-means algorithm uses minimizing intra-cluster distances and maximizing inter-cluster distances as its objective functions. Through repeated allocation of data points and adjustment of centroid positions, the clustering results are continuously optimized, yielding the second cluster centroids. Based on the second cluster centroids, silhouette coefficient analysis is performed to quantify the quality of the current clustering results and determine the optimal number of clusters. The silhouette coefficient evaluates the distance relationship between data points within their own cluster and other clusters, providing an objective basis for selecting a reasonable number of clusters. Based on the optimal number of clusters, the battery performance features are finally clustered. Following this clustering, the entropy method is used to calculate the information entropy and weight coefficient of each feature, evaluating the contribution of each feature to the overall feature space and obtaining a feature importance score. The entropy method ensures the scientific and objective nature of weight allocation by measuring the distributional differences of features among samples. The feature importance score is then weighted and combined with the clustering results to generate the final feature representation. To optimize the feature encoding strategy, a locality-sensitive hash function (LSH) is used to generate a hash encoding strategy. LSH efficiently maps high-dimensional features to a low-dimensional space, maintaining the relative similarity between features while reducing computational complexity. The hash encoding strategy is used to reduce the dimensionality of features, obtaining the feature association strength. Based on the feature association strength, high-order feature extraction and combination are performed on multiple first-stage battery performance features and multiple second-stage battery performance features. Through nonlinear feature combination and interaction effect analysis, potential high-order relationships in the original feature space are mined, constructing richer feature representations. Based on the high-order feature extraction, multiple battery performance feature vectors are finally formed through appropriate feature fusion techniques.

[0032] 103. Input multiple battery performance feature vectors into the battery reuse potential prediction model set for prediction, and output the comprehensive battery reuse potential prediction result.

[0033] Specifically, the high-dimensional feature space is decomposed into several independent feature subspaces by subsegmenting multiple battery performance feature vectors. Feature selection and dimensionality reduction techniques, such as principal component analysis, linear discriminant analysis, or t-SNE, are employed to ensure that the partitioned feature subspaces are independent while retaining sufficient feature information. Through subspace partitioning, complex high-dimensional features are effectively decomposed into multiple low-dimensional subspaces, thereby improving the computational efficiency and predictive ability of the model. The independent feature subspaces are then input into a battery reuse potential prediction model set for prediction. This model set consists of multiple battery reuse potential prediction sub-models, each with an independent structure and optimized for a specific subspace. The prediction sub-model comprises three parts: a feature extraction layer, a feature mapping layer, and a prediction output layer. The feature extraction layer consists of three fully connected layers, each using BatchNormalization and the ReLU activation function to ensure balanced distribution of feature data and enhanced nonlinear modeling capabilities. BatchNormalization stabilizes the training process and accelerates model convergence, while the ReLU activation function introduces nonlinearity to improve the model's ability to represent complex feature relationships. The feature extraction and feature mapping layers consist of two residual connection blocks, each containing a two-layer perceptron and skip connections. The residual connection design mitigates the vanishing gradient problem in deep models by directly short-circuiting paths, while enhancing the network's sensitivity to input features and its learning ability. The prediction output layer uses a sigmoid activation function, mapping the sub-model's predictions to a range of 0 to 1, facilitating a probabilistic interpretation of the battery reuse potential. Each sub-model outputs a prediction based on its corresponding independent feature subspace, and the predictions of all sub-models collectively form the sub-model prediction result set. Dynamic weight allocation is applied to the sub-model prediction results to optimize the final comprehensive prediction result. The weight allocation process is based on a dynamic weighting method, combining the historical performance and current prediction accuracy of each sub-model, and dynamically calculating the fusion weight coefficients through Bayesian optimization or an attention-based weight adjustment strategy. Based on the fusion weight coefficients, a weighted analysis is performed on the sub-model prediction result set. By weighted summing the prediction results of each sub-model with their corresponding weights, a comprehensive battery reuse potential prediction result is generated. 104. Based on the prediction results of the comprehensive battery reuse potential, the evaluation node parameters are set, and the evaluation node parameters are adaptively sampled and dynamically optimized to obtain a multi-stage evaluation and control scheme for the battery.

[0034] Specifically, the predicted results of the comprehensive battery reuse potential are input into the first-layer variational Bayesian inference network for data distribution modeling. Through this process, the variational expectation-maximization algorithm iteratively calculates the posterior distribution of the latent variables, deriving the distribution parameters of battery performance. The variational Bayesian inference network, with its nonlinear modeling capabilities, handles complex, high-dimensional battery performance data distribution problems, ensuring that the calculated distribution parameters accurately reflect the actual situation. The battery performance distribution parameters are then input into the second-layer variational Bayesian inference network to analyze the evolution of battery performance. Combining random field theory, the network calculates the transition probability matrix of the battery state, describing the probability distribution of battery performance transitioning from one state to another. The state transition probability matrix reveals the inherent laws of battery performance degradation, providing a basis for assessing its performance evolution trend throughout its entire life cycle. Using random field theory for probabilistic modeling of state transitions adapts to the nonlinear and random characteristics of battery performance changes, ensuring the reliability and scientific validity of the analysis results. Finally, the battery performance degradation laws are input into the third-layer variational Bayesian inference network to probabilistically model the parameters of the evaluation nodes. By refining the distribution of node parameters, the network outputs a set of high-precision evaluation node parameters. These parameters reflect key performance indicators of the battery at different stages, including capacity, internal resistance, and remaining life. Based on the evaluation node parameters, a multi-stage evaluation control strategy is designed and optimized. The optimization process comprehensively considers the dynamic nature of battery performance characteristics and generates an optimized evaluation control strategy by establishing a mapping relationship between evaluation nodes and battery states. To improve the applicability and flexibility of the evaluation strategy, adaptive sampling is performed on the optimized evaluation control strategy to obtain an adaptive sampling scheme. By dynamically adjusting the sampling frequency and sampling range, the evaluation strategy can adapt to changes in battery performance in real time. Time-series feature extraction is performed on the adaptive sampling scheme to obtain a dynamic adjustment matrix for evaluation parameters. The dynamic adjustment matrix provides a basis for parameter adjustment for multi-stage evaluation by quantifying the trend of battery performance changes over time. Based on this matrix, evaluation accuracy and evaluation efficiency are jointly optimized to ensure the scientificity and practicality of the evaluation scheme. In the joint optimization process, a multi-objective optimization method is adopted, taking the improvement of evaluation accuracy and the reduction of resource consumption as optimization objectives, and achieving a balance between the two through iterative adjustments. Combining the outputs of all steps, a complete multi-stage battery evaluation control scheme is generated. This solution integrates historical data and future trends of battery performance, enabling dynamic adjustment of evaluation strategies to adapt to changes in battery performance, thereby achieving scientific management of the entire battery lifecycle.

[0035] In this embodiment of the invention, by deploying a distributed detection device network and combining a multi-level data acquisition and processing mechanism, the invention achieves efficient acquisition and analysis of full lifecycle data of batteries in battery swapping cabinets, significantly improving the completeness and accuracy of data acquisition. An evaluation method combining deep learning and variational Bayesian inference is employed, improving the accuracy of battery performance evaluation through multi-level feature extraction and dynamic optimization strategies. A tiered utilization decision mechanism based on multi-objective optimization, combined with scene feature matching and resource optimization allocation strategies, improves the matching accuracy of battery reuse. Through adaptive sampling and dynamic optimization control, the invention can automatically adjust the evaluation strategy according to changes in battery performance. The introduction of a multi-layer variational Bayesian inference network, combined with analysis of battery performance degradation patterns, enhances the reliability of the evaluation results. By integrating multiple prediction sub-models and a dynamic weight allocation mechanism, the invention achieves a comprehensive evaluation of battery reuse potential, improving prediction accuracy and exhibiting strong generalization ability.

[0036] In one specific embodiment, the process of performing step 101 may specifically include the following steps:

[0037] Based on geographical location information and usage environment parameters, the battery cluster of the battery swapping cabinet is grouped and distributed nodes are optimized to obtain a distributed detection device network.

[0038] Data sampling period synchronization and timestamp alignment are performed on the distributed detection device network to obtain synchronization sampling control parameters;

[0039] Based on the synchronous sampling control parameters, the battery voltage, current and temperature parameters are sampled at high frequency to obtain battery operating condition data. Then, time domain feature analysis and frequency domain feature analysis are performed on the battery operating condition data to obtain battery dynamic characteristic data.

[0040] Voltage and current response curves are constructed based on battery dynamic characteristic data, and the battery internal resistance change is calculated through the voltage and current response curves. At the same time, the internal resistance parameters of the battery are obtained by combining the temperature compensation coefficient.

[0041] The charging and discharging process of the batteries in the battery cluster of the battery swapping cabinet is recorded. The charging and discharging capacity is calculated by the charge integration method, and the number of cycles is counted by the coulomb counting method to obtain battery usage record data.

[0042] Battery operating condition data, battery dynamic characteristic data, battery internal resistance parameters, and battery usage record data are time-series merged and data standardization processed to obtain the original battery dataset.

[0043] Specifically, the battery swapping cabinet clusters are grouped and distributed nodes are optimized based on geographical location information and usage environment parameters to establish an efficient distributed detection device network. Assume that a city's battery swapping cabinet network is distributed across multiple areas, and the geographical location information of each area is represented by coordinates (x...). i ,y i ), where i is the number of the battery swapping cabinet. Environmental parameters such as temperature (T) are also considered. i ), humidity (H) i ) and load frequency (f i To optimize distributed nodes, an objective function is constructed. Where w i d represents the weight. i Let P be the distance between the battery swapping cabinets, λ be the regularization parameter, and P be the distance between the battery swapping cabinets. i This represents the load parameters of the nodes. By solving this optimization problem, the battery swapping cabinets are effectively grouped, and the optimal deployment locations of the testing devices are determined, ensuring maximum network coverage and data acquisition efficiency. After constructing the distributed testing device network, its data sampling period is synchronized and timestamps are aligned to ensure consistency of data from different testing nodes on the same time base. Assume the sampling period for each node is T. s The timestamp is t i Synchronous sampling control parameters are obtained through Δt = t i -t j (where i and j are different nodes) are adjusted so that all nodes satisfy t i +kT s =t j +mT s Where k,m∈Z. This method eliminates data offset between different nodes and achieves global synchronization. After synchronization, key parameters of the battery in the battery swapping cabinet are sampled at high frequency based on control parameters, including voltage (V), current (I), and temperature (T). Let the sampling frequency be f. s Then, the sampled data at time t is represented as {V(t), I(t), T(t)}. After time-domain and frequency-domain feature analysis, the collected battery operating data is used to extract key dynamic characteristics. For example, the time-domain characteristics of voltage and current are obtained by calculating the mean. and standard deviation (Where X represents V or I) is obtained. In the frequency domain, Fast Fourier Transform is used to analyze the signal's spectral characteristics, extracting the dominant frequency component and high-frequency harmonic information. These analysis results are used to construct a battery dynamic characteristic dataset. Based on the battery dynamic characteristic data, voltage-current response curves are plotted. Assume the voltage and current sampling data are... The response curve was obtained by fitting the formula V = RI + V0 (where R is the internal resistance and V0 is the open-circuit voltage). The result was calculated using the least squares optimization method. Since temperature has a significant effect on internal resistance, a temperature compensation coefficient α is introduced. T The internal resistance value is corrected to obtain the corrected internal resistance parameter R. T =R(1+α) T (TT ref )), where T ref This is the reference temperature. For the recording of the charging and discharging process, the charging and discharging capacity is calculated using the charge integration method. The battery's charging capacity Q over time [t1, t2] is... c and discharge capacity Q d They are respectively and Using the coulomb counting method, the number of battery cycles, n, is calculated using the following formula: Q rated Here, k represents the rated capacity, and k is the cycle number. Battery operating condition data, battery dynamic characteristic data, battery internal resistance parameters, and battery usage record data are merged in a time series. Interpolation methods are used to unify data from different time axes onto a global time axis, and the merged data is then standardized. For example, min-max normalization is used to map the data to the [0,1] range, and the normalization formula is: The standardized dataset is the original battery dataset, which can comprehensively reflect the battery's performance status under different times and operating conditions.

[0044] In one specific embodiment, the process of performing step 102 may specifically include the following steps:

[0045] Based on the original battery dataset, the voltage, current and temperature parameters are reduced in dimensionality to obtain the battery parameter dimensionality reduction matrix. Then, time-series correlation analysis is performed on the battery parameter dimensionality reduction matrix to obtain the time-series feature weight coefficients.

[0046] By performing a weighted fusion operation on the time-series feature weight coefficients and the battery parameter dimension reduction matrix, multiple first battery performance features are obtained.

[0047] The capacity data in the original battery dataset is decomposed into a time series, and the capacity decay curve is fitted by a double exponential function. The parameters of the capacity decay model are then obtained by combining the temperature stress coefficient for correction.

[0048] A battery capacity prediction function is constructed based on the parameters of the capacity decay model. The capacity decay rate under different operating conditions is calculated by Taylor expansion to obtain the predicted capacity decay value.

[0049] The internal resistance data in the original battery dataset is segmented, and an internal resistance change function is established using a piecewise linear regression method. The parameters are then optimized by combining historical data to obtain the coefficients of the internal resistance change model.

[0050] The internal resistance change model coefficients are input into the prediction function, and the predicted internal resistance values ​​at different time points are obtained through recursive calculation. The internal resistance change is then corrected by combining the temperature correction coefficient to obtain the predicted internal resistance change value.

[0051] By performing feature vectorization on the predicted values ​​of capacity decay and internal resistance change, multiple second battery performance characteristics are obtained.

[0052] Multiple first battery performance features and multiple second battery performance features are classified and vector-encoded to obtain multiple battery performance feature vectors.

[0053] Specifically, dimensionality reduction is performed on the voltage parameter V(t), current parameter I(t), and temperature parameter T(t) in the original battery dataset. Assuming the dataset contains N samples, X = [V, I, T] is an N×3 matrix, where each column represents the time series of a parameter. Principal component analysis is used to calculate the dimensionality-reduced matrix Z = XW, where W is the eigenvector matrix, defining the projection direction of the original parameters in the reduced dimensionality space. The dimensionality-reduced matrix Z retains the main features of the voltage, current, and temperature parameters while reducing data dimensionality to minimize redundancy. Time-series correlation analysis is then performed on the battery parameter dimensionality-reduced matrix Z to extract the time-series feature weight coefficients. Let Z = [z1, z2, ..., z...]. m The symbol represents the feature components after dimensionality reduction. The autocorrelation function is calculated for each component. The temporal correlation weights of the feature components are determined based on different lag step sizes τ. These weight coefficients are then combined to construct the temporal feature weight vector w. t The time-series feature weight coefficients are weighted and fused with the dimensionality-reduced matrix Z, and the calculation formula is F1 = w t T Z is used to obtain multiple first-stage battery performance characteristics. The capacity data C(t) in the original battery dataset is decomposed into a time series, and its changing trend is modeled by fitting the capacity decay curve using a double exponential function. Capacity decay is expressed as C(t) = C0e^(-t / C0). -αt +C r e -βt Where C0 is the initial capacity, α and β are the decay rate constants, and C r This is the residual capacity. The fitting parameters α, β, C0, and C are optimized using the nonlinear least squares method. r Since temperature has a significant impact on capacity decay, combined with the temperature stress coefficient γ T The modified capacity model is C. T (t)=C(t)(1+γ T (TT ref )), where T ref This is the reference temperature. A battery capacity prediction function is constructed based on the parameters of the capacity decay model. The capacity decay rate under different operating conditions is approximated using Taylor expansion. The expression for the prediction function is C. p (t)≈C T (t)-(αC0e -αt +

[0054] βC r e -βt The predicted capacity decay value at future time points is derived using this prediction function, thereby assessing the battery's lifespan and remaining performance. Simultaneously, the internal resistance data R(t) in the original battery dataset is segmented to capture the changing trend of internal resistance at different stages. Assuming the internal resistance data is divided into k stages, the internal resistance trend in each stage is expressed as R(t) = a i t+b i ,t∈[t i-1 ,t i ], where a i and b i These are the linear regression coefficients for stage i. The least squares method is used to fit (a... i ,b i The model is then optimized using historical data to obtain complete coefficients for the internal resistance variation model. These coefficients are then input into the internal resistance prediction function R. p (t)=R(t)+ΔR T , where ΔR T =γ R (TT ref The ) represents the temperature correction term, which ultimately yields the predicted value R of the internal resistance change. p (t). The predicted values ​​of capacity decay and internal resistance change are vectorized to generate multiple second battery performance features. These features, represented in high dimension, uniformly describe the dynamic changes in battery performance across both capacity and internal resistance dimensions. The first battery performance feature F1 and the second battery performance feature F2 are then classified and vectorized. All features are clustered using K-means or peak density clustering algorithms to obtain feature classification labels. These labels are then combined to vectorize the features, for example, by embedding the features into a low-dimensional space, forming the final battery performance feature vector.

[0055] In one specific embodiment, the process of performing feature classification and vector encoding on multiple first battery performance features and multiple second battery performance features to obtain multiple battery performance feature vectors can specifically include the following steps:

[0056] The performance characteristics of multiple first batteries and multiple second batteries are normalized, and the initial value of the cluster center in the feature space is calculated by the density peak detection algorithm to obtain the first cluster centroid.

[0057] The first cluster centroid is input into the K-means algorithm, and the second cluster centroid is obtained by iteratively calculating the objective function of minimizing intra-cluster distance and maximizing inter-cluster distance.

[0058] Silhouette coefficient analysis is performed on the centroid of the second cluster to obtain the optimal number of clusters. Clustering is then performed based on the optimal number of clusters. The information entropy and weight coefficient of each feature are calculated using the entropy method to obtain the feature importance score.

[0059] The feature importance score is weighted and combined with the clustering results, a hash encoding strategy is generated by the locality-sensitive hash function, and a dimension reduction mapping is performed according to the hash encoding strategy to obtain the feature association strength.

[0060] Based on the feature correlation strength, high-order feature extraction and combination are performed on multiple first battery performance features and multiple second battery performance features to obtain multiple battery performance feature vectors.

[0061] Specifically, for multiple first battery performance characteristics F1 = [f 11 ,f 12 ,…,f 1m ] and multiple second battery performance characteristics F2=[f 21 ,f 22 ,…,f 2n Normalization was performed using the minimum-maximum normalization formula. Mapping all feature values ​​to the range [0,1] eliminates dimensional differences between feature values, improving the accuracy of subsequent clustering and analysis. Initial estimation of cluster centers in the feature space is performed based on the density peak detection algorithm. Assume the normalized feature matrix is ​​F = [F1; F2], where each row represents a feature sample and each column represents a normalized feature. In the density peak detection algorithm, the distance matrix D = [d...] between feature samples is calculated. ij ], where d ij =||f i -f j || represents the Euclidean distance between samples i and j. The local density of each sample is defined. Where δ is the distance kernel width parameter. Simultaneously, the minimum distance δ from each sample to samples with higher density is calculated. i =min j:ρj>ρi (d ij ). ρ i and δ i Combining density and distance scores, the sample with the highest combined score is selected as the cluster center, resulting in the first cluster centroid. This first centroid is then used as the initial condition input to the K-means clustering algorithm to optimize the cluster centers. K-means optimizes the cluster centers by minimizing the sum of squared intra-cluster distances. Iterative adjustment of cluster centers c using the objective function k In each iteration, samples are reassigned to the nearest cluster centers until the objective function converges. After iterative optimization, the second cluster centroid is obtained, representing a more accurate clustering result. Silhouette coefficient analysis is performed on the second cluster centroid to determine the optimal number of clusters K. * The profile coefficient S(i) is defined as follows: Where a(i) is the average distance from sample i to other samples in its cluster, and b(i) is the average distance from sample i to its nearest cluster. K is chosen by calculating the average silhouette coefficient of all samples. * This maximizes the profile coefficient and is based on K. * The final clustering is then performed. After clustering, the information entropy and weight coefficients of each feature are calculated using the entropy method. The information entropy formula is H(f j )=-∑ i p ij log(p ij ),in This represents the normalized probability distribution of feature j. Weight coefficients are calculated based on information entropy. Feature importance scores are obtained. These scores are then weighted and combined with the clustering results to generate the final feature representation. To optimize feature processing and reduce dimensionality, a locality-sensitive hashing (LSH) strategy is employed to generate the hash encoding. The LSH is used to perform dimensionality reduction mapping on the features, with the mapping formula being: Where a k It is a random vector, b k This is the offset; the generated hash code achieves efficient dimensionality reduction by preserving the local similarity of similar feature samples. The feature association strength matrix S after dimensionality reduction represents the similarity relationship between features. Based on the feature association strength, higher-order features are extracted and combined from the first and second battery performance features. Higher-order features are extracted through interaction terms and multi-level nonlinear transformations, generating multiple battery performance feature vectors. For example, suppose the feature vector is f. i =[f i1 ,f i2 ,f i3 ], by generating a secondary interaction term f ij ·f ik and nonlinear mapping φ(f i j)=ReLU(Wf ij +b), forming a higher-order feature matrix.

[0062] In one specific embodiment, the process of performing step 103 may specifically include the following steps:

[0063] Subspace partitioning is performed based on multiple battery performance feature vectors to obtain multiple independent feature subspaces;

[0064] Multiple independent feature subspaces are input into a battery reuse potential prediction model set. The battery reuse potential prediction model set includes multiple battery reuse potential prediction sub-models. Each battery reuse potential prediction sub-model includes a feature extraction layer, a feature mapping layer, and a prediction output layer. The feature extraction layer contains three fully connected layers, each using BatchNormalization and ReLU activation functions. The feature mapping layer contains two residual connected blocks, each containing a two-layer perceptron and skip connections. The prediction output layer uses the Sigmoid activation function.

[0065] The independent feature subspaces are input into the corresponding battery reuse potential prediction sub-models to obtain the sub-model prediction result set. The sub-model prediction result set is then dynamically weighted to obtain the fusion weight coefficients.

[0066] The prediction results of the sub-models are weighted and analyzed based on the fusion weight coefficients to output the comprehensive prediction results of battery reuse potential.

[0067] Specifically, based on multiple battery performance feature vectors F = [f1, f2, ..., f N The high-dimensional feature space is decomposed into multiple independent feature subspaces {S1, S2, ..., S} using a subspace partitioning algorithm. K Subspace partitioning is performed using principal component analysis or kernel methods. For example, principal component analysis calculates the covariance matrix. The eigenvalues ​​and eigenvectors are used to project the original features onto the first K principal components, forming a low-dimensional subspace matrix S. k =XW k W k This is the projection matrix of the corresponding subspace. Independent feature subspaces are input into an ensemble of battery reuse potential prediction models. The model ensemble consists of multiple sub-models, each focusing on an independent feature subspace. Each prediction sub-model consists of a feature extraction layer, a feature mapping layer, and a prediction output layer. In the feature extraction layer, three fully connected layers are used, with each layer calculated using the formula z. (l) =ReLU(BatchNorm(W (l) z (l-1) +b (l ))), where z (l) It is the output of the l-th layer, W (l) and b (l)These are the weight matrix and bias vector, respectively. BatchNorm represents the batch normalization operation, and ReLU(x) = max(0,x) is the activation function. This structure can effectively extract the nonlinear relationships of subspace features and improve training stability through normalization. The feature mapping layer consists of two residual connection blocks, each including a two-layer perceptron and skip connections. Assume the input is h. in The output of the residual block is h out =h in +MLP(h in MLP(h) = ReLU(W2ReLU(W1h+b1)+b2) is a two-layer perceptron. This structure alleviates the gradient vanishing problem through skip connections while enhancing feature extraction capabilities. In the prediction output layer, a sigmoid activation function is used to map the model's output values ​​to the [0,1] interval, outputting the predicted reuse potential value for each sub-model. in This is the prediction result of the k-th sub-model. By inputting all independent feature subspaces into their respective prediction sub-models, we obtain the set of sub-model prediction results. Dynamic weights are assigned to the prediction results set of the sub-models to reflect the relative importance of different sub-models. The weight assignment is performed using a weighted formula. Calculate, where MSE k ω is the mean squared error of the k-th sub-model on the validation data. k These are the corresponding fusion weight coefficients. The dynamic weight allocation method automatically adjusts the weights, allowing the better-performing sub-models to have a larger share in the overall prediction. Based on the fusion weight coefficients, a weighted analysis is performed on the set of sub-model prediction results to calculate the overall battery reuse potential prediction result. The formula for calculating the overall prediction result is as follows: in This represents the final comprehensive prediction result. The weighted approach achieves a comprehensive assessment of the battery reuse potential by fusing predictions from multiple sub-models.

[0068] In one specific embodiment, the process of performing step 104 may specifically include the following steps:

[0069] The comprehensive battery reuse potential prediction results are input into the first layer variational Bayesian inference network to model the data distribution. The posterior distribution of the latent variables is calculated by the variational expectation-maximization algorithm to obtain the battery performance distribution parameters.

[0070] The battery performance distribution parameters are input into the second-layer variational Bayesian inference network to analyze the battery performance evolution. The state transition probability matrix is ​​calculated by combining random field theory to obtain the battery performance degradation law.

[0071] The battery performance degradation law is input into the third layer variational Bayesian inference network to perform probabilistic modeling of the evaluation node parameters and obtain the evaluation node parameters.

[0072] Multi-stage evaluation and optimization are performed based on the evaluation node parameters to obtain the evaluation and optimization control strategy. Adaptive sampling is then performed on the evaluation and optimization control strategy to obtain the adaptive sampling scheme.

[0073] The adaptive sampling scheme is subjected to time-series feature extraction to obtain the dynamic adjustment matrix of evaluation parameters. Based on the dynamic adjustment matrix of evaluation parameters, the evaluation accuracy and evaluation efficiency are jointly optimized to obtain a multi-stage evaluation control scheme for the battery.

[0074] Specifically, the predicted result y of the comprehensive battery reuse potential is input into the first-layer variational Bayesian inference network for data distribution modeling. It is assumed that the generation of y is controlled by the latent variable z, and its joint probability distribution is p(y,z)=p(y|z)p(z). To estimate the posterior distribution q(z) of the latent variable, a variational Bayesian framework is used to approximate the posterior distribution as q(z)=Π i q i (z i The variational parameters are iteratively optimized using the variational expectation-maximization (EM) algorithm. The variational lower bound L(q) = E is calculated in the E-step. q(z) [logp(y,z)]-E q(z) [logq(z)], in the M-step, the model parameters are updated by maximizing L(q) to obtain the battery performance distribution parameters θ={μ,Σ}, where μ and Σ are the mean and covariance of the latent variables, respectively. The battery performance distribution parameters are then input into the second-layer variational Bayesian inference network to analyze the evolution of battery performance. Combining random field theory, it is assumed that the battery performance state is represented as a time series {s}. t}, where the transition of each state is determined by the conditional probability p(s). t+1 |s t Control. Using the battery performance distribution parameter θ, a random field model is used to define the state transition probability matrix P, whose elements P ij =p(s t+1 =j|s t =i) represents the transition probability from state i to state j. The expected value E[P] of the matrix elements is calculated using Bayesian inference. ijThe battery performance degradation law is obtained. This law describes the evolution trend of battery performance in the form of state transition probabilities. In the third-layer variational Bayesian inference network, the battery performance degradation law is input into the model to perform probabilistic modeling of the evaluation node parameters. The posterior distribution of the evaluation node parameter q(p) is optimized by combining the performance degradation law and the node observation data. Assuming that the prior distribution of the node parameter is p(p) and the observation data is o, its posterior distribution is q(p) ∝ p(o|p)p(p). After optimization by variational inference, the probability representation of the node parameter p={p1,p2,…,p k Each node parameter represents a weight for evaluating battery performance at different stages. Based on the evaluation node parameters, the multi-stage battery evaluation is optimized to generate an evaluation optimization control strategy. The optimization objective function is defined as follows: Where s t Let be the state distribution at stage t, and the optimization goal is to maximize the overall evaluation benefit. The optimal control strategy is solved using gradient descent or dynamic programming. Adaptive sampling is then performed on the optimized control strategy, generating an adaptive sampling scheme by adjusting the sampling frequency and strategy. Assuming the sampling strategy is π(t), the optimization problem is defined as minimizing the sampling cost. Where c t It is the sampling cost of stage t, while also satisfying the evaluation accuracy constraint. An adaptive sampling scheme that balances sampling cost and evaluation accuracy is generated through dynamic optimization. After generating the adaptive sampling scheme, temporal features are extracted to generate a dynamically adjusted evaluation parameter matrix A. The elements A of matrix A are... ij This represents the dynamic adjustment weight of parameter j at time i, and the specific calculation formula is as follows: The weights of the evaluation parameters are dynamically adjusted using this matrix. Based on the dynamic adjustment matrix A of the evaluation parameters, the evaluation accuracy and efficiency are jointly optimized to obtain a multi-stage evaluation control scheme for the battery. The optimization objective is defined as O = α·Precision(p) - β·Cost(p), where α and β are trade-off coefficients, and accuracy and cost are expressed as Precision(p) = E[||p] true -p|| 2 Cost(p) = ∑ t c t The optimal evaluation parameter configuration is solved using multi-objective optimization algorithms, such as genetic algorithms or particle swarm optimization.

[0075] The process involves inputting the battery performance degradation law into a third-layer variational Bayesian inference network to probabilistically model the evaluation node parameters, obtaining the evaluation node parameters; performing multi-stage evaluation optimization based on the evaluation node parameters to obtain an evaluation optimization control strategy; and adaptively sampling the evaluation optimization control strategy to obtain an adaptive sampling scheme. This includes: segmenting the battery performance degradation law; analyzing the battery capacity decay rate through a first battery performance evaluation unit to obtain a first evaluation parameter; analyzing the internal resistance change rate through a second battery performance evaluation unit to obtain a second evaluation parameter, thus obtaining segmented battery performance data; inputting the first and second evaluation parameters into a parameter correlation processing unit based on the segmented battery performance data; quantitatively analyzing the mutual influence between the parameters to obtain a parameter correlation index; constructing an evaluation reference model based on the parameter correlation index; obtaining a first optimization target value through quantitative index analysis within a first evaluation interval; and obtaining a second optimization target value through qualitative index analysis within a second evaluation interval. The evaluation reference standard is obtained; the evaluation reference standard is input into the multi-stage evaluation control unit, and the evaluation control boundary conditions are obtained by setting the battery performance degradation threshold, evaluation time threshold, and resource consumption threshold; based on the evaluation control boundary conditions, the first optimization target value and the second optimization target value are prioritized, and the first control parameter is obtained through multi-objective collaborative optimization, and the second control parameter is obtained through hierarchical progressive optimization, thus obtaining the optimized control strategy; the optimized control strategy is input into the adaptive sampling unit, and the first control parameter is dynamically adjusted in the time dimension to obtain the first sampling sequence, and the second control parameter is dynamically adjusted in the spatial dimension to obtain the second sampling sequence; an evaluation quality feedback mechanism is established based on the first sampling sequence and the second sampling sequence, and the sampling accuracy is calculated by the first evaluation unit to obtain the quality score, and the sampling efficiency is calculated by the second evaluation unit to obtain the resource utilization rate; the quality score and resource utilization rate are input into the optimization adjustment unit, and the sampling parameters are corrected and dynamically compensated online, and the adaptive sampling scheme is obtained through feedback iterative calculation.

[0076] In one specific embodiment, the battery reuse evaluation method for battery swapping cabinets further includes the following steps:

[0077] A target layer was constructed for a multi-stage evaluation and control scheme for batteries. Three primary indicators, namely capacity decay rate, internal resistance change rate and cycle life, were set by the analytic hierarchy process to obtain an evaluation index system.

[0078] Based on the evaluation index system, a judgment matrix is ​​constructed, the importance of the indicators is compared pairwise, and matrix normalization is performed to obtain the indicator weight vector.

[0079] A consistency check is performed on the indicator weight vector, and the consistency ratio is calculated using the eigenvalue calculation method. When the consistency ratio is less than the preset target value, the target weight coefficient is obtained.

[0080] Battery performance is weighted and calculated based on target weight coefficients to obtain battery grading results. Scene feature matching is then performed based on the battery grading results to obtain a scene matching degree matrix.

[0081] The optimal solution is searched using the scenario matching degree matrix to obtain the scenario priority ranking. Based on the scenario priority ranking, the allocation scheme is checked through constraint conditions to ensure that it meets the performance requirements and safety standards of each scenario, thus obtaining the battery reuse execution scheme.

[0082] Specifically, a target layer is constructed for the multi-stage evaluation and control scheme of the battery, setting three primary indicators: capacity decay rate, internal resistance change rate, and cycle life. These three indicators represent the rate of battery performance degradation, the impact of internal resistance change on battery efficiency and heat generation, and the actual service life of the battery, respectively, comprehensively characterizing the battery's health status. In the analytic hierarchy process (AHP), the evaluation index system needs to clearly define the target layer, criterion layer, and scheme layer. The target layer assesses the battery's reuse potential, the criterion layer consists of the three primary indicators, and the scheme layer represents the actual selectable scenarios. Based on the established evaluation index system, a judgment matrix A = [a ij ], where a ij This indicates the importance of indicator i relative to indicator j. After construction, the judgment matrix is ​​a positive reciprocal matrix, satisfying a ij =1 / a ji And a ii =1. For the three primary indicators, the judgment matrix is ​​in the form of:

[0083]

[0084] Calculate the weight vector w = [w1, w2, w3] for each indicator using matrix normalization. T The normalization process involves summing the elements in each column, then dividing each element by the sum of its column, resulting in a normalized matrix A′ = [a′]. ij ],in After normalization, the mean of each row is calculated to obtain the weight vector. Here, n is the number of indicators. A consistency check is performed on the calculated indicator weight vector to ensure the rationality of the judgment matrix. The consistency ratio is calculated, and the largest eigenvalue λ of the judgment matrix is ​​determined. max Through formula Calculate, where (Aw) i It is the i-th element of the matrix-vector product. Then, the consistency index (Cl) is calculated using the formula: The final consistency rate is Here, RI is the random consistency index, and its value depends on the order of the matrix. When CR < 0.1, the consistency of the judgment matrix is ​​considered acceptable, and the weight vector is the target weight coefficient. Based on the target weight coefficient, the battery performance is weighted and calculated to obtain the battery grading result. Let the battery performance index matrix be P = [p ij ], where p ij The value of the i-th battery on the j-th index represents the battery grading result S = [s1, s2, ..., sj]. m [Through formula s] i =∑ j w j p ij Calculate, where s i This represents the overall performance score of the i-th battery. Based on the battery grading results, scene feature matching is performed to generate a scene matching degree matrix. Assume the scene feature matrix is ​​C = [c kj ], where c kj The scenario matching degree matrix M represents the demand weight of scenario k for indicator j. ik [Through formula m] ik =∑ j p ij c kj Calculate, where m ik This represents the degree of matching between battery i and scene k. Using the scene matching degree matrix, an optimization algorithm is used to search for the optimal solution to determine the scene priority ranking. The optimization objective function is defined as Q = ∑ i,k x ik m ik , where x ik It is an allocation variable; if battery i is allocated to scene k, then x ik =1, otherwise x ik =0. The optimization process needs to meet constraints, such as each battery can only be assigned to one scenario: And the total battery performance for each scenario must not be lower than the safety requirements: Where τ k This represents the performance threshold for scenario k. The optimization problem is solved using linear programming or integer programming to obtain a scenario priority ranking. Based on the scenario priority ranking and constraint checks, the generated battery reuse execution plan is ensured to meet all performance requirements and safety standards, thus obtaining the battery reuse execution plan.

[0085] After obtaining the battery reuse implementation plan, the process also includes:

[0086] The batteries in the battery reuse implementation plan are classified for tiered utilization. By analyzing the remaining capacity, internal resistance growth rate, and cycle count, the batteries are divided into Class A tiered utilization group, Class B tiered utilization group, and recycling group, resulting in a tiered classification result. Based on the tiered classification result, backup power supply adaptability tests are conducted on the Class A tiered utilization group batteries. Adaptability parameters for a 48V communication base station backup power supply are obtained through 1C constant current discharge and 0.5C constant current charge tests. Low-voltage energy storage system adaptability tests are also conducted on the Class A tiered utilization group batteries. Adaptability parameters for a 220V home energy storage system are obtained through 0.3C constant power charge / discharge and voltage consistency tests. Photovoltaic energy storage adaptability tests are conducted on the Class B tiered utilization group batteries. Adaptability parameters for a 220V home energy storage system are obtained through 0.2C intermittent charge / discharge and environmental adaptability tests. The system includes: adapting parameters for photovoltaic energy storage systems; pre-treating the recycled batteries before dismantling them through forced discharge, insulation testing, and pressure relief to obtain dismantling safety parameters; automating the dismantling of the recycled batteries using mechanical separation and material sorting to obtain four types of recycled components: positive electrode material, negative electrode material, separator, and electrolyte; extracting metal elements from the positive electrode material and separating and purifying nickel, cobalt, and manganese through hydrometallurgical processes to obtain battery metal raw materials; purifying the negative electrode material using graphite to obtain carbon material raw materials; chemically treating the separator and electrolyte through solvent extraction and high-temperature pyrolysis processes to obtain renewable organic materials; establishing recycling records for battery metal raw materials, carbon material raw materials, and renewable organic materials; determining recovery rate indicators through material balance calculations to form a closed-loop recycling system.

[0087] The battery reuse evaluation method for battery swapping cabinets in embodiments of the present invention has been described above. The battery reuse evaluation system for battery swapping cabinets in embodiments of the present invention will be described below. Please refer to [link / reference]. Figure 2 One embodiment of the battery reuse evaluation system for battery swapping cabinets in this invention includes:

[0088] The data acquisition module 201 is used to deploy a distributed detection device network in the battery cluster of the battery swapping cabinet and to collect the raw battery dataset of the battery cluster.

[0089] Encoding module 202 is used to perform feature classification and vector encoding on the original battery dataset to obtain multiple battery performance feature vectors;

[0090] Prediction module 203 is used to input multiple battery performance feature vectors into a set of battery reuse potential prediction models for prediction and output a comprehensive battery reuse potential prediction result.

[0091] The calculation module 204 is used to set the evaluation node parameters based on the prediction results of the comprehensive battery reuse potential, and to perform adaptive sampling and dynamic optimization calculation on the evaluation node parameters to obtain a multi-stage evaluation and control scheme for the battery.

[0092] Through the collaborative efforts of the aforementioned components, and by deploying a distributed detection network combined with a multi-level data acquisition and processing mechanism, this invention achieves efficient acquisition and analysis of battery lifecycle data for battery swapping cabinets, significantly improving the completeness and accuracy of data acquisition. Employing an evaluation method combining deep learning and variational Bayesian inference, and utilizing multi-level feature extraction and dynamic optimization strategies, the accuracy of battery performance evaluation is improved. A multi-objective optimization-based tiered utilization decision mechanism, combined with scenario feature matching and resource optimization allocation strategies, improves the matching accuracy for battery reuse. Through adaptive sampling and dynamic optimization control, this invention can automatically adjust the evaluation strategy according to changes in battery performance. The introduction of a multi-layer variational Bayesian inference network, combined with analysis of battery performance degradation patterns, enhances the reliability of the evaluation results. By integrating multiple prediction sub-models and a dynamic weight allocation mechanism, this invention achieves a comprehensive evaluation of battery reuse potential, improving prediction accuracy and exhibiting strong generalization ability.

[0093] above Figure 2 The battery reuse evaluation system for the battery swapping cabinet in this embodiment of the invention is described in detail from the perspective of modular functional entities. The battery reuse evaluation equipment for the battery swapping cabinet in this embodiment of the invention is described in detail from the perspective of hardware processing.

[0094] Figure 3 This is a schematic diagram of the structure of a battery recycling evaluation device for a battery swapping cabinet provided in an embodiment of the present invention. The battery recycling evaluation device 300 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors) and a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) for storing application programs 333 or data 332. The memory 320 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the battery recycling evaluation device 300. Furthermore, the processor 310 may be configured to communicate with the storage media 330 and execute the series of instruction operations in the storage media 330 on the battery recycling evaluation device 300 to implement the steps of the aforementioned battery recycling evaluation method for the battery swapping cabinet.

[0095] The battery recycling evaluation device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The battery recycling evaluation device structure shown does not constitute a limitation on the battery recycling evaluation device provided by the present invention. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0096] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the battery reuse evaluation method for the battery swapping cabinet.

[0097] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0098] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0099] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for evaluating the reuse of batteries in a battery swapping cabinet, characterized in that, The method includes: A distributed detection device network is deployed in the battery cluster of the battery swapping cabinet, and the raw battery dataset of the battery cluster is collected. The original battery dataset is subjected to feature classification and vector encoding to obtain multiple battery performance feature vectors; The process involves inputting multiple battery performance feature vectors into a set of battery reuse potential prediction models for prediction, and outputting a comprehensive battery reuse potential prediction result. This includes: partitioning the multiple battery performance feature vectors into subspaces to obtain multiple independent feature subspaces; inputting these independent feature subspaces into the set of battery reuse potential prediction models, which includes multiple battery reuse potential prediction sub-models. Each sub-model comprises a feature extraction layer, a feature mapping layer, and a prediction output layer. The feature extraction layer contains three fully connected layers, each using BatchNormalization and ReLU activation functions; the feature mapping layer contains two residual connection blocks, each containing a two-layer perceptron and skip connections; and the prediction output layer uses a Sigmoid activation function. The independent feature subspaces are then input into their corresponding battery reuse potential prediction sub-models to obtain a set of sub-model prediction results. Dynamic weight allocation is performed on this set of sub-model prediction results to obtain fusion weight coefficients. Finally, a weighted analysis is performed on the sub-model prediction results set based on these fusion weight coefficients to output a comprehensive battery reuse potential prediction result. Based on the comprehensive battery reuse potential prediction results, evaluation node parameters are set, and adaptive sampling and dynamic optimization calculations are performed on the evaluation node parameters to obtain a multi-stage battery evaluation control scheme. This includes: inputting the comprehensive battery reuse potential prediction results into a first-layer variational Bayesian inference network for data distribution modeling; calculating the posterior distribution of latent variables using a variational expectation-maximization algorithm to obtain battery performance distribution parameters; inputting the battery performance distribution parameters into a second-layer variational Bayesian inference network for battery performance evolution analysis; calculating the state transition probability matrix using random field theory to obtain the battery performance degradation law; inputting the battery performance degradation law into a third-layer variational Bayesian inference network for probability modeling to obtain evaluation node parameters; performing multi-stage evaluation optimization based on the evaluation node parameters to obtain an evaluation optimization control strategy; adaptively sampling the evaluation optimization control strategy to obtain an adaptive sampling scheme; extracting time-series features from the adaptive sampling scheme to obtain a dynamic adjustment matrix for evaluation parameters; and jointly optimizing evaluation accuracy and evaluation efficiency based on the dynamic adjustment matrix to obtain a multi-stage battery evaluation control scheme.

2. The battery reuse evaluation method for battery swapping cabinets according to claim 1, characterized in that, The deployment of a distributed detection device network in the battery cluster of the battery swapping cabinet, and the collection of the raw battery dataset of the battery cluster, includes: Based on geographical location information and usage environment parameters, the battery cluster of the battery swapping cabinet is grouped and distributed nodes are optimized to obtain a distributed detection device network. The distributed detection device network is synchronized with data sampling period and timestamp alignment to obtain synchronization sampling control parameters; Based on the synchronous sampling control parameters, the battery voltage, current and temperature parameters are sampled at high frequency to obtain battery operating condition data. Then, the battery operating condition data is subjected to time domain feature analysis and frequency domain feature analysis to obtain battery dynamic characteristic data. Based on the battery dynamic characteristic data, a voltage and current response curve is constructed, and the battery internal resistance change value is calculated through the voltage and current response curve. At the same time, the internal resistance parameter of the battery is obtained by combining the temperature compensation coefficient. The charging and discharging process of the batteries in the battery cluster of the battery swapping cabinet is recorded. The charging and discharging capacity is calculated by the charge integration method, and the number of cycles is counted by the coulomb counting method to obtain battery usage record data. The battery operating condition data, battery dynamic characteristic data, battery internal resistance parameters, and battery usage record data are time-series merged and data standardization processed to obtain the original battery dataset.

3. The battery reuse evaluation method for battery swapping cabinets according to claim 2, characterized in that, The process of performing feature classification and vector encoding on the original battery dataset yields multiple battery performance feature vectors, including: Based on the original battery dataset, the voltage, current, and temperature parameters are reduced in dimensionality to obtain a battery parameter dimensionality reduction matrix. Then, a time-series correlation analysis is performed on the battery parameter dimensionality reduction matrix to obtain time-series feature weight coefficients. The time-series feature weight coefficients are weighted and fused with the battery parameter dimension reduction matrix to obtain multiple first battery performance features. The capacity data in the original battery dataset is decomposed into a time series, and the capacity decay curve is fitted by a double exponential function. The temperature stress coefficient is then used for correction to obtain the capacity decay model parameters. Based on the parameters of the capacity decay model, a battery capacity prediction function is constructed, and the capacity decay rate under different operating conditions is calculated by Taylor expansion to obtain the predicted capacity decay value. The internal resistance data in the original battery dataset is segmented, and an internal resistance change function is established by a piecewise linear regression method. The parameters are then optimized by combining historical data to obtain the internal resistance change model coefficients. The internal resistance change model coefficients are input into the prediction function, and the predicted internal resistance values ​​at different time points are obtained through recursive calculation. The predicted internal resistance change values ​​are then corrected by combining the temperature correction coefficient. The predicted values ​​of capacity decay and internal resistance change are processed by feature vectorization to obtain multiple second battery performance characteristics. The plurality of first battery performance features and the plurality of second battery performance features are classified and vector encoded to obtain a plurality of battery performance feature vectors.

4. The battery reuse evaluation method for battery swapping cabinets according to claim 3, characterized in that, The step of performing feature classification and vector encoding on the plurality of first battery performance features and the plurality of second battery performance features to obtain a plurality of battery performance feature vectors includes: The performance characteristics of the plurality of first batteries and the performance characteristics of the plurality of second batteries are normalized, and the initial value of the cluster center of the feature space is calculated by the density peak detection algorithm to obtain the first cluster centroid. The first cluster centroid is input into the K-means algorithm, and the second cluster centroid is obtained by iterative calculation using the optimization objective function of minimizing intra-cluster distance and maximizing inter-cluster distance; The silhouette coefficient analysis is performed on the second cluster centroid to obtain the optimal number of cluster categories. Based on the optimal number of cluster categories, clustering is performed. The information entropy and weight coefficient of each feature are calculated by the entropy method to obtain the feature importance score. The feature importance score is weighted and combined with the clustering results, a hash encoding strategy is generated using a locality-sensitive hash function, and a dimensionality reduction mapping is performed based on the hash encoding strategy to obtain the feature association strength. Based on the feature correlation strength, high-order feature extraction and combination are performed on the plurality of first battery performance features and the plurality of second battery performance features to obtain a plurality of battery performance feature vectors.

5. The battery reuse evaluation method for battery swapping cabinets according to claim 1, characterized in that, The battery reuse evaluation method for the battery swapping cabinet also includes: The target layer of the multi-stage evaluation and control scheme for the battery is constructed. Three primary indicators, namely capacity decay rate, internal resistance change rate and cycle life, are set by the analytic hierarchy process to obtain the evaluation index system. Based on the evaluation index system, a judgment matrix is ​​constructed, the importance of the indicators is compared pairwise, and matrix normalization is performed to obtain the indicator weight vector. A consistency check is performed on the indicator weight vector, and the consistency ratio is calculated using the eigenvalue calculation method. When the consistency ratio is less than the preset target value, the target weight coefficient is obtained. The battery performance is weighted and calculated based on the target weight coefficients to obtain the battery grading results. Then, scene feature matching is performed based on the battery grading results to obtain the scene matching degree matrix. The optimal solution is searched using the scenario matching degree matrix to obtain the scenario priority ranking. Based on the scenario priority ranking, the allocation scheme is checked through constraint conditions to ensure that it meets the performance requirements and safety standards of each scenario, thus obtaining the battery reuse execution scheme.

6. A battery reuse evaluation system for battery swapping cabinets, characterized in that, The system for performing the battery reuse evaluation method for battery swapping cabinets as described in any one of claims 1-5, the system comprising: The data acquisition module is used to deploy a distributed detection device network in the battery cluster of the battery swapping cabinet and to collect the raw battery dataset of the battery cluster. The encoding module is used to perform feature classification and vector encoding on the original battery dataset to obtain multiple battery performance feature vectors. The prediction module is used to input the multiple battery performance feature vectors into the battery reuse potential prediction model set for prediction, and output the comprehensive battery reuse potential prediction result. The calculation module is used to set evaluation node parameters based on the comprehensive battery reuse potential prediction results, and to perform adaptive sampling and dynamic optimization calculation on the evaluation node parameters to obtain a multi-stage battery evaluation and control scheme.

7. A battery recycling evaluation device for battery swapping cabinets, characterized in that, The battery recycling evaluation device for the battery swapping cabinet includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the battery swapping cabinet battery reuse evaluation device to perform the battery swapping cabinet battery reuse evaluation method as described in any one of claims 1-5.

8. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the battery reuse evaluation method for battery swapping cabinets as described in any one of claims 1-5.

Images