Feature-driven method and system for estimating state of health of retired power battery

Abstract

The application provides a feature-driven method and system for estimating the health state of a retired power battery, and relates to the field of battery health state estimation.The method comprises the following steps: constructing an algorithm library and presetting the capability vector of each algorithm in the algorithm library; acquiring the voltage and current of the retired power battery at the current time, combining a preset battery equivalent circuit model to acquire the feature vector at the current time; based on the feature vector, calculating the matching score of each algorithm in the algorithm library through a dynamic weighted scoring mechanism, and determining the current execution algorithm according to the matching score; using the current execution algorithm to perform parameter identification on the battery equivalent circuit model, obtaining the optimal model parameters, and calculating the current health state of the retired power battery according to the optimal model parameters; and calculating the convergence performance of the current execution algorithm to update the historical credibility of the algorithm in the algorithm library, so as to be used for the matching score calculation at the next time.The application can adapt to various working conditions, and improve the estimation accuracy and convergence speed.

Description

Technical Field

[0001] This invention relates to the field of battery health state estimation, and in particular to a feature-driven method and system for estimating the health state of retired power batteries. Background Technology

[0002] With the increasing popularity of electric vehicles, the number of retired power batteries is surging year by year. Retired batteries typically face the challenge of exhibiting unique characteristics, with complex internal aging mechanisms and model parameters (such as ohmic internal resistance and polarization resistance / capacitance) changing nonlinearly with temperature, state of charge (SOC), and degree of aging. Existing SOH estimation methods mostly employ parameter identification methods based on equivalent circuit models (ECM). However, current technologies typically use a single optimization algorithm (such as Standard Particle Swarm Optimization, PSO), but no single algorithm can adapt to all the complex and varied operating conditions of retired batteries. Using a single algorithm indefinitely often leads to unstable estimation accuracy, slow convergence speed, or even divergence when facing different scenarios such as drastic changes in operating conditions, stable operation, or high noise interference. Summary of the Invention

[0003] The main objective of this invention is to propose a feature-driven method and system for estimating the health status of retired power batteries, which can be adapted to various operating conditions and improve estimation accuracy and convergence speed.

[0004] This invention is achieved through the following technical solution:

[0005] The feature-driven method for estimating the health status of decommissioned power batteries includes the following steps:

[0006] Step S1: Construct an algorithm library and pre-set the capability vectors of each algorithm in the library. The algorithm library shall contain at least global exploratory algorithms and local development algorithms.

[0007] Step S2: Obtain the voltage and current of the retired power battery at the current moment, and combine it with the preset battery equivalent circuit model to obtain the feature vector at the current moment. The feature vector includes the prediction residual of the battery equivalent circuit model, data volatility and algorithm stagnation factor.

[0008] Step S3: Based on the feature vector, calculate the matching score of each algorithm in the algorithm library through a dynamic weighted scoring mechanism. When the highest matching score is greater than the matching score of the currently executed algorithm, select the algorithm with the highest matching score as the new currently executed algorithm; otherwise, the currently executed algorithm remains unchanged.

[0009] Step S4: Use the current execution algorithm to identify the parameters of the battery equivalent circuit model, obtain the optimal model parameters, and calculate the current health status of the retired power battery based on the optimal model parameters.

[0010] Step S5: Calculate the convergence performance of the currently executed algorithm to update the historical reputation of the algorithm in the algorithm library for use in the matching score calculation at the next time step.

[0011] Furthermore, in step S1, the algorithm library includes global exploratory algorithms, local exploitation algorithms, and noise-resistant robust algorithms. The global exploratory algorithms include the adaptive spiral flight sparrow search algorithm, the local exploitation algorithms include the transient triangular Harris Eagle optimization algorithm, and the noise-resistant robust algorithms include the hybrid genetic particle swarm optimization algorithm.

[0012] Furthermore, in step S1, the capability vector of the k-th algorithm is represented as follows: , Let represent the weight factors of the k-th algorithm in global exploration, local exploitation, and diversity maintenance, respectively. , If the k-th algorithm is a global exploratory algorithm, then The maximum, if the k-th algorithm is a locally exploitable algorithm, then The maximum, if the k-th algorithm is a diversity-maintaining algorithm, then maximum.

[0013] Furthermore, in step S2, the predicted residual is expressed as... The data volatility is the standard deviation of the current over a predetermined time period prior to the current moment, and the algorithm stall factor is determined based on the change in the objective function value of the currently executing algorithm. The theoretical voltage value is obtained based on the battery equivalent circuit model. To obtain the voltage at the current time t, Open circuit voltage, , The electrochemical polarization voltage and concentration polarization voltage at the current time t are given by the electrochemical polarization resistance parameter identified at the previous time. Electrochemically polarized capacitors Concentration polarization resistance Concentration polarization capacitor get, The ohmic resistance identified in the previous moment. To obtain the current at the current time t.

[0014] Furthermore, step S1 also includes the initialization of feature vectors and historical reputation scores. In the initialization of feature vectors, the prediction residual is set to a maximum value, the data volatility and the algorithm stagnation factor are both set to 0, and the historical reputation scores of each algorithm in the algorithm library are set to the same value.

[0015] Furthermore, in step S3, the matching score of the k-th algorithm in the algorithm library is determined according to the formula... Calculate, where, As a weight for global search demand, Weighting based on local development needs. Weighting is required for noise resistance robustness. Let be the historical reputation score of the k-th algorithm at the current time t. The historical reputation weight is set.

[0016] Furthermore, in step S3, the global search demand weight is: , The preset error threshold, The stagnation factor of the algorithm. , For adjustment coefficients, The weight of the local development requirement is The noise resistance robustness requirement weight is , This is a sequence of current data over a predetermined time period, starting from the current moment. To calculate the current data sequence standard deviation The reference volatility is set.

[0017] Furthermore, in step S5, when the k-th algorithm is the algorithm executed at the previous time step, the historical reputation of the k-th algorithm at the current time t is... When the k-th algorithm is not the algorithm executed in the previous time step, the historical reputation score of the k-th algorithm at the current time t is the same as the historical reputation score at the previous time step. Forgetting factor, Based on the historical credibility of the previous moment, The performance score of the algorithm executed at the previous time step. The root mean square error of the algorithm executed in the previous moment. This is the maximum value of the root mean square error. The computation time taken to execute the algorithm in the previous moment. This is the maximum set computation time. , for.

[0018] Furthermore, the battery equivalent circuit model adopts a second-order RC equivalent circuit model.

[0019] This invention is also achieved through the following technical solutions:

[0020] A feature-driven health state estimation system for decommissioned power batteries is provided to implement any of the feature-driven health state estimation methods for decommissioned power batteries described above, including:

[0021] Algorithm library construction module: used to build algorithm libraries and pre-set the capability vectors of each algorithm in the algorithm library. The algorithm library must contain at least global exploratory algorithms and local development algorithms;

[0022] Feature vector acquisition module: used to acquire the current voltage and current of the retired power battery, and combine it with the preset battery equivalent circuit model to acquire the feature vector at the current moment. The feature vector includes the prediction residual of the battery equivalent circuit model, data volatility and algorithm stagnation factor.

[0023] Algorithm confirmation module: Based on feature vectors, it calculates the matching score of each algorithm in the algorithm library through a dynamic weighted scoring mechanism, and selects the algorithm with the highest matching score as the current execution algorithm;

[0024] Parameter identification module: Used to identify the parameters of the battery equivalent circuit model using the currently executed algorithm, obtain the optimal model parameters, and calculate the current health status of the retired power battery based on the optimal model parameters;

[0025] Historical reputation calculation module: Used to record the convergence performance parameters of the currently executing algorithm, so as to update the historical reputation of each algorithm in the algorithm library for use in the matching score calculation at the next moment.

[0026] As can be seen from the above description of the present invention, compared with the prior art, the present invention has the following beneficial effects:

[0027] This invention constructs an algorithm library containing at least global exploratory algorithms and local development algorithms. Based on the feature vector of the retired power battery at the current moment, a dynamic weighted scoring mechanism is used to calculate the matching score of each algorithm in the library. The algorithm with the highest matching score is selected as the current execution algorithm. The current execution algorithm is used to identify the parameters of the battery's equivalent circuit model to obtain the current health status of the retired power battery. The historical reputation of the current execution algorithm is updated based on its convergence performance for use in calculating the matching score at the next moment. This achieves automatic selection of appropriate estimation algorithms according to specific operating conditions, thus adapting to various operating conditions and effectively addressing the challenges of large parameter drift and unknown initial values ​​in retired power batteries, improving estimation accuracy and convergence. Furthermore, it can automatically switch to lightweight algorithms under stable operating conditions, thereby significantly reducing the computational load on the BMS. By expanding the types or number of algorithms in the algorithm library, it can be applied to different types of batteries (such as LFP / NCM). Attached Figure Description

[0028] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] Figure 1 This is a flowchart of the present invention.

[0030] Figure 2This is a detailed flowchart of the present invention.

[0031] Figure 3 The curves show the SOH estimation error comparison between the present invention and the comparative method under dynamic operating conditions. Detailed Implementation

[0032] The present invention will be further described below through specific embodiments.

[0033] like Figure 1 and Figure 2 As shown, the feature-driven method for estimating the health status of decommissioned power batteries includes the following steps:

[0034] Step S1: Construct an algorithm library and pre-set the capability vectors of each algorithm in the library. The algorithm library shall contain at least global exploratory algorithms and local development algorithms.

[0035] In this embodiment, the algorithm library includes global exploration algorithms, local exploitation algorithms, and noise-resistant robust algorithms. The global exploration algorithm includes the Adaptive Spiral Flying Sparrow Search Algorithm (ASFSSA), which has strong global search capabilities and is suitable for stages where initial values ​​are unknown. The local exploitation algorithm includes the Transient Triangular Harris Eagle Optimization Algorithm (IHHO), which has strong local convergence capabilities and is suitable for the parameter fine-tuning stage. The noise-resistant robust algorithm includes the Mixed Genetic Particle Swarm Optimization Algorithm (Mix-PSO), which has strong noise resistance and is suitable for stages with large current fluctuations. Under other operating conditions, the Gaussian Mutation White Whale Optimization Algorithm can also be added to the algorithm library. This algorithm has strong versatility and is suitable for escaping local optima.

[0036] The capability vector of the k-th algorithm is represented as: The capability matrix is ​​composed of the capability vectors of all algorithms. Let represent the weight factors of the k-th algorithm in global exploration, local exploitation, and diversity maintenance, respectively. , If the k-th algorithm is a global exploratory algorithm, then Maximum, specifically can be set to If the k-th algorithm is a locally exploitable algorithm, then Maximum, specifically can be set to If the k-th algorithm is a diversity-maintaining algorithm, then Maximum, specifically can be set to .

[0037] Step S2: Obtain the voltage and current of the retired power battery at the current moment, and combine it with the preset battery equivalent circuit model to obtain the feature vector at the current moment. The feature vector includes the prediction residual of the battery equivalent circuit model, data volatility and algorithm stagnation factor.

[0038] The predicted residual is expressed as ,in, The theoretical voltage value is obtained based on the battery equivalent circuit model, which is a second-order RC equivalent circuit model. To obtain the terminal voltage at the current time t, Open circuit voltage, , The electrochemical polarization voltage and concentration polarization voltage at the current time t are given by the electrochemical polarization resistance parameter identified at the previous time. Electrochemically polarized capacitors Concentration polarization resistance Concentration polarization capacitor get, The ohmic resistance identified in the previous moment. To obtain the current at the current time t.

[0039] Data volatility is the standard deviation of the current over a specified time period (e.g., 30 seconds) prior to the current moment. Data volatility reflects the severity of the operating conditions; models are prone to failure under severe conditions, requiring algorithms with strong anti-interference capabilities.

[0040] The algorithm stall factor is determined based on the change in the objective function value of the currently executing algorithm. If the change in the objective function value is extremely small (close to 0) in the past N iterations, it means that the algorithm is "stuck" and the extracted algorithm stall factor will be very small.

[0041] At t=0, there is no historical data, no parameters from the previous time step, and no historical reputation score. Therefore, the initial state needs to be considered; that is, in step S1, the feature vector and historical reputation score also need to be initialized. During feature vector initialization, the prediction residual is forcibly set to a maximum value (e.g., 100mV). This "tricks" the system into thinking the current error is huge, thus forcibly triggering an increase in the global search weight. The data volatility and algorithm stagnation factor are both set to 0, and the historical reputation scores of all algorithms in the algorithm library are set to the same value (e.g., all 0.5) to ensure fair competition. This implementation ensures that the system will directly select a global exploratory algorithm (e.g., ASFSSA) in the first second to determine the parameter baseline value.

[0042] Step S3: Based on the feature vector, calculate the matching score of each algorithm in the algorithm library through a dynamic weighted scoring mechanism. When the highest matching score is greater than the matching score of the currently executed algorithm, select the algorithm with the highest matching score as the new currently executed algorithm; otherwise, the currently executed algorithm remains unchanged.

[0043] The matching score of the k-th (k=1,2,3) algorithm in the algorithm library is calculated according to the formula. Calculate, where, As a weight for global search demand, Weighting based on local development needs. Weighting is required for noise resistance robustness. Let be the historical reputation score of the k-th algorithm at the current time t. The historical reputation weight is set (e.g., 0.3).

[0044] The global search demand weight is , The preset error threshold (e.g., 20mV). The stagnation factor of the algorithm. , For adjustment coefficient and (e.g., both are 0.5), used to balance the importance of error and stagnation factors.

[0045] The calculation logic is as follows: when the error When the function value approaches 0, it indicates that the error is controllable and a global search is not needed; when the error... When the function value rapidly approaches 1, it indicates that the error is too large, and a global search must be initiated. This is a "soft switch" that smoothly activates the global algorithm by increasing the weights when the error exceeds the limit.

[0046] The computational logic is as follows: if the algorithm keeps evolving, It's very big. If the algorithm gets stuck (into a local optimum), it will not trigger a global search. ,but The weights are increased. When the existing algorithm is found to be unable to solve the problem, a forced switch to a global algorithm is made to "disrupt" the process and break the infinite loop.

[0047] The weight of local development requirements is When the voltage error is very small and has not stalled, the system determines that the current parameters are relatively accurate. At this point... To improve performance, the system prioritizes the fastest local development algorithm for fine-tuning to save computing power.

[0048] The weight of noise resistance robustness requirement is It measures the ratio of current current fluctuations to the "maximum permissible fluctuation". A sequence of current data over a predetermined time period, starting from the current time. , To calculate the current data sequence The standard deviation (if the vehicle is cruising at a constant speed and the current is stable, the standard deviation is very small; if the vehicle is frequently accelerating and decelerating rapidly, the standard deviation is very large). The reference volatility is set as an empirical constant, such as based on test data of this vehicle model under aggressive driving conditions. .

[0049] Historical credibility This indicates the algorithm's past performance. To prevent frequent switching, although algorithm k has a slightly lower theoretical matching degree, it has always performed very stably and reliably in the past. The system may continue to use it due to the reputation bonus, thus ensuring the smoothness of the system.

[0050] The algorithm with the highest matching score is compared with the currently executing algorithm based on the formula. conduct, The preset switching hysteresis threshold typically ranges from 0.05 to 0.15. The introduction of this method aims to avoid the switching oscillation phenomenon caused by frequent fluctuations in the scores of various algorithms in the algorithm library due to microwave operating conditions, thus ensuring the smoothness and computational efficiency of the estimation system. In this embodiment, The preferred setting is 0.1.

[0051] Step S4: Use the current execution algorithm to identify the parameters of the battery equivalent circuit model, obtain the optimal model parameters, and calculate the current health status of the retired power battery based on the optimal model parameters.

[0052] The specific process of parameter identification using the current execution algorithm is existing technology.

[0053] Step S5: Calculate the convergence performance of the currently executed algorithm to update the historical reputation of the algorithm in the algorithm library for use in the matching score calculation at the next time step.

[0054] When the k-th algorithm is the algorithm executed in the previous time step, the historical reputation of the k-th algorithm at the current time t is: When the k-th algorithm is not the algorithm executed in the previous time step, the historical reputation score of the k-th algorithm at the current time t is the same as the historical reputation score at the previous time step. The forgetting factor indicates whether more emphasis is placed on recent performance or long-term historical performance, and its value ranges from [value range missing]. This is used to adjust the algorithm's preference for historical and recent performance when updating reputation. Based on the historical credibility accumulated in the previous moment, The performance score of the algorithm executed at the previous time step. The root mean square error of the algorithm executed in the previous moment. This is the maximum value of the root mean square error. The computation time taken to execute the algorithm in the previous moment. This is the maximum set computation time. , The preset performance evaluation weighting coefficients satisfy... , Used to adjust the degree of influence of estimation accuracy on reputation updates. Used to adjust the impact of algorithm computation time on reputation updates. and The specific calculation process is based on existing technology.

[0055] Forgetting factor If one tends to prioritize long-term historical performance to filter out transient noise, It is advisable to take a smaller value, such as If one prioritizes recent performance for rapid response to drastic changes in operating conditions, It is advisable to take a larger value, such as In this embodiment, to balance the stability of the algorithm evaluation with the agility of the response, The preferred setting is 0.3. This means the current performance score. The latest reputation score accounts for 30% of the weight, while historical accumulation accounts for 70%, thus ensuring the smoothness of the evaluation while giving the system the sensitivity to the performance degradation of the recognition algorithm.

[0056] Feature-driven health status estimation systems for retired power batteries include:

[0057] Algorithm library construction module: used to build algorithm libraries and pre-set the capability vectors of each algorithm in the algorithm library. The algorithm library must contain at least global exploratory algorithms and local development algorithms;

[0058] Feature vector acquisition module: used to acquire the current voltage and current of the retired power battery, and combine it with the preset battery equivalent circuit model to acquire the feature vector at the current moment. The feature vector includes the prediction residual of the battery equivalent circuit model, data volatility and algorithm stagnation factor.

[0059] Algorithm confirmation module: Based on feature vectors, it calculates the matching score of each algorithm in the algorithm library through a dynamic weighted scoring mechanism, and selects the algorithm with the highest matching score as the current execution algorithm;

[0060] Parameter identification module: Used to identify the parameters of the battery equivalent circuit model using the currently executed algorithm, obtain the optimal model parameters, and calculate the current health status of the retired power battery based on the optimal model parameters;

[0061] Reputation calculation module: Used to record the convergence performance parameters of the currently executing algorithm, so as to update the reputation of each algorithm in the algorithm library for use in the matching score calculation at the next time step.

[0062] like Figure 3 The error comparison curves of the present invention and the comparison method (using a single estimation algorithm) are shown. Table 1 shows the process of the present invention automatically switching algorithms according to the working conditions:

[0063] Table 1

[0064]

[0065] like Figure 3 As shown in Table 1, compared with the comparative method, the average error of SOH estimation of the method of the present invention is reduced from 1.45% to 0.82%, and the total calculation time is reduced by 42%, which proves the dual advantages of the present invention in terms of accuracy and efficiency.

[0066] In this invention, the terms "first," "second," and "third," etc., are used only to distinguish similar objects and are not necessarily used to describe a specific order or sequence, nor should they be construed as indicating or implying relative importance. The use of terms such as "upper," "lower," "left," "right," "front," and "rear" to indicate orientation or positional relationships is based on the orientation or positional relationships shown in the accompanying drawings and is only for the convenience of describing the invention, not to indicate or imply that the device referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation on the scope of protection of this invention. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0067] Furthermore, in the description of this application, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0068] The above are merely specific embodiments of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial modifications made to the present invention using this concept shall be considered as infringing upon the protection scope of the present invention.

Claims

1. A feature-driven method for estimating the health status of decommissioned power batteries, characterized in that: Includes the following steps: Step S1: Construct an algorithm library and pre-set the capability vectors of each algorithm in the library. The algorithm library shall contain at least global exploratory algorithms and local development algorithms. Step S2: Obtain the voltage and current of the retired power battery at the current moment, and combine it with the preset battery equivalent circuit model to obtain the feature vector at the current moment. The feature vector includes the prediction residual of the battery equivalent circuit model, data volatility and algorithm stagnation factor. Step S3: Based on the feature vector, calculate the matching score of each algorithm in the algorithm library through a dynamic weighted scoring mechanism. When the highest matching score is greater than the matching score of the currently executed algorithm, select the algorithm with the highest matching score as the new currently executed algorithm; otherwise, the currently executed algorithm remains unchanged. Step S4: Use the current execution algorithm to identify the parameters of the battery equivalent circuit model, obtain the optimal model parameters, and calculate the current health status of the retired power battery based on the optimal model parameters. Step S5: Calculate the convergence performance of the currently executed algorithm to update the historical reputation of the algorithm in the algorithm library for use in the matching score calculation at the next time step.

2. The feature-driven method for estimating the health status of decommissioned power batteries according to claim 1, characterized in that: In step S1, the algorithm library includes global exploratory algorithms, local exploitation algorithms, and noise-resistant robust algorithms. The global exploratory algorithms include the adaptive spiral flight sparrow search algorithm, the local exploitation algorithms include the transient triangular Harris Eagle optimization algorithm, and the noise-resistant robust algorithms include the hybrid genetic particle swarm optimization algorithm.

3. The feature-driven method for estimating the health status of decommissioned power batteries according to claim 2, characterized in that: In step S1, the capability vector of the k-th algorithm is represented as follows: , Let represent the weight factors of the k-th algorithm in global exploration, local exploitation, and diversity maintenance, respectively. , If the k-th algorithm is a global exploratory algorithm, then The maximum, if the k-th algorithm is a locally exploitable algorithm, then The maximum, if the k-th algorithm is a diversity-maintaining algorithm, then maximum.

4. The feature-driven method for estimating the health status of decommissioned power batteries according to claim 3, characterized in that: In step S2, the predicted residual is expressed as The data volatility is the standard deviation of the current over a predetermined time period prior to the current moment, and the algorithm stall factor is determined based on the change in the objective function value of the currently executing algorithm. The theoretical voltage value is obtained based on the battery equivalent circuit model. To obtain the voltage at the current time t, Open circuit voltage, , The electrochemical polarization voltage and concentration polarization voltage at the current time t are given by the electrochemical polarization resistance parameter identified at the previous time. Electrochemically polarized capacitors Concentration polarization resistance Concentration polarization capacitor get, The ohmic resistance identified in the previous moment. To obtain the current at the current time t.

5. The feature-driven method for estimating the health status of decommissioned power batteries according to claim 4, characterized in that: Step S1 also includes the initialization of feature vectors and historical reputation scores. In the initialization of feature vectors, the prediction residual is set to a maximum value, the data volatility and the algorithm stagnation factor are both set to 0, and the historical reputation scores of each algorithm in the algorithm library are set to the same value.

6. The feature-driven method for estimating the health status of decommissioned power batteries according to claim 5, characterized in that: In step S3, the matching score of the k-th algorithm in the algorithm library is calculated according to the formula. Calculate, where, As a weight for global search demand, Weighting based on local development needs. Weighting is required for noise resistance robustness. Let be the historical reputation score of the k-th algorithm at the current time t. The historical reputation weight is set.

7. The feature-driven method for estimating the health status of decommissioned power batteries according to claim 6, characterized in that: In step S3, the global search demand weight is: , The preset error threshold, The stagnation factor of the algorithm. , For adjustment coefficients, The weight of the local development requirement is The noise resistance robustness requirement weight is , This is a sequence of current data over a predetermined time period, starting from the current moment. To calculate the current data sequence standard deviation The reference volatility is set.

8. The feature-driven method for estimating the health status of decommissioned power batteries according to claim 7, characterized in that: In step S5, when the k-th algorithm is the algorithm executed at the previous time step, the historical reputation of the k-th algorithm at the current time t is: When the k-th algorithm is not the algorithm executed in the previous time step, the historical reputation score of the k-th algorithm at the current time t is the same as the historical reputation score at the previous time step. Forgetting factor, Based on the historical credibility of the previous moment, The performance score of the algorithm executed at the previous time step. The root mean square error of the algorithm executed in the previous moment. This is the maximum value of the root mean square error. The computation time taken to execute the algorithm in the previous moment. This is the maximum set computation time. , These are the preset evaluation weight coefficients.

9. The feature-driven method for estimating the health status of decommissioned power batteries according to any one of claims 1 to 8, characterized in that: The battery equivalent circuit model adopts a second-order RC equivalent circuit model.

10. A feature-driven health state estimation system for retired power batteries, used to implement the feature-driven health state estimation method for retired power batteries as described in any one of claims 1 to 9, comprising: Algorithm library construction module: used to build algorithm libraries and pre-set the capability vectors of each algorithm in the algorithm library. The algorithm library must contain at least global exploratory algorithms and local development algorithms; Feature vector acquisition module: used to acquire the current voltage and current of the retired power battery, and combine it with the preset battery equivalent circuit model to acquire the feature vector at the current moment. The feature vector includes the prediction residual of the battery equivalent circuit model, data volatility and algorithm stagnation factor. Algorithm confirmation module: Based on feature vectors, it calculates the matching score of each algorithm in the algorithm library through a dynamic weighted scoring mechanism, and selects the algorithm with the highest matching score as the current execution algorithm; Parameter identification module: Used to identify the parameters of the battery equivalent circuit model using the currently executed algorithm, obtain the optimal model parameters, and calculate the current health status of the retired power battery based on the optimal model parameters; Historical reputation calculation module: Used to record the convergence performance parameters of the currently executing algorithm, so as to update the historical reputation of each algorithm in the algorithm library for use in the matching score calculation at the next moment.

Images