Intelligent evaluation and gradient recovery method for retired power batteries of hybrid architecture

Abstract

The invention provides an intelligent evaluation and gradient recovery method for an ex-service power battery of a hybrid architecture, and belongs to the technical field of energy, the method is used for realizing accurate evaluation and graded recovery decision of the health state of the ex-service power battery of an electric vehicle, the framework extracts battery degradation characteristics through a CNN network, and captures time sequence dynamic characteristics in combination with a GRU network; and performing hyper-parameter optimization by adopting a WOA algorithm, and finally constructing a multi-stage recovery strategy according to an SOH evaluation result, so as to formulate differentiated recovery paths such as regeneration utilization, echelon utilization or harmless treatment for the retired batteries with different health grades.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy, and specifically, relates to an intelligent evaluation and gradient recovery method for retired power batteries with a hybrid architecture. Background Art

[0002] Under the background of the intensifying global climate crisis, as the core component of the large-scale application of electric vehicles, the retired power batteries have increasingly prominent resource and environmental problems caused by their retirement tide. According to industry predictions, the total amount of retired power batteries in China will exceed 780,000 tons in 2025. If not properly handled, it will lead to the loss of strategic metal resources and the risk of heavy metal pollution.

[0003] The existing battery state of health (SOH) evaluation system faces dual challenges. On the one hand, although the degradation model based on electrochemical mechanism has high theoretical accuracy, due to the complexity of the internal reaction mechanism of the battery, it is difficult to comprehensively capture the aging process of the battery in practical applications, resulting in limited prediction accuracy of the model. On the other hand, although the data-driven method can construct a prediction model by using multi-source heterogeneous operation data, there are still significant technical bottlenecks. The "data hunger" of the deep learning framework requires a large amount of high-quality data to train the model, but in practice, it is often difficult to obtain high-quality data. The problem of time-series correlation attenuation makes it difficult for the model to effectively capture the dynamic changes of battery performance over time; the hyperparameter sensitivity leads to unstable performance of the model on different data sets and is difficult to be widely applicable.

[0004] In view of the above problems, the present invention proposes an intelligent evaluation and gradient recovery method for retired power batteries with a hybrid architecture. By constructing a collaborative mechanism of feature extraction, time-series modeling, and parameter optimization, it breaks through the limitations of traditional methods in terms of accuracy and efficiency, improves the accuracy and efficiency of SOH evaluation, provides reliable technical support for the gradient recovery of retired batteries, and helps to achieve the efficient utilization of resources and the sustainable development of the environment. Summary of the Invention

[0005] The embodiment of the present invention provides an intelligent evaluation and gradient recovery method for retired power batteries with a hybrid architecture, which solves the problem of the limitations of traditional battery state evaluation methods in terms of accuracy and efficiency.

[0006] In view of the above problems, the technical solution proposed by the present invention is:

[0007] The present invention provides an intelligent evaluation and gradient recovery method for retired power batteries with a hybrid architecture, including the following steps:

[0008] S1, determine the basic data for battery health evaluation according to the charge and discharge historical data of the discarded power batteries;

[0009] S2. Extract the core degradation characteristics of discharge capacity, internal resistance, constant current charging time, and constant voltage charging time from the charge-discharge historical data of waste power batteries;

[0010] S3. Integrate the CNN-GRU network optimized by the WOA algorithm. Extract spatial features through the CNN network, model the temporal decay law with the GRU network, and globally optimize the network structure and hyperparameters with the WOA algorithm;

[0011] S4. Divide the dataset into a training set and a validation set. Use the training set to train the model established by the CNN network and validate it on the validation set;

[0012] S5. Use the test set to evaluate the trained model and evaluate the health status of waste power batteries according to the output of the model;

[0013] S6. Based on the quantitative evaluation results of the SOH of retired power batteries, construct a multi-level gradient recovery decision-making mechanism and process the batteries according to the decision.

[0014] As a preferred technical solution of the present invention, the WOA algorithm is used for the global optimization of model parameters. The WOA algorithm in step S3 optimizes the hyperparameters of the CNN network and the GRU network through the following update formula:

[0015] X t+1 = X t − 2·A·|X t − P b | + P b

[0016] where X t is the parameter position at the t-th iteration, A is a factor controlling the search range, A is an amplification factor that gradually decreases with the number of iterations, controlling the contraction of the search range, P b is the currently found best position, and all whales will tend to this optimal solution, gradually converging to the optimal parameter configuration through the above formula.

[0017] As a preferred technical solution of the present invention, the CNN network in step S3 is used to extract spatial features from the multi-dimensional time series data of the battery. Through multi-layer convolution and pooling operations, identify the change patterns of data such as voltage, current, and temperature during the battery degradation process. The CNN network performs convolution operations on the time series data through the following formula:

[0018] Z (l) = f(W (l) * X (l―1) + b (l) )

[0019] where: W (l)is the convolutional kernel, X (l―1) is the output of the previous layer, b (l) is the bias term, and f is the activation function. In the battery life prediction task, the CNN network identifies effective life degradation patterns from a large amount of complex and high-dimensional battery monitoring data.

[0020] As a preferred technical solution of the present invention, the GRU network in step S3 is used to process the long sequence data of the battery and capture long-term dependencies. The GRU network includes an update gate and a reset gate. The GRU processes long dependencies in the time series through the following formula:

[0021] Update gate:

[0022] z t = σ(W z · [h t―1 , x t + b z )

[0023] Reset gate:

[0024] r t = σ(W r · [h t―1 , x t + b r )

[0025] where z t and r t respectively represent the outputs of the update gate and the reset gate, W z , W r , b z , b r are learnable parameters, σ is the sigmoid activation function, and h t―1 is the hidden state;

[0026] Based on the calculation results of the update gate and the reset gate, the GRU network calculates the candidate hidden state including the candidate hidden state:

[0027]

[0028] where is the candidate hidden state adjusted according to the input x t at the current time and the reset gate r t . Finally, according to the output z t of the update gate, the GRU network calculates the hidden state h t at the current time. The final hidden state is expressed as:

[0029]

[0030] Among them, the formula shows how the hidden state at the current moment is obtained by a weighted combination of the hidden state at the previous moment and the candidate hidden state, where ⊙ represents element-wise multiplication.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0032] (1) By constructing a deep learning framework of WOA, CNN, and GRU, the present invention improves the accuracy and timeliness of the SOH evaluation of discarded electric vehicle power batteries, provides reliable technical support for the screening and grading and value evaluation of battery cascade utilization. The model can process complex time series data and improve the stability and robustness of the model by optimizing hyperparameters, and is applicable to different types of discarded batteries.

[0033] The above description is only an overview of the technical solution of the present invention. In order to understand the technical means of the present invention more clearly, it can be implemented according to the content of the specification. In order to make the above and other purposes, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention are given below. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is a schematic flow chart of an intelligent evaluation and gradient recovery method for retired power batteries with a hybrid architecture disclosed by the present invention;

[0035] Figure 2 is a recovery flow chart of retired power batteries for an intelligent evaluation and gradient recovery method for retired power batteries with a hybrid architecture disclosed by the present invention;

[0036] Figure 3 is a model framework flow chart of an intelligent evaluation and gradient recovery method for retired power batteries with a hybrid architecture disclosed by the present invention;

[0037] Figure 4 is a comparative analysis chart of different model results for an intelligent evaluation and gradient recovery method for retired power batteries with a hybrid architecture disclosed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention fall within the protection scope of the present invention.

[0039] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0040] It should be noted that similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0041] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0042] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, "a plurality" means two or more unless otherwise specifically defined.

[0043] Embodiment 1

[0044] Referring to the attached Figure 1-2 As shown, the present invention provides a technical solution: an intelligent evaluation and gradient recovery method for retired power batteries with a hybrid architecture, specifically including the following steps:

[0045] S1. Determine the basic data for battery health assessment according to the charge and discharge historical data of the discarded power battery.

[0046] S2. Extract core degradation features such as discharge capacity, internal resistance, constant current charging time, and constant voltage charging time from the charge and discharge historical data of the discarded power battery.

[0047] S3. Integrate a CNN-GRU network optimized by the WOA algorithm. Extract spatial features through the CNN network, model the temporal decay law through the GRU network, and globally optimize the network structure and hyperparameters by the WOA algorithm, specifically including:

[0048] S31, the WOA algorithm is used as a pre-training step to optimize hyperparameters such as CNN filter size, GRU hidden units, and learning rate. The WOA algorithm evaluates multiple parameter combinations by simulating the search process of whale predation behavior and determines the optimal parameter set by verifying the model performance on the dataset;

[0049] S32, after determining the optimal parameters, builds a CNN-GRU architecture. The model first extracts spatial features from the input time series data (such as voltage, current, and temperature curves) through CNN layers. These convolutional layers are enhanced by pooling operations to capture local patterns and reduce dimensionality while retaining key information.

[0050] At step S33, the output of the CNN layer is flattened and passed to the GRU layer. The GRU layer is used to model sequential dependencies in the data, such as charge and discharge history and degradation trends. The GRU's gating mechanism can effectively manage long-term and short-term information and is particularly well suited for processing time series data. This mechanism can identify the most relevant patterns in the data (such as performance mutations) and enhance the interpretability of predictions.

[0051] Finally, the fully connected layer S34 aggregates the features extracted by the CNN and GRU modules and maps them to the output to provide the predicted SOH value. This model integrates optimization, deep feature extraction, and time series modeling, providing an efficient framework for battery health status prediction.

[0052] S4, divide the data set into a training set and a validation set, use the training set to train the model established by the CNN network, and validate it on the validation set to prevent overfitting;

[0053] S5, use the test set to evaluate the trained model. According to the output of the model, the established model first optimizes the parameters through the WOA algorithm, then uses the CNN network to extract the spatiotemporal features of the battery data, and finally uses the GRU network to process long sequence dependencies and output the RUL prediction results of the battery to evaluate the health status of the discarded power battery. The WOA-CNN-GRU model framework is as follows: Figure 3 As shown in Figure 1, the model combines the WOA algorithm with the CNN-GRU network architecture to predict the battery health status. The model training includes:

[0054] During the model training process, 80% of the cyclic data was used to train the WOA-CNN-GRU model. This part of the data was used to optimize the hyperparameters of the CNN and GRU parts through WOA to improve the prediction accuracy of the model. At the same time, in order to prevent overfitting, cross-validation and early stopping strategies were adopted during the training process. After the training was completed, the remaining 20% of the data was used as the test set to evaluate the performance of the model and test its prediction ability on unknown data. During the test process, the model predicted the SOH of the battery based on the input features and compared it with the actual SOH value. The accuracy and stability of the model were measured through evaluation metrics (such as root mean square error RMSE, mean absolute error MAE, etc.);

[0055] S6. Based on the quantitative evaluation results of the SOH of retired power batteries, a multi-level gradient recovery decision-making mechanism is constructed. The multi-level gradient recovery decision-making mechanism specifically includes:

[0056] According to the health status of the battery, it is divided into different levels, high, medium, and low health status, and different recovery strategies are formulated for each level. The key to gradient recovery is to allocate the battery to different application scenarios or recovery paths according to its health status and remaining value. For example:

[0057] High health status (higher SOH, 60 < SOH < 80%): The battery can continue to be used in energy storage systems or low-power application scenarios;

[0058] Medium health status (medium SOH, 40 < SOH ≤ 60%): After the battery is disassembled, it is used for material recovery or low-value secondary utilization;

[0059] Low health status (lower SOH, 0 < SOH ≤ 40%): The battery is directly subjected to material recovery or scrapping treatment.

[0060] The embodiments of the present invention are also implemented through the following technical solutions.

[0061] In the embodiments of the present invention, the WOA algorithm is used for the global optimization of model parameters. The WOA algorithm simulates three behaviors of whales during the predation process, including surrounding prey, searching for prey, and bubble net attack. The WOA algorithm optimizes the hyperparameters of the CNN network and GRU network through the following update formula:

[0062] X t+1 =X t ―2·A·∣X t ―P b ∣+Pb

[0063] Or

[0064] X t+1 =X t ―2·A·|Xt -P l | + rand()·(P b -P l )

[0065] where X t is the parameter position at the t-th iteration, A is a factor controlling the search range, A is a magnification factor that gradually decreases with the number of iterations, controlling the contraction of the search range, P b is the currently found best position, and all whales will tend to this optimal solution, P l is any local best position, and rand() is a random number in the interval [0, 1];

[0066] Among them, through the WOA algorithm, it is possible to effectively search the model parameter space globally, quickly find potential optimal parameter configurations, thereby improving the learning efficiency and prediction performance of the battery health status prediction model. In practical applications, the WOA algorithm is used to optimize the weights and other hyperparameter settings of the CNN or GRU model in the pre-training stage, in order to obtain more accurate battery life prediction results;

[0067] Gradually converge to the optimal parameter configuration through the above formula.

[0068] In the embodiments of the present invention, the CNN network is used to extract spatial features from the multi-dimensional time series data of the battery. Through multi-layer convolution and pooling operations, it identifies the change patterns of data such as voltage, current, and temperature during the battery degradation process. The CNN network performs convolution operations on the time series data through the following formula:

[0069] Z (l) = f(W (l) * X (l―1) + b (l) )

[0070] where: W (l) is the convolution kernel, X (l―1) is the output of the previous layer, b (l) is the bias term, and f is the activation function. In the battery life prediction task, the CNN network identifies effective life degradation patterns from a large amount of complex and high-dimensional battery monitoring data, which are difficult to capture by traditional statistical methods or simple models. After parameter optimization combined with the WOA algorithm, the CNN network can better mine the effective information in the data, thereby improving the prediction accuracy.

[0071] In an embodiment of the present invention, the GRU network is used to process the long sequence data of the battery and capture long-term dependencies. Especially when the sequence length is relatively long, the GRU effectively avoids the problem of gradient disappearance through the gating mechanism. The GRU network includes an update gate and a reset gate, and the GRU processes the long dependencies in the time series through the following formula:

[0072] Update gate:

[0073] z t = σ(W z · [h t―1 , x t + b z )

[0074] Reset gate:

[0075] r t = σ(W r · [h t―1 , x t + b r )

[0076] Where z t and r t respectively represent the outputs of the update gate and the reset gate, W z , W r , b z , b r are learnable parameters, σ is the sigmoid activation function, and these gates control how the previous hidden state h t―1 should be updated and forgotten;

[0077] Based on the calculation results of the update gate and the reset gate, the GRU network calculates the candidate hidden state including the candidate hidden state:

[0078]

[0079] Where, is the candidate hidden state adjusted according to the input x t at the current time and the reset gate r t . Finally, according to the output z t of the update gate, the GRU network calculates the hidden state h t at the current time. The final hidden state is expressed as:

[0080]

[0081] Where the formula represents how the hidden state at the current time is weighted and combined from the hidden state at the previous time and the candidate hidden state, where ⊙ represents element-wise multiplication;

[0082] Therefore, as a gated recurrent neural network, the GRU network can help the model better understand and utilize the internal relationships of battery time series data in the task of predicting battery health status, thereby improving the accuracy and reliability of the prediction.

[0083] Specifically, in this embodiment, the charge and discharge historical data of the discarded power batteries in step S1 uses the dataset of CALCE lithium batteries of the University of Maryland, USA, including 4 batteries with lithium cobalt oxide (LiCoO2) as the positive electrode material and a rated capacity of 1.1 Ah: CS2_35 (CS35), CS2_36 (CS36), CS2_37 (CS37) and CS2_38 (CS38). At the same room temperature (25 °C), the batteries are charged at a constant current rate of 0.5C. When the voltage reaches 4.2V, the batteries are charged at a constant voltage. When the charging current of the battery drops below 50 mA, the charging stops. After charging, a constant current discharge is carried out at a rate of 1C until the voltage drops to 2.7V;

[0084] In evaluating the battery health status, in step S2, five features including discharge capacity, state of health (SOH) of the battery, internal resistance, constant current charging time, and constant voltage charging time are extracted from the original battery test data (i.e., historical data);

[0085] Discharge capacity: The discharge capacity can be calculated by accumulating the charge transfer amount in the discharge stage. The formula is as follows:

[0086] Qdischarge=i=1∑nIi·Δt i

[0087] where I i is the average current in the i-th time interval (unit: A), and Δt i is the i-th time interval (unit: s). First, calculate the product of the current difference between adjacent time points and the time difference, and then accumulate to obtain the total discharge capacity (unit conversion to Ah). The change in battery capacity directly indicates its degradation during the charge and discharge cycle. Therefore, capacity can be used as a direct health factor to evaluate the performance degradation of the battery, and thus evaluate the SOH of the power battery;

[0088] State of health (SOH) of the battery: The SOH is estimated as follows:

[0089]

[0090] where Capacity 3,4V is the remaining capacity of the battery at the cut-off voltage of 3.4V, and Capacity rated is the rated capacity of the battery. The SOH is approximately calculated by selecting the capacity difference corresponding to the voltage range between 3.8V and 3.4V;

[0091] Internal resistance: The internal resistance (IR) is calculated using the average value recorded in historical data:

[0092]

[0093] where R i is the internal resistance value measured in the i-th discharge step, and N is the number of measurements;

[0094] Constant current charging time: The constant current charging time (CCCT) is equal to the difference between the end time and the start time of the constant current charging stage:

[0095]

[0096] This time period is obtained by comparing the maximum and minimum timestamps in the constant current charging step;

[0097] Constant voltage charging time: The constant voltage charging time (CVCT) is calculated in the same way as CCCT:

[0098]

[0099] The duration of the constant voltage charging stage is obtained by taking the difference between the maximum and minimum timestamps in the constant voltage charging step.

[0100] In step S5, the model automatically optimizes key hyperparameters such as the number of convolutional filters, the number of GRU hidden units, and the learning rate using WOA to ensure a globally optimal configuration. This method eliminates the need for manual tuning and reduces the likelihood of getting stuck in local optima, thus significantly improving the training efficiency and model performance. The CNN component is particularly good at automatically extracting local spatial features from time series data (such as voltage and current patterns) without manual feature engineering. The GRU layer complements this function by capturing time dependencies, being able to model short-term fluctuations and long-term degradation trends, and having higher computational efficiency compared to the LSTM layer. The synergistic effect of these components forms a comprehensive battery health modeling method that integrates spatial, temporal, and attention-based insights.

[0101] Table 1: The detailed structure table of the neural network architecture of this model is as follows:

[0102]

[0103] The main evaluation metrics during the model training process in step S5 include the mean absolute error (MAE) and the root mean square error (RMSE), and the calculation formulas are as follows:

[0104]

[0105] where n is the number of samples; y iis the true value of the i-th sample; is the predicted value of the i-th sample; is the average value of all true values y i ;

[0106] Experimental Results and Analysis:

[0107] Four models were used in this application to evaluate the battery SOH: WOA-CNN-GRU, CNN-GRU, WOA-CNN-LSTM, and CNN-LSTM. These models were optimized on the training set and their performance was evaluated by predicting the test set. The comparative evaluation results of the four models are as Figure 4 shown. The WOA-CNN-GRU model performs better in terms of fitting ability, accurately predicting most samples. Especially in the test set, it can effectively capture the trend and detailed features of SOH changes. The CNN-GRU and CNN-LSTM models can predict the overall trend of SOH, but due to the difficulty in capturing complex non-linear features, the prediction curves fluctuate greatly and the errors are high. The performance of the WOA-CNN-LSTM model is between the above two, which improves the global fitting well, but is still inferior to the WOA-CNN-GRU model in capturing local details;

[0108] Table 2: The detailed error data table of the models is as follows:

[0109]

[0110] The error indicators in Table 2 show that the performance of the WOA-CNN-GRU model is better than other models, demonstrating the best accuracy and fitting degree. In contrast, the performance of the CNN-LSTM model is the worst, and other models (such as WOA-CNN-LSTM) show medium performance. These results highlight the excellent effectiveness of the WOA-CNN-GRU model in predicting SOH.

[0111] The hybrid architecture integrating WOA-CNN-GRU can comprehensively utilize the spatio-temporal information and features in time series data and effectively capture the dependencies between sequences, thereby improving the prediction accuracy and robustness of the battery health status. By combining the CNN and GRU optimized by WOA, the evaluation accuracy of the power battery SOH is improved and the computational complexity is reduced, which is applicable to high-precision evaluations under different battery states and working conditions.

[0112] The above is only the preferred embodiment of the present invention and is not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0113] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of the present disclosure. The appended method claims present the elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy recited.

[0114] In the above detailed description, various features are combined in a single embodiment to simplify the present disclosure. This method of disclosure should not be interpreted as reflecting an intention that the embodiments of the claimed subject matter require more features than are expressly recited in each claim. Rather, as reflected by the appended claims, the invention lies in less than the full scope of the features of the single disclosed embodiment. Accordingly, the appended claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

[0115] Those skilled in the art should also understand that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments herein can be implemented as electronic hardware, computer software, or combinations thereof. To clearly illustrate the interchangeability of hardware and software, the various illustrative components, blocks, modules, circuits, and steps have been generally described in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as departing from the scope of the present disclosure.

[0116] The steps of a method or algorithm described in connection with the embodiments herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination thereof. The software module may be located in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium well known in the art. An exemplary storage medium is coupled to the processor such that the processor can read information from, and write information to, the storage medium. Of course, the storage medium may also be integral to the processor. The processor and the storage medium may be located in an ASIC. The ASIC may be located in a user terminal. Of course, the processor and the storage medium may also exist as discrete components in a user terminal.

[0117] For software implementation, the techniques described in this application can be implemented by modules (e.g., procedures, functions, etc.) that perform the functions described in this application. These software codes can be stored in a memory unit and executed by a processor. The memory unit can be implemented within the processor or outside the processor. In the latter case, it is communicatively coupled to the processor via various means, which are well known in the art.

[0118] The above description includes examples of one or more embodiments. Of course, it is not possible to describe all possible combinations of components or methods for the purpose of describing the above embodiments, but those of ordinary skill in the art should recognize that the various embodiments can be further combined and arranged. Therefore, the embodiments described herein are intended to cover all such changes, modifications, and variations that fall within the scope of the appended claims. Additionally, with respect to the term "comprising" used in the specification or claims, the word is intended to be construed in a manner similar to the term "including" as that term is interpreted when used as a transitional word in a claim. Further, any use of the term "or" in the specification or claims is intended to mean "non-exclusive or".

Claims

1. An intelligent evaluation and gradient recovery method for retired power batteries with a hybrid architecture, characterized in that, Including the following steps: S1. Determine the basic data for battery health assessment based on the charge-discharge historical data of the discarded power battery. S2. Extract the core degradation characteristics of discharge capacity, internal resistance, constant current charging time, and constant voltage charging time from the charge-discharge historical data of the discarded power battery. S3. Integrate the CNN-GRU network optimized by the WOA algorithm. Extract spatial features through the CNN network, model the temporal decay law through the GRU network, and globally optimize the network structure and hyperparameters by the WOA algorithm. S4. Divide the dataset into a training set and a validation set. Use the training set to train the model established by the CNN network and validate it on the validation set. S5. Use the test set to evaluate the trained model and evaluate the health status of the discarded power battery according to the output of the model. S6. Based on the quantitative evaluation results of the SOH of the retired power battery, construct a multi-level gradient recovery decision-making mechanism and process the battery according to the decision.

2. The intelligent evaluation and gradient recovery method for retired power batteries with a hybrid architecture according to claim 1, characterized in that The WOA algorithm in step S3 is used for the global optimization of model parameters. The WOA algorithm optimizes the hyperparameters of the CNN network and the GRU network through the following update formula: X t+1 = X t − 2·A·|X t − P b | + P b Among them, X t is the parameter position at the t-th iteration, A is a factor controlling the search range, and A is a magnification factor that gradually decreases with the number of iterations, controlling the contraction of the search range. P b is the currently found best position, and all whales will tend to this optimal solution, gradually converging to the optimal parameter configuration through the above formula.

3. A method for intelligent evaluation and gradient recovery of retired power batteries with a hybrid architecture according to claim 1, characterized in that The CNN network in step S3 is used to extract spatial features from the multi-dimensional time series data of the battery. Through multi-layer convolution and pooling operations, identify the change patterns of data such as voltage, current, and temperature during the battery degradation process. The CNN network performs convolution operations on the time series data through the following formula: Z (l) = f(W (l) * X (l―1) + b (l) ) Where: W (l) is the convolutional kernel, X (l―1) is the output of the previous layer, b (l) is the bias term, and f is the activation function. In the battery life prediction task, the CNN network identifies effective life degradation patterns from a large amount of complex and high-dimensional battery monitoring data.

4. A method for intelligent evaluation and gradient recovery of retired power batteries with a hybrid architecture according to claim 1, characterized in that The GRU network in step S3 is used to process the long sequence data of the battery and capture long-term dependencies. The GRU network includes an update gate and a reset gate. The GRU processes the long dependencies in the time series through the following formula: Update gate: z t = σ(W z · [h t―1 , x t + b z ) Reset gate: r t = σ(W r · [h t―1 , x t + b r ) where z t and r t represent the outputs of the update gate and the reset gate respectively, W z , W r , b z , b r are learnable parameters, σ is the sigmoid activation function, and h t―1 is the hidden state; Based on the calculation results of the update gate and the reset gate, the GRU network calculates a candidate hidden state including a candidate hidden state: Among them, is the candidate hidden state adjusted according to the input x t at the current moment and the reset gate r t Finally, according to the output z t of the update gate, the GRU network calculates the hidden state h t at the current moment, and the final hidden state is expressed as: Among them, the formula represents how the hidden state at the current moment is weighted and combined by the hidden state at the previous moment and the candidate hidden state, where ⊙ represents element-wise multiplication.

Images