A configuration method of an energy storage system suitable for the cascade utilization of retired batteries and a classification controller

Abstract

The application discloses a kind of energy storage system configuration methods suitable for retired battery echelon utilization, belong to new energy retired battery echelon utilization and energy storage system technical field, utilize battery cycle life test and new energy automobile retired battery cloud platform historical data, the aging state of retired battery is predicted, after completion prediction, state similar battery is selected to carry out development and design of energy storage echelon utilization energy storage system.For improving the evaluation of key evaluation parameters of retired battery health state and remaining useful life, the application discloses a kind of enhanced battery state prediction network, after the pre-training of model is completed using battery cycle life test data, the model is trained cooperatively using the cloud platform data of retired battery to improve the key evaluation parameters of retired battery, especially the prediction accuracy of remaining useful life RUL, reduce the configuration difficulty when developing echelon utilization energy storage system, provide guarantee for echelon utilization energy storage system design of retired battery.

Description

Technical Field

[0001] This invention belongs to the field of cascade utilization of power batteries for new energy vehicles and the development of energy storage systems, specifically involving the configuration and development method of an energy storage system for the cascade utilization of retired batteries. Background Technology

[0002] New energy vehicles, represented by pure electric and plug-in hybrid electric vehicles, have become an inevitable trend in the development of my country's automotive industry. As one of the core components of new energy vehicles, the power battery, after a large number of power batteries are retired, makes the development of industrial and commercial energy storage systems using these retired power battery packs an effective way to utilize batteries in a tiered manner.

[0003] Based on current lithium battery technology and automotive usage requirements, when the State of Health (SOH) of a battery decays to below 80% of its rated capacity, the power and range requirements of new energy vehicles will no longer be met, necessitating its retirement. When the SOH falls below 40%, it essentially ceases to have commercial applicability and requires scrapping or recycling.

[0004] The secondary use of retired batteries requires accurate assessment of battery status, such as State of Health (SOH), Remaining Life (RUL), and internal resistance, to achieve better charge and discharge control, maximize the value of secondary use of retired battery packs for new energy vehicles, and ensure the safety of the system.

[0005] Lithium-ion batteries, SOH (State of Health) refers to the battery's healthy lifespan, generally defined as the ratio of usable capacity to rated capacity. RUL (Rest Useful Life) refers to the battery's remaining lifespan, i.e., the estimated time it can be used under its current condition, typically quantified by the number of battery cycles.

[0006] As the number of charge and discharge cycles increases, the battery gradually ages, and its State of Harm (SOH) value continuously decreases. Generally, when the SOH of a battery decreases to around 80%, the power battery for new energy vehicles will be retired; when the SOH decreases to around 40%, the battery will be completely discarded and enter the battery material recycling process.

[0007] Currently, prediction methods for key evaluation parameters of retired batteries, such as State of Health (SOH), Rullow Resistance (RUL), and internal resistance, can be broadly categorized into model-based methods and data-driven methods. Model-based methods analyze the internal aging mechanisms of the battery and establish electrochemical or equivalent circuit models to illustrate its aging behavior. Electrochemical models utilize electrothermal dynamics and lithium-ion movement to establish matrix equations, providing a detailed description of internal electrochemical reactions. However, these methods typically involve multiple complex parameters, limiting their practical application. Equivalent circuit models use multiple circuit elements to simulate battery charge-discharge characteristics; their structure is relatively simple, but their accuracy is relatively poor. Data-driven methods utilize large amounts of historical data (such as voltage, current, and temperature) collected by battery sensors to train the model, establishing input-output mapping relationships to accurately estimate key battery state evaluation parameters. Data-driven prediction methods feature high dynamic accuracy and strong generalization ability, but require a large amount of data, computation, and model training time.

[0008] Existing methods for sorting batteries for secondary use generally borrow testing methods from new batteries. Batteries intended for secondary use are recharged and discharged, as described in patent documents such as CN103439665B, CN105665309B, and CN110752410B. The test data is then used for sorting. However, these test results mainly reflect the current state of the battery and cannot reflect its historical degradation history and internal mechanisms, thus failing to accurately assess the health status of lithium batteries. This invention combines historical data of retired batteries from new energy vehicles with a data-driven prediction method to improve the prediction accuracy of key evaluation parameters for retired batteries.

[0009] Meanwhile, when designing energy storage systems for the secondary use of retired batteries, the main assessment focuses on the impact of the retired batteries' State of Health (SOH) on the system design. Patent documents such as CN112305442B, compared with this invention, do not fully consider the differences in the Remaining Life (RUL) of retired batteries under the same SOH due to differences in their operating conditions before retirement. As a result, the synchronicity of battery degradation varies among batteries in the actual secondary use energy storage system, increasing the later operation and maintenance costs of the secondary use energy storage system.

[0010] To address the aforementioned issues, this invention proposes an enhanced battery state prediction network that utilizes lithium battery cycle life test data and historical data from a retired battery cloud platform to predict key evaluation parameters for retired batteries. After prediction and calculation, the predicted data for key evaluation parameters such as the retired battery's RUL (Relative Usage Limit), DC internal resistance, and median voltage are used for battery sorting, thereby improving the battery degradation synchronization and the safety of the energy storage system for secondary use. Summary of the Invention

[0011] This invention proposes a model using an enhanced battery state prediction network to predict key evaluation parameters of retired batteries. After prediction and calculation, the key evaluation parameters are used to sort retired batteries, thus completing the configuration and design of a cascaded energy storage system for their reuse. This provides a more effective battery state prediction method for the cascaded utilization of retired batteries, improving the safety and economic efficiency of the energy storage system during this process.

[0012] To achieve the above objectives, this invention first proposes a prediction method for key evaluation parameters of retired batteries, comprising:

[0013] Battery cycle life test data and retired battery cloud platform data acquisition and processing. After the cycle life test is completed, based on the battery voltage changes, charge and discharge current changes, and other data recorded in the test, key evaluation parameters for assessing the battery's health status can be obtained, including the battery's State of Health (SOH), median discharge voltage, and DC internal resistance. Before retired batteries are decommissioned, data on their use in new energy vehicles can be obtained from the automotive cloud platform.

[0014] Battery cycle life test data and data from a retired battery cloud platform are input into an enhanced prediction network for collaborative training. After training, the trained enhanced prediction network is used to predict the battery state of each retired battery at the time of retirement, obtaining effective state information for each retired battery.

[0015] Furthermore, key evaluation parameters related to battery aging are defined and data preprocessing is completed.

[0016] 1) Battery health status

[0017] During the cycle life test, the first The SOH value of the battery after the first cycle can be calculated, i.e., using the first cycle... The ratio of the discharge energy of the second cycle to the discharge energy of the first cycle.

[0018]

[0019] In the formula, Indicates the cell in the first... Discharge energy of the next cycle This indicates the discharge energy of the battery cell during its first cycle.

[0020] 2) Cell DC internal resistance

[0021] DC internal resistance can generally be calculated based on hybrid pulse power characteristic (HPPC) tests at different time scales. As the battery ages, DC internal resistance includes contact resistance and SEI film resistance (ohmic polarization), charge transfer resistance (electrochemical polarization), and lithium-ion migration resistance (concentration polarization). DC internal resistance is calculated as the ratio of voltage change to current change under HPPC current excitation.

[0022] 3) Median discharge voltage

[0023] The median discharge voltage refers to the median value of the battery voltage change during the battery discharge process. As the battery cycles through charge and discharge, the efficiency of the internal chemical reaction gradually decreases, and electrochemical polarization intensifies, leading to a decrease in the median discharge voltage.

[0024] Based on battery aging status data obtained from battery cycle life testing and cloud platform data from retired batteries during use, the above three key evaluation parameters are extracted for each battery. To effectively remove data noise and eliminate outliers, Discrete wavelet transform (DWT) is used to process the data. DWT mainly uses wavelet signals of different scales to decompose noisy signals into principal components and detail components. By utilizing the difference between the noise signal and the effective signal, the principal and detail components are selectively reconstructed, thereby reducing the impact of noise on the signal.

[0025] This allows us to obtain key evaluation parameters for the health status of each battery. , , ,in Indicates the battery's first Next discharge cycle.

[0026] Cycle life test data can be taken from the start of the battery discharge cycle to the [number]th cycle. The test data from this test, This represents the number of discharge cycles from the point when the battery completes its cycle life test until it is no longer in use. The equivalent number of discharge cycles for a retired battery at the time of retirement is... Take the front This data is used for collaborative training. Historical usage data of retired batteries can be obtained through a cloud platform, and after data processing, the current performance of each retired battery is obtained. Data on cyclic discharge.

[0027] Furthermore, a data model is constructed for pre-training an enhanced battery state prediction network;

[0028] The enhanced battery state prediction network is a conventional prediction network that is extended with a sparse output matrix generated by random variables to improve the nonlinear fitting ability of the prediction network and thus enhance the learning performance of the model.

[0029] The enhanced battery state prediction network consists of three parts: an input layer, a hidden layer, and an output layer.

[0030] The input layer is for input data and is defined as follows:

[0031] (1)

[0032] in , This indicates the number of key evaluation parameters used to assess battery status. This indicates the number of time-delayed data points, i.e., the number of data points used to calculate the first time delay. When using key battery state evaluation parameters for each cycle number, use The delay data (the first) One to the first The calculation is performed using a cyclic data set as input, taking the time series delay state as an example. ,Right now and At that moment, we received:

[0033] Hidden layers define the feature vectors of the entire model. :

[0034] (2)

[0035] in,

[0036] Regular section , ;

[0037]

[0038] (3)

[0039] Extended section , ;

[0040] (4)

[0041] Here, This represents the leakage coefficient of the prediction model, used to control the linearity of states over time. tanh(*) is the activation function. Coefficients characterizing the magnitude of the state in the prediction model. It is the input weight matrix. It is the state weight matrix of the reserve pool. and These are the number of neurons in the input layer and the sparse matrix, respectively, from which we can obtain... .

[0042] The output layer is calculated. :

[0043] (5)

[0044] Output weight matrix , During the training process , That is, the number of key evaluation parameters.

[0045] Define the loss function during training. :

[0046] (6)

[0047] in, Represents trace calculation, Representation matrix The transpose operation, It is about diagonal matrix, , For the output matrix No. Line number Column elements The scaling factor, and The penalty coefficients are set between 0 and 1. , and This is the loss coefficient;

[0048] The definition is as follows:

[0049] (7)

[0050] The definition is as follows:

[0051] (8)

[0052] in, The output weight matrix calculated by the least squares method is the first... Line number Column elements ;

[0053] Furthermore, the pre-training of the enhanced battery state prediction model was completed.

[0054] The battery cycle life test data was divided into two datasets. and , ,in This indicates the number of batteries used in the completed cycle life test for each of the two datasets. This indicates the number of key evaluation parameters for battery status, which is 3 in this case, representing the aforementioned key battery evaluation parameters. , and , This indicates the number of discharge cycles from the point when the battery completes its lifespan test until it is no longer in use; also, regarding the number of battery discharge cycles, there are... These represent the percentage of battery health at the time of battery retirement and the number of battery discharge cycles, respectively. These represent the percentage of battery health at the time the battery is no longer in use and the number of battery discharge cycles, respectively. ,and , .

[0055] Here, datasets are used respectively. and of One battery Training was performed using data from cyclic lifetime tests, and... The cyclical data is used to evaluate the prediction results of key evaluation parameters.

[0056] In the actual model training process, Perform iterative updates. This represents the output weight matrix after each iteration update.

[0057] (7)

[0058] (8)

[0059] Finish After iterative updates, two enhanced battery state prediction networks are obtained. and prediction networks .

[0060] The loss function (6) is solved using the adaptive elastic ball algorithm to obtain the optimal output weight estimation matrix. The specific calculation steps are as follows:

[0061] Step 1: Initialize the total number of individual bouncy balls Maximum number of iterations Maximum upper limit of space vector and lower limit ;

[0062] Step 2: Define the iteration count vector ;set up ;

[0063] Step 3: Initialize the space vectors of all bouncy balls , ;

[0064] Step 4: Calculate the fitness of all bouncy balls using the loss function (6). , ;

[0065] Step 5: Update the following calculation process. All parameters during iteration:

[0066] Step A: Considering the effect of temperature on the bouncy ball, update the following formula: The spatial vector of an elastic sphere:

[0067] (11)

[0068] in, For the first A bouncy ball in Space vector during iteration, It is the first The optimal space vector during iteration; For the first Temperature changes during iteration Random coefficients between 0 and 1 The thermal coefficient is . Calculated for the exponent;

[0069] Determined by the following formula:

[0070] (12)

[0071] in, and The first During iteration and the first Temperature during iteration;

[0072] Step B: Considering the repulsive force between the bouncy balls, continue updating the equation using the following formula. The spatial vector of an elastic sphere:

[0073] (14)

[0074] in, For the first A bouncy ball in Space vector during iteration, ,and ; In the first In the next iteration, other elastic balls... The attractive force of an elastic ball is determined by the following formula:

[0075] (15)

[0076] in, Initial attraction parameters, No. The elastic ball and the first The elastic ball in the first... Spatial vector distance at the next iteration ,and ;

[0077] Step C: Passing the maximum upper limit of space vectors and lower limit The space vector of all elastic balls is restricted.

[0078] Step D: Update the best space vector for the current iteration. ;

[0079] Step 6: Determine the current iteration number Has the maximum number of iterations S been reached? If not, then proceed to step 5.

[0080] Step 7: Output the globally optimal space vector Optimal output , , and the estimated optimal output weight matrix .

[0081] Furthermore, the collaborative training process of the augmented battery state prediction network is as follows:

[0082] The historical data of retired batteries from the cloud platform was divided into two equally sized datasets. and , ,in This represents the number of retired batteries used in each of the two datasets. This indicates the number of key evaluation parameters for battery status. This represents the equivalent number of discharge cycles for each retired battery at the time of retirement. Defined here... The equivalent number of discharge cycles used for collaborative training for each retired battery is given by [reference to number of cycles]. . [1, ]The cyclic data is used as input data, and the [[...] of each retired battery , [This is used as evaluation data.]

[0083] Dataset Input the pre-trained prediction network respectively and prediction networks The prediction network 1 pairs the data of retired batteries. The predicted results ( arrive (in loop) The most confident data (positive examples) One and a counterexample (1) and its result, namely the selected prediction result SP1, and The datasets were merged; the prediction network 2 was used to combine the data from retired batteries. The predicted results ( arrive (in loop) The most confident data (positive examples) A counterexample (1) and its result, namely the selected prediction result SP2, and Merge the datasets. After completing the above operations, put these... Data from dataset Removed from the middle.

[0084] Repeat the above process until... For an empty set, complete the collaborative training.

[0085] Furthermore, after the prediction is completed, the key state parameters of the retired batteries are sorted.

[0086] By using historical data from the battery cloud platform, key evaluation parameters for the battery status of a single retired battery are initialized, resulting in... , , Indicates the first A retired battery.

[0087] The key evaluation parameters of the battery state of a single retired battery are input into an enhanced battery state prediction network that has completed co-training, to obtain the equivalent number of cycles for that retired battery at the time of retirement. Until the battery is no longer in use Predicted values ​​of key battery state assessment parameters This represents the percentage of battery health at the time a retired battery is no longer in use. The corresponding number of battery cycles, and has .

[0088] The remaining service life of a single retired battery is calculated. Here, the percentage decay of SOH is defined as... When a battery is retired and no longer in use, the RUL (Relative Usage Limit) can be calculated using the following formula.

[0089] (16)

[0090] Predict key evaluation parameters for retired batteries, and obtain the parameters for each retired battery from the time of retirement. to Key battery state evaluation parameters obtained through model prediction and calculation Here .

[0091] Based on the key evaluation parameters of each retired battery obtained through prediction and calculation, and in accordance with the principle of similar state, a normalized classification of excellent, good, medium, and poor is completed, thereby completing the configuration of the cascade utilization energy storage system.

[0092] Furthermore, the design of a classification controller for the tiered utilization of retired batteries was completed;

[0093] The retired battery sorting controller includes the following modules:

[0094] The data input module is used to obtain key evaluation parameters of battery status.

[0095] The model pre-training module uses battery cycle life test data to complete the pre-training of the model;

[0096] Model co-training: The model is co-trained using historical data from the retired battery cloud platform.

[0097] The model loss function is used to calculate the loss during the training process.

[0098] The retired battery classification module uses a trained model to perform normalized classification of a single retired battery.

[0099] The data output module outputs normalized classification results for the classification of retired batteries.

[0100] After completing the module design of the retired battery classification controller, a method for predicting and evaluating the battery status of retired batteries is used, based on battery cycle life test data and historical data from the cloud platform before retirement. This method further completes the configuration and development of a cascaded energy storage system using retired batteries.

[0101] The enhanced battery state prediction network training model was built and initialized on the local computer. Battery cycle life test data and retired battery historical data were input into the local computer to complete the localized training of the enhanced prediction network.

[0102] For each retired battery, the data is input into a localized model for prediction, yielding key evaluation parameters for the battery's condition.

[0103] Using the predicted key evaluation parameters of battery status, retired batteries are classified in a normalized manner, defining different retired battery statuses of excellent, good, medium, and poor.

[0104] Furthermore, the configuration method and system design of an energy storage system for the cascade utilization of sorted retired batteries;

[0105] After the retired batteries are classified, the entire pack of retired batteries is used for the development and design of energy storage systems for secondary use.

[0106] Without disassembling the retired battery, the energy storage management system (EMS) is directly connected to the power battery management system (BMS) using its original CAN communication interface, and the retired battery BMS is upgraded with adapted software.

[0107] Energy conversion between the power battery and the power grid is achieved through a power control system (PCS). The power battery has a voltage range of 310V-482V and a rated voltage of 415V. It is directly connected to the 380V / 50Hz AC power grid via the PCS's DC / AC module, enabling bidirectional charging and discharging.

[0108] The energy storage thermal management module uses the same water-cooling solution as the original power battery pack.

[0109] Other modules required for cascaded energy storage systems include fire protection modules, monitoring modules, and lighting modules.

[0110] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects:

[0111] This invention presents a method for predicting key evaluation parameters of the decommissioned state of retired batteries. This method utilizes an enhanced battery state prediction network to predict the key state parameters of each retired battery. To improve system performance, this invention incorporates battery cycle life test data for pre-training and uses data from a retired battery cloud platform for collaborative training, enabling accurate prediction of the retired battery's state. Simultaneously, by classifying retired batteries based on their key evaluation parameters, the invention facilitates the design of cascaded energy storage systems for batteries with similar decommissioned states. This improves the safety and economy of cascaded energy storage systems, laying the foundation for large-scale cascaded utilization of retired batteries. Attached Figure Description

[0112] Figure 1 This is a schematic diagram of a process for predicting the state of retired batteries and configuring a cascaded energy storage system using an enhanced battery state prediction network, as provided in an embodiment of the present invention.

[0113] Figure 2 This invention provides a method for pre-training using cycle life test data.

[0114] Figure 3 This invention provides a method for collaborative training and prediction using data from a retired battery cloud platform.

[0115] Figure 4 This invention provides a method for configuring a cascaded energy storage system for the reuse of retired batteries.

[0116] Figure 5 This is the internal structure of the retired battery state prediction controller provided in the embodiments of the present invention; Detailed Implementation

[0117] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0118] Example: Figure 1 The diagram shown is a schematic flowchart of a configuration method for the secondary use of retired batteries provided by an embodiment of the present invention. The method includes the following steps:

[0119] Step S1: Obtain key evaluation parameters for battery health status through battery cycle life testing and data from the retired battery cloud platform.

[0120] Step S2: Use data processing methods to preprocess the data, remove data noise, and eliminate abnormal data.

[0121] Step S3: Train the enhanced battery state prediction network using key evaluation parameters of battery health status.

[0122] Define the battery cycle life test dataset as follows and They are fed into the prediction network as input 1 and input 2, respectively. and prediction networks Pre-training is performed within this process. Before using this method... Training was performed using cycle life test data. to Evaluation is performed using cyclical data. This represents the number of discharge cycles corresponding to the percentage of battery health at the time of battery retirement. This indicates the number of discharge cycles from the point when the battery completes its cycle life test until it is no longer in use.

[0123] For the pre-trained augmented prediction network, the retired battery dataset is used. and Conduct collaborative training.

[0124] Step S4: Use the enhanced battery state prediction network that has completed co-training to predict the key evaluation parameters of the state of each retired battery, and obtain... , , Indicates the first A retired battery.

[0125] Here, it is necessary to convert the SOH data of the retired battery into RUL data of the retired battery. ,in It is calculated from equation (16).

[0126] Step S5: Use a clustering algorithm to normalize and classify the retired batteries that have been predicted.

[0127] Step S6: Combine retired batteries with similar classifications to construct a retired battery cascade utilization energy storage system.

[0128] like Figure 2 The model was pre-trained using battery cycle life test data, as shown.

[0129] like Figure 3 The model was collaboratively trained using data from a retired battery cloud platform, as shown.

[0130] like Figure 4 The diagram illustrates a configuration method and system for designing a cascaded energy storage system using retired batteries. Retired batteries are classified using a battery classification controller, and then batteries in similar condition are grouped together for cascaded energy storage system development.

[0131] By using battery cycle life test data and cloud platform data of retired batteries, a localized model is trained. Then, the key evaluation parameters of the battery state of a single retired battery are predicted. After the prediction is completed, retired batteries with similar states are reassembled into a tiered energy storage system.

[0132] Example 2: As Figure 5As shown, the main modules of the retired battery classification controller include a data input module that receives battery cycle life test data and historical data from the retired battery cloud platform. After data processing, the enhanced battery state prediction network uses the data to complete the model's pre-training and co-training.

[0133] Localized equipment can use a trained enhanced battery state prediction network to calculate and predict key battery state parameters for each retired battery. Then, the output module outputs normalized classification results to guide the design and configuration development of cascaded energy storage systems.

[0134] It should be noted that, depending on the implementation needs, the various steps / components described in this application can be broken down into more steps / components, or two or more steps / components or parts of the operation of steps / components can be combined into new steps / components to achieve the purpose of this invention.

[0135] The methods described above according to the invention can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD, ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code originally stored on a remote recording medium or a non-transitory machine-readable medium and subsequently stored on a local recording medium, downloaded via a network. Thus, the methods described herein can be stored as software processing on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components (e.g., RAM, ROM, flash memory, etc.) capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the processing methods described herein. Furthermore, when a general-purpose computer accesses the code used to implement the processing shown herein, the execution of the code transforms the general-purpose computer into a dedicated computer for performing the processing shown herein.

[0136] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for configuring an energy storage system suitable for the cascade utilization of retired batteries, characterized in that, The method includes: Obtain key evaluation parameters for battery aging during battery cycle life testing; including: Battery health status Battery internal resistance and median discharge voltage ,in Indicates the number of battery discharge cycles; Obtain historical data uploaded to the cloud platform during the normal use of retired batteries before their retirement; The enhanced battery state prediction network was pre-trained using key battery state evaluation parameters from battery cycle life testing. Specifically, the relevant parameters of the enhanced battery state prediction network were initialized. The entire battery prediction network consists of an input layer, hidden layers, and an output layer, which are defined as follows: The input layer contains input data, totaling... The input data is defined as follows: (1) in, , This indicates the number of key evaluation parameters used to assess battery status. This indicates the number of time-delayed data points, i.e., the number of data points used to calculate the first time delay. When using key evaluation parameters for battery state of being after a certain number of cycles, use The calculation is performed using the delayed data as input; Hidden layers define the feature vectors of the entire model. It includes a regular section and an extended section: (2) in, Regular section The calculation formula is as follows: (3) Extended section Its formula is expressed as: (4) Here, This represents the internal scaling factor of the prediction model, and tanh(*) is the activation function. Characterizing the external scaling factor, It is the input weight matrix. It is the state weight matrix. and These represent the number of neurons in the input layer and the reserve pool, respectively; [then]... ; The number of neurons in the output layer is Here , which is the number of key evaluation parameters, representing the prediction result at the next time point; To calculate : (5) Pre-trained output weight matrix , During the training process , ; Collaborative training of an enhanced battery state prediction network was completed using historical data from a retired battery cloud platform. For each retired battery, an enhanced battery state prediction network that has completed pre-training and co-training is used to predict the key evaluation parameters of retired power batteries for new energy vehicles and calculate the remaining service life of the retired battery. Based on the key evaluation parameters of each retired battery obtained from prediction and calculation, they are classified according to the principle of similar status (excellent, good, medium, poor) for the configurable development of cascade utilization energy storage systems.

2. The energy storage system configuration method for the cascade utilization of retired batteries according to claim 1, characterized in that, Obtain key evaluation parameters for battery aging and historical usage data before battery retirement. After obtaining the data of the above key battery status evaluation parameters, preprocessing is performed, including data denoising and outlier data removal; Among them, the health status of the battery During the cycle life test, the first The SOH value of the battery after the first cycle is calculated, i.e., using the first cycle... The ratio of the discharge energy of the second cycle to the discharge energy of the first cycle. In the formula, Indicates the cell in the first... Discharge energy of the next cycle This indicates the discharge energy of the battery cell during its first cycle.

3. The energy storage system configuration method for the cascade utilization of retired batteries according to claim 1, characterized in that, The enhanced battery state prediction network was pre-trained using key battery state evaluation parameter data from battery cycle life testing. This pre-training process was used to initially build the prediction network. The calculation steps are as follows: Step 1: Define the dataset for pre-training; The battery cycle life test data was divided into two datasets. and , ,in This indicates the number of batteries used in the completed cycle life test for each of the two datasets. This indicates the number of key battery status evaluation parameters, with a value of 3, representing the aforementioned key battery evaluation parameters. , and ;in, This indicates the number of discharge cycles from the point when a battery completes a cycle life test until it is no longer in use; at the same time, These represent the percentage of battery health at the time of battery retirement and the number of battery discharge cycles, respectively. These represent the percentage of battery health at the moment the battery is no longer in use and the number of battery discharge cycles, respectively. ,and , ; Step 2: Initialize the relevant parameters of the enhanced battery state prediction network. Step 3: Define the loss function during training. : (6) in, Represents trace calculation, Representation matrix The transpose operation, It is about diagonal matrix, , For the output matrix No. Line 1 Column elements Scaling factor, and The penalty coefficients are set between 0 and 1; this parameter is determined through ten-fold cross-validation. , and This is the loss coefficient; The definition is as follows: (7) The definition is as follows: (8) in, The output weight matrix calculated by the least squares method is the first... Line 1 Column elements ; Step 4: Using the dataset and Pre-train the enhanced battery state prediction network; Here, datasets are used respectively. and of One battery Training was performed using data from cyclic lifetime tests, and... The cyclical data is used to evaluate the prediction results of key evaluation parameters; During the pre-training process of this model, Perform iterative updates. This represents the output weight matrix after each iteration update; (9) (10) Finish After iterative updates, two enhanced battery state prediction networks are obtained. and prediction networks Their output matrices are respectively and .

4. The energy storage system configuration method for the cascade utilization of retired batteries according to claim 3, characterized in that, The loss function is solved using the adaptive elastic ball algorithm to obtain the optimal output weight estimation matrix. The specific calculation steps are as follows: Step 1: Initialize the total number of individual bouncy balls Maximum number of iterations Maximum upper limit of space vector and lower limit ; Step 2: Define the iteration count vector ;set up ; Step 3: Initialize the space vectors of all bouncy balls , ; Step 4: Calculate the fitness of all bouncy balls using the loss function formula (6). , ; Step 5: Update the following calculation process. All parameters during iteration: Step A: Considering the effect of temperature on the bouncy ball, update the following formula: The spatial vector of an elastic sphere: (11) in, For the first A bouncy ball in Space vector during iteration, It is the first The optimal space vector during iteration; For the first Temperature changes during iteration The random coefficients are between 0 and 1. The thermal coefficient is . ; Calculated for the exponent; Determined by the following formula: (12) in, and The first During iteration and the first Temperature during iteration; Step B: Considering the repulsive force between the bouncy balls, continue updating the equation using the following formula. The spatial vector of an elastic sphere: (14) in, For the first A bouncy ball in Space vector during iteration, ,and ; In the first In the next iteration, other elastic balls... The attractive force of an elastic ball is determined by the following formula: (15) in, Initial attraction parameters, Indicates the first The elastic ball and the first The elastic ball in the first... Spatial vector distance at the next iteration ,and ; Step C: Passing the maximum upper limit of space vectors and lower limit The space vector of all elastic balls is restricted. Step D: Update the best space vector for the current iteration. ; Step 6: Determine the current iteration number Has the maximum number of iterations S been reached? If not, then jump to step 5. Step 7: Output the globally optimal space vector Optimal output , , and the estimated optimal output weight matrix .

5. The energy storage system configuration method for the cascade utilization of retired batteries according to claim 1, characterized in that, The collaborative training of the model was completed using data from the retired battery cloud platform, as detailed below: Step 1: Define the dataset for retired batteries and Used for predictive networks and prediction networks Collaborative training; The historical data of retired batteries from the cloud platform was divided into two equally sized datasets. and , ,in This indicates the number of retired batteries used in each of the two datasets. This indicates the number of key evaluation parameters for battery status. This represents the equivalent number of discharge cycles for each retired battery at the time of retirement. definition The equivalent number of discharge cycles used for collaborative training for each retired battery is given by [reference to number of cycles]. ; Dataset and In the middle, each retired battery [1, Recurrent data is used as input data for the prediction network for each decommissioned battery. , The recurrent data is used as evaluation data for the prediction network; Step 2: Complete the prediction network using retired battery data and prediction networks Collaborative training; Dataset Input the pre-trained prediction network respectively and prediction networks Prediction network Data on retired batteries The predicted results are as follows The most confident data and its results, i.e., the selected prediction result SP1, and The datasets were merged; the prediction network was combined. Data on retired batteries The predicted results are as follows The most confident data and its results, i.e., the selected prediction result SP2, and Merge the datasets; after completing the above operations, remove these data from the datasets. Remove from; Step 3: Repeat step 2 until... It is an empty set; Step 4: Complete the co-training to obtain the prediction network. Depend on Updated and prediction networks Depend on Updated ; After completing the above training, a prediction network can be obtained. and prediction networks A predictive model for key evaluation parameters of the health status of retired batteries, which enables joint decision-making.

6. The energy storage system configuration method for the cascade utilization of retired batteries according to claim 1, characterized in that, Using the trained enhanced battery state prediction network, the key evaluation parameters of each retired power battery for new energy vehicles are predicted. The calculation steps for calculating the remaining service life of the retired battery are as follows: Step 1: Initialize the key evaluation parameters of the battery status of a single retired battery using historical data from the battery cloud platform, and obtain... , , Indicates the first One retired battery; Step 2: Input the key evaluation parameters of the battery state of a single retired battery into the battery state prediction network that has completed co-training, and obtain the equivalent number of cycles at the time of retirement for that retired battery. Until the battery is no longer in use Predicted values ​​of key battery state assessment parameters. This represents the percentage of battery health at the time a retired battery ceases use. The corresponding number of battery cycles, and has ; Step 3: Calculate the remaining service life of a single retired battery. ; Here, the percentage decay of SOH is defined as... When a battery is retired and no longer in use, the RUL (Relative Usage Limit) is calculated using the following formula: (16) Step 4: Predict the key evaluation parameters of the retired batteries, obtaining the parameters for each retired battery from the time of retirement. to Key battery state evaluation parameters obtained through model prediction and calculation Here .

7. The energy storage system configuration method for the cascade utilization of retired batteries according to claim 1, characterized in that, Based on the predicted and calculated key evaluation parameters for each retired battery, Based on the principle of similar status, complete the normalization classification of excellent, good, average and poor.

8. A classification controller for an energy storage system suitable for the secondary use of retired batteries, characterized in that, For implementing the energy storage system configuration method for the secondary utilization of retired batteries as described in any one of claims 1-7, the controller includes: The data input module is used to obtain key evaluation parameters of battery status. The model pre-training module uses battery cycle life test data to complete the pre-training of the model; The model co-training module uses historical data from the retired battery cloud platform to complete the co-training of the model; The model loss function module is used to calculate the loss function during the training process; The retired battery classification module uses a trained model to perform normalized classification of a single retired battery. The data output module outputs normalized classification results for the classification of retired batteries.

9. The classification controller for an energy storage system configuration suitable for the secondary utilization of retired batteries according to claim 8, characterized in that: After receiving the input data processed by the input module, the model pre-training module completes the pre-training of the model using battery cycle life testing. The initialization of the augmented battery state prediction network has a total of [number] input data for its input layer. The input data is defined as follows: ,in , This indicates the number of key evaluation parameters used to assess battery status. This indicates the number of time-delayed data points, i.e., the number of data points used to calculate the first time delay. When using key evaluation parameters for battery state of being after a certain number of cycles, use The calculation is performed using the delayed data as input; Hidden layers define the feature vectors of the entire model. Including the regular part and extensions ; Regular section , ; Extended section , ,in This represents the internal scaling factor of the prediction model, used to control the linearity of states over time. tanh(*) is the activation function. Characterizing the external scaling factor, It is the input weight matrix. It is the state weight matrix. and These are the number of neurons in the input layer and the reserve pool, respectively. ; The number of neurons in the output layer is Here , which is the number of key evaluation parameters, used to represent the prediction result at the next time point; The pre-trained output weight matrix is ​​obtained. and , During the training process , .

10. The classification controller for an energy storage system configuration suitable for the cascade utilization of retired batteries according to claim 9, characterized in that: After receiving the input data processed by the input module, the model co-training module uses cloud platform data from retired batteries to complete the co-training of the model. The historical data of retired batteries from the cloud platform was divided into two equally sized datasets. and , ,in This indicates the number of retired batteries used in each of the two datasets. This indicates the number of key evaluation parameters for battery status. This represents the equivalent number of discharge cycles for each retired battery at the time of retirement; definition The equivalent number of discharge cycles used for collaborative training for each retired battery is given by [reference to number of cycles]. ; Dataset and In the middle, each retired battery [1, Recurrent data is used as input data for the prediction network for each decommissioned battery. , The recurrent data is used as evaluation data for the prediction network; Dataset Input the pre-trained prediction network respectively and prediction networks The prediction network 1 pairs the data of retired batteries. The predicted results are as follows The most confident data and its results, i.e., the selected prediction result SP1, and The datasets were merged; the prediction network 2 was used to combine the data from retired batteries. The predicted results are as follows The most confident data and its results, i.e., the selected prediction result SP2, and Merge the datasets; after completing the above operations, remove these data from the datasets. Remove from; Repeat the above process until... The set is empty; the prediction network is obtained. Depend on Updated and prediction networks Depend on Updated Thus, a prediction network is obtained. and prediction networks A predictive model for key evaluation parameters of the health status of retired batteries, which enables joint decision-making; The model loss function module is used to calculate the loss function during training, specifically to solve for the following loss function: in, Represents trace calculation, Representation matrix The transpose operation, It is about diagonal matrix, , For the output matrix The element in row i and column j Scaling factor, and The penalty coefficients are set between 0 and 1. , and This is the loss coefficient; The retired battery classification module, its normalized classification, For each retired battery, cloud platform data collected during its usage is used as input to the retired battery status classification controller. After analysis and prediction by the model within the classification controller, the key evaluation parameters of the retired battery are obtained. , , Indicates the first One retired battery; Formula (16) is used to obtain the contents of Key evaluation parameters of battery status for each retired battery in the data. , ; By using normalized classification, the state of each retired battery is determined as excellent, good, medium, or poor. Based on the state of the retired batteries after classification, the configuration and development of an energy storage system for the cascade utilization of retired batteries are completed.

Images