Intelligent method and system for dynamic parameter weight adjustment in battery cell sorting process

Abstract

The invention belongs to the technical field of battery sorting, and discloses an intelligent method and system for dynamic parameter weight adjustment in the battery cell sorting process. The method comprises the steps of deploying sensors in an electric vehicle and a charging station, and collecting cell data in real time; performing cleaning and normalization processing on the battery cell data to obtain standard battery cell data; performing correlation analysis on the standard battery cell data to obtain battery cell sorting parameters; an initial weight is given to each cell sorting parameter; splicing the initial weights of the cell sorting parameters as an initial weight vector, and optimizing the initial weight vector based on a natural heuristic optimization algorithm to obtain the optimal weight of the cell sorting parameters; according to the method, the weight can be dynamically updated according to real-time monitoring data, the sorting accuracy is always ensured, the overall quality and reliability of battery cell sorting are greatly improved, and a solid guarantee is provided for efficient and stable operation of a two-wheeled vehicle and a three-wheeled vehicle.

Description

Technical Field

[0001] The present invention relates to the technical field of battery sorting, and more specifically, to an intelligent method and system for adjusting dynamic parameter weights during battery cell sorting. Background Art

[0002] Compared with electric vehicles, the batteries used in two-wheelers and three-wheelers differ in specifications, charge and discharge characteristics, and usage environment. Traditional battery cell sorting methods are often designed based on the characteristics of electric vehicle batteries, and fail to fully consider the characteristics of two-wheeler and three-wheeler batteries, resulting in sorting results that cannot meet their actual needs.

[0003] The Chinese patent application with publication number CN117872147A discloses a retired battery sorting method based on SOA-FCM, which specifically includes: S1, constructing SOA-FCM, the input parameters of the model include static parameters and dynamic parameters, the static parameters include voltage, internal resistance, SOC and SOH, and the dynamic parameters include voltage change rate ΔU, SOC change rate ΔSOC and SOH change rate ΔSOH; initializing the parameters, determining the weight and threshold encoding of FCM; S2, calculating the fitness value of each seagull, comparing the fitness values ​​between individuals, and finding the individual optimal and overall optimal values ​​of the current population; S3, updating the position of the seagull, and updating the new position of the seagull after the migration behavior by cyclically judging the optimization target; S4, updating the final position; S5, comparing the fitness between the individuals in the current population again, and finding the global optimal value; S6, judging whether the algorithm reaches the termination condition during execution, if not, the calculation goes to steps S4~S6 to continue the position update; S7, outputting the overall optimal position and fitness value of the SOA algorithm. This invention solves the problem that the traditional unsupervised clustering FCM algorithm is prone to fall into local optimality, and effectively suppresses the life decay rate after the single cells are combined into a power battery system.

[0004] Although the above method can meet most scenarios, research and practical application of the above method and existing technology have found that the above method and existing technology have at least the following defects: In the retired battery sorting method based on SOA-FCM, although the characteristic parameters of the battery, such as rated capacity, ohmic internal resistance, open circuit voltage, SOC, SOH and their change rates are dynamically clustered, there is no clear mechanism for flexibly adjusting the weights of each dynamic parameter; and once the model structure and parameter initialization settings are determined in this method, in the subsequent sorting process, targeted adjustments cannot be made in time for new special situations or changes in data distribution, resulting in a decrease in the sorting effect.

[0005] In view of this, the present invention proposes an intelligent method and system for adjusting dynamic parameter weights during battery cell sorting to solve the above problems. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention provides the following technical solution: an intelligent method for adjusting dynamic parameter weights during cell sorting, comprising the following steps: Deploy sensors in electric vehicles and charging stations to collect battery cell data in real time; Clean and normalize the battery cell data to obtain standard battery cell data; Conduct correlation analysis on standard battery cell data to obtain battery cell sorting parameters; Assign initial weights to each cell sorting parameter; The initial weights of the cell sorting parameters are concatenated as the initial weight vector, which is optimized based on the nature-inspired optimization algorithm to obtain the optimal weights of the cell sorting parameters.

[0007] Furthermore, the method for obtaining the optimal weight of the cell sorting parameters includes: Step 1: Use the initial weight vector as the prairie dog individual and randomly generate weight vectors, and the initial weight vector is summed up as the population size The initial population of Step 2: Construct a fitness function for the population based on performance indicators; the performance indicators include driving range, operating temperature, and lifespan; Step 3: In each iteration, for each prairie dog individual in the population, calculate the corresponding fitness value; find the optimal individual in the population in the current iteration based on the population update strategy; Step 4: Real-time monitoring of the performance indicators of the battery during actual use and charging; Step 5: When the preset conditions are met or the set number of iterations is reached, the obtained optimal weight vector is applied to the weight adjustment of the cell sorting parameters.

[0008] Furthermore, the method for constructing the fitness function includes: Step 2.1: Define the state space ; Step 2.2: Define the action space ; Step 2.3: Design reward function ; Step 2.4, initialize a Q value table for storing the value of state-action pairs, wherein the rows of the Q value table correspond to states and the columns correspond to actions, and initialize all Q values ​​in the Q value table to 0; Step 2.5: Extract the cell sorting parameters from the cell data to obtain training data. , according to the current state , select an action , perform the action After that, observe the next state and rewards received , based on a preset experience replay buffer , used to record the time step The corresponding quadruple ; Step 2.6: Replay from the Experience Buffer Extraction The corresponding Q value is calculated based on the Q-Learning algorithm: Step 2.7, repeat step 2.6 until the predetermined number of training times is reached, select the input state The action with the highest Q value ;in, For the The status corresponding to each cell sorting parameter; Step 2.8: The state corresponding to each cell sorting parameter and the action with the highest Q value As the input of the performance prediction model, the predicted performance index is obtained; the mean square error between the predicted performance index and the actual performance index is compared with the execution action The inverse of the product of the changes in the Q value is used as the fitness function.

[0009] Furthermore, the method for obtaining the reward function includes: constructing a reward function based on the sum of the change in cruising range, the change in operating temperature and the change in life; wherein, a positive reward is given when the cruising range is improved, the operating temperature is reduced or the life is extended after the selected action is performed compared with before the selected action is performed, and a negative reward is given when the cruising range is reduced, the operating temperature is increased or the life is shortened.

[0010] Furthermore, the training method of the performance prediction model includes: Collect training data in advance, the training data includes The state corresponding to the cell sorting parameters, the action with the highest Q value and the corresponding performance indicators are grouped; the training data is divided into a training set, a validation set and a test set according to a preset ratio; the training set is used as the input of the performance prediction model, and the parameters of the performance prediction model are adjusted by the gradient descent algorithm to minimize the loss function of the performance prediction model on the training set; the loss function is the mean square error between the corresponding predicted performance indicator and the actual corresponding performance indicator, and the validation set is used to monitor the prediction mean absolute error of the performance prediction model during the training process. When the prediction mean absolute error of the performance prediction model on the validation set reaches the preset prediction mean absolute error threshold, the test set is used to evaluate the prediction accuracy of the performance prediction model. When the prediction accuracy reaches the preset accuracy threshold, the corresponding performance prediction model is obtained.

[0011] Furthermore, the population update strategy is as follows: The population is sorted in descending order according to the fitness value, and the individuals ranked in the top WA% are marked as elite individuals; the individuals ranked WA%~WB% are marked as middle individuals, and the individuals ranked after WB% are marked as ordinary individuals; among them, WA and WB are both fitness division thresholds; a conservative update strategy is used to update the elite individuals; a conventional update strategy is used to update the middle individuals; and an aggressive update strategy is used to update the ordinary individuals; among them, the conservative update strategy combines the foraging adjustment coefficient and the adaptive coefficient to update the elite individuals, the conventional update strategy combines the adaptive coefficient to update the middle individuals, and the aggressive update strategy combines the adaptive coefficient and the random trigger probability to update the ordinary individuals. A memory mechanism is introduced in the updating process of the elite individuals, the middle individuals and the ordinary individuals to record the weight vector corresponding to the historical optimal fitness value of the individuals in the population, and guide the individuals in the population to approach the global optimal weight vector in each iteration.

[0012] Furthermore, the battery cell data includes voltage, current, ambient temperature, operating temperature, battery cell internal resistance, battery cell capacity, charge state data, health state data, self-discharge rate, charge and discharge times, charging time and discharging time.

[0013] Furthermore, the method for obtaining the standard battery cell data includes: The cell data is compared with the corresponding preset range threshold, abnormal cell data exceeding the preset range threshold is eliminated, the cell data after the abnormal cell data is eliminated is normalized, and the normalized cell data is used as the standard cell data.

[0014] Furthermore, the method for obtaining the cell sorting parameters includes: The Pearson correlation coefficient between the standard battery cell data and the performance index is calculated and compared with the preset correlation threshold. When the Pearson correlation coefficient exceeds the preset correlation threshold, the corresponding standard battery cell data is used as the battery cell sorting parameter.

[0015] Furthermore, the method of assigning an initial weight to each cell sorting parameter includes: An initial weight is assigned to each cell sorting parameter in a random manner.

[0016] An intelligent system for dynamic parameter weight adjustment during battery cell sorting, and an intelligent method for implementing the dynamic parameter weight adjustment during battery cell sorting, comprising: Data collection module: used to deploy sensors in electric vehicles and charging stations to collect battery cell data in real time; Data processing module: used to clean and normalize the battery cell data to obtain standard battery cell data; Data analysis module: used to perform correlation analysis on standard battery cell data and obtain battery cell sorting parameters; Weight assignment module: used to assign initial weights to each cell sorting parameter; Dynamic optimization module: used to concatenate the initial weights of the cell sorting parameters as the initial weight vector, optimize the initial weight vector based on the nature-inspired optimization algorithm, and obtain the optimal weights of the cell sorting parameters.

[0017] Technical effects and advantages of the intelligent method and system for dynamic parameter weight adjustment in the cell sorting process of the present invention: The present invention monitors the dynamic parameters of the battery cells of two-wheeled and three-wheeled vehicles under different working conditions in real time. The intelligent algorithm can quickly and accurately adjust the weights of various parameters according to preset sorting targets and real-time data characteristics, thereby improving the performance matching of the battery cells in actual complex usage scenarios, and effectively avoiding the problem of substandard battery cell performance caused by unreasonable parameter weights. Moreover, during long-term use, as the battery cell performance naturally decays, the method can dynamically update the weights based on real-time monitoring data, always ensuring the accuracy of sorting, greatly improving the overall quality and reliability of battery cell sorting, and providing a solid guarantee for the efficient and stable operation of two-wheeled and three-wheeled vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic flow chart of an intelligent method for adjusting dynamic parameter weights in a cell sorting process of the present invention; Figure 2 A schematic flow chart of a method for obtaining optimal weights of battery cell sorting parameters according to the present invention; Figure 3 A block diagram of an intelligent system for dynamic parameter weight adjustment in the cell sorting process of the present invention; Figure 4Schematic diagram of the intelligent system interface for dynamic parameter weight adjustment in the cell sorting process of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] Example 1

[0021] See also Figure 1 As shown, the intelligent method for adjusting dynamic parameter weights in the cell sorting process described in this embodiment includes the following steps: Deploy sensors in electric vehicles and charging stations to collect battery cell data in real time; The battery cell data includes voltage, current, ambient temperature, operating temperature, battery cell internal resistance, battery cell capacity, charge state data, health status data, self-discharge rate, number of charge and discharge times, charging time and discharge time; among them, the charge state data is the ratio of the current remaining power of the battery cell to the power in the fully charged state; the health status data is the ratio of the current capacity of the battery cell to the initial capacity; the self-discharge rate is the rate at which the battery cell naturally consumes power when not in use. Voltage data can be used to determine the charge state and charge and discharge degree of the battery cell, providing a basis for considering the immediate working state and remaining power of the battery cell in the subsequent dynamic parameter weight adjustment; current data can reflect the power output or input of the battery cell during the charging and discharging process, and help dynamically adjust the weights of power-related parameters to adapt to the energy conversion requirements under different working conditions; ambient temperature affects the performance of the battery cell, and plays an important reference role in the later construction of fitness functions, dynamic weight adjustment, and population update strategies, and also helps to understand the actual working state of the battery cell in different environments; operating temperature data is crucial for monitoring whether the working environment of the battery cell is normal, and is the key basis for considering the thermal stability of the battery cell and the degree of influence of temperature on performance in the dynamic parameter weight adjustment; the internal resistance data of the battery cell can reflect the energy loss inside the battery cell, and is used to measure the efficiency and heat generation of the battery cell during the charging and discharging process in the dynamic parameter weight adjustment; the battery cell capacity data is directly related to the endurance of the electric vehicle, and is used in the sorting process. It is an important reference for adjusting the weights of parameters related to battery life to match application scenarios with different battery life requirements; the state of charge data can accurately indicate the current remaining power ratio of the battery cell, and provide the real-time power information of the battery cell for the dynamic parameter weight adjustment, so as to reasonably arrange the charging and discharging strategy; the health status data is a key indicator for evaluating the degree of battery cell performance degradation, and is used to judge the service life and reliability of the battery cell in the dynamic parameter weight adjustment to ensure the quality of the sorted battery cells; the self-discharge rate data can reflect the power loss characteristics of the battery cell when it is stationary, and is used to consider the battery cell storage performance and power retention ability in long-term use when adjusting the dynamic parameter weight; the charge and discharge number data is an important factor in measuring the life of the battery cell, and provides information on the degree of battery cell aging for the dynamic parameter weight adjustment, so that the sorting process can consider the remaining effective charge and discharge cycle number of the battery cell; the charging time and discharge time data can reflect the efficiency and actual use of the battery cell during the charging and discharging process, and can optimize the parameter weights related to the charging and discharging performance in the dynamic parameter weight adjustment.

[0022] Clean and normalize the battery cell data to obtain standard battery cell data; Methods for obtaining standard battery cell data include: Compare the cell data with the corresponding preset range threshold, remove abnormal cell data that exceeds the preset range threshold, and normalize the cell data after removing the abnormal cell data ,in, For the Kind The cell data after normalization of the cell data; For the Kind Battery cell parameters; and Respectively Kind The maximum and minimum values ​​of the battery cell parameters are obtained; the normalized battery cell data are used as the standard battery cell data.

[0023] Conduct correlation analysis on standard battery cell data to obtain battery cell sorting parameters; Methods for obtaining cell sorting parameters include: The Pearson correlation coefficient between the standard battery cell data and the performance indicators is calculated and compared with the preset correlation threshold. When the Pearson correlation coefficient exceeds the preset correlation threshold, the corresponding standard battery cell data is used as the battery cell sorting parameter; the performance indicators include cruising range, operating temperature and life.

[0024] Assign initial weights to each cell sorting parameter; The method of assigning an initial weight to each cell sorting parameter includes: Assign initial weights to each cell sorting parameter in a random manner ,satisfy ;in, For the The initial weight of each cell sorting parameter, The number of cell sorting parameters.

[0025] Will The initial weights of the cell sorting parameters are spliced ​​as the initial weight vector, and the initial weight vector is optimized based on the nature-inspired optimization algorithm to obtain the optimal weights of the cell sorting parameters.

[0026] Reference Figure 2 , the method of obtaining the optimal weights of the cell sorting parameters includes: Step 1: Initial weight vector As a prairie dog individual, randomly generated Weight vector , and the initial weight vector is summarized as the population size The initial population of is the current; is the ambient temperature; is the operating temperature; is the internal resistance of the battery cell; is the cell capacity; is the state of charge data; Health status data; is the self-discharge rate; is the number of charge and discharge times; For charging time; is the discharge time; Step 2: Construct a fitness function for the population based on performance indicators; Constructing the fitness function The methods include: Step 2.1: Define the state space : ;in, For the Cell sorting parameters; is the number of cell sorting parameters; Step 2.2: Define the action space : ;in, For the Weight adjustment value of each cell sorting parameter; Step 2.3: Design reward function : ;in, and They are the mileage before and after performing the selected action; and They are the operating temperature fluctuations before and after performing the selected action, respectively; and are the lifespans before and after performing the selected action, respectively; , and For the corresponding , and The weight coefficient of for The value of for The value of for The value of Step 2.4, initialize a Q value table for storing the value of state-action pairs, wherein the rows in the Q value table correspond to states and the columns correspond to actions, and initialize all Q values ​​in the Q value table to 0; Step 2.5: Extract the collected cell sorting parameters to obtain training data. , according to the current state , select an action , perform the action After that, observe the next state and the rewards received , based on a preset experience replay buffer , used to record the time step The corresponding quadruple ; Step 2.6: Replay from the Experience Buffer Extraction The corresponding Q value is calculated based on the Q-Learning algorithm: ; in, for The Q value update value of For the status Next action Q value; is the learning rate; is the time step The corresponding reward function value; is the discount factor; Status All actions The maximum value of the medium Q value; Step 2.7, repeat step 2.6 until the predetermined number of training times is reached, select the input state The action with the highest Q value ;in, For the The status corresponding to each cell sorting parameter; Step 2.8: The state corresponding to each cell sorting parameter and the action with the highest Q value As the input of the performance prediction model, the predicted performance index is obtained; the mean square error between the predicted performance index and the actual performance index is compared with the execution action The inverse of the product of the changes in the Q value is used as the fitness function.

[0027] Methods for obtaining predictive performance indicators include: The first The state corresponding to each cell sorting parameter and the action with the highest Q value As an input to the performance prediction model, a prediction performance index is obtained; The training methods of the performance prediction model include: Collect training data in advance, the training data includes The state corresponding to the cell sorting parameters, the action with the highest Q value and the corresponding performance indicators are grouped; the training data is divided into a training set, a validation set and a test set according to a preset ratio; the training set is used as the input of the performance prediction model, and the parameters of the performance prediction model are adjusted by the gradient descent algorithm to minimize the loss function of the performance prediction model on the training set; the loss function is the mean square error between the corresponding predicted performance indicator and the actual corresponding performance indicator, and the validation set is used to monitor the prediction mean absolute error of the performance prediction model during the training process. When the prediction mean absolute error of the performance prediction model on the validation set reaches the preset prediction mean absolute error threshold, the test set is used to evaluate the prediction accuracy of the performance prediction model. When the prediction accuracy reaches the preset accuracy threshold, the corresponding performance prediction model is obtained.

[0028] Get the fitness function ;in, is the mean square error between the predicted performance index and the actual performance index; is the balance coefficient; To perform actions The change in the post-Q value; Step 3: In each iteration For each individual in the population , that is, the value of the corresponding weight vector, calculate the corresponding fitness value; find the optimal individual in the population in the current number of iterations based on the population update strategy .

[0029] The population update strategy is as follows: The population is sorted in descending order according to the fitness value, and the individuals ranked in the top WA% are marked as elite individuals; the individuals ranked in the range of WA% to WB% are marked as middle-level individuals, and the individuals ranked after WB% are marked as ordinary individuals; WA and WB are both fitness division thresholds; in this embodiment, WA is 10 and WB is 70; The conservative update strategy is used to update the elite individuals. The conservative update strategy combines the foraging adjustment coefficient and the adaptive coefficient to update the elite individuals. Specifically, it includes: In foraging behavior, the update formula is: :in, For the elite The first iteration The value of the weight vector; For the elite The first iteration The value of the weight vector; for A random number between For the foraging adjustment coefficient, the elite individuals take smaller steps when approaching the optimal solution to avoid destroying the existing excellent solution structure; For the elite The value of the weight vector corresponding to the optimal fitness value obtained in the iteration; is the value of the weight vector corresponding to the historical optimal fitness value; is the adaptive coefficient. As the number of iterations increases, Gradually increases, so that the elite individuals can refer to their own historical optimal solutions while approaching the global optimal solution, improving search efficiency; based on random trigger probability Trigger alert behavior. When the alert behavior is triggered, the elite individuals are updated according to the following formula: :in, for A random number between To trigger alert behavior, the elite The first iteration The value of the weight vector; is the warning adjustment coefficient; For the The value of the random weight vector at iteration .

[0030] The conventional update strategy is used to update the individuals in the middle layer. The conventional update strategy is combined with the adaptive coefficient to update the individuals in the middle layer, which specifically includes: In foraging behavior, the update formula is: :in, The middle layer The first iteration The value of the weight vector; The middle layer The first iteration The value of the weight vector; The middle layer The value of the weight vector corresponding to the optimal fitness value obtained in the iteration; based on the random trigger probability Triggering alert behavior. When the alert behavior is triggered, the middle-level individuals are updated according to the following formula: :in, To trigger the alert behavior, the middle layer The first iteration The value of the weight vector.

[0031] The radical update strategy is used to update the common layer individuals. The radical update strategy combines the adaptive coefficient and the random trigger probability to update the common layer individuals. Specifically, it includes: In foraging behavior, the update formula is: :in, For the ordinary layer The first iteration The value of the weight vector; For the ordinary layer The first iteration The value of the weight vector; For the ordinary layer The value of the weight vector corresponding to the optimal fitness value obtained in the iteration; based on the random trigger probability Trigger the alert behavior. When the alert behavior is triggered, the common layer individuals are updated according to the following formula: :in, To trigger the alert behavior, the first The first iteration The value of the weight vector.

[0032] Step 4: Real-time monitoring of the performance indicators of the battery during actual use and charging; Step 5: When the preset conditions (such as algorithm convergence) are met or the set number of iterations is reached, the obtained optimal weight vector is applied to the weight adjustment of the cell sorting parameters.

[0033] Example 2

[0034] This embodiment provides an intelligent method for adjusting dynamic parameter weights in the battery cell sorting process, which also includes the following steps: According to the chemical reaction equation corresponding to the actual power supply in the battery cell, the concentration of the substances involved in the actual power supply process is determined, and the reaction entropy at different times is calculated based on the collected battery cell data; Methods for calculating reaction entropy include: Obtain the chemical reaction equation of the electrode reaction involved in the actual power supply process ,in, , , and It is the substance that participates in the actual power supply process; , , and Material , , and Corresponding coefficients; Calculate the corresponding reaction entropy ,in, , , and Material , , and The corresponding concentration; Calculation of theoretical voltage based on Nernst equation : , calculate the theoretical voltage change. Among them, is the standard electrode potential; is the gas constant; is the absolute temperature; is the number of electrons transferred in the reaction; ' is the reaction entropy; is the Faraday constant; the deviation between the theoretical voltage calculation and the actual voltage ;in, is the actual voltage; Combining theoretical voltage and deviation to fitness function Make corrections to obtain the corrected fitness function ; Physical constraints are introduced to ensure that the value of each weight vector is not less than 0 and the sum of all weight vectors is 1.

[0035] The optimal weights of the cell sorting parameters are further optimized based on the modified fitness function.

[0036] Example 3

[0037] See also Figure 3 As shown, the intelligent system for dynamic parameter weight adjustment in the cell sorting process described in this embodiment includes: Data collection module: used to deploy sensors in electric vehicles and charging stations to collect battery cell data in real time; Data processing module: used to clean and normalize the battery cell data to obtain standard battery cell data; Data analysis module: used to perform correlation analysis on standard battery cell data and obtain battery cell sorting parameters; Weight assignment module: used to assign initial weights to each cell sorting parameter; Dynamic optimization module: used to concatenate the initial weights of the cell sorting parameters as the initial weight vector, optimize the initial weight vector based on the nature-inspired optimization algorithm, and obtain the optimal weights of the cell sorting parameters.

[0038] Reference Figure 4, the system interface presents multi-dimensional information of the battery cell. The sensor status is good, of which 128 online devices have a data collection rate of 99.8%, 256K data has been processed, and the processing efficiency has reached 95.2%. The system has clearly performed 32 optimizations and obtained 12 optimal solutions. Battery cell data collection covers multiple indicators such as voltage, current, ambient temperature and operating temperature, and has abnormal data and normalization functions. The correlation analysis part shows that factors such as cruising range are closely related. The weight configuration is 0.35, 0.45 and 0.20 for the operating temperature, battery cell temperature and self-discharge rate, respectively. The current optimization parameters are set to have a population size of 100, a number of iterations of 1000, and a learning rate of 0.01, which are used for battery cell status monitoring and performance evaluation optimization.

[0039] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

[0040] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent method for dynamic parameter weight adjustment during cell sorting, characterized in that: The steps include: Collect battery cell data in real time; Clean and normalize the battery cell data to obtain standard battery cell data; Conduct correlation analysis on standard battery cell data to obtain battery cell sorting parameters; Assign initial weights to each cell sorting parameter; The initial weights of the cell sorting parameters are concatenated as the initial weight vector, which is optimized based on the nature-inspired optimization algorithm to obtain the optimal weights of the cell sorting parameters.

2. The intelligent method for adjusting dynamic parameter weights in the cell sorting process according to claim 1, characterized in that: The method for obtaining the optimal weight of the cell sorting parameters comprises: Step 1: Use the initial weight vector as the prairie dog individual and randomly generate weight vectors, and the initial weight vector is summed up as the population size The initial population of Step 2: Construct a fitness function for the population based on performance indicators; the performance indicators include driving range, operating temperature, and lifespan; Step 3: In each iteration, for each prairie dog individual in the population, calculate the corresponding fitness value; find the optimal individual in the population in the current iteration based on the population update strategy; Step 4: Real-time monitoring of the performance indicators of the battery during actual use and charging; Step 5: When the preset conditions are met or the set number of iterations is reached, the obtained optimal weight vector is applied to the weight adjustment of the cell sorting parameters.

3. The intelligent method for adjusting dynamic parameter weights in the cell sorting process according to claim 2, characterized in that: The method for constructing the fitness function includes: Step 2.1: Define the state space ; Step 2.2: Define the action space ; Step 2.3: Design reward function ; Step 2.4, initialize a Q value table for storing the value of state-action pairs, wherein the rows in the Q value table correspond to states and the columns correspond to actions, and initialize all Q values ​​in the Q value table to 0; Step 2.5: Extract the cell sorting parameters from the cell data to obtain training data. , according to the current state , select an action , perform the action After that, observe the next state and rewards received , based on a preset experience replay buffer , used to record the time step The corresponding quadruple ; Step 2.6: Replay from the Experience Buffer Extraction The corresponding Q value is calculated based on the Q-Learning algorithm: Step 2.7, repeat step 2.6 until the predetermined number of training times is reached, select the input state The action with the highest Q value ;in, For the The status corresponding to each cell sorting parameter; Step 2.8: The state corresponding to each cell sorting parameter and the action with the highest Q value As the input of the performance prediction model, the predicted performance index is obtained; the mean square error between the predicted performance index and the actual performance index is compared with the execution action The inverse of the product of the changes in the Q value is used as the fitness function.

4. The intelligent method for adjusting dynamic parameter weights in the cell sorting process according to claim 3, characterized in that: The method for obtaining the reward function includes: constructing a reward function based on the sum of the change in cruising range, the change in operating temperature and the change in life; wherein, when the cruising range is improved, the operating temperature is reduced or the life is extended after the selected action is performed compared with before the selected action is performed, a positive reward is given, and a negative reward is given when the cruising range is reduced, the operating temperature is increased or the life is shortened.

5. The intelligent method for adjusting dynamic parameter weights in the cell sorting process according to claim 3, characterized in that: The training method of the performance prediction model includes: Collect training data in advance, the training data includes The state corresponding to the cell sorting parameters, the action with the highest Q value and the corresponding performance indicators are grouped; the training data is divided into a training set, a validation set and a test set according to a preset ratio; the training set is used as the input of the performance prediction model, and the parameters of the performance prediction model are adjusted by the gradient descent algorithm to minimize the loss function of the performance prediction model on the training set; the loss function is the mean square error between the corresponding predicted performance indicator and the actual corresponding performance indicator, and the validation set is used to monitor the prediction mean absolute error of the performance prediction model during the training process. When the prediction mean absolute error of the performance prediction model on the validation set reaches the preset prediction mean absolute error threshold, the test set is used to evaluate the prediction accuracy of the performance prediction model. When the prediction accuracy reaches the preset accuracy threshold, the corresponding performance prediction model is obtained.

6. The intelligent method for adjusting dynamic parameter weights in the battery cell sorting process according to claim 2, characterized in that: The population update strategy is as follows: The population is sorted in descending order according to the fitness value, and the individuals ranked in the top WA% are marked as elite individuals; the individuals ranked WA%~WB% are marked as middle individuals, and the individuals ranked after WB% are marked as ordinary individuals; among them, WA and WB are both fitness division thresholds; a conservative update strategy is used to update the elite individuals; a conventional update strategy is used to update the middle individuals; and an aggressive update strategy is used to update the ordinary individuals; among them, the conservative update strategy combines the foraging adjustment coefficient and the adaptive coefficient to update the elite individuals, the conventional update strategy combines the adaptive coefficient to update the middle individuals, and the aggressive update strategy combines the adaptive coefficient and the random trigger probability to update the ordinary individuals. A memory mechanism is introduced in the updating process of the elite individuals, the middle individuals and the ordinary individuals to record the weight vector corresponding to the historical optimal fitness value of the individuals in the population, and guide the individuals in the population to approach the global optimal weight vector in each iteration.

7. The intelligent method for adjusting dynamic parameter weights in the battery cell sorting process according to claim 1, characterized in that: The battery cell data includes voltage, current, ambient temperature, operating temperature, battery cell internal resistance, battery cell capacity, charge state data, health state data, self-discharge rate, charge and discharge times, charging time and discharging time.

8. The intelligent method for adjusting dynamic parameter weights in the battery cell sorting process according to claim 1, characterized in that: The method for obtaining the standard battery cell data includes: The cell data is compared with the corresponding preset range threshold, abnormal cell data exceeding the preset range threshold is eliminated, the cell data after the abnormal cell data is eliminated is normalized, and the normalized cell data is used as the standard cell data.

9. The intelligent method for adjusting dynamic parameter weights in the battery cell sorting process according to claim 1, characterized in that: The method for obtaining the cell sorting parameters includes: The Pearson correlation coefficient between the standard battery cell data and the performance index is calculated and compared with the preset correlation threshold. When the Pearson correlation coefficient exceeds the preset correlation threshold, the corresponding standard battery cell data is used as the battery cell sorting parameter.

10. The intelligent method for dynamic parameter weight adjustment in the battery cell sorting process according to claim 1, characterized in that: The method of assigning an initial weight to each cell sorting parameter comprises: An initial weight is assigned to each cell sorting parameter in a random manner.

11. An intelligent system for dynamic parameter weight adjustment during battery cell sorting, implementing the intelligent method for dynamic parameter weight adjustment during battery cell sorting according to any one of claims 1 to 10, characterized in that: include: Data collection module: used to deploy sensors in electric vehicles and charging stations to collect battery cell data in real time; Data processing module: used to clean and normalize the battery cell data to obtain standard battery cell data; Data analysis module: used to perform correlation analysis on standard battery cell data and obtain battery cell sorting parameters; Weight assignment module: used to assign initial weights to each cell sorting parameter; Dynamic optimization module: used to concatenate the initial weights of the cell sorting parameters as the initial weight vector, optimize the initial weight vector based on the nature-inspired optimization algorithm, and obtain the optimal weights of the cell sorting parameters.

Images