CN116683588B

CN116683588B - Lithium ion battery charge and discharge control method and system

Info

Publication number: CN116683588B
Application number: CN202310961087.8A
Authority: CN
Inventors: 罗加鹏; 刘伟强
Original assignee: Shenzhen Shentong World Technology Co ltd
Current assignee: Shenzhen Shentong World Technology Co ltd
Priority date: 2023-08-02
Filing date: 2023-08-02
Publication date: 2024-01-26
Anticipated expiration: 2043-08-02
Also published as: CN116683588A

Abstract

The invention relates to the technical field of lithium battery control, in particular to a method and a system for controlling charge and discharge of a lithium ion battery. The method comprises the following steps: monitoring and recording state parameters of the lithium ion battery and carrying out parameter preprocessing through an internet of things (IoT) device to obtain battery preprocessing data; performing battery health condition assessment on the battery pretreatment data so as to obtain battery health scoring data; generating a charge-discharge strategy according to the battery health scoring data, and acquiring an initial charge-discharge strategy model; performing strategy optimization on the initial charge-discharge strategy model to obtain an optimized charge-discharge strategy model; and carrying out charge and discharge operation through the cloud platform according to the optimized charge and discharge strategy model, and collecting data in real time to obtain cloud synchronous data so as to update the optimized model of the optimized charge and discharge strategy model. According to the invention, through continuous evaluation of the battery health condition and optimization of the charge-discharge strategy, the service life of the battery can be effectively prolonged, so that the frequency of battery replacement and related cost are reduced.

Description

Lithium ion battery charge and discharge control method and system

Technical Field

The invention relates to the technical field of lithium battery control, in particular to a method and a system for controlling charge and discharge of a lithium ion battery.

Background

The charge and discharge control method of the lithium ion battery generally relates to how to manage the charge and discharge processes of the lithium ion battery through a hardware and software control system so as to protect the battery, prolong the service life of the battery and improve the efficiency of the battery. This typically involves various aspects of monitoring the state of the battery, management of the amount of power, formulation of charge and discharge strategies, assessment of the state of health, etc. Conventional charging methods are typically based on a fixed charging strategy, often without consideration of the actual state of the battery (e.g., battery health, temperature, electrochemical characteristics), which may lead to reduced battery life or performance.

Disclosure of Invention

The invention provides a method and a system for controlling charge and discharge of a lithium ion battery to solve at least one technical problem.

The application provides a charge and discharge control method of a lithium ion battery, which comprises the following steps:

step S1: monitoring and recording state parameters of the lithium ion battery and performing parameter preprocessing through an internet of things (IoT) device, so as to obtain battery preprocessing data;

step S2: performing battery health condition assessment on the battery pretreatment data so as to obtain battery health scoring data;

Step S3: generating a charge-discharge strategy according to the battery health scoring data, thereby acquiring an initial charge-discharge strategy model;

step S4: performing strategy optimization on the initial charge-discharge strategy model so as to obtain an optimized charge-discharge strategy model;

step S5: and carrying out charge and discharge operation through the cloud platform according to the optimized charge and discharge strategy model and collecting data in real time, so as to obtain cloud synchronous data, and updating the optimized charge and discharge strategy model.

According to the invention, through continuous evaluation of the battery health condition and optimization of the charge-discharge strategy, the service life of the battery can be effectively prolonged, so that the frequency of battery replacement and related cost are reduced. The optimized charge-discharge strategy can improve the performance of the battery, such as the charge efficiency and the discharge stability, thereby improving the overall performance of the device. The system can adjust the charge and discharge strategy according to the health condition of the battery, so that the safety problem caused by over-charge and discharge of the battery or other improper operation can be avoided. Advanced internet technology, internet of things (IoT) devices and artificial intelligence algorithms are utilized to make battery management more intelligent, capable of responding to changes in battery status in real time and making adjustments accordingly. The cloud data synchronized in real time can be used for continuously optimizing the model, so that the management and the use of the battery are more efficient, and meanwhile energy conservation and environmental protection are facilitated.

Preferably, step S1 is specifically:

step S11: deploying and configuring through the IoT device, thereby obtaining device configuration data;

step S12: real-time parameter monitoring is carried out according to the equipment configuration data, so that original monitoring data are obtained;

step S13: performing data quality evaluation on the original monitoring data so as to obtain monitoring quality scoring data;

step S14: detecting and processing abnormal values of the monitoring quality scoring data, so as to obtain abnormal value processing data;

step S15: and carrying out data standardization and aggregation processing on the abnormal value processing data so as to obtain battery pretreatment data.

According to the invention, through real-time parameter monitoring, data quality evaluation, abnormal value detection and processing, and data standardization and aggregation processing, the quality of data can be improved, so that the subsequent battery health evaluation and charge-discharge strategy optimization are more accurate and effective. Since the initial step is the deployment and configuration of IoT devices, the devices can be flexibly configured according to actual needs to meet different application scenarios. By preprocessing the data, such as data normalization and aggregation, the complexity of the data can be reduced and the efficiency of subsequent data processing and analysis can be improved. Outlier detection and handling can avoid systematic errors caused by data anomalies, thereby improving the robustness of the system.

Preferably, the data quality evaluation is performed by an evaluation process by a data quality evaluation calculation formula, wherein the data quality evaluation calculation formula is specifically:

;

for monitoring quality score data, < >>For time item->Dimension data for original monitoring data, +.>For the order item of the original monitoring data, +.>For monitoring quality score base constant term, +.>For the original monitoring data, ++>Is->Temperature data in the individual raw monitoring data, +.>Is->Humidity data in the individual raw monitoring data, +.>Is->The pressure data in the individual raw monitoring data,and (5) smoothing the degree term for the original monitoring data.

The invention constructs a data quality evaluation calculation formula, which evaluates the quality of multi-dimensional data and considers the smoothness and the change trend of the data. Monitoring quality score dataThis is the output data quality score. The higher the value, the better the data quality. Time item->Representing the change of data quality over a certain period of time by means of the pair +.>And deriving to obtain the variation trend of the data quality score. Dimension data of original monitoring data +.>This value represents usHow many pieces of independent monitoring data are available. Monitoring quality score base constant term- >This value may adjust the base value of the score, normalizing or normalizing the score. Original monitoring data smoothness term->This value is used to adjust the smoothness of the data quality score. The first part of the formula is to find the limit of the logarithmic function, which has certain smoothness and can be used for smoothing the data distribution and reducing the noise of the data. The second part of the formula is the euclidean distance (square root of sum of squares) of three parameters, which can reflect the change of the data. The third part of the formula is to find the smoothness of the data, which can be used to measure the smoothness of the data. By +.>Is the derivative of (2) to obtain the data quality score +.>This rate of change can be used to observe the change in data quality score over time.

Preferably, step S2 is specifically:

step S21: carrying out data extraction on the battery pretreatment data so as to obtain battery condition basic data and battery operation basic data;

step S22: respectively carrying out feature extraction on the battery condition basic data and the battery operation basic data so as to obtain battery condition feature data and battery operation feature data;

Step S23: building a health evaluation model for the battery condition characteristic data and the battery operation characteristic data, so as to build an initial health evaluation model;

step S24: model training is carried out on the initial health assessment model, so that a trained health assessment model is obtained;

step S25: performing model verification on the trained health evaluation model so as to obtain a model verification result;

step S26: performing model adjustment on the trained health assessment model by using a model verification result, so as to obtain an adjusted health assessment model;

step S27: and monitoring and evaluating the battery pretreatment data by using the adjusted health evaluation model, so as to obtain battery monitoring and scoring data.

According to the invention, the running state of the battery is comprehensively acquired, and the health condition of the battery is comprehensively estimated through feature extraction and model construction, so that the abnormal condition of the battery can be found in time, and the service life and safety of the battery are improved. All the sub-steps of the step S2, including the construction, training, verification and adjustment of the model, are all in fine operation, so that the accuracy of the evaluation model can be ensured, and the reliability of the evaluation is improved. Through continuous model verification and adjustment, the evaluation model is ensured to adapt to the change of the battery state, and continuous monitoring and evaluation of the battery health condition are realized. The battery pretreatment data is monitored and evaluated by utilizing the adjusted health evaluation model, so that battery monitoring scoring data is obtained, the change of the battery health condition can be found in time, early warning is provided, and serious problems of the battery are prevented.

Preferably, step S24 is specifically:

step S241: acquiring a historical data set, and dividing the historical data set by utilizing battery condition basic data and battery operation basic data so as to acquire a historical training data set and a historical verification data set;

step S242: extracting features of the historical training data set to obtain historical battery condition feature data and historical battery operation feature data, and constructing a feature matrix of the historical battery condition feature data and the historical battery operation feature data to obtain a feature matrix data set;

step S243: performing target vector conversion on the historical verification data set so as to obtain verification target vector data;

step S244: performing model iterative training on the initial health assessment model by utilizing the feature matrix data set and a preset early-stop strategy, so as to acquire initial health assessment model parameter data;

step S245: and reversely training the initial health assessment model parameter data by using the verification target vector data and a preset early-stop strategy, so as to obtain a trained health assessment model.

According to the method, the historical data set is obtained, and the data set is divided, so that the data used in the training and verification process of the model are not repeated, the model is prevented from being fitted excessively, and the generalization capability of the model is improved. Through feature extraction and feature matrix construction, the training process can be performed in an effective feature space, and the training effect is improved. Through target vector conversion and a preset early-stop strategy, the training quality of the model can be ensured, and meanwhile, the training efficiency of the model can be greatly improved. Through reverse training, model parameters can be continuously adjusted in the model training process, so that the model can be converged more quickly, and the training stability is ensured. Step S24 not only extracts the historical battery condition and the running characteristic data, but also further improves the accuracy of the health evaluation model through matrix construction and parameter iterative training. Reverse training can be performed according to the verification target vector data, and dynamic adjustment of model parameters is achieved, so that the health condition of the battery can be estimated more accurately.

Preferably, the monitoring evaluation in step S27 is evaluated by a battery monitoring evaluation calculation formula, wherein the battery monitoring evaluation calculation formula is specifically:

；

scoring data for battery monitoring->For the estimated battery life item +.>Scoring a base constant term for battery health, +.>For the number of charge-discharge cycles, +.>In the (th)>Open circuit voltage value at sub-cycle, +.>Is the open circuit voltage value of the battery in the initial state, < >>In the (th)>Charge-discharge current value at sub-cycle, +.>Is the charge-discharge current value of the battery in the initial state, < >>In the (th)>Temperature value at sub-cycle, +.>Is a temperature value of the battery in an initial state.

The invention constructs a battery monitoring evaluation calculation formula which aims at obtaining the health condition score of the battery and reflects the overall performance condition of the battery.Reflecting the change of the voltage of the battery in the charge-discharge cycle, the voltage of the batteryDegradation generally means degradation of the battery. />Reflecting the change in current of the battery during charge and discharge cycles, instability in current may lead to degradation in battery performance. />Reflecting the change in temperature of the battery during charge and discharge cycles, an increase in the battery temperature may cause a decrease in battery performance. Predicting battery life term- >With the increase of the charge and discharge cycle times of the battery, < + >>The value will change reflecting the health and life of the battery. Battery health score base constant term->By varying the base, the sensitivity of the health score can be adjusted. The current state of health of the battery can be intuitively reflected by quantitatively evaluating the changes of the voltage, the current and the temperature to obtain a battery health score. The formula can dynamically reflect the health condition of the battery along with the increase of the charge and discharge cycle times, and is beneficial to finding and preventing possible battery problems. The service life estimation term in the formula reflects the expected performance of the battery in the future, and is beneficial to predicting the service life of the battery.

Preferably, step S3 is specifically:

step S31: acquiring battery decision environment definition data through the acquisition of the parameters of the IoT device;

step S32: selecting a reinforcement learning algorithm according to the battery decision environment definition data, so as to obtain reinforcement learning strategy data;

step S33: acquiring historical experience interaction data, wherein the historical experience interaction data are generated interactively according to preset lithium ion battery charging and discharging basic strategy data;

step S34: performing strategy updating on the historical experience interaction data by utilizing the reinforcement learning strategy data so as to obtain charging and discharging strategy updating data;

Step S35: and carrying out iterative training on the charging and discharging strategy updating data so as to obtain an initial charging and discharging strategy model.

According to the invention, through the parameter collection and the decision environment definition of the IoT device, the charge and discharge strategy can be customized according to the actual state and the environmental condition of the battery, the charge and discharge efficiency is improved, and the service life of the battery is prolonged. By utilizing the reinforcement learning algorithm, the charge-discharge strategy can be continuously optimized according to the historical experience interaction data, so that the charge-discharge strategy is more in line with the actual application requirements. The charge-discharge strategy is continuously adjusted and optimized by carrying out iterative training on the charge-discharge strategy so as to adapt to the change of the battery state and the environmental condition. By using the IoT device and the reinforcement learning algorithm, intelligent management of the battery charge-discharge strategy is realized, the battery state can be predicted more accurately, the battery performance is prevented from being reduced or damaged in advance, and the service life and the stability of the battery are greatly improved.

Preferably, the reinforcement learning strategy data includes reinforcement learning type selection data, learning parameter setting data, exploration strategy data and model structure selection data, and the step S32 is specifically:

step S321: carrying out decision environment description according to the battery decision environment definition data so as to obtain decision environment description data;

Step S322: performing reinforcement learning type selection according to the decision environment description data so as to obtain reinforcement learning type selection data;

step S323: acquiring charge and discharge control demand data, and determining learning parameters of reinforcement learning type selection data by utilizing battery decision environment definition data and the charge and discharge control demand data, so as to acquire learning parameter setting data;

step S324: and generating exploration strategy data according to the learning parameter setting data, and generating model structure selection data according to the battery decision environment definition data.

According to the invention, the reinforcement learning strategy suitable for the current environment and the battery state can be generated through reinforcement learning type selection and learning parameter setting according to the decision environment definition data of the battery. This dynamic adaptation mechanism makes the learning strategy more flexible and efficient. By acquiring the charge and discharge control requirement data and the battery decision environment definition data, the learning parameters suitable for the current environment and requirement can be automatically determined, the workload and the error of manually setting the parameters are reduced, and the learning efficiency and accuracy are improved. The search strategy is generated according to the learning parameter setting data, so that excellent strategies can be searched and found more effectively in the learning process, and the optimization speed and quality of the charge-discharge strategy are improved. Model structure selection data is generated according to battery decision environment definition data, and a proper model structure can be selected according to actual environment and requirements, so that flexibility and expansibility of the reinforcement learning model are provided.

Preferably, step S4 is specifically:

step S41: performing actual environment strategy effect evaluation on the initial charge-discharge strategy model so as to obtain strategy evaluation result data;

step S42: acquiring strategy feedback data in the strategy executing process;

step S43: performing strategy feedback analysis on strategy feedback data through preset expert rules, so as to obtain expert feedback analysis report data;

step S44: performing depth policy feedback analysis on the policy feedback data so as to obtain depth feedback analysis report data;

step S45: determining strategy improvement points of the initial charge-discharge strategy model and the strategy evaluation result data by using expert feedback analysis report data and depth feedback analysis report data, thereby obtaining strategy improvement point data;

step S46: generating policy improvement data according to the policy improvement point data;

step S47: and carrying out strategy optimization on the initial charge-discharge strategy model by utilizing strategy improvement data, thereby obtaining an optimized charge-discharge strategy model.

According to the invention, by evaluating the actual environment strategy effect of the initial charge-discharge strategy model, the applicability and effect of the strategy model in the actual environment can be confirmed, and the reliability of the strategy model is increased. The strategy feedback data in the strategy execution process can provide real-time and accurate basis for strategy improvement and can also provide powerful support for the next strategy optimization. The strategy feedback analysis is carried out through the preset expert rules, and the depth strategy feedback analysis is combined, so that the strategy feedback data can be comprehensively and deeply analyzed, the defects and advantages of the strategy are identified and understood, and a more comprehensive and deep basis is provided for strategy improvement. Generating strategy improvement data according to the strategy improvement point data, and optimizing an initial charge-discharge strategy model by utilizing the strategy improvement point data, wherein the process provides opportunities for strategy continuous improvement and optimization, and further improves the performance and effect of the strategy model.

Preferably, the present application further provides a lithium ion battery charge and discharge control system, including:

the battery data acquisition model is used for monitoring and recording state parameters of the lithium ion battery through the IoT device and carrying out parameter preprocessing so as to acquire battery preprocessing data;

the battery health condition evaluation module is used for evaluating the battery health condition of the battery pretreatment data so as to acquire battery health scoring data;

the charge-discharge strategy generation module is used for generating charge-discharge strategies according to the battery health scoring data so as to acquire an initial charge-discharge strategy model;

the strategy optimization module is used for carrying out strategy optimization on the initial charge-discharge strategy model so as to obtain an optimized charge-discharge strategy model;

and the real-time data collection module is used for carrying out charge and discharge operation through the cloud platform according to the optimized charge and discharge strategy model and collecting data in real time, so that cloud synchronous data are obtained, and the optimized charge and discharge strategy model is updated.

The invention has the beneficial effects that: the state parameters of the lithium ion battery are monitored and recorded through the IoT device and are subjected to parameter preprocessing, so that the working state of the battery can be obtained in real time and accurately, the complex battery state parameters can be subjected to proper preprocessing, and accurate and high-quality data support is provided for subsequent battery health condition evaluation and charge-discharge strategy generation. The battery health condition evaluation is carried out on the battery pretreatment data, so that the health condition of the battery can be known in time, and the charging and discharging strategy can be dynamically adjusted according to the health grading data of the battery, so that the charging and discharging strategy can be better adapted to the actual condition of the battery, the battery is protected, and the service life of the battery is prolonged. The charging and discharging strategy is generated according to the battery health scoring data, and the strategy is self-adaptive, can be correspondingly adjusted according to the health condition of the battery, so that the charging and discharging efficiency is improved, the battery is protected, and the abrasion of the battery is reduced. By collecting cloud synchronous data in real time and updating an optimized charge-discharge strategy model, the charge-discharge strategy can be quickly and accurately adjusted according to the real-time data, the flexibility and adaptability of the charge-discharge strategy are improved, and the continuously-changing battery state and running environment requirements are met. The cloud platform is used for carrying out charge and discharge operation and collecting data in real time, so that the collection and processing of the data are more convenient and quicker, meanwhile, the strong computing capacity and the storage capacity of the cloud platform also provide possibility for large data processing and complex model computing, and the generation and optimization of a charge and discharge strategy are more efficient and accurate. The monitoring capability of the IoT device, the evaluation capability of the battery health scoring model, the strategy generation capability of reinforcement learning and the data processing capability of the cloud platform are fully utilized, so that the charge and discharge management of the battery is more accurate, efficient and intelligent, the service efficiency and the service life of the battery are greatly improved, the loss of the battery is reduced, and the method has important value for the management and maintenance of the battery.

Drawings

Other features, objects and advantages of the present application will become more apparent upon reading of the detailed description of non-limiting implementations made with reference to the following drawings in which:

fig. 1 is a flowchart showing steps of a method for controlling charge and discharge of a lithium ion battery according to an embodiment;

FIG. 2 shows a step flow diagram of step S1 of an embodiment;

FIG. 3 shows a step flow diagram of step S2 of an embodiment;

FIG. 4 shows a step flow diagram of step S24 of an embodiment;

FIG. 5 shows a step flow diagram of step S3 of an embodiment;

FIG. 6 shows a step flow diagram of step S32 of an embodiment;

fig. 7 shows a step flow diagram of step S4 of an embodiment.

Detailed Description

The following is a clear and complete description of the technical method of the present patent in conjunction with the accompanying drawings, and it is evident that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, are intended to fall within the scope of the present invention.

Furthermore, the drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus a repetitive description thereof will be omitted. Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically separate entities. The functional entities may be implemented in software or in one or more hardware modules or integrated circuits or in different networks and/or processor methods and/or microcontroller methods.

It will be understood that, although the terms "first," "second," etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another element. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element, without departing from the scope of example embodiments. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.

Referring to fig. 1 to 6, the present application provides a method for controlling charge and discharge of a lithium ion battery, comprising the following steps:

in particular, for example, ioT devices (e.g., embedded sensors) may be used to monitor and record various state parameters of the battery, such as voltage, current, temperature. The preprocessing comprises the steps of filtering, missing value filling, noise reduction and outlier processing, so that the data quality and the accuracy of subsequent analysis are ensured.

specifically, the battery pre-processing data is assessed for health using, for example, a machine learning model (e.g., support vector machine, random forest or neural network, etc.). The battery health score may be based on a number of indicators of battery usage, charge/discharge patterns, environmental factors.

specifically, an initial charge-discharge strategy model may be generated using a reinforcement learning algorithm (e.g., Q-learning, deep Q Networks, etc.), for example, based on battery health score data. The policy model dynamically adjusts the charge and discharge policy based on current battery health and battery history data.

specifically, the initial charge-discharge strategy model is optimized, for example, based on feedback data (e.g., battery performance data, device operation data, etc.) in the actual environment. This may involve optimizing parameters of the model or modifying the structure of the model to accommodate new environments and requirements.

Specifically, for example, the charging and discharging operation is performed through a cloud platform, and the platform may integrate functions of remote monitoring, data analysis, intelligent control and the like. Meanwhile, the platform can collect and synchronize the operation data of the battery in real time. These data are used to further optimize the charge-discharge strategy model, thereby forming an adaptive, continuous learning and optimization system.

Preferably, step S1 is specifically:

specifically, ioT devices are deployed, for example, by installing and configuring some necessary hardware devices (e.g., battery state monitoring sensors), and corresponding system configurations are performed, such as setting data acquisition frequencies, setting alarm thresholds.

specifically, various operating parameters of the battery, such as voltage, current, temperature, and charge and discharge times, are monitored in real time by the monitoring device, for example, according to the device configuration data.

specifically, raw monitoring data is evaluated, for example, by some standard data quality evaluation method (e.g., data integrity check, data consistency check, etc.), to determine the reliability and availability of the data.

specifically, for example, a statistical method (such as a box plot method, a z-score method) is used to detect an outlier, and processing is performed according to circumstances, such as deleting the outlier, replacing with an average value or a median value.

Specifically, for example, data normalization may be performed by a method of subtracting an average value and then dividing by a standard deviation or the like to eliminate the dimensional and scale influence of data; the aggregation treatment can adopt methods such as average value, median, maximum value, minimum value and the like according to requirements so as to facilitate subsequent analysis and treatment.

;

The invention constructs a data quality evaluation calculation formula, which evaluates the quality of multi-dimensional data and considers the smoothness and the change trend of the data. Monitoring quality score dataThis is the output data quality score. The higher the value, the better the data quality. Time item->Representing the change of data quality over a certain period of time by means of the pair +.>And deriving to obtain the variation trend of the data quality score. Dimension data of original monitoring data +.>This value represents how many independent pieces of monitoring data we have. Monitoring quality score base constant term- >This value may adjust the base value of the score, normalizing or normalizing the score. Original monitoring data smoothness term->This value is used to adjust the smoothness of the data quality score. The first part of the formula is to find the limit of the logarithmic function, which has certain smoothness and can be used for smoothing the data distribution and reducing the noise of the data. The second part of the formula is the euclidean distance (square root of sum of squares) of three parameters, which can reflect the change of the data. The third part of the formula is to find the smoothness of the data, which can be used to measure the smoothness of the data. By +.>Is the derivative of (2) to obtain the data quality score +.>This rate of change can be used to observe the change in data quality score over time.

Preferably, step S2 is specifically:

specifically, for example, different information is extracted from the battery pretreatment data according to actual needs, for example, the battery condition basic data may include the charge and discharge times of the battery, the open circuit voltage, and the battery operation basic data may include the temperature and the current of the battery.

in particular, different statistical or machine learning methods may be utilized, for example, in the feature extraction process. For example, for the charge and discharge times of the battery, feature extraction may be performed by calculating the mean value and standard deviation thereof; for current and temperature, it is possible to extract its spectral features using fourier transform methods.

specifically, for example, in constructing the health evaluation model, an appropriate model, such as a linear regression model, a neural network model, may be selected, with battery condition feature data and battery operation feature data as inputs, and the health of the battery as an output.

specifically, the initial health assessment model is trained, for example, by using a suitable optimization algorithm (such as gradient descent method and newton method), so as to obtain a trained health assessment model.

specifically, for example, by using a part of battery data as a verification set, the trained health evaluation model is verified, and evaluation indexes such as errors between the prediction result and the actual value of the model are obtained.

specifically, for example, according to the model verification result, parameters of the model are adjusted or the model structure is changed, so that model optimization is performed. Such as verification results using the model, such as how the model performs on unseen data, such as accuracy, precision, recall, F1 values of the model. Based on these criteria, it is determined whether to adjust the parameters of the model or change the structure of the model. For example, if the model is found to perform poorly on the validation set, possibly due to overfitting caused by the model being too complex, then reducing the complexity of the model, such as reducing the number of layers of the neural network, reducing the number of neurons, increasing the regularization term, may be considered. If the model performs poorly on the training set, possibly due to under-fitting caused by too simple a model, increasing the complexity of the model, such as increasing the number of layers of the neural network, increasing the number of neurons, may be considered. In addition, methods of optimizing models using different optimization algorithms, adjusting learning rates, changing activation functions are also contemplated.

Specifically, for example, the adjusted health evaluation model is used to predict new data of the battery to obtain health score data of the battery, which can be used for monitoring the health condition of the battery. For example, the health score may be a value between 0 and 1, with higher values indicating better health of the battery and conversely worse health of the battery. When the health score is below a certain preset threshold, an alarm may be raised, indicating that maintenance or replacement of the battery is required. The possible problems of the battery are found in advance, and equipment faults or safety accidents caused by the battery problems are avoided.

Preferably, step S24 is specifically:

specifically, the historical data set is divided into a training data set and a verification data set, for example, using random division or in a time division manner. For example, it is possible to use 70% of the data as training set and 30% as validation set.

specifically, for example, feature extraction is performed using a fourier transform, principal component analysis method based on battery condition basis data and battery operation basis data, and then the extracted features are organized into a matrix form, thereby constructing a feature matrix data set.

Specifically, for example, in a history verification data set, data is marked as positive (healthy) or negative (unhealthy) according to the health condition of a battery, and these marks are organized into vectors, and verification target vector data is constructed.

specifically, for example, the feature matrix dataset is input into an initial health assessment model for training. In the training process, when the performance of the verification set is not improved or the improvement degree is smaller than a certain threshold value, the training is stopped, and the overfitting is prevented.

Specifically, for example, during training, the predicted result of the model is compared with verification target vector data, a loss function is calculated, and the parameters of the model are adjusted using a back propagation algorithm to reduce the value of the loss function. Also, during training, an early stop strategy may be used to prevent overfitting.

；

Scoring data for battery monitoring->For the estimated battery life item +.>Scoring a base constant term for battery health, +.>For the number of charge-discharge cycles, +.>In the (th)>Open circuit voltage value at sub-cycle, +.>Is the open circuit voltage value of the battery in the initial state, < >>In the (th)>Charge and discharge during sub-cycleCurrent value->Is the charge-discharge current value of the battery in the initial state, < >>In the (th)>Temperature value at sub-cycle, +.>Is a temperature value of the battery in an initial state.

The invention constructs a battery monitoring evaluation calculation formula which aims at obtaining the health condition score of the battery and reflects the overall performance condition of the battery.A decrease in battery voltage generally means a decrease in battery performance, reflecting a change in battery voltage during charge and discharge cycles. />Reflecting the change in current of the battery during charge and discharge cycles, instability in current may lead to degradation in battery performance. />Reflecting the change in temperature of the battery during charge and discharge cycles, an increase in the battery temperature may cause a decrease in battery performance. Predicting battery life term->With the increase of the charge and discharge cycle times of the battery, < + >>The value will change reflecting the health and life of the battery. Battery health score base constant term- >By changingThe base may adjust the sensitivity of the health score. The current state of health of the battery can be intuitively reflected by quantitatively evaluating the changes of the voltage, the current and the temperature to obtain a battery health score. The formula can dynamically reflect the health condition of the battery along with the increase of the charge and discharge cycle times, and is beneficial to finding and preventing possible battery problems. The service life estimation term in the formula reflects the expected performance of the battery in the future, and is beneficial to predicting the service life of the battery.

Preferably, step S3 is specifically:

specifically, real-time operating parameters of the battery, such as battery voltage, current, temperature, etc., are acquired, for example, using sensors of the IoT device, defining a decision-making environment for the battery.

specifically, for example, a reinforcement learning algorithm suitable for a battery decision environment, such as Q-learning, deep Q-Network (DQN), is selected for subsequent charge-discharge strategy generation. The battery decision context definition data may include factors such as the type of battery, the usage environment, the operation mode, etc., which may affect the optimal charge and discharge strategy. Therefore, the most appropriate reinforcement learning algorithm is selected based on these factors. For example, if the state space and the action space of the battery are small and can be explicitly modeled, a table-type reinforcement learning algorithm, such as Q-learning, can be selected for use; if the state space and the action space of the battery are large or cannot be explicitly modeled, a reinforcement learning algorithm using function approximation, such as Deep Q-Network (DQN), may be selected.

specifically, historical empirical interaction data is formed, for example, based on past charge-discharge strategies and feedback results of the battery. For example, the state change of the battery after each charge or discharge operation is recorded.

specifically, the charge-discharge strategy is updated based on historical empirical interaction data and reinforcement learning strategy data, for example, using a reinforcement learning algorithm, such as Q-learning or DQN. This process may include an update to the Q value (Q-learning) or an approximation to the value function (DQN).

Specifically, for example, after the charge-discharge strategy update data is obtained, the charge-discharge strategy can be gradually optimized through multiple iterative training. Each iteration may include executing a current charge-discharge strategy, observing feedback from the battery, and then updating the strategy based on the feedback. Through the iterative process, an optimized model which can give an optimal charge-discharge strategy in various states can be finally obtained.

specifically, for example, the operating environment and operating conditions of the battery, such as the operating voltage of the battery, the current range, and the ambient temperature, are described, thereby defining a state space for the reinforcement learning model.

specifically, the most appropriate reinforcement learning algorithm may be selected, for example, based on decision context description data. For example, if the environment description data indicates that both the state and action space are large, deep Q-Network (DQN) may be selected for use. Some conventional tabular reinforcement learning algorithms, such as Q-learning, sarsa, or value iteration, are used. Such algorithms can find accurate solutions on small scale problems and with relatively low computational complexity. For example, Q-learning is a well-known tabular reinforcement learning algorithm. The algorithm corresponds each state-action to a Q value that represents the expected return that can be obtained by performing an action in a state. By interacting with the environment in the environment and updating the Q value, Q-learning can learn the optimal strategy gradually. Q-learning is a very practical choice when neither the state nor the action space is large, since it only needs to save the Q value of one state-action pair. However, when the state or action space is large, a tabular reinforcement learning algorithm may encounter difficulties because the value of each state-action pair needs to be saved, which may not be feasible on a large scale. In this case, a reinforcement learning algorithm of function approximation, such as a Deep Q Network (DQN), is used.

specifically, learning parameters such as a learning rate, a damping factor are determined, for example, according to the charge-discharge demand of the battery and the environmental state.

Specifically, for example, an exploration strategy, such as an epsilon-greedy strategy, is generated, and the structure of the reinforcement learning model is determined. For example, depending on the complexity and nature of the battery decision environment, it is chosen whether a deep neural network is required, as well as the number of layers and nodes of the neural network.

Preferably, step S4 is specifically:

specifically, the effect evaluation is performed, for example, by a set KPI (key performance indicator), such as battery life, battery efficiency, charging time.

Step S42: acquiring strategy feedback data in the strategy executing process;

in particular, this feedback data may be obtained from the actual environment, such as the real-time voltage, current, temperature of the battery, for example.

in particular, expert rules, for example, may be set by basic knowledge and empirical rules of the battery, such as an excessive temperature or an excessive voltage of the battery, which may affect the life and performance of the battery.

specifically, for example, the depth policy feedback analysis may be performed with a deep learning algorithm, for example, using RNN (recurrent neural network) to analyze time series data of the battery.

in particular, drawbacks and improvement points of the initial strategy, such as too low a voltage when the battery is charged or too high a temperature when the battery is discharged, can be found, for example, from feedback analysis reports.

in particular, policy parameters may be modified or policies may be changed, such as adjusting charging current or changing charging policies, for example, according to policy improvement points.

Specifically, the policy model may be optimized, for example, by iterative learning or optimization algorithms (such as gradient descent methods).

The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

The foregoing is only a specific embodiment of the invention to enable those skilled in the art to understand or practice the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. The lithium ion battery charge and discharge control method is characterized by comprising the following steps:

Step S3: generating a charge-discharge strategy according to the battery health scoring data, thereby acquiring an initial charge-discharge strategy model; the step S3 specifically comprises the following steps:

step S32: selecting a reinforcement learning algorithm according to the battery decision environment definition data, so as to obtain reinforcement learning strategy data; the reinforcement learning strategy data includes reinforcement learning type selection data, learning parameter setting data, exploration strategy data and model structure selection data, and the step S32 specifically includes:

step S324: generating exploration strategy data according to the learning parameter setting data, and generating model structure selection data according to the battery decision environment definition data;

step S35: performing iterative training on the charging and discharging strategy updating data so as to obtain an initial charging and discharging strategy model;

step S4: performing strategy optimization on the initial charge-discharge strategy model so as to obtain an optimized charge-discharge strategy model; the step S4 specifically comprises the following steps:

step S42: acquiring strategy feedback data in the strategy executing process;

step S47: performing strategy optimization on the initial charge-discharge strategy model by utilizing strategy improvement data so as to obtain an optimized charge-discharge strategy model;

2. The method according to claim 1, wherein step S1 is specifically:

3. The method according to claim 2, wherein the data quality assessment is assessed by a data quality assessment calculation formula, wherein the data quality assessment calculation formula is specifically:

;

For monitoring quality score data, < >>For time item->Dimension data for original monitoring data, +.>For the order item of the original monitoring data, +.>For monitoring quality score base constant term, +.>For the original monitoring data, ++>Is->Temperature data in the individual raw monitoring data, +.>Is->Humidity data in the individual raw monitoring data, +.>Is->Pressure data in the individual raw monitoring data, +.>And (5) smoothing the degree term for the original monitoring data.

4. The method according to claim 1, wherein step S2 is specifically:

5. The method according to claim 4, wherein step S24 is specifically:

6. The method according to claim 4, wherein the monitoring evaluation in step S27 is evaluated by a battery monitoring evaluation calculation formula, wherein the battery monitoring evaluation calculation formula is specifically:

；

7. A lithium ion battery charge and discharge control system for performing the lithium ion battery charge and discharge control method according to claim 1, comprising: