A lithium ion battery fast charging method based on a strategy library

Abstract

The application discloses a kind of lithium ion battery fast charging method based on strategy library, belong to the field of lithium ion battery, including: by the grid strategy library construction method based on meta reinforcement learning, for different battery model, health state, state of charge, temperature and charging time weight combination Pre-training and store optimization charging strategy and its corresponding voltage, temperature confidence interval and predicted charging time.Before charging, system according to battery real-time parameter matching strategy library, and provide intuitive charging time selection interface, so that user can easily select personalized charging scheme.During charging, system real-time monitoring battery state, if voltage or temperature exceeds confidence interval, then dynamically adjust strategy or call online model to generate special strategy.The application effectively reduces the computing burden, improves the real-time adaptability and safety of charging process, and at the same time, realizes fast, safe, personalized lithium ion battery charging through user-friendly interaction mode.

Description

Technical Field

[0001] This invention belongs to the field of lithium-ion batteries, and particularly relates to a fast charging method for lithium-ion batteries based on a strategy library. Background Technology

[0002] As a core energy storage component in new energy vehicles, lithium-ion batteries have always faced a technical challenge in balancing charging speed, safety, and lifespan. Traditional charging strategies, such as the constant current-constant voltage (CC-CV) method, are simple and safe, but suffer from low charging efficiency and a lack of multi-objective optimization capabilities. To improve charging performance, existing technologies have developed offline optimization strategies (such as multi-stage constant current, Taguchi optimization, and dynamic programming) and online optimization strategies (such as model predictive control and deep reinforcement learning).

[0003] However, while offline strategies offer higher security, they cannot dynamically adjust based on real-time battery status; online strategies, while highly adaptable, impose a heavy computational burden, pose risks of overheating and overvoltage, and are difficult to popularize in practical applications. Furthermore, existing personalized solutions often require users to set specialized parameters, resulting in complex operations and a poor user experience. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a fast charging method for lithium-ion batteries based on a strategy library, comprising:

[0005] An offline fast charging strategy library is constructed. The strategy library includes charging strategies trained by weighted combinations of different battery models, health status, state of charge, current temperature and charging time, as well as the voltage and temperature confidence intervals and expected charging time corresponding to each strategy.

[0006] Before charging, obtain the current parameters of the battery to be charged, match the corresponding charging strategy from the strategy library according to the current parameters, and determine the final charging strategy according to the charging time selected by the user.

[0007] During the charging process, the battery voltage and temperature are monitored in real time. If the monitored data exceeds the confidence interval corresponding to the selected strategy, a new charging strategy is matched from the strategy library based on the current parameters or a dedicated charging strategy is generated by calling the online model until charging is complete.

[0008] Optionally, the construction of the offline fast charging strategy library includes:

[0009] Based on the gridded task space, for each combination of battery type, health status, state of charge, current temperature and charging time weights, the corresponding charging strategy is obtained by training a meta-reinforcement learning model.

[0010] For each charging strategy, perform multiple simulated charging operations, statistically analyze the confidence intervals of voltage and temperature changes over time, and calculate the average charging time.

[0011] The charging strategy is associated with and stored with its corresponding confidence interval and average charging time to form the offline fast charging strategy library.

[0012] Optionally, the task space is composed of a combination of battery type, state of charge range, health status range, temperature range, and charging time weight.

[0013] Optionally, determining the final charging strategy based on the user-selected charging time includes:

[0014] Based on the current parameters, extract the estimated charging time corresponding to all available charging time weights from the strategy library, and display the shortest charging time, longest charging time, and default charging time to the user.

[0015] Receive the user's desired charging time selected between the shortest and longest charging times;

[0016] The corresponding charging time weight is determined based on the expected charging time, and the charging strategy that matches the weight is selected as the final charging strategy.

[0017] Optionally, the real-time monitoring of battery voltage and temperature includes:

[0018] The battery's real-time voltage and temperature data are continuously collected during the charging process;

[0019] The real-time data is compared with the voltage and temperature confidence intervals corresponding to the selected charging strategy;

[0020] If the real-time data exceeds the confidence interval, a policy adjustment process is triggered.

[0021] Optionally, the triggering strategy adjustment process includes:

[0022] If the confidence interval is exceeded for the first time, a new charging strategy is re-matched from the strategy library based on the current battery parameters;

[0023] If the confidence interval is exceeded again within the preset time, the charging time weight is reduced and the charging strategy is rematched.

[0024] If the confidence interval is exceeded for the third time within a preset time, the current parameters are input into the meta-reinforcement learning model to generate a dedicated charging strategy online.

[0025] Optionally, the charging strategy outputs a normalized charging current command through a continuous action space, and the actual charging current is the product of the command and the maximum allowable charging current.

[0026] Optionally, the reward function of the meta-reinforcement learning model is a weighted combination of charging time, energy efficiency, and battery aging penalty, and safety constraints of voltage and temperature are applied during training.

[0027] On the other hand, the present invention also provides an electronic device including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0028] On the other hand, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0029] Compared with the prior art, the present invention has the following advantages and technical effects:

[0030] This invention proposes a lithium battery fast charging method that combines an offline fast charging strategy library trained on a Meta-RL model based on a grid policy library with online supervision. First, the ample offline policy library avoids the high computational cost of online methods. Second, the integration with an online supervision system addresses the limitation of offline methods in flexibly adjusting based on current battery data. Finally, it provides users with the option to select the estimated charging time, catering to individual user needs. Any user can intuitively select the ideal optimization weights without any prior knowledge. This invention retains the advantages of both offline and online charging strategies while avoiding their key drawbacks, proposing a lithium battery fast charging approach that combines offline and online modes, and also providing solutions adapted to individual user needs. Attached Figure Description

[0031] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0032] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0033] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0034] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0035] Example 1

[0036] This embodiment provides a fast charging method for lithium-ion batteries based on a strategy library, including:

[0037] An offline fast charging strategy library is constructed. The strategy library includes charging strategies trained by weighted combinations of different battery models, health status, state of charge, current temperature and charging time, as well as the voltage and temperature confidence intervals and expected charging time corresponding to each strategy.

[0038] Before charging, obtain the current parameters of the battery to be charged, match the corresponding charging strategy from the strategy library according to the current parameters, and determine the final charging strategy according to the charging time selected by the user.

[0039] During the charging process, the battery voltage and temperature are monitored in real time. If the monitored data exceeds the confidence interval corresponding to the selected strategy, a new charging strategy is matched from the strategy library based on the current parameters or a dedicated charging strategy is generated by calling the online model until charging is complete.

[0040] The lithium-ion battery fast charging strategy proposed in this embodiment mainly includes three modules: offline strategy library construction, optimization strategy selection, and online supervision. First, using a Meta-RL model based on a grid strategy library, a separate offline fast charging strategy is trained for each different battery model, different stages of SOH and SOC, current temperature, and different combinations of charging time weights. Through multiple simulated charging operations, the confidence intervals for voltage and temperature, as well as the estimated charging time, are calculated for each strategy. These strategies, along with their corresponding confidence intervals and charging times, constitute the offline strategy library. Then, when the user is charging, the various parameters of the lithium battery are first acquired. Based on these parameters, a matching strategy is found from the strategy library, providing the shortest, longest, and default estimated charging times under different charging time weights. The user can freely choose between the shortest and longest estimated charging times. Finally, the charging strategy corresponding to the most suitable charging time weight is selected as the initial fast charging strategy to begin charging. Finally, during charging, the current voltage and temperature of the battery are monitored in real time to see if they exceed the confidence interval of the strategy. If the interval is exceeded, a new charging strategy is selected from the strategy library based on the current parameters. If the voltage or temperature exceeds the interval multiple times in a short period of time, the current battery parameters are directly input into the Meta-RL model to calculate a dedicated charging strategy and simulated and estimated confidence intervals for voltage and temperature to replace the offline strategy until charging is complete.

[0041] (1) Construction of offline policy library;

[0042] First, a Meta-RL model based on a grid-based policy library is built to train offline fast charging policies. This model discretizes the continuous task space into a structured grid and trains an independent, fixed, and fully validated policy at each grid point, constructing a predictable, verifiable, efficient, and queryable policy library. This method combines the generalization ability of Meta-RL with the deterministic and security advantages of a policy library, making it suitable for application scenarios with high security requirements and structured state spaces in lithium battery fast charging policies.

[0043] To cover all scenarios during actual charging, the offline policy library needs to include combinations of weights for each battery type, SOH, SOC, current temperature, and charging time, defining the task. As shown in formula (1):

[0044] (1)

[0045] in This represents a set of battery models. Represents a set of different SOC intervals. Represents the set of different SOH intervals. This indicates the current temperature range of different batteries. This represents the set of different charging time weights, with larger ones... This indicates a priority towards shorter charging times. The task grid is defined accordingly. for:

[0046] (2)

[0047] For each individual task ,Strategy The expected cumulative return is defined as a weighted combination of the three optimization objectives: charging time, energy efficiency, and battery aging. The reward function is defined as follows: As shown in formula (3):

[0048] (3)

[0049] in Weight the charging time for this task. The charging time required to achieve the charging target; This is an energy efficiency indicator. Energy efficiency weighting; As punishment for aging This represents the weighting of aging factors. In other words, it aims to minimize charging time while maximizing energy efficiency and minimizing battery aging.

[0050] At the same time, the strategy must satisfy security constraints, and at any given time... voltage The voltage limit must not be exceeded. At any time Battery temperature The temperature limit must not be exceeded. As shown in formula (4):

[0051] (4)

[0052] When a strategy violates the above security constraints, it will be considered a serious violation and subject to severe penalties.

[0053] Based on the definitions of task space, reward function, and security constraints mentioned above, a fast charging policy model for Meta-RL based on a grid policy library is constructed, with the state space defined as follows:

[0054] (5)

[0055] in They represent Battery voltage, temperature, and SOC at any given moment are the core dynamic state variables; These represent the health status, cumulative cycle count, and battery model, respectively. These parameters are considered static during a charging cycle.

[0056] To achieve smooth and safe current control, this model adopts a continuous action space:

[0057] (6)

[0058] This action represents a normalized charging current command, the actual charging current. for:

[0059] (7)

[0060] in This represents the maximum permissible charging current. Continuous operation space generates a smooth charging curve, avoiding battery stress caused by sudden current changes in traditional multi-stage constant current charging strategies.

[0061] Based on the above definition, a fast charging policy is trained for each grid point in the task space using reinforcement learning. All strategies constitute an offline strategy library. ,Right now:

[0062] (8)

[0063] in, For the task space, It is a task The trained fast charging strategy. Then, for each trained fast charging strategy... ,conduct Each charging simulation records the voltage and temperature trajectories, as well as the total charging time, obtained from each simulation. The calculations are performed at each time point. The average of all analog voltages and temperatures:

[0064] (9)

[0065] in and Indicates the first In this simulation The voltage and battery temperature measured at all times, and Indicates in The sample mean values ​​of voltage and battery temperature at each time point. Confidence intervals are constructed by calculating the upper and lower quantiles of each time point using the mean values.

[0066] (10)

[0067] in and It is the width of the confidence interval for voltage and temperature. and They are respectively in The confidence intervals for voltage and temperature at any given time.

[0068] At the same time for each task Calculate the average charging time across all charging simulations as the expected charging time for this task:

[0069] (11)

[0070] in Indicates the first Charging time in this simulation For the task Expected charging time.

[0071] For each task Train an optimal fast charging strategy offline. Combined with the confidence intervals of its voltage and temperature and and expected charging time This ultimately forms the offline fast charging strategy library. As shown in Formula 12:

[0072] (12)

[0073] Strategy library This is a complete offline fast charging strategy library, which includes a combination of weights for each battery type, SOC, SOH, temperature, and charging time, as well as the corresponding fast charging strategy, confidence intervals for voltage and temperature, and estimated charging time.

[0074] Before actually charging, the most suitable strategy needs to be selected from the strategy library based on the user's current battery data, while also giving the user some choice.

[0075] When a user connects a lithium battery to the charging system, the system first obtains the battery's model, SOH, SOC, and current temperature. Based on these parameters, it retrieves all charging time weights for that parameter combination from the strategy library. The strategy with the highest charging time weight is then selected. The corresponding estimated charging time serves as the shortest charging time that users can choose. Lowest charging time weight The corresponding estimated charging time serves as the maximum charging time that the user can choose. At the same time, it provides a most balanced default weight. , Expected charging time corresponding to the strategy This is the default charging time provided to the user. Users can... You can freely choose an ideal charging time. After selection, the system will determine the most suitable charging time weight based on that time. The strategy corresponding to this weight and battery data is determined as the initial fast charging strategy.

[0076] In addition, when a user selects a charging duration, the system will prompt the user about the positive or negative effects of the selection compared to the default duration.

[0077] After selecting the fast charging strategy, the system begins to charge the lithium battery according to the strategy. In order to ensure that the actual charging situation is significantly different from the offline simulation situation and to prevent the voltage and temperature from exceeding the safety threshold, the current temperature and voltage of the battery need to be continuously monitored during charging.

[0078] When an unexpected situation occurs where the voltage or temperature exceeds the confidence interval, the system immediately re-acquires the battery's current SOC, voltage, and temperature. Based on these parameters, it selects the most suitable strategy from the offline strategy library to continue charging. If the voltage or temperature exceeds the upper interval a second time within a short period, the system re-acquires the parameters, reduces the charging time weight by one level, and then acquires a new strategy to continue charging. If the voltage or temperature exceeds the upper interval a third time within a short period, the system acquires the current parameters, reduces the charging time weight, and then directly passes them to the Meta-RL model for calculation, determining a completely new, dedicated strategy independent of the strategy library to continue charging until charging is complete. Furthermore, if an unexpected situation occurs where the voltage or temperature suddenly reaches a safety threshold, the system immediately terminates charging, informs the user of the specific anomaly, and prohibits further charging of the battery.

[0079] Therefore, this invention designs an offline policy library construction method for Meta-RL based on a grid policy library. A separate charging policy is trained for each battery type, SOH, SOC, current temperature, and charging time weight combination. A confidence interval for voltage and temperature during charging, as well as the estimated charging time, are calculated through simulated charging. The policies for all parameter combinations, along with their voltage and temperature confidence intervals and estimated charging times, constitute a complete offline policy library.

[0080] This invention proposes a fast charging method combining offline and online modes. It acquires battery parameters and searches an offline policy library for the best-matching strategy for charging. During charging, it monitors online whether the voltage and temperature exceed confidence intervals. If they do, it reacquires the current parameters and retrieves a new offline strategy. If the voltage and temperature repeatedly exceed the confidence interval within a short period, it reduces the weight of charging time or directly trains a dedicated strategy online, depending on the situation. This method combines the advantages of both online and offline modes, minimizing their respective drawbacks.

[0081] This invention proposes a personalized adaptation scheme that allows users to freely select the ideal charging time from the estimated charging time corresponding to different time weights. Users do not need to understand the technical terms such as priority and weight; they can simply select the ideal strategy by adjusting the ideal charging time. It is a simple, easy-to-understand, and efficient solution to meet the personalized needs of different users.

[0082] On the other hand, this embodiment also provides an electronic device, including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0083] On the other hand, this embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0084] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for fast charging lithium-ion batteries based on a strategy library, characterized in that, include: An offline fast charging strategy library is constructed. The strategy library includes charging strategies trained by weighted combinations of different battery models, health status, state of charge, current temperature and charging time, as well as the voltage and temperature confidence intervals and expected charging time corresponding to each strategy. Before charging, obtain the current parameters of the battery to be charged, match the corresponding charging strategy from the strategy library according to the current parameters, and determine the final charging strategy according to the charging time selected by the user. During the charging process, the battery voltage and temperature are monitored in real time. If the monitored data exceeds the confidence interval corresponding to the selected strategy, a new charging strategy is matched from the strategy library according to the current parameters or a dedicated charging strategy is generated by calling the online model until charging is completed. The construction of the offline fast charging strategy library includes: Based on the gridded task space, for each combination of battery type, health status, state of charge, current temperature and charging time weights, the corresponding charging strategy is obtained by training a meta-reinforcement learning model. For each charging strategy, perform multiple simulated charging operations, statistically analyze the confidence intervals of voltage and temperature changes over time, and calculate the average charging time. The charging strategy is associated with its corresponding confidence interval and average charging time and stored to form the offline fast charging strategy library; The process of determining the final charging strategy based on the user-selected charging time includes: Based on the current parameters, extract the estimated charging time corresponding to all available charging time weights from the strategy library, and display the shortest charging time, longest charging time, and default charging time to the user. Receive the user's desired charging time selected between the shortest and longest charging times; The corresponding charging time weight is determined based on the expected charging time, and the charging strategy that matches the weight is selected as the final charging strategy.

2. The method according to claim 1, characterized in that, The task space is composed of a combination of battery type, state of charge range, health status range, temperature range, and charging time weight.

3. The method according to claim 1, characterized in that, The real-time monitoring of battery voltage and temperature includes: The battery's real-time voltage and temperature data are continuously collected during the charging process; The real-time voltage and temperature data are compared with the voltage and temperature confidence intervals corresponding to the selected charging strategy; If the real-time voltage and temperature data exceed the confidence interval, the strategy adjustment process is triggered.

4. The method according to claim 3, characterized in that, The triggering strategy adjustment process includes: If the confidence interval is exceeded for the first time, a new charging strategy is re-matched from the strategy library based on the current battery parameters; If the confidence interval is exceeded again within the preset time, the charging time weight is reduced and the charging strategy is rematched. If the confidence interval is exceeded for the third time within a preset time, the current parameters are input into the meta-reinforcement learning model to generate a dedicated charging strategy online.

5. The method according to claim 1, characterized in that, The charging strategy outputs a normalized charging current command through a continuous action space, and the actual charging current is the product of the command and the maximum allowable charging current.

6. The method according to claim 1, characterized in that, The reward function of the meta-reinforcement learning model is a weighted combination of charging time, energy efficiency, and battery aging penalty, and safety constraints of voltage and temperature are applied during training.

7. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, characterized in that, When the processor executes the computing program, it implements the method of any one of claims 1-6.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-6.

Images