Training method and device of battery fast charge control information prediction model, computer equipment, readable storage medium and program product

Abstract

The invention relates to a training method and device of a battery fast charge control information prediction model, computer equipment, a computer readable storage medium and a computer program product. The method comprises the steps that state information at the current moment serves as target state information, the target state information is input into a deep reinforcement learning network of a target battery, and charging control information of the target battery at the current moment is obtained; performing single-step simulation based on the charging control information at the current moment, determining state information of the target battery at the next moment and a target return value at the current moment, and updating target state information based on the state information at the next moment; the target return value is generated based on cathode potential information of the target battery; and training a deep reinforcement learning network according to the state information of each moment, the target return value and the charging control information and the state information of the next moment corresponding to each moment to obtain a battery fast charging control information prediction model. By adopting the method, quick charging can be realized.

Description

Technical Field

[0001] The present application relates to the field of battery management technology, and in particular to a training method, apparatus, computer equipment, computer-readable storage medium, and computer program product for a battery fast-charging control information prediction model. Background Art

[0002] Traditional fuel vehicles only need a few minutes to recharge, which can be achieved quickly. Compared with traditional fuel vehicles, electric vehicles still take a long time to recharge, and slow charging can take up to several hours, which seriously affects the user's charging experience and affects the large-scale popularization of electric vehicles.

[0003] Therefore, there is an urgent need for a training method for a battery fast charging control information prediction model that can achieve fast charging. Summary of the invention

[0004] Based on this, it is necessary to provide a training method, device, computer equipment, computer-readable storage medium and computer program product for a battery fast charging control information prediction model that can achieve fast charging in response to the above technical problems.

[0005] In a first aspect, the present application provides a method for training a battery fast charging control information prediction model, comprising:

[0006] Acquire the current state information of the target battery, use the current state information as the target state information, input the target state information into the deep reinforcement learning network of the target battery, and obtain the charging control information of the target battery at the current moment;

[0007] Performing a single-step simulation based on the charging control information at the current moment, determining the state information of the target battery at the next moment and the target reward value at the current moment, updating the target state information based on the state information at the next moment, and returning to the step of inputting the target state information into the deep reinforcement learning network of the target battery to obtain the charging control information of the target battery at the current moment; the target reward value is generated based on the negative electrode potential information of the target battery;

[0008] According to the state information, target reward value and charging control information at each moment, as well as the state information of the next moment corresponding to each moment, the deep reinforcement learning network is trained to obtain a battery fast charging control information prediction model.

[0009] In one embodiment, the single-step simulation based on the charging control information at the current moment to determine the state information of the target battery at the next moment and the target return value at the current moment includes:

[0010] Performing a single-step simulation based on the charging control information at the current moment to obtain the state information and hidden information of the target battery at the next moment;

[0011] The target reward value of the target battery at the current moment is determined according to the state information and hidden information at the next moment.

[0012] In one embodiment, the state information includes negative electrode potential information; or the hidden information includes negative electrode potential information; and determining the target reward value of the target battery at the current moment according to the state information and the hidden information at the next moment includes:

[0013] Calculating a negative electrode potential violation penalty value of the target battery at the current moment according to the negative electrode potential information at the next moment;

[0014] Calculating a non-negative electrode potential violation penalty value of the target battery at the current moment according to the state information and hidden information at the next moment;

[0015] The target reward value of the target battery at the current moment is weightedly calculated according to the negative electrode potential violation penalty value and the non-negative electrode potential violation penalty value.

[0016] In one embodiment, the calculating the negative electrode potential violation penalty value of the target battery at the current moment according to the negative electrode potential information at the next moment includes:

[0017] When the power level of the target battery at the next moment is less than or equal to the preset power level threshold, the negative electrode potential violation penalty value of the target battery at the current moment is calculated according to the preset negative electrode potential penalty coefficient and critical value, and the negative electrode potential information at the next moment;

[0018] When the power level of the target battery at the next moment is greater than the preset power level threshold, the preset penalty value is used as the negative electrode potential violation penalty value of the target battery at the current moment.

[0019] In one embodiment, the deep reinforcement learning network includes a policy network, and the step of inputting the target state information into the deep reinforcement learning network of the target battery to obtain the charging control information of the target battery at the current moment includes:

[0020] Based on the preset attenuation function, determine the exploration threshold corresponding to the current moment;

[0021] If the random number generated at the current moment is greater than or equal to the exploration threshold, the target state information is input into the strategy network to obtain the charging control information of the target battery at the current moment;

[0022] If the random number generated at the current moment is less than the exploration threshold, the charging control information of the target battery at the current moment is generated based on the uniform random action generation function.

[0023] In one embodiment, the deep reinforcement learning network includes a policy network and a Q network, and the deep reinforcement learning network is trained according to the state information, target reward value and charging control information at each moment, and the state information of the next moment corresponding to each moment to obtain a battery fast charging control information prediction model, including:

[0024] For each moment in each charging process, a set of charging experience information is generated based on the state information, target reward value and charging control information at the moment, and the state information of the next moment corresponding to the moment;

[0025] When the preset model training conditions are met, m groups of charging experience information are randomly selected every preset time period;

[0026] Inputting the m groups of charging experience information into the strategy network to obtain current prediction values ​​of the charging control information corresponding to the m groups;

[0027] updating the strategy network based on the m groups of charging experience information, the current predicted values ​​of the charging control information corresponding to the m groups, and the Q network;

[0028] When the preset training completion conditions are met, the current strategy network is used as a battery fast charging control information prediction model.

[0029] In a second aspect, the present application provides a battery fast charging method, comprising:

[0030] Get the status information of the target battery;

[0031] Inputting the state information into a pre-trained battery fast charge control information prediction model to obtain the charging control information of the target battery;

[0032] Controlling charging of the target battery according to the charging control information;

[0033] Among them, the battery fast charging control information prediction model is trained by the battery fast charging control information prediction model training method described in the first aspect above.

[0034] In a third aspect, the present application also provides a training device for a battery fast charging control information prediction model, including:

[0035] A first determination module is used to obtain the current state information of the target battery, and use the current state information as the target state information, input the target state information into the deep reinforcement learning network of the target battery, and obtain the charging control information of the target battery at the current moment;

[0036] a second determination module, configured to perform a single-step simulation based on the charging control information at the current moment, determine the state information of the target battery at the next moment, and the target reward value at the current moment, and update the target state information based on the state information at the next moment, and return to the step of inputting the target state information into the deep reinforcement learning network of the target battery to obtain the charging control information of the target battery at the current moment; the target reward value is generated based on the negative electrode potential information of the target battery;

[0037] The training module is used to train the deep reinforcement learning network according to the state information, target reward value and charging control information at each moment, as well as the state information of the next moment corresponding to each moment, so as to obtain a battery fast charging control information prediction model.

[0038] In a fourth aspect, the present application also provides a battery fast charging device, comprising:

[0039] An acquisition module, used to acquire status information of a target battery;

[0040] A prediction module, used to input the state information into a pre-trained battery fast charge control information prediction model to obtain the charging control information of the target battery;

[0041] A control module, used for controlling charging of the target battery according to the charging control information;

[0042] Among them, the battery fast charging control information prediction model is trained by the battery fast charging control information prediction model training method described in the first aspect above.

[0043] In a fifth aspect, the present application further provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps described in the first aspect or the second aspect when executing the computer program.

[0044] In a sixth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps described in the first aspect or the second aspect.

[0045] In a seventh aspect, the present application further provides a computer program product, comprising a computer program, which implements the steps described in the first aspect or the second aspect when executed by a processor.

[0046] The training method, device, computer equipment, computer-readable storage medium and computer program product of the above-mentioned battery fast charging control information prediction model obtain the state information of the target battery at the current moment, and use the state information at the current moment as the target state information, input the target state information into the deep reinforcement learning network of the target battery, and obtain the charging control information of the target battery at the current moment; perform single-step simulation based on the charging control information at the current moment, determine the state information of the target battery at the next moment, and the target reward value at the current moment, and update the target state information based on the state information at the next moment, return to the step of inputting the target state information into the deep reinforcement learning network of the target battery, and obtain the charging control information of the target battery at the current moment; the target reward value is generated based on the negative electrode potential information of the target battery; according to the state information, target reward value and charging control information at each moment, and the state information at the next moment corresponding to each moment, train the deep reinforcement learning network to obtain the battery fast charging control information prediction model. In this way, an interactive closed-loop feedback framework of the battery single-step simulation environment and the deep reinforcement learning network is established. The deep reinforcement learning network predicts the charging control information at the current moment based on the current state information obtained by the battery single-step simulation environment. The battery single-step simulation environment simulates the battery state information at the next moment based on the charging control information output by the deep reinforcement learning network, and generates a target reward value based on the negative electrode potential information of the battery. The deep reinforcement learning network is trained based on the state information at each moment, the target reward value and the charging control information, as well as the state information at the next moment corresponding to each moment, to obtain a battery fast charging control information prediction model. The model can predict the battery charging control information in real time based on the battery status information, adjust the battery charging rate in real time, and achieve fast charging. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0048] FIG1 is a schematic flow chart of a method for training a battery fast charging control information prediction model in one embodiment;

[0049] FIG2 is a flow chart of a step of performing a single-step simulation based on the charging control information at the current moment to determine the state information of the target battery at the next moment and the target reward value at the current moment in one embodiment;

[0050] FIG3 is a flow chart of a step of determining a target reward value of a target battery at a current moment according to state information and hidden information at a next moment in one embodiment;

[0051] FIG4 is a flow chart of a step of calculating a negative electrode potential violation penalty value of a target battery at a current moment according to negative electrode potential information at a next moment in one embodiment;

[0052] FIG5 is a flowchart of steps for inputting target state information into a deep reinforcement learning network of a target battery to obtain charging control information of the target battery at the current moment in one embodiment;

[0053] FIG6 is a flowchart of steps for training a deep reinforcement learning network to obtain a battery fast charging control information prediction model according to state information, target reward value and charging control information at each moment, and state information at the next moment corresponding to each moment in one embodiment;

[0054] FIG7 is a schematic diagram of a power battery fast charging control strategy framework based on deep reinforcement learning in one embodiment;

[0055] FIG8 is a schematic diagram of a current variation curve during charging control using a battery fast charging control information prediction model in one embodiment;

[0056] FIG9 is a schematic diagram of a voltage variation curve during charging control using a battery fast charging control information prediction model in one embodiment;

[0057] FIG10 is a schematic diagram of a curve showing a change in the negative electrode potential during charging control using a battery fast charging control information prediction model in one embodiment;

[0058] FIG11 is a schematic diagram of a curve showing a change in SOC during charging control using a battery fast charging control information prediction model in one embodiment;

[0059] FIG12 is a schematic diagram of a temperature variation curve during charging control using a battery fast charging control information prediction model in one embodiment;

[0060] FIG13 is a schematic diagram of a process flow of a battery fast charging method in one embodiment;

[0061] FIG14 is a structural block diagram of a training device for a battery fast charging control information prediction model in one embodiment;

[0062] FIG15 is a structural block diagram of a battery fast charging device in one embodiment;

[0063] FIG. 16 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0065] In one embodiment, as shown in FIG1 , a method for training a battery fast charging control information prediction model is provided. This embodiment uses the method applied to a terminal as an example for illustration. It is understandable that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. Among them, the terminal can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, projection devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The head-mounted device can be a virtual reality (VR) device, an augmented reality (AR) device, smart glasses, etc. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services. In this embodiment, the method includes the following steps:

[0066] Step 101, obtain the current state information of the target battery, and use the current state information as the target state information, input the target state information into the deep reinforcement learning network of the target battery, and obtain the charging control information of the target battery at the current moment.

[0067] In an embodiment of the present application, the target battery is a battery to be charged or fast-charged, and may be a battery for an electric vehicle. The target battery may be a lithium-ion battery, such as a lithium iron phosphate battery, a nickel-cobalt-manganese (NI CO Mn, NCM) series ternary battery, and a nickel-cobalt-aluminum (NI CO Al, NCA) series ternary battery. The deep reinforcement learning network may be a soft actor-critic (Soft Actor-Critic, SAC) network, a deep deterministic policy gradient (Deep Deterministic Policy Gradient, DDPG) network, and a twin-delayed deep deterministic policy gradient (Twin Delayed Deep Deterministic policy gradient, TD3) network. The deep reinforcement learning network includes an actor (Actor) subnetwork and a critic (Critic) subnetwork. The actor (Actor) subnetwork may be a fully connected neural network, a time series network, and a convolutional neural network. The critic (Critic) subnetwork may be a fully connected neural network, a time series network, and a convolutional neural network. The state information is used to represent the state of the target battery, including at least one of the battery temperature, the state of charge (State of Charge, SOC), and the terminal voltage, and may also include negative electrode potential information. The battery temperature is the temperature inside the battery. The power level indicates the ratio of the available power in the battery to the nominal capacity, usually expressed as a percentage. The negative electrode potential information is used to indicate the negative electrode potential state. The charging control information includes but is not limited to the charging rate, and may also include the cooling flow rate. In this way, electric-thermal coordinated fast charging control can be achieved.

[0068] In one example, the terminal uses the state information as the input state of the deep reinforcement learning network and the charging control information as the action of the deep reinforcement learning network to obtain the state-action pair of the deep reinforcement learning network. For example, the terminal uses the battery temperature, power and terminal voltage as the input state of the deep reinforcement learning network and the charging rate as the action of the deep reinforcement learning network to obtain the state-action pair of the deep reinforcement learning network.

[0069] Step 102, perform single-step simulation based on the charging control information at the current moment, determine the state information of the target battery at the next moment, and the target reward value at the current moment, and update the target state information based on the state information at the next moment, return to input the target state information into the deep reinforcement learning network of the target battery, and obtain the charging control information step of the target battery at the current moment.

[0070] The target return value is generated based on the negative electrode potential information of the target battery.

[0071] In the embodiment of the present application, the terminal performs a single-step simulation based on the charging control information at the current moment in a pre-established single-step simulation environment of the target battery to determine the state information of the target battery at the next moment and the target reward value at the current moment. Then, the terminal updates the target state information to the state information at the next moment, and returns to the step of inputting the target state information into the deep reinforcement learning network of the target battery to obtain the charging control information of the target battery at the current moment.

[0072] In one example, the terminal establishes a battery model of a target battery. Then, the terminal performs model state initialization processing on the battery model to obtain a battery model after state initialization. Then, the terminal establishes a single-step simulation environment for the target battery based on the battery model after state initialization. Specifically, the terminal discretizes the state equation of the battery model after state initialization according to a preset single-step simulation time to obtain a single-step simulation environment for the target battery. Among them, the battery model includes but is not limited to a neural network model, an electrochemical-thermal-lithium precipitation coupling model, a reduced-order electrochemical model, and a hierarchical equivalent circuit model. The battery model includes a main reaction model and a side reaction model.

[0073] In one example, the terminal uses a data-driven model to fuse mechanism information, establishes a battery negative electrode potential state estimation model based on a physical constraint neural network, and uses the battery negative electrode potential state estimation model based on a physical constraint neural network as a battery model. In this way, the battery temperature, power, terminal voltage and negative electrode potential data are generated through the mechanism model to meet the dynamic constraints, which can improve the accuracy of the battery model, improve the accuracy of the battery state information obtained by simulation, and then improve the accuracy of the battery fast charging control information prediction model, and improve the accuracy and efficiency of battery charging.

[0074] In one example, the terminal charges the target battery to full power at a preset initial ambient temperature, discharges it, and charges it to the initial power. Then, the terminal leaves the target battery for a preset time, waits for the target battery to reach a balanced state, and uses the battery temperature, power, and terminal voltage at this time as the initial state information of the target battery, and uses the negative electrode potential information and whether the target power is reached as hidden information. The initial ambient temperature includes but is not limited to low temperature, room temperature, and high temperature. The initial power includes but is not limited to 10%, 20%, 30%, 40%, and 50%.

[0075] In one embodiment, the terminal establishes an electrochemical-thermal-lithium precipitation coupling model for the high-energy lithium-ion battery LG INR 21700 M50, with a positive electrode composed of 80% nickel, 10% cobalt and 10% manganese (NCM811) and a silicon-carbon (SiOx+C) negative electrode, and calls the public parameter set of the target battery provided by the Python Battery Mathematical Modelling (pybamm) platform to initialize the parameters of the battery model. Then, the terminal verifies the validity of the battery model after parameter initialization and obtains a verified battery model.

[0076] In one example, the terminal tests the single-step simulation environment of the target battery to verify the validity of establishing the single-step simulation environment of the target battery.

[0077] Step 103, based on the state information, target reward value and charging control information at each moment, as well as the state information at the next moment corresponding to each moment, a deep reinforcement learning network is trained to obtain a battery fast charging control information prediction model.

[0078] In an embodiment of the present application, for each moment in each charging process, the terminal generates a set of charging experience information based on the state information, target reward value and charging control information at that moment, and the state information of the next moment corresponding to that moment. When the preset model training conditions are met, the terminal randomly extracts m groups of charging experience information every preset time period. Then, the terminal updates the deep reinforcement learning network based on the m groups of charging experience information. When the preset training completion conditions are met, the terminal determines the battery fast charging control information prediction model based on the current deep reinforcement learning network. Among them, the training completion condition can be whether the number of network updates exceeds the preset network update number threshold.

[0079] In the training method of the above-mentioned battery fast charge control information prediction model, a closed-loop feedback framework of the interaction between the battery single-step simulation environment and the deep reinforcement learning network is established. The deep reinforcement learning network predicts the charging control information at the current moment according to the current state information obtained by the battery single-step simulation environment. The battery single-step simulation environment simulates the state information of the battery at the next moment according to the charging control information output by the deep reinforcement learning network, and generates a target reward value based on the negative electrode potential information of the battery. According to the state information, target reward value and charging control information at each moment, and the state information at the next moment corresponding to each moment, the deep reinforcement learning network is trained to obtain a battery fast charge control information prediction model, which can predict the battery charging control information in real time according to the battery state information, adjust the battery charging rate in real time, and realize fast charging. In addition, this method adopts a closed-loop feedback optimization method based on the interaction between the battery environment and the intelligent agent to adjust the battery charging rate in real time, which is more in line with the actual situation, can improve the real-time and stability of the charging strategy execution, and improve the accuracy of fast charging. In addition, this method optimizes the battery fast charge strategy based on the self-learning ability of the deep reinforcement learning framework and the neural network, and can improve the adaptive ability of the charging strategy. Moreover, this method generates a target return value based on the negative electrode potential information of the battery, and considers not only the external characteristics of the battery, but also the impact of the internal negative electrode potential state of the battery on the charging rate. This can avoid the lithium plating side reaction of the battery, and avoid the accelerated capacity decay caused by the lithium plating side reaction. It can improve the safety of battery fast charging, slow down the battery capacity decay, and develop a fast charging strategy that can ensure both charging safety and durability. In addition, compared with the traditional charging protocol developed for specific batteries under specific conditions, this method learns and optimizes the battery's fast charging strategy through the interaction between the deep reinforcement learning network, a charging agent, and the battery environment. It does not need to rely on a large number of test experiments and calibrations, and solves the problems of long test time and high cost, and lack of key status information. At the same time, it is applicable to a wide range of batteries and has good migration capabilities.

[0080] In an exemplary embodiment, as shown in FIG2 , a specific process of performing a single-step simulation based on the charging control information at the current moment to determine the state information of the target battery at the next moment and the target reward value at the current moment includes the following steps:

[0081] Step 201 , performing a single-step simulation based on the charging control information at the current moment, and obtaining the state information and hidden information of the target battery at the next moment.

[0082] In the embodiment of the present application, hidden information is also used to indicate the state of the target battery, but is not used as an input to the deep reinforcement learning network. The state information includes the negative electrode potential information, or the hidden information includes the negative electrode potential information. For example, the state information includes the battery temperature, power, terminal voltage and negative electrode potential information, and the hidden information includes whether the target power is reached. Or the state information includes the battery temperature, power and terminal voltage, and the hidden information includes the negative electrode potential information and whether the target power is reached.

[0083] Step 202, determining the target reward value of the target battery at the current moment according to the state information and hidden information at the next moment.

[0084] In an embodiment of the present application, the terminal calculates the target reward value of the target battery at the current moment based on a preset reward function, the state information and hidden information at the next moment. The target reward value at the current moment is the reward value of executing the charging control information at the current moment at the current moment. The reward function includes but is not limited to the negative electrode potential violation penalty function.

[0085] In the training method of the battery fast charging control information prediction model, a single-step simulation is performed based on the charging control information at the current moment to obtain the state information and hidden information of the target battery at the next moment; based on the state information and hidden information at the next moment, the target reward value of the target battery at the current moment is determined. In this way, the information used to calculate the target reward value is divided into state information that is used as the input of the deep reinforcement learning network and hidden information that is not used as the input of the deep reinforcement learning network, which can reduce the size and computational complexity of the deep reinforcement learning network.

[0086] In an exemplary embodiment, the state information includes negative electrode potential information. Or the hidden information includes negative electrode potential information. As shown in FIG3 , the specific process of determining the target return value of the target battery at the current moment according to the state information and hidden information at the next moment includes the following steps:

[0087] Step 301, calculating the negative electrode potential violation penalty value of the target battery at the current moment according to the negative electrode potential information at the next moment.

[0088] In an embodiment of the present application, the terminal calculates the negative electrode potential violation penalty value of the target battery at the current moment according to the negative electrode potential information at the next moment and the negative electrode potential violation penalty function included in the preset reward function.

[0089] Step 302, calculating the non-negative electrode potential violation penalty value of the target battery at the current moment according to the state information and hidden information at the next moment.

[0090] In an embodiment of the present application, the terminal calculates the non-negative electrode potential violation penalty value of the target battery at the current moment based on the state information and hidden information at the next moment, and the non-negative electrode potential violation penalty function contained in the preset reward function. Among them, the non-negative electrode potential violation penalty function includes but is not limited to a temperature violation penalty function, a target power violation penalty function, and a terminal voltage violation penalty function. The non-negative electrode potential violation penalty value includes but is not limited to a temperature violation penalty function value, a target power violation penalty function value, and a terminal voltage violation penalty function value. The reward function includes but is not limited to a negative electrode potential violation penalty function, a temperature violation penalty function, a target power violation penalty function, and a terminal voltage violation penalty function. The non-negative electrode potential violation penalty function corresponds to the state information and the hidden information.

[0091] In one example, the non-negative electrode potential violation penalty function value includes but is not limited to the temperature violation penalty function value, the target power violation penalty function value, and the terminal voltage violation penalty function value. The terminal calculates the temperature violation penalty function value, the target power violation penalty function value, and the terminal voltage violation penalty function value of the target battery at the current moment according to the battery temperature, power, terminal voltage, and whether the target power is reached at the next moment, as well as the temperature violation penalty function, the target power violation penalty function, and the terminal voltage violation penalty function included in the preset reward function.

[0092] Step 303 , weightedly calculating the target reward value of the target battery at the current moment according to the negative electrode potential violation penalty value and the non-negative electrode potential violation penalty value.

[0093] In an embodiment of the present application, the terminal performs weighted summation based on the negative electrode potential violation penalty value, the negative electrode potential weight, each non-negative electrode potential violation penalty value, and the weights corresponding to each non-negative electrode potential violation penalty value to obtain the target reward value of the target battery at the current moment.

[0094] In one embodiment, step 303 can be expressed as:

[0095] reward=reward T *w T +reward F *w F +reward Vt *w Vt +reward phi *w phi

[0096] Among them, reward is the target reward value, reward T is the temperature violation penalty function value, w T is the temperature weight corresponding to the temperature violation penalty function value, reward F is the penalty function value for failure to reach the target power, wT The reward is the power weight corresponding to the penalty function value for not reaching the target power. Vt is the terminal voltage violation penalty function value, w Vt is the terminal voltage weight corresponding to the terminal voltage violation penalty function value, reward phi is the penalty value for negative electrode potential violation, w phi is the negative electrode potential weight. For example, the temperature weight is 1, the charge weight is 2, the terminal voltage weight is 1, and the negative electrode potential weight is 3.

[0097] In the training method of the battery fast charging control information prediction model, the negative electrode potential violation penalty value of the target battery at the current moment is calculated based on the negative electrode potential information at the next moment; the non-negative electrode potential violation penalty value of the target battery at the current moment is calculated based on the state information and hidden information at the next moment; and the target reward value of the target battery at the current moment is weightedly calculated based on the negative electrode potential violation penalty value and the non-negative electrode potential violation penalty value. In this way, the target reward function includes multiple penalty functions, and the target reward value is generated not only based on the battery negative electrode potential information, but also based on the battery temperature, SOC and terminal voltage. It comprehensively considers multi-state constraint factors, comprehensively optimizes the battery's fast charging strategy, solves the problem of incomplete constraint conditions, can avoid thermal runaway and accelerated decay caused by fast charging, and ensures the charging speed while ensuring the safety and durability of the battery during fast charging, providing strong support for the safe and rapid energy replenishment of electric vehicles. Moreover, the negative electrode potential information can be used as the input state of the charging agent of the deep reinforcement learning network, and can also be hidden as penalty information in the single-step charging return. The negative electrode potential state is used as a hidden constraint condition of the charging agent instead of as the state input of the agent, so as to avoid excessive state information affecting strategy training and reduce the size and computational complexity of the deep reinforcement learning network.

[0098] In an exemplary embodiment, as shown in FIG4 , the specific process of calculating the negative electrode potential violation penalty value of the target battery at the current moment according to the negative electrode potential information at the next moment includes the following steps:

[0099] Step 401, when the power of the target battery at the next moment is less than or equal to the preset power threshold, the negative electrode potential violation penalty value of the target battery at the current moment is calculated according to the preset negative electrode potential penalty coefficient and critical value, and the negative electrode potential information at the next moment.

[0100] In an embodiment of the present application, when the power of the target battery at the next moment is less than or equal to the preset power threshold, the terminal uses the absolute value of the difference between the negative electrode potential information at the next moment and the preset critical value as the negative electrode potential critical distance. Then, the terminal uses the product of the preset negative electrode potential penalty coefficient and the negative electrode potential critical distance as the negative electrode potential violation penalty value of the target battery at the current moment. Among them, the state information of the target battery at the next moment may include the power of the target battery at the next moment. The critical value can be a number slightly larger than 0, for example, 0.01. The power threshold can be 0.45, and the negative electrode potential penalty coefficient can be -10.

[0101] In one embodiment, the terminal calculates the negative electrode potential violation penalty value of the target battery at the current moment, which can be expressed as: negative electrode potential violation penalty value = penalty coefficient * abs (negative electrode potential value - critical value), where abs() is an absolute value.

[0102] Step 402 , when the power level of the target battery at the next moment is greater than the preset power level threshold, the preset penalty value is used as the negative electrode potential violation penalty value of the target battery at the current moment.

[0103] In the embodiment of the present application, the penalty value is a fixed penalty value, which may be -1.

[0104] In the training method of the battery fast charge control information prediction model, when the target battery's power at the next moment is less than or equal to the preset power threshold, the target battery's negative electrode potential violation penalty value at the current moment is calculated based on the preset negative electrode potential penalty coefficient and critical value, as well as the negative electrode potential information at the next moment; when the target battery's power at the next moment is greater than the preset power threshold, the preset penalty value is used as the target battery's negative electrode potential violation penalty value at the current moment. In this way, this method uses negative electrode potential drive in the low SOC range, and can achieve faster charging speed when the negative electrode potential approaches the critical value. It uses target SOC drive in the high SOC range to avoid exceeding the cutoff voltage, forming a dual-drive fast charging strategy, which can further improve the battery charging speed.

[0105] In an exemplary embodiment, as shown in FIG5 , the deep reinforcement learning network includes a policy network, and the target state information is input into the deep reinforcement learning network of the target battery, and the specific process of obtaining the charging control information of the target battery at the current moment includes the following steps:

[0106] Step 501: Determine the exploration threshold corresponding to the current moment based on a preset attenuation function.

[0107] In an embodiment of the present application, the terminal determines the function value corresponding to the current moment based on a preset decay function, and uses the function value as the exploration threshold corresponding to the current moment. The decay function may be a linear decay function. The exploration threshold is used to determine the way to generate the action (i.e., charging control information) of the deep reinforcement learning network.

[0108] Step 502: If the random number generated at the current moment is greater than or equal to the exploration threshold, the target state information is input into the strategy network to obtain the charging control information of the target battery at the current moment.

[0109] Step 503: If the random number generated at the current moment is less than the exploration threshold, the charging control information of the target battery at the current moment is generated based on the uniform random action generation function.

[0110] In the training method of the above-mentioned battery fast-charge control information prediction model, the exploration threshold corresponding to the current moment is determined based on a preset decay function; if the random number generated at the current moment is greater than or equal to the exploration threshold, the target state information is input into the strategy network to obtain the charging control information of the target battery at the current moment; if the random number generated at the current moment is less than the exploration threshold, the charging control information of the target battery at the current moment is generated based on a uniform random action generation function. In this way, a battery single-step simulation environment and a deep reinforcement learning network interaction framework are established, and a uniform random action generation function is initially adopted to enhance the exploration ability of the deep reinforcement learning network. With the continuous optimization of the actor network in the deep reinforcement learning network, the probability of randomly generated actions is reduced, and the probability of using the actor network to generate actions is increased. The exploration and utilization of the deep reinforcement learning network are balanced, and the action network of the deep reinforcement learning network is further optimized, which can improve the training effect of the battery fast-charge control information prediction model and improve the accuracy of the battery fast-charge control information prediction model.

[0111] In an exemplary embodiment, as shown in FIG6 , the deep reinforcement learning network includes a policy network and a Q network. According to the state information, target reward value and charging control information at each moment, and the state information at the next moment corresponding to each moment, the deep reinforcement learning network is trained to obtain the specific process of the battery fast charging control information prediction model, including the following steps:

[0112] Step 601, for each moment in each charging process, a set of charging experience information is generated based on the state information, target reward value and charging control information at the moment, and the state information of the next moment corresponding to the moment.

[0113] In the embodiment of the present application, for each moment in each charging process, the terminal generates a set of charging experience information including the state information, target reward value and charging control information of the moment, and the state information of the next moment corresponding to the moment. Among them, the process of charging the battery from the initial power to the target power is called a charging process, also known as an episode. A charging process includes at least one charging time step.

[0114] In one example, for each moment in each charging process, the terminal combines the state information, target reward value and charging control information of the moment, and the state information of the next moment corresponding to the moment, into a set of charging experience information.

[0115] In another example, for each moment in each charging process, the terminal combines the state information, target reward value and charging control information of the moment, the state information of the next moment corresponding to the moment, and other information of the moment to form a set of charging experience information. Among them, other information includes but is not limited to whether the current round exceeds the preset maximum number of charging steps and whether the current round is charged to the target power.

[0116] Step 602: When the preset model training conditions are met, m groups of charging experience information are randomly selected every preset time period.

[0117] In an embodiment of the present application, each time a set of charging experience information is generated, the terminal saves the set of charging experience information to an experience replay pool. When the preset model training conditions are met, the terminal randomly extracts m sets of charging experience information from the experience replay pool every preset time period. Wherein, m is a positive integer. The model training condition may be that the number of groups of charging experience information reaches a preset threshold of the number of groups of charging experience information. The m sets of charging experience information may be charging experience information of different rounds.

[0118] Step 603: input the m groups of charging experience information into the strategy network to obtain the current predicted value of the charging control information corresponding to the m groups.

[0119] In the embodiment of the present application, the policy network is an actor sub-network, and the Q network is a critic sub-network.

[0120] Step 604 : updating the strategy network based on the m groups of charging experience information, the current predicted values ​​of the charging control information corresponding to the m groups, and the Q network.

[0121] In an embodiment of the present application, the terminal calculates the target state action value based on the m-group charging experience information through the Q network. The terminal calculates the first state action value and the second state action value based on the m-group charging experience information and the current predicted value of the charging control information corresponding to the m-group through the Q network. Then, the terminal uses the minimum value of the first state action value and the second state action value as the predicted state action value. Then, the terminal updates the strategy network according to the difference between the target state action value and the predicted state action value.

[0122] Step 605: When the preset training completion condition is met, the current strategy network is used as a battery fast charging control information prediction model.

[0123] In an embodiment of the present application, when the preset training completion condition is met, the terminal stops updating the policy network and uses the current policy network as a battery fast charging control information prediction model.

[0124] In the training method of the battery fast charge control information prediction model, for each moment in each charging process, a group of charging experience information is generated based on the state information, target reward value and charging control information at the moment, and the state information of the next moment corresponding to the moment; when the preset model training conditions are met, m groups of charging experience information are randomly extracted every preset time period; the m groups of charging experience information are input into the policy network to obtain the current prediction value of the charging control information corresponding to the m groups; based on the m groups of charging experience information, the current prediction value of the charging control information corresponding to the m groups, and the Q network, the policy network is updated; when the preset training completion conditions are met, the current policy network is used as the battery fast charge control information prediction model. In this way, the charging control information prediction at each moment, the state information simulation at the next moment and the target reward value calculation at the current moment are performed through the closed-loop feedback framework of the battery single-step simulation environment and the deep reinforcement learning network interaction, and the charging experience information is accumulated. When the preset model training conditions are met, m groups of charging experience information are regularly extracted, combined with the value function, and the policy network is updated to realize the training of the battery fast charge control information prediction model. The trained battery fast charge control information prediction model has good real-time computing capabilities.

[0125] In one embodiment, a power battery fast charging control strategy framework based on deep reinforcement learning is shown in Figure 7. The power battery fast charging control strategy framework includes a single-step simulation environment of the battery (referred to as: battery environment) and a deep reinforcement learning network. The deep reinforcement learning network is a flexible actor-critic network, including: an actor subnetwork (also known as a strategy network) and a critic subnetwork (also known as a Q network). The single-step simulation environment includes the main reaction model and the side reaction model of the battery. The Q network includes the Q1 network and the Q2 network. The terminal obtains the current state information of the target battery from the single-step simulation environment , and use the current state information as the target state information, input the target state information into the deep reinforcement learning network of the target battery, and obtain the charge rate of the target battery at the current moment Then, the terminal charges based on the current charging rate Perform single-step simulation to determine the state information of the target battery at the next moment , and the target return value (also known as reward) at the current moment , and update the target state information based on the state information at the next moment, return to the step of inputting the target state information into the deep reinforcement learning network of the target battery, and obtain the charging control information step of the target battery at the current moment. For each moment in each charging process, the terminal will , target return value and charging rate , the state information of the next moment corresponding to this moment , and other information info at that moment, constitute a set of charging experience information. Each time a set of charging experience information is generated, the terminal saves the set of charging experience information to the experience replay pool (replaybuffer) (also known as experience cache). Under the condition of meeting the preset model training conditions, the terminal randomly extracts m sets of charging experience information from the experience replay pool every preset time period. Then, the terminal inputs the m sets of charging experience information into the strategy network to obtain the current prediction value of the charging control information corresponding to the m sets Then, the terminal updates the Q1 network, Q2 network and policy network based on the m group charging experience information, the current predicted value of the charging control information corresponding to the m group, and the Q network, that is, updates the deep reinforcement learning network. Specifically, the terminal updates the network of the Q1 network Update to Q target network , the network of Q2 network Update to Q target network , the network of the policy network Update to policy target network .in, express The state of the moment, express The charging rate at the moment, Indicated in Always perform actions The return value, Indicates the state at time t+1, and info indicates other information, for example, whether the current round exceeds the preset maximum number of charging steps, or whether the current round is charged to the target SOC. is the charging strategy network to be optimized, Represents the strategy target network, which can stabilize the strategy optimization process. and Represents the state-action value network. Using two networks can avoid over-estimation of the state-action value. and Represents the state-action-value target network, which can play the role of stabilizing the state-action-value network.

[0126] In one embodiment, the terminal analyzes and tests the actor network generated by training, observes whether the temperature of the charging battery violates the constraints at each step, whether the terminal voltage violates the constraints, whether the negative electrode potential violates the constraints, and whether the battery is finally charged from the initial SOC to the target SOC, and determines the effect of the battery fast charging control information prediction model. Then, the terminal sets different initial ambient temperatures, and analyzes and tests the trained battery fast charging control information prediction model in a manner similar to the training method of the above-mentioned battery fast charging control information prediction model to determine the adaptability of the battery fast charging control information prediction model to different temperatures.

[0127] In one embodiment, 25°C is used as the initial ambient temperature and 20% is used as the initial SOC. In the process of charging control using the battery fast charging control information prediction model, the current change curve, the voltage change curve, the negative electrode potential change curve, the SOC change curve and the temperature change curve are shown in Figures 8-12 respectively. It can be seen that the charging time is 1260s, about 20min, which greatly reduces the battery charging time. The battery temperature meets the maximum temperature constraint of 50°C, the battery voltage does not exceed 4.2V, the battery SOC is 0.2~0.8, reaching 80% of the setting, and the negative electrode potential is above the critical negative electrode potential of 0, avoiding the accelerated capacity decay and safety problems triggered by lithium plating.

[0128] In one embodiment, as shown in FIG13 , a battery fast charging method is provided. This embodiment uses the method applied to a terminal as an example. It is understandable that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. Among them, the terminal can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart car devices, projection devices, etc. Portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The head-mounted device can be a virtual reality (VR) device, an augmented reality (AR) device, smart glasses, etc. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services. It is understandable that the application scenarios of the battery fast charging method and the training method of the above-mentioned battery fast charging control information prediction model can be the same or different, and this application does not limit it. In this embodiment, the method includes the following steps:

[0129] Step 1301, obtaining status information of a target battery.

[0130] Step 1302, input the status information into a pre-trained battery fast charging control information prediction model to obtain the charging control information of the target battery.

[0131] Step 1303, controlling the charging of the target battery according to the charging control information.

[0132] Among them, the battery fast charging control information prediction model is trained by any one of the above-mentioned battery fast charging control information prediction model training methods.

[0133] In an embodiment of the present application, the terminal controls the charging of the target battery at the current moment according to the charging rate at the current moment.

[0134] In the above-mentioned battery fast charging method, the battery fast charging control information prediction model trained by the above-mentioned battery fast charging control information prediction model training method is used to predict the charging control information in real time, adjust the battery charging rate in real time, and control the charging of the target battery in real time, so as to achieve fast charging.

[0135] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0136] Based on the same inventive concept, the embodiment of the present application also provides a training device for a battery fast charging control information prediction model for implementing the training method of the battery fast charging control information prediction model involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiments of the training device for one or more battery fast charging control information prediction models provided below can be found in the above limitations on the training method for the battery fast charging control information prediction model, and will not be repeated here.

[0137] In an exemplary embodiment, as shown in FIG. 14 , a training device 1400 for a battery fast charge control information prediction model is provided, comprising: a first determination module 1410 , a second determination module 1420 , and a training module 1430 , wherein:

[0138] A first determination module 1410 is used to obtain the current state information of the target battery, and use the current state information as the target state information, input the target state information into the deep reinforcement learning network of the target battery, and obtain the charging control information of the target battery at the current moment;

[0139] A second determination module 1420 is used to perform a single-step simulation based on the charging control information at the current moment, determine the state information of the target battery at the next moment, and the target reward value at the current moment, and update the target state information based on the state information at the next moment, and return to the step of inputting the target state information into the deep reinforcement learning network of the target battery to obtain the charging control information of the target battery at the current moment; the target reward value is generated based on the negative electrode potential information of the target battery;

[0140] The training module 1430 is used to train the deep reinforcement learning network according to the state information, target reward value and charging control information at each moment, as well as the state information of the next moment corresponding to each moment, to obtain a battery fast charging control information prediction model.

[0141] Optionally, the second determining module 1420 is specifically configured to:

[0142] Performing a single-step simulation based on the charging control information at the current moment to obtain the state information and hidden information of the target battery at the next moment;

[0143] The target reward value of the target battery at the current moment is determined according to the state information and hidden information at the next moment.

[0144] Optionally, the state information includes cathode potential information; or the hidden information includes cathode potential information; the second determination module 1420 is specifically configured to:

[0145] Calculating a negative electrode potential violation penalty value of the target battery at the current moment according to the negative electrode potential information at the next moment;

[0146] Calculating a non-negative electrode potential violation penalty value of the target battery at the current moment according to the state information and hidden information at the next moment;

[0147] The target reward value of the target battery at the current moment is weightedly calculated according to the negative electrode potential violation penalty value and the non-negative electrode potential violation penalty value.

[0148] Optionally, the second determining module 1420 is specifically configured to:

[0149] When the power level of the target battery at the next moment is less than or equal to the preset power level threshold, the negative electrode potential violation penalty value of the target battery at the current moment is calculated according to the preset negative electrode potential penalty coefficient and critical value, and the negative electrode potential information at the next moment;

[0150] When the power level of the target battery at the next moment is greater than the preset power level threshold, the preset penalty value is used as the negative electrode potential violation penalty value of the target battery at the current moment.

[0151] Optionally, the deep reinforcement learning network includes a policy network, and the first determination module 1410 is specifically used to:

[0152] Based on the preset attenuation function, determine the exploration threshold corresponding to the current moment;

[0153] If the random number generated at the current moment is greater than or equal to the exploration threshold, the target state information is input into the strategy network to obtain the charging control information of the target battery at the current moment;

[0154] If the random number generated at the current moment is less than the exploration threshold, the charging control information of the target battery at the current moment is generated based on the uniform random action generation function.

[0155] Optionally, the deep reinforcement learning network includes a policy network and a Q network, and the training module 1430 is specifically used for:

[0156] For each moment in each charging process, a set of charging experience information is generated based on the state information, target reward value and charging control information at the moment, and the state information of the next moment corresponding to the moment;

[0157] When the preset model training conditions are met, m groups of charging experience information are randomly selected every preset time period;

[0158] Inputting the m groups of charging experience information into the strategy network to obtain current prediction values ​​of the charging control information corresponding to the m groups;

[0159] updating the strategy network based on the m groups of charging experience information, the current predicted values ​​of the charging control information corresponding to the m groups, and the Q network;

[0160] When the preset training completion conditions are met, the current strategy network is used as a battery fast charging control information prediction model.

[0161] Based on the same inventive concept, the embodiment of the present application also provides a battery fast charging device for implementing the battery fast charging method involved above. The implementation solution provided by the device to solve the problem is similar to the implementation solution recorded in the above method, so the specific limitations in one or more battery fast charging device embodiments provided below can refer to the limitations on the battery fast charging method above, and will not be repeated here.

[0162] In an exemplary embodiment, as shown in FIG. 15 , a battery fast charging device 1500 is provided, including: an acquisition module 1510 , a prediction module 1520 , and a control module 1530 , wherein:

[0163] An acquisition module 1510 is used to acquire status information of a target battery;

[0164] A prediction module 1520, configured to input the state information into a pre-trained battery fast charge control information prediction model to obtain the charging control information of the target battery;

[0165] A control module 1530, configured to control charging of the target battery according to the charging control information;

[0166] Wherein, the battery fast charging control information prediction model is trained by any one of the above-mentioned battery fast charging control information prediction model training methods.

[0167] The training device of the battery fast charging control information prediction model and each module in the battery fast charging device can be implemented in whole or in part by software, hardware and a combination thereof. The above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0168] In an exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be shown in FIG16. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, the memory, and the input / output interface are connected via a system bus, and the communication interface, the display unit, and the input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a training method for a battery fast charging control information prediction model or a battery fast charging method is implemented. The display unit of the computer device is used to form a visually visible picture, and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball, or touchpad provided on the computer device housing, or an external keyboard, touchpad, or mouse, etc.

[0169] Those skilled in the art will understand that the structure shown in FIG. 16 is merely a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0170] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0171] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0172] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0173] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0174] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0175] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0176] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A training method for a battery fast charging control information prediction model, characterized in that: The method comprises: Acquire the current state information of the target battery, use the current state information as the target state information, input the target state information into the deep reinforcement learning network of the target battery, and obtain the charging control information of the target battery at the current moment; Performing a single-step simulation based on the charging control information at the current moment, determining the state information of the target battery at the next moment and the target reward value at the current moment, updating the target state information based on the state information at the next moment, and returning to the step of inputting the target state information into the deep reinforcement learning network of the target battery to obtain the charging control information of the target battery at the current moment; the target reward value is generated based on the negative electrode potential information of the target battery; According to the state information at each moment, the target reward value and the charging control information, as well as the state information at the next moment corresponding to each moment, the deep reinforcement learning network is trained to obtain a battery fast charging control information prediction model.

2. The method according to claim 1, characterized in that: The single-step simulation is performed based on the charging control information at the current moment to determine the state information of the target battery at the next moment and the target return value at the current moment, including: Performing a single-step simulation based on the charging control information at the current moment to obtain the state information and hidden information of the target battery at the next moment; The target reward value of the target battery at the current moment is determined according to the state information and hidden information at the next moment.

3. The method according to claim 2, characterized in that The state information includes negative electrode potential information; or the hidden information includes negative electrode potential information; and determining the target reward value of the target battery at the current moment according to the state information and the hidden information at the next moment includes: Calculating a negative electrode potential violation penalty value of the target battery at the current moment according to the negative electrode potential information at the next moment; Calculating a non-negative electrode potential violation penalty value of the target battery at the current moment according to the state information and hidden information at the next moment; The target reward value of the target battery at the current moment is weightedly calculated according to the negative electrode potential violation penalty value and the non-negative electrode potential violation penalty value.

4. The method according to claim 3, characterized in that The calculating, according to the negative electrode potential information at the next moment, the negative electrode potential violation penalty value of the target battery at the current moment includes: When the power level of the target battery at the next moment is less than or equal to the preset power level threshold, the negative electrode potential violation penalty value of the target battery at the current moment is calculated according to the preset negative electrode potential penalty coefficient and critical value, and the negative electrode potential information at the next moment; When the power level of the target battery at the next moment is greater than the preset power level threshold, the preset penalty value is used as the negative electrode potential violation penalty value of the target battery at the current moment.

5. The method according to claim 1, characterized in that The deep reinforcement learning network includes a policy network, and the target state information is input into the deep reinforcement learning network of the target battery to obtain the charging control information of the target battery at the current moment, including: Based on the preset attenuation function, determine the exploration threshold corresponding to the current moment; If the random number generated at the current moment is greater than or equal to the exploration threshold, the target state information is input into the strategy network to obtain the charging control information of the target battery at the current moment; If the random number generated at the current moment is less than the exploration threshold, the charging control information of the target battery at the current moment is generated based on the uniform random action generation function.

6. The method according to claim 1, characterized in that The deep reinforcement learning network includes a policy network and a Q network. The deep reinforcement learning network is trained according to the state information, target reward value and charging control information at each moment, and the state information of the next moment corresponding to each moment to obtain a battery fast charging control information prediction model, including: For each moment in each charging process, a set of charging experience information is generated based on the state information, target reward value and charging control information at the moment, and the state information of the next moment corresponding to the moment; When the preset model training conditions are met, m groups of charging experience information are randomly selected every preset time period; Inputting the m groups of charging experience information into the strategy network to obtain current prediction values ​​of the charging control information corresponding to the m groups; updating the strategy network based on the m groups of charging experience information, the current predicted values ​​of the charging control information corresponding to the m groups, and the Q network; When the preset training completion conditions are met, the current strategy network is used as a battery fast charging control information prediction model.

7. A battery fast charging method, characterized in that: The method comprises: Get the status information of the target battery; Inputting the state information into a pre-trained battery fast charge control information prediction model to obtain the charging control information of the target battery; Controlling charging of the target battery according to the charging control information; Wherein, the battery fast charging control information prediction model is trained by the battery fast charging control information prediction model training method described in any one of claims 1 to 6.

8. A training device for a battery fast charging control information prediction model, characterized in that: The device comprises: A first determination module is used to obtain the current state information of the target battery, and use the current state information as the target state information, input the target state information into the deep reinforcement learning network of the target battery, and obtain the charging control information of the target battery at the current moment; a second determination module, configured to perform a single-step simulation based on the charging control information at the current moment, determine the state information of the target battery at the next moment, and the target reward value at the current moment, and update the target state information based on the state information at the next moment, and return to the step of inputting the target state information into the deep reinforcement learning network of the target battery to obtain the charging control information of the target battery at the current moment; the target reward value is generated based on the negative electrode potential information of the target battery; The training module is used to train the deep reinforcement learning network according to the state information, target reward value and charging control information at each moment, as well as the state information of the next moment corresponding to each moment, so as to obtain a battery fast charging control information prediction model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 or claim 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method of any one of claims 1 to 6 or claim 7 are implemented.

Images