Flexible actor and commentator fast charging method based on electrothermal aging coupling model

Abstract

The invention relates to the technical field of electric vehicle charging, in particular to a flexible actor commentator fast charging method based on an electrothermal aging coupling model, and the method comprises the steps: building an electrothermal aging coupling model of a battery; acquiring the temperature, the terminal voltage and the terminal current of the load at the t moment; a relational expression of the state of charge and the open-circuit voltage of the battery is determined through formula fitting; respectively identifying an electric model parameter, a thermal model parameter and an aging model parameter by adopting a recursive least square method and a genetic algorithm; determining an optimal charging strategy by using a flexible actor commentator algorithm; carrying out simulation verification on the optimal charging strategy by using PyBaMM; according to the method, multiple factors influencing the charging safety of the vehicle lithium ion power battery are analyzed in a multi-dimensional mode, the electric heating aging coupling model considering aging is established, and the dynamic behavior of the battery in the charging process is accurately described and predicted; and the safety and robustness of the flexible actor commentator fast charging method based on the electrothermal aging coupling model are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electric vehicle charging, and in particular to a flexible actor-critic fast charging method based on an electrothermal aging coupling model. Background Art

[0002] As the core component of electric vehicles, rechargeable batteries need to have the characteristics of long life, environmental protection, high energy density, high discharge rate, and low maintenance cost. Compared with other commonly used rechargeable batteries such as lead-acid, nickel-cadmium, and nickel-metal hydride batteries, lithium-ion batteries have become the most popular choice for electric vehicles due to their advantages. Although lithium-ion batteries perform well as secondary batteries and have a long charging time, range anxiety and long charging time have become obstacles to user experience and have seriously restricted the development of the electric vehicle industry. Therefore, how to improve charging efficiency, reduce charging time, and ensure charging safety is the key to alleviating user concerns and improving the practical applicability of electric vehicles.

[0003] Chinese Patent Publication No.: CN117559387A discloses a VPP internal energy optimization method and system based on deep reinforcement learning dynamic pricing. The invention discloses a VPP internal energy optimization method and system based on deep reinforcement learning dynamic pricing. First, basic data is input to establish a two-layer day-ahead energy management framework: the framework includes two layers, the upper layer, the VPP operator sends an internal price signal to the DER; in the lower layer, the DER receives the price signal from the upper-layer VPP operator, makes scheduling decisions on the premise of maximizing personal profits, and transmits the decision signal back to the VPP operator; establishes a reinforcement learning model: adopts a flexible action-criteria algorithm (SAC) as a model-free reinforcement learning (RL) algorithm, and performs optimal multidimensional continuous decision variables in a multidimensional state environment through online learning; finally, VPP operations are performed based on the reinforcement learning model to achieve optimal scheduling of internal energy.

[0004] China Patent Publication No.: CN118017521A discloses a new power system voltage control method and device taking into account electric vehicle charging. This invention discloses a new power system voltage control method and device taking into account electric vehicle charging, which belongs to the field of new power system voltage control. The method includes constructing a new power system voltage control system; Markov modeling of the new power system voltage control system to obtain a system voltage control model; according to the system voltage control model, using the SAC flexible action-evaluation algorithm to solve, and obtaining a new power system voltage control scheme taking into account electric vehicles. This invention solves the problems of random load fluctuations and voltage instability in new power systems.

[0005] It can be seen that the above technical solutions have the following problems: dependence on a large amount of previous data, limitations in the policy generalization ability, and inadequate direct processing of physical safety constraints in some cases, the computational complexity of high-precision models, the difficulty of parameter identification, and the limited generalization ability of the model. Summary of the invention

[0006] To this end, the present invention provides a flexible actor-critic fast charging method based on an electrothermal aging coupling model to overcome the problems in the prior art of reliance on a large amount of preliminary data, the limitation of strategy generalization capability, the insufficiency of direct processing of physical safety constraints in some cases, the computational complexity of high-precision models, the difficulty of parameter identification and the limited generalization capability of models.

[0007] To achieve the above object, the present invention provides a flexible actor-critic fast charging method based on an electrothermal aging coupling model, comprising:

[0008] Establishing an electrothermal aging coupling model of a battery, determining parameters of the electrothermal aging coupling model, discretizing the established electrothermal aging coupling model parameters and constructing a state space equation;

[0009] Using a K-type thermocouple, a voltage sensor and a current sensor to collect the actual temperature, actual terminal voltage and actual terminal current of the battery at time t;

[0010] Using a lower charging rate method to collect the state of charge and open circuit voltage of the battery, and determining the relationship expression between the state of charge and the open circuit voltage of the battery by formula fitting;

[0011] Establishing a discrete equation for identifying the parameters of the electrothermal aging coupling model, inputting the actual temperature, the actual terminal voltage and the actual terminal current at the time t, using the recursive least squares method to identify the electrical model parameters and the thermal model parameters, and using the genetic algorithm to identify the aging model parameters;

[0012] Define the Gym environment according to the state-space equations, simulate the battery charging process, formulate a multi-constraint objective function, and use a flexible actor-critic algorithm to balance and explore the action space to determine the optimal charging strategy;

[0013] Use PyBaMM to simulate and verify the optimal charging strategy;

[0014] The electrothermal aging coupling model parameters include the electrical model parameters, the thermal model parameters and the aging model parameters; the constraints of the multi-constraint objective function include charging time, battery temperature, charging voltage, charging current and degradation rate.

[0015] Furthermore, an electrothermal aging coupling model of the battery is established, wherein:

[0016] A coupling model of the electrical model and the thermal model of the battery is established according to Kirchhoff's voltage and current laws, a second-order Rc equivalent circuit model and a lumped parameter two-state thermal model;

[0017] Establishing the aging model of the battery according to the Arrhenius equation and the Puckett equation;

[0018] The electrothermal aging coupling model includes the electrical model, the thermal model and the aging model.

[0019] Furthermore, the state of charge of the battery is estimated and calculated using an ampere-hour integration method.

[0020] Furthermore, the open circuit voltage values ​​corresponding to different state of charge values ​​of the battery in a fully discharged state during the charging process are recorded, and a relationship expression between the state of charge and the open circuit voltage is obtained by fitting the state of charge value and the open circuit voltage value.

[0021] Furthermore, the steps of establishing a discrete equation for identifying the parameters of the electrothermal aging coupling model, inputting the actual temperature, the actual terminal voltage and the actual terminal current at the time t, using the recursive least squares method to identify the electrical model parameters and the thermal model parameters, and using the genetic algorithm to identify the aging model parameters include:

[0022] Construct discrete equations for electrical model identification;

[0023] Construct discrete equations for thermal model identification;

[0024] Using the recursive least square method with a forgetting factor to respectively identify the parameters of the electrical model and the thermal model;

[0025] The genetic algorithm is used to perform parameter identification on the aging model.

[0026] Furthermore, the steps of respectively identifying the parameters of the electrical model and the thermal model using the recursive least square method with a forgetting factor include:

[0027] Initialize the error covariance and parameter identification vector;

[0028] Inputting the value of the actual terminal current collected at the time t into the electrical model and the thermal model to obtain a predicted voltage and a predicted temperature respectively;

[0029] Comparing the actual temperature collected at the time t with the predicted temperature to obtain a temperature difference value, and comparing the actual terminal voltage collected at the time t with the predicted voltage to obtain a voltage difference value;

[0030] An error covariance matrix is ​​adjusted according to the temperature difference value, the voltage difference value, and the forgetting factor.

[0031] Further, a Gym environment is defined according to the state space equation, the battery charging process is simulated, and the multi-constraint objective function is formulated, wherein:

[0032] balancing the charging time, the battery temperature, the charging voltage, the charging current, and the degradation rate by minimizing a cost function,

[0033] The state space is a set of states, including the charge state, voltage state and temperature state of the battery.

[0034] Further, the flexible actor-critic algorithm is used to balance and explore the action space to determine the optimal charging strategy, wherein:

[0035] A soft penalty mechanism is used to guide the RL agent to impose soft penalties on behaviors that deviate from the optimal charging strategy by adjusting the weighting factors of the charging time, the battery temperature, the charging voltage, the charging current, and the degradation rate in the cost function.

[0036] The weighting factors of the charging voltage and the battery temperature are three orders of magnitude higher than the weighting factors of the charging time, the charging current, and the degradation rate. The RL agent is an entity that performs actions and interacts with the environment in reinforcement learning. The soft penalty includes extending the charging time, increasing the charging current fluctuation, increasing the battery temperature, increasing the charging voltage, and accelerating the degradation rate.

[0037] Further, the degree of deviation of the behavior of the RL agent from the behavior of the optimal charging strategy is quantified by minimizing the cost function, wherein,

[0038] The reward function is designed to be the negative of the cost function, and negative cost feedback is provided to the RL agent, and the RL agent learns and executes the charging strategy with the minimum deviation through the negative cost feedback.

[0039] Furthermore, the flexible actor-critic algorithm includes a strategy network and a dual Q-value estimation calculation network, wherein:

[0040] The policy network consists of a decision network and an evaluation network.

[0041] Compared with the prior art, the beneficial effect of the present invention lies in that the present invention combines the model-based method and the data-driven method. It first relies on the established multi-physics field coupling model that takes aging into consideration to simulate the operating state of the battery, ensuring the accurate description of the dynamic characteristics of the battery. On this solid basis, the charging strategy is further optimized to achieve both high efficiency and safety in the charging process. In fact, the data-driven method makes up for the descriptive limitations of the pure model method in this framework, and provides necessary supplements for those complex behaviors or nonlinear effects that are difficult to directly reflect through the model. Therefore, the hybrid method that combines the strengths of the two not only strengthens the grasp of the battery operation mechanism at the theoretical level, but also effectively improves the safety and robustness of the flexible actor critic fast charging method based on the electrothermal aging coupling model.

[0042] Furthermore, the present invention analyzes the intrinsic mechanism of charging affecting safety and attenuation, analyzes and decouples the various factors affecting the charging safety of automotive lithium-ion power batteries in multiple dimensions, and establishes an electro-thermal aging coupling model that takes aging into consideration. The model can more accurately describe and predict the dynamic behavior of the battery during the charging process, and provide a scientific basis for the formulation of battery management strategies. The core action mechanism of different battery electro-thermal systems is identified, aiming to reveal basic principles with universal applicability and to clarify the intrinsic connection between battery performance attenuation and its internal state, in order to suppress the attenuation process by controlling the internal state of the battery, thereby further improving the safety and robustness of the flexible actor critic fast charging method based on the electro-thermal aging coupling model.

[0043] Furthermore, the present invention constructs a state valuation calculation model and a parameter identification method to address the problems of complex structure and large computational complexity of existing electrochemical mechanism models, reconstructs the original complex internal electrochemical mechanism model, optimizes the model structure to reduce the computational load without sacrificing accuracy, and designs targeted parameter identification methods. By optimizing existing measurement technologies and data processing algorithms, the speed and accuracy of parameter identification are improved, providing technical support for accurately estimating the internal state of the battery, and further improving the safety and robustness of the flexible actor critic fast charging method based on the electrothermal aging coupling model.

[0044] Furthermore, the present invention develops a safe fast charging algorithm based on state observation and optimal control, and uses the flexible actor critic algorithm to design a reasonable input parameter combination scheme to maximize or even improve the overall performance level of the battery while ensuring the shortest charging time. To this end, we will define an objective function that includes multiple dimensional evaluation indicators such as charging time, charging safety, and capacity decay rate, and set appropriate constraints to prevent exceeding the safety range, so that the thermal safety and life decay of the battery can be taken into account during charging control. Through the learning ability of the flexible actor critic algorithm, the global optimal solution or suboptimal solution set is found to achieve dynamic optimization of the charging strategy, further improving the safety and robustness of the flexible actor critic fast charging method based on the electrothermal aging coupling model. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart of the flexible actor critic fast charging method based on the electrothermal aging coupling model of the present invention;

[0046] Figure 2 Schematic diagram of an electrothermal aging coupling model according to an embodiment of the present invention;

[0047] Figure 3 A graph showing the relationship between the state of charge estimation result and time of a battery according to an embodiment of the present invention;

[0048] Figure 4 is a graph showing the relationship between the state of charge estimation error and time of a battery according to an embodiment of the present invention;

[0049] Figure 5 A flow chart of optimizing a charging curve according to an embodiment of the present invention;

[0050] Figure 6 A schematic diagram of the state of charge of a battery according to a PyBaMM simulation verification result of an embodiment of the present invention;

[0051] Figure 7 A schematic diagram of the battery voltage of the PyBaMM simulation verification result according to an embodiment of the present invention;

[0052] Figure 8 A schematic diagram of the internal temperature of a battery according to a PyBaMM simulation verification result of an embodiment of the present invention;

[0053] Fig. 9 A schematic diagram of the heat generation rate of a battery according to a PyBaMM simulation verification result of an embodiment of the present invention;

[0054] Fig.10 It is a schematic diagram of the health status of the battery according to the PyBaMM simulation verification result of an embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0056] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0057] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0058] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0059] See also Figure 1 As shown, it is a flow chart of a flexible actor critic fast charging method based on an electrothermal aging coupling model of the present invention, a flexible actor critic fast charging method based on an electrothermal aging coupling model, comprising:

[0060] Step S1, establishing an electrothermal aging coupling model of a battery, determining parameters of the electrothermal aging coupling model, discretizing the established electrothermal aging coupling model parameters and constructing a state space equation;

[0061] Step S2, using a K-type thermocouple, a voltage sensor, and a current sensor to collect the actual temperature, actual terminal voltage, and actual terminal current of the battery at time t;

[0062] Step S3, using a lower charging rate method to collect the state of charge and open circuit voltage of the battery, and determining the relationship expression between the state of charge and the open circuit voltage of the battery by formula fitting;

[0063] Step S4, establishing a discrete equation for identifying parameters of the electrothermal aging coupling model, inputting the actual temperature, actual terminal voltage and actual terminal current at time t, using the recursive least squares method to identify the electrical model parameters and the thermal model parameters, and using the genetic algorithm to identify the aging model parameters;

[0064] Step S5, defining the Gym environment according to the state space equation, simulating the battery charging process, formulating a multi-constraint objective function, and using a flexible actor-critic algorithm to balance and explore the action space to determine the optimal charging strategy;

[0065] Step S6, using PyBaMM to simulate and verify the optimal charging strategy;

[0066] Among them, the parameters of the electrothermal aging coupling model include electrical model parameters, thermal model parameters and aging model parameters; the constraints of the multi-constraint objective function include charging time, battery temperature, charging voltage, charging current and degradation rate.

[0067] Specifically, the present invention combines the model-based method and the data-driven method. It first relies on the established multi-physics field coupling model that takes aging into consideration to simulate the operating state of the battery, ensuring an accurate description of the dynamic characteristics of the battery. On this solid basis, the charging strategy is further optimized to achieve both high efficiency and safety in the charging process. In fact, the data-driven method makes up for the descriptive limitations of the pure model method in this framework, and provides necessary supplements for those complex behaviors or nonlinear effects that are difficult to directly reflect through the model. Therefore, the hybrid method that combines the strengths of the two not only strengthens the grasp of the battery operation mechanism at the theoretical level, but also effectively improves the safety and robustness of the flexible actor critic fast charging method based on the electrothermal aging coupling model.

[0068] See also Figure 2 As shown, it is a schematic diagram of the electrothermal aging coupling model of an embodiment of the present invention. In step S1, the electrothermal aging coupling model of the battery is established, wherein:

[0069] A coupling model of the battery's electrical model and thermal model is established based on Kirchhoff's voltage and current laws, a second-order Rc equivalent circuit model, and a lumped parameter two-state thermal model;

[0070] Establish a battery aging model based on the Arrhenius equation and the Puckett equation;

[0071] Among them, the electrothermal aging coupling model includes electrical model, thermal model and aging model.

[0072] In the implementation, in order to study the internal reaction of the battery under fast charging conditions, a thermoelectric coupling model considering battery aging is established. According to Kirchhoff's voltage and current laws, the second-order Rc equivalent circuit model can be expressed as follows.

[0073]

[0074]

[0075] Vt =V ocv +IR0+V1+V2 (4)

[0076] Among them, SOC represents the available state of the remaining power in the battery; I is the charging current of the battery, (A ampere); Q is the nominal capacity of the battery, (Ah ampere-hour); V t Finger tip voltage, (V volts); V ocv is the open circuit voltage of the battery (nonlinearly related to SOC), (V volts); V1 and V2 are the voltages across the two Rc networks, (V volts); R0 is the ohmic internal resistance, (Ω ohms); R1 and R2 are the polarization resistances of the two Rc networks, (Ω ohms); C1 and C2 are the polarization capacitances of the two Rc networks, (μF microfarads); t is time, (s seconds).

[0077]

[0078] Among them, T s 、T c 、T amb and T a Represent the surface temperature, core temperature, ambient temperature and average temperature of the battery, respectively (K Kelvin); R u , R c are convection thermal resistance and thermal conduction thermal resistance, (Ω ohm); C s , C c are surface heat capacity and internal heat capacity, respectively (J / K joule per Kelvin); the calorific value H includes Ohmic heat, polarization heat and irreversible entropy heat, (W watt); V t Finger tip voltage, (V volts); V ocv is the open circuit voltage of the battery (nonlinearly related to SOC), (V volts); I is the charging current of the battery, (A amperes); is the entropy heat coefficient, where the heat generated by entropy heat is ignored

[0079] The capacity loss based on the Arrhenius equation is:

[0080]

[0081] E a (C) = (31700-370.3·C) (11)

[0082] Among them, ΔQ n is the percentage of capacity loss, E a (C) is the activation energy, (J / mol joules per mole); R is the ideal gas constant, (J / (mol·K) joules per mole per Kelvin); T ambrepresents the ambient temperature of the battery, (K Kelvin); z is the power law coefficient equal to 0.55; C is the ratio of the charge and discharge current per unit time to the nominal capacity of the battery; A(C,Ta) is the cumulative ampere-hour throughput, (Ah ampere-hour); a Puckett equation is used to describe the degradation rate at different C rates, which is expressed as:

[0083] λ(C)=e plnC+q (12)

[0084] Where p and q are the parameters to be identified.

[0085] It is understood that C is C-rate, which is a standard term widely used in the battery industry to indicate the charge and discharge rate relative to the rated capacity of the battery. The C-rate is defined as the ratio of the charge and discharge current per unit time to the nominal capacity of the battery. For example, 1C means a full charge and discharge in one hour, while 0.5C means a full charge and discharge in two hours.

[0086] Generally speaking, when the battery capacity loss reaches 20%, we define it as reaching the scrap age. The corresponding cumulative ampere-hour throughput A(C, Ta) and the total number of cycles N(C, Ta) of the battery reaching the scrap age are as follows:

[0087]

[0088] Where A(C,Ta) is the cumulative ampere-hour throughput, (Ah); E a (c) is the activation energy, (J / mol Joules per mole); R is the ideal gas constant, (J / (mol·K) Joules per mole per Kelvin); T a represents the average temperature of the battery, (K Kelvin); z is the power law coefficient equal to 0.55; Q is the nominal capacity of the battery, (Ah ampere-hour).

[0089] Then, to represent the capacity decay of the battery, SOH can be given by formula (15):

[0090]

[0091] Among them, SOH is the health state of the battery; I is the charging current of the battery, (A ampere); N(C,Ta) is the total number of cycles before the battery reaches its scrap limit; Q is the nominal capacity of the battery, (Ah ampere-hour).

[0092] The dynamic characteristics of the battery can be described as:

[0093]

[0094] Wherein, SOC represents the available state of the remaining charge in the battery; k represents a discrete time index, k = 0, 1, 2, ..., N; k identifies the kth time step in the simulation process; each time step corresponds to an actual time t = kΔt, where Δt is a fixed time step, (s seconds); therefore, k is used to track the position in the discrete time series, not directly representing time, but as an iteration number or step number to advance the process of numerical simulation, z is a power law coefficient equal to 0.55; Q is the nominal capacity of the battery, (Ah ampere-hour); I is the charging current of the battery, (A ampere); V1 and V2 are the voltages across the two Rc networks, respectively, (V volts); C1 and C2 are the polarization capacitances of the two Rc networks, respectively, (μF microfarads); R1 and R2 are the polarization resistances of the two Rc networks, respectively, (Ω ohms).

[0095] According to the principle of conservation of energy, we can get:

[0096]

[0097] Among them, T s is the surface temperature of the battery, (K Kelvin); T amb is the ambient temperature of the battery, (K Kelvin); T a is the average temperature of the battery, (K Kelvin); T c is the core temperature of the battery, (K Kelvin); R u , R c are convection thermal resistance and thermal conduction thermal resistance, (Ω ohm); C s , C c are the surface heat capacity and internal heat capacity, (J / K joule per Kelvin); Δt is the fixed time step, (s seconds); V1 and V2 are the voltages across the two Rc networks, (V volts); I is the charging current of the battery, (A amperes); R0 is the ohmic internal resistance, (Ω ohms).

[0098] The state space model is established using formulas (15)-(16), where the state variables are V1, V2, SOC, T s and T c , the system input is I, the system output is V t .

[0099] Where V1 and V2 are the voltages across the two Rc networks, (V volts); SOC represents the available state of charge remaining in the battery; T s is the surface temperature of the battery, (K Kelvin); T c is the core temperature of the battery, (K Kelvin); I is the charging current of the battery, (A ampere); V t Finger tip voltage, (V volts).

[0100] The state space of the electrothermal aging coupling model considering the aging factor can be expressed in non-parametric form as follows:

[0101]

[0102] y=h(x,u)(18)

[0103] Where x = [x1 x2 x3 x4 x5] T , x1=SOC, x2=V1, x3=V2, x4=T s , x5=T c , And the control input u=[I] T ∈R. Output vector h(x,u)=[V OCV +x2+x3+uR0] T ∈R,

[0104]

[0105] SOC represents the available state of charge remaining in the battery; V1 and V2 are the voltages across the two Rc networks, (V volts); T s is the surface temperature of the battery, (K Kelvin); T c is the core temperature of the battery, (K Kelvin); T a is the average temperature of the battery, (K Kelvin); R is a real number set; V ocv is the open circuit voltage of the battery (nonlinearly related to SOC), (V volts); R0 is the ohmic internal resistance, (Ω ohms); R1 and R2 are the polarization resistances of the two Rc networks, (Ω ohms); C1 and C2 are the polarization capacitances of the two Rc networks, (μF microfarads); T amb is the ambient temperature of the battery, (K Kelvin); R u , R c are convection thermal resistance and thermal conduction thermal resistance, (Ω ohm); C s , C c are surface heat capacity and internal heat capacity, respectively (J / K joule per Kelvin); V t Finger tip voltage, (V volts); V ocv is the open circuit voltage of the battery (non-linearly related to SOC), (V volts).

[0106] Based on this model, we can gain a deeper understanding of the battery's behavioral characteristics during fast charging, and provide a solid theoretical foundation for the design of subsequent optimization strategies.

[0107] Specifically, in step S3, the battery state of charge is estimated and calculated using the ampere-hour integration method.

[0108] In the implementation, the battery SOC is estimated using the ampere-hour integration method.

[0109]

[0110] Where Q is the nominal capacity of the battery, (Ah ampere-hours); I(k) is the current value at time k, (A amperes); and t is the total time, (s seconds).

[0111] Specifically, in step S3, the open circuit voltage values ​​corresponding to different state of charge values ​​of the battery in a fully discharged state during the charging process are recorded, and the relationship expression between the state of charge and the open circuit voltage is obtained by fitting the state of charge value and the open circuit voltage value.

[0112] In practice, to fit SOC to V ocv The battery was discharged until it was completely discharged, and then charged at 1 / 2C for 6 minutes for 20 times. The V ocv Finally, through fitting, we can get V ocv The relationship between and SOC.

[0113] V ocv =P0SOC 5 +P1SOC 4 +P2SOC 3 +P3SOC 2 +P4SOC+P5(20)

[0114] Among them, P0, P1, P2, P3, P4, and P5 are parameters to be fitted; SOC represents the available state of the remaining power in the battery; V ocv is the open circuit voltage of the battery (non-linearly related to SOC), (V volts).

[0115] Please cooperate Figure 3 See also Figure 4 As shown, they are respectively a graph of the relationship between the state of charge estimation result and time of the battery of the embodiment of the present invention and a graph of the relationship between the state of charge estimation error and time of the battery of the embodiment of the present invention. In step S4, a discrete equation for identifying the parameters of the electric and thermal aging coupling model is established, and the actual temperature, the actual terminal voltage and the actual terminal current at time t are input. The steps of using the recursive least squares method to identify the electric model parameters and the thermal model parameters, and using the genetic algorithm to identify the aging model parameters include:

[0116] Step S401, constructing a discrete equation for electrical model identification;

[0117] Step S402, constructing a discrete equation for thermal model identification;

[0118] Step S403, using a recursive least square method with a forgetting factor to identify parameters of the electrical model and the thermal model respectively;

[0119] Step S404, using a genetic algorithm to identify parameters of the aging model;

[0120] Among them, the electrothermal coupling model includes an electrical model and a thermal model.

[0121] In the implementation, the discrete equations for electrical model identification are constructed:

[0122] y(k)=θ(k)φ T (k)(21)

[0123] Where y(k) = V t (k)-V ocv (k); θ(k) is the parameter vector to be identified at time k; θ(k) = [θ1(k), θ2(k), θ3(k), θ4(k), θ5(k)]; φ(k) is the observation matrix at time k; φ(k) = [y(k-1), y(k-2), I(k), I(k-1), I(k-2)]; V t Finger tip voltage, (V volts); V ocv is the open circuit voltage of the battery (nonlinearly related to SOC), (V volts); k represents a discrete time index, k = 0, 1, 2, ..., N.

[0124] Construct the identification parameter vector:

[0125]

[0126] Among them, a=R0; b=τ1τ2; C=τ1+τ2; d=R0+R1+R2; e=R0(τ1+τ2)+R1τ2+R2τ1. Among them, τ1 is the value of C1, τ2 is the value of C2, Δt is the fixed time step; C1 and C2 are the polarized capacitances of the two Rc networks, respectively, (μF microfarads); R0 is the ohmic internal resistance, (Ω ohms);

[0127] R1 and R2 are the polarization resistances of the two Rc networks, respectively (Ω ohms); C is the ratio of the charge and discharge current per unit time to the nominal capacity of the battery.

[0128] Construct the discrete equations for thermal model identification:

[0129] y(k)=θ(k)φ T (k)(22)

[0130] k represents the discrete time index, k = 0, 1, 2, ..., N; θ is the parameter to be identified, φ is the regression variable, y(k) = T s (k+2)-T s(k+1),θ(k)=[θ1(k),θ2(k),θ3(k)],φ(k)=[H(k),T f (k)-T s (k),T s (k+1)-T s (k)]; T s is the surface temperature of the battery, (K Kelvin).

[0131] Construct the identification parameter vector:

[0132]

[0133] Among them, R u , R c are convection thermal resistance and thermal conduction thermal resistance, (Ω ohm); C s , C c are the surface heat capacity and internal heat capacity, respectively (J / K joule per Kelvin).

[0134] It can be understood that each model in the established electrothermal aging coupling model can be selected and combined. For example, the electrical model has an equivalent circuit model, an electrochemical model, etc., the thermal model can be divided into one-dimensional, two-dimensional, and three-dimensional thermal models, etc. The aging model based on the Arrhenius equation can construct the coefficients of the equation by itself.

[0135] Specifically, in step S403, the steps of using the recursive least square method with a forgetting factor to respectively identify the parameters of the electrical model and the thermal model include:

[0136] Step S4031, initializing the error covariance and parameter identification vector;

[0137] Step S4032, inputting the actual terminal current value collected at time t into the electrical model and the thermal model to obtain the predicted voltage and the predicted temperature respectively;

[0138] Step S4033, comparing the actual temperature collected at time t with the predicted temperature to obtain a temperature difference value, and comparing the actual terminal voltage collected at time t with the predicted voltage to obtain a voltage difference value;

[0139] Step S4034, adjusting the error covariance matrix according to the temperature difference value, the voltage difference value and the forgetting factor.

[0140] In the implementation, the voltage and current data are collected online to clarify the input φ(k) and output y(k) of the model at time k.

[0141] Calculate the estimated error:

[0142]

[0143] Update system gain:

[0144] K(k)=P(k-1)φ(k)[λ+φ(k) T P(k-1)φ(k)] -1 (25)

[0145] Parameter vector without considering bias compensation:

[0146]

[0147] Update the error covariance:

[0148]

[0149] in, is the parameter vector without considering deviation compensation; k represents the discrete time index, k = 0, 1, 2, ..., N; λ is the forgetting factor, and the general value range of λ is 0.95 to 1.

[0150] It can be understood that the genetic algorithm first encodes the parameters to form an initial population, and then uses the fitness function to evaluate the fitness of each individual in the population. Next, the algorithm simulates the three genetic operations of selection, crossover and mutation in biological evolution, and screens according to the fitness value of the individual, that is, individuals with lower fitness are gradually eliminated during the evolution process. As the iteration proceeds, the overall fitness of the population gradually increases until the predetermined stopping standard is reached. After that, the best individual is decoded to obtain the global optimal solution of the parameters.

[0151] It can be understood that the parameter identification of the battery aging model through the genetic algorithm (GA) ensures the high accuracy and reliability of the model. The parameters such as crossover operator, mutation operator, fitness function, population size, maximum evolutionary generation, crossover probability and mutation probability listed in the following table jointly optimize the performance of GA, thereby improving the accuracy and efficiency of aging model parameter identification. The appropriate selection of these aging model parameters provides a solid foundation for the optimization of fast charging strategies based on the flexible actor-critic (SAC) algorithm, enabling the agent to learn in an environment close to the actual battery behavior and balance the charging speed with the battery health status.

[0152]

[0153] Specifically, the present invention analyzes the intrinsic mechanism of charging affecting safety and attenuation, analyzes and decouples the various factors affecting the charging safety of automotive lithium-ion power batteries in multiple dimensions, and establishes an electro-thermal aging coupling model that takes aging into consideration. The model can more accurately describe and predict the dynamic behavior of the battery during the charging process, and provide a scientific basis for the formulation of battery management strategies. The core action mechanism of different battery electro-thermal systems is identified, aiming to reveal basic principles with universal applicability and to clarify the intrinsic connection between battery performance attenuation and its internal state, in order to suppress the attenuation process by controlling the internal state of the battery, and further improve the safety and robustness of the flexible actor critic fast charging method based on the electro-thermal aging coupling model.

[0154] See also Figure 5 As shown, it is a flow chart of optimizing the charging curve according to an embodiment of the present invention. In step S5, the Gym environment is defined according to the state space equation, the battery charging process is simulated, and a multi-constraint objective function is formulated, wherein:

[0155] By minimizing the cost function to balance the charging time, battery temperature, charging voltage, charging current and degradation rate,

[0156] The state space is a collection of states, including the battery's charge state, voltage state, and temperature state.

[0157] It can be understood that the Gym environment is defined according to the state-space equations, and the battery charging process is simulated, involving Python code, using the gym library. In the simulation of the battery charging process, the battery state at the next moment is calculated based on the current applied and the existing state by applying the constructed state-space equations. These equations describe the voltage response and dynamic electrical behavior, thermal effects, and degradation trends of the battery, and truly reflect the behavioral changes of the battery during the charging process.

[0158] It is understandable that the algorithm selection for optimizing the charging curve includes some control algorithms (model predictive control algorithm), some optimization algorithms (particle swarm algorithm), reinforcement learning that can handle continuous control tasks, machine learning, etc.

[0159] Specifically, in step S5, the flexible actor-critic algorithm is used to balance and explore the action space to determine the optimal charging strategy, where:

[0160] A soft penalty mechanism is used to guide the RL agent to impose soft penalties on behaviors that deviate from the optimal charging strategy by adjusting the weighting factors of charging time, battery temperature, charging voltage, charging current, and degradation rate in the cost function.

[0161] The weighting factors of charging voltage and battery temperature are three orders of magnitude higher than those of charging time, charging current, and degradation rate. The RL agent is an entity that performs actions and interacts with the environment in reinforcement learning. Soft penalties include extending the charging time, increasing the fluctuation size of the charging current, increasing the battery temperature, increasing the charging voltage, and accelerating the degradation rate.

[0162] It is understandable that the agent plays a core role in the SAC optimized fast charging algorithm based on the electrothermal aging coupling model. Its main task is to optimize the control of the charging current through the reinforcement learning mechanism to achieve an efficient and safe battery charging process while minimizing battery aging and thermal damage. To this end, the agent relies on a second-order equivalent circuit model to accurately simulate the voltage response and dynamic electrical characteristics of the battery, and uses a two-state thermal model to predict the temperature distribution and its changing law. In addition, the Arrhenius equation and the Puckett equation are combined to evaluate the battery aging rate, so as to fully understand the complex physical and chemical processes inside the battery.

[0163] The agent continuously explores different charging strategies through interaction with the environment, and dynamically adjusts its behavior based on immediate feedback (such as reward or penalty signals) obtained from the environment. The environmental state not only covers the current state parameters of the battery, but also includes the future state evolution predicted by the electrothermal aging coupling model. The agent aims to find an optimal strategy that can complete charging in the shortest time while ensuring that the battery aging is not accelerated due to overheating or excessive use.

[0164] To achieve this goal, the SAC algorithm adopts an Actor-Critic architecture, in which the "Actor" is responsible for selecting the optimal action (i.e., determining the charging current) based on the current state, while the "Critic" evaluates the effect of the action and provides feedback to guide the Actor to improve the decision-making process. In addition, SAC introduces an entropy regularization term, which not only promotes the pursuit of high-return paths (such as fast charging), but also maintains a certain degree of exploratory nature, preventing the algorithm from falling into a local optimal solution, thereby more effectively exploring the global optimal charging path.

[0165] It is understandable that soft penalties include extending the charging time, increasing the charging current fluctuation, increasing the battery temperature, increasing the charging voltage, and accelerating the degradation rate. Although these phenomena will not cause battery failure in the short term, they will lead to battery performance degradation and shortened life in the long run, so they must be weighed and managed when designing a charging strategy.

[0166] Specifically, in step S5, the degree of deviation of the behavior of the RL agent from the behavior of the optimal charging strategy is quantified by minimizing the cost function, where

[0167] The reward function is designed to be the negative of the cost function, providing negative cost feedback to the RL agent, which learns and executes the charging strategy with the least deviation through negative cost feedback.

[0168] It can be understood that in the SAC optimized fast charging algorithm based on the electrothermal aging coupling model, minimizing the cost function not only accurately quantifies the degree of deviation between the agent's behavior and the optimal charging strategy, but also guides the agent to gradually adjust its decision through the feedback mechanism to minimize these deviations, thereby achieving an efficient and safe charging process.

[0169] The application of the minimization cost function in this framework is not only a measure of the degree of deviation from the ideal charging path, but more importantly, it provides the agent with a clear learning goal and optimization direction. By quantifying a variety of unfavorable factors (such as extended charging time, excessive current fluctuations, increased battery temperature, increased charging voltage, and accelerated degradation rate), the minimization cost function can comprehensively evaluate the combined impact of different charging strategies. High cost values ​​reflect behaviors that have a negative impact on battery health or charging efficiency, prompting the agent to avoid such behaviors in the future; while low cost values ​​encourage strategies that can shorten charging time while ensuring battery health.

[0170] In addition, minimizing the cost function promotes a balance between short-term and long-term goals. It not only focuses on the immediate performance indicator of charging speed, but also attaches importance to the long-term health of the battery, ensuring that the agent does not accelerate battery aging or compromise its safety while pursuing fast charging. This mechanism enables the agent in the SAC algorithm to continuously optimize its charging strategy in a complex electrothermal aging environment, ultimately achieving a charging effect that is both efficient and safe, providing a more reliable and durable solution for applications that rely on battery power.

[0171] Specifically, the present invention constructs a state estimation model and a parameter identification method to address the problems of complex structure and large computational complexity of existing electrochemical mechanism models, reconstructs the original complex internal electrochemical mechanism model, optimizes the model structure to reduce the computational load without sacrificing accuracy, and designs targeted parameter identification methods. By optimizing existing measurement techniques and data processing algorithms, the speed and accuracy of parameter identification are improved, providing technical support for accurately estimating the internal state of the battery, and further improving the safety and robustness of the flexible actor critic fast charging method based on the electrothermal aging coupling model.

[0172] Specifically, in step S5, the flexible actor-critic algorithm includes a strategy network and a dual Q-value estimation calculation network, wherein:

[0173] The policy network consists of a decision network and an evaluation network.

[0174] Specifically, the present invention develops a safe fast charging algorithm based on state observation and optimal control, and uses the flexible actor critic algorithm to design a reasonable input parameter combination scheme to maximize or even improve the overall performance level of the battery while ensuring the shortest charging time. To this end, we will define an objective function that includes multiple dimensional evaluation indicators such as charging time, charging safety, and capacity decay rate, and set appropriate constraints to prevent exceeding the safety range, so that the thermal safety and life decay of the battery can be taken into account during charging control. Through the learning ability of the flexible actor critic algorithm, the global optimal solution or suboptimal solution set is found to achieve dynamic optimization of the charging strategy, further improving the safety and robustness of the flexible actor critic fast charging method based on the electrothermal aging coupling model.

[0175] Please cooperate Figure 6 , Figure 7 , Figure 8 , Fig. 9 See also Fig.10 As shown, they are respectively a schematic diagram of the charge state of the battery according to the PyBaMM simulation verification results of an embodiment of the present invention, a schematic diagram of the voltage of the battery according to the PyBaMM simulation verification results of an embodiment of the present invention, a schematic diagram of the internal temperature of the battery according to the PyBaMM simulation verification results of an embodiment of the present invention, a schematic diagram of the heat generation rate of the battery according to the PyBaMM simulation verification results of an embodiment of the present invention, and a schematic diagram of the health status of the battery according to the PyBaMM simulation verification results of an embodiment of the present invention.

[0176] As you can understand, PyBaMM is an open source software package for mathematical modeling of batteries, which is used here to simulate the charging behavior of lithium-ion batteries.

[0177] In the implementation, by defining the battery electrothermal aging model and setting boundary conditions, and using the charging strategy optimized by SAC to simulate charging, PyBaMM can simulate the battery's charging and discharging behavior, and finally obtain SOC, temperature, voltage and other graphs. The final charging strategy was simulated and verified using PyBaMM, and the simulation results fully reflected the dynamic change trends of charging time, battery aging rate, heat generation and temperature.

[0178] In practice, the design of the charging scheme should achieve the best balance between the three mutually constrained goals of improving charging speed, ensuring thermal safety, and extending battery life. Here, a method is proposed to achieve this balance by minimizing the cost function, thereby optimizing the control strategy and achieving the ideal charging effect. The defined minimization cost function is as follows:

[0179] J t =ω1C volt +ω2C soc +ω3C heat +ω4C soh +ω5Csmooth (28)

[0180] In the formula, ω1, ω2, ω3, ω4, and ω5 represent the weights of different charging targets. In order to effectively avoid the cost increase caused by overvoltage and overheating during the reinforcement learning training process, and in order not to destroy the exploration process of reinforcement learning, a soft penalty mechanism can be used to smoothly guide the RL agent. This can not only prevent accidental interruptions during training, but also help the agent learn safer and more stable strategies.

[0181]

[0182] Among them, V p,t Represents the terminal voltage at time t, V max 、V min Indicates the upper and lower limits of the terminal voltage (V volts); T t 、T max Represents the current battery surface temperature and the upper limit of the temperature, (K Kelvin). In order to ensure the safety of lithium batteries in practical applications, overvoltage and overheating should be avoided. Therefore, weighting factors three orders of magnitude higher than other factors are used for voltage and temperature items to ensure that the voltage and temperature remain within a safe range.

[0183] C soc The item describes the battery charging time as follows:

[0184] C soc =|SOC max -SOC t |(31)

[0185] Among them, SOC represents the available state of the remaining power in the battery; SOC max Indicates the target SOC, that is, the expected SOC when charging is completed, SOC t Indicates the current SOC.

[0186] C soh The term describes the degree of SOH reduction caused by charging at the current moment:

[0187] C soh =|ΔSOH t |(32)

[0188] Among them, SOH is the health state of the battery.

[0189] In actual charging applications, in order to ensure the stability and safety of the charging process, it is hoped that the charging current can be smoothly controlled to avoid equipment damage or other safety hazards caused by excessive current fluctuations. smooth The items are as follows:

[0190] C smooth =|I t -I t-1 |(33)

[0191] In the formula, I t Indicates the current value at the current time t, (A ampere), I t-1 Indicates the current value at the previous moment t-1 second, (A ampere).

[0192] C volt describes the charging voltage, C heat Describes the battery temperature.

[0193] In order to convert this optimal control problem into a form suitable for deep reinforcement learning (DRL), we design the reward function to minimize the negative of the cost function. Maximizing the reward function is to minimize the cost function.

[0194] r(s t ,a t )=-J t (34)

[0195] The state space is a collection of states, including the charge state, voltage state and temperature state of the battery, where Ts can be directly measured, V1 and V2 can be calculated by formula (4), and SOC and Tc can be estimated online using the method in step S4.

[0196] S={SOC,V1,V2,T s ,T c}(35)

[0197] The flexible actor-critic (SAC) algorithm can continuously control the charging current, and the action space can be defined as:

[0198] A={C t |C t ∈(0,6C)}(36)

[0199] Among them, C t represents the charging current (A ampere); C is the ratio of the charging and discharging current per unit time to the nominal capacity of the battery; s represents the state; a represents the action; T s is the surface temperature of the battery, (K Kelvin); V1 and V2 are the voltages across the two Rc networks, (V volts); SOC represents the available state of charge remaining in the battery; T c is the core temperature of the battery, (K Kelvin).

[0200] In implementation, Soft Actor-Critic (SAC) is an offline deep reinforcement learning algorithm. The core feature of the Soft Actor-Critic algorithm is entropy regularization. The strategy is trained to maximize the trade-off between expected return and entropy. The entropy reward term encourages strategy diversity, enabling the agent to explore more stably in uncertain environments and achieve higher performance and robustness in continuous control tasks.

[0201] The main goal in reinforcement learning is to maximize the cumulative reward r, which can usually be expressed as:

[0202]

[0203] Among them, τ and π are the trajectory and strategy, γ is the discount factor, t represents the current time, (s seconds); s t is the state at time t; a t is the action at time t; s t+1 is the state at time t+1.

[0204] Assume that x is a random variable with a probability density function of P. The entropy H of x is calculated based on its probability density distribution P. The calculation formula is:

[0205]

[0206] According to formula (37) and formula (38), the general form of the objective function of the flexible actor-critic algorithm can be obtained, as shown in the following formula:

[0207]

[0208] Among them, α is the exploration coefficient; τ and π are the trajectory and strategy, γ is the discount factor, t represents the current time, (s seconds); s t is the state at time t; a t is the action at time t; s t+1 is the state at time t+1; Combining the above formulas, we can get the strategy Q π The Bellman equation is shown in formula (40):

[0209]

[0210] The above strategy Q π The Bellman equation can be approximated by samples, as shown in formula (41):

[0211] Q π (s,a)≈r+γ(Q π (s′,a′)-αlogπ(a′∣s′)),a′~π(·∣s′)(41)

[0212] Among them, π is the trajectory and strategy, γ is the discount factor, t represents the current time (s seconds); s represents the state; a represents the action; r is the maximum cumulative reward; H is the entropy; V π is the state value function, which is a mapping that maps each state s in the environment to a real value, which represents the expected cumulative reward that can be obtained by acting according to the strategy π starting from this state; a′ is the action resampled from the strategy network in the s′ state, not the historical data in the experience pool, while r and s′ are the historical reward values ​​and states from the experience pool.

[0213] It can be understood that, referring to equations (35) and (36), in reinforcement learning, s′ represents the next state to which the environment transfers after taking action a in the current state s. a′ represents the action taken in the next state s′.

[0214] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0215] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A flexible actor-critic fast charging method based on an electrothermal aging coupling model, characterized in that: include: Establishing an electrothermal aging coupling model of a battery, determining parameters of the electrothermal aging coupling model, discretizing the established electrothermal aging coupling model parameters and constructing a state space equation; Using a K-type thermocouple, a voltage sensor and a current sensor to collect the actual temperature, actual terminal voltage and actual terminal current of the battery at time t; Using a lower charging rate method to collect the state of charge and open circuit voltage of the battery, and determining the relationship expression between the state of charge and the open circuit voltage of the battery by formula fitting; Establishing a discrete equation for identifying the parameters of the electrothermal aging coupling model, inputting the actual temperature, the actual terminal voltage and the actual terminal current at the time t, using the recursive least squares method to identify the electrical model parameters and the thermal model parameters, and using the genetic algorithm to identify the aging model parameters; Define the Gym environment according to the state-space equations, simulate the battery charging process, formulate a multi-constraint objective function, and use a flexible actor-critic algorithm to balance and explore the action space to determine the optimal charging strategy; Use PyBaMM to simulate and verify the optimal charging strategy; The electrothermal aging coupling model parameters include the electrical model parameters, the thermal model parameters and the aging model parameters; the constraints of the multi-constraint objective function include charging time, battery temperature, charging voltage, charging current and degradation rate.

2. The flexible actor-critic fast charging method based on the electrothermal aging coupling model according to claim 1 is characterized in that: The electric and thermal aging coupling model of the battery is established, wherein: A coupling model of the electrical model and the thermal model of the battery is established according to Kirchhoff's voltage and current laws, a second-order Rc equivalent circuit model and a lumped parameter two-state thermal model; Establishing the aging model of the battery according to the Arrhenius equation and the Puckett equation; The electrothermal aging coupling model includes the electrical model, the thermal model and the aging model.

3. The flexible actor-critic fast charging method based on the electrothermal aging coupling model according to claim 2 is characterized in that: The state of charge of the battery is estimated and calculated using an ampere-hour integration method.

4. The flexible actor-critic fast charging method based on the electrothermal aging coupling model according to claim 3 is characterized in that: The open circuit voltage values ​​corresponding to different state of charge values ​​of the battery in a fully discharged state during the charging process are recorded, and a relationship expression between the state of charge and the open circuit voltage is obtained by fitting the state of charge value and the open circuit voltage value.

5. The flexible actor-critic fast charging method based on the electrothermal aging coupling model according to claim 4 is characterized in that: The steps of establishing a discrete equation for identifying the parameters of the electrothermal aging coupling model, inputting the actual temperature, the actual terminal voltage and the actual terminal current at the time t, using the recursive least squares method to identify the electrical model parameters and the thermal model parameters, and using the genetic algorithm to identify the aging model parameters include: Construct discrete equations for electrical model identification; Construct discrete equations for thermal model identification; Using the recursive least square method with a forgetting factor to respectively identify the parameters of the electrical model and the thermal model; The genetic algorithm is used to perform parameter identification on the aging model.

6. The flexible actor-critic fast charging method based on the electrothermal aging coupling model according to claim 5 is characterized in that: The steps of respectively identifying the parameters of the electrical model and the thermal model using the recursive least square method with a forgetting factor include: Initialize the error covariance and parameter identification vector; Inputting the value of the actual terminal current collected at the time t into the electrical model and the thermal model to obtain a predicted voltage and a predicted temperature respectively; Comparing the actual temperature collected at the time t with the predicted temperature to obtain a temperature difference value, and comparing the actual terminal voltage collected at the time t with the predicted voltage to obtain a voltage difference value; An error covariance matrix is ​​adjusted according to the temperature difference value, the voltage difference value, and the forgetting factor.

7. The flexible actor-critic fast charging method based on the electrothermal aging coupling model according to claim 6 is characterized in that: The Gym environment is defined according to the state space equation, the battery charging process is simulated, and the multi-constraint objective function is formulated, wherein: balancing the charging time, the battery temperature, the charging voltage, the charging current, and the degradation rate by minimizing a cost function, The state space is a set of states, including the charge state, voltage state and temperature state of the battery.

8. The flexible actor-critic fast charging method based on the electrothermal aging coupling model according to claim 7 is characterized in that: The flexible actor-critic algorithm is used to balance and explore the action space to determine the optimal charging strategy, wherein: A soft penalty mechanism is used to guide the RL agent to impose soft penalties on behaviors that deviate from the optimal charging strategy by adjusting the weighting factors of the charging time, the battery temperature, the charging voltage, the charging current, and the degradation rate in the cost function. The weighting factors of the charging voltage and the battery temperature are three orders of magnitude higher than the weighting factors of the charging time, the charging current, and the degradation rate. The RL agent is an entity that performs actions and interacts with the environment in reinforcement learning. The soft penalty includes extending the charging time, increasing the charging current fluctuation, increasing the battery temperature, increasing the charging voltage, and accelerating the degradation rate.

9. The flexible actor-critic fast charging method based on the electrothermal aging coupling model according to claim 8 is characterized in that: The degree to which the behavior of the RL agent deviates from the behavior of the optimal charging strategy is quantified by minimizing the cost function, where The reward function is designed to be the negative of the cost function, and negative cost feedback is provided to the RL agent, and the RL agent learns and executes the charging strategy with the minimum deviation through the negative cost feedback.

10. The flexible actor-critic fast charging method based on the electrothermal aging coupling model according to claim 9 is characterized in that: The flexible actor-critic algorithm includes a strategy network and a dual Q-value estimation calculation network, wherein: The policy network consists of a decision network and an evaluation network.

Images