A lithium iron phosphate battery fast charging strategy optimization method and system based on reinforcement learning

Abstract

The application provides a lithium iron phosphate battery fast charging strategy optimization method and system based on reinforcement learning, and belongs to the technical field of battery management. The application extracts three-factor features and capacity attenuation features, generates gate fusion features in combination with potential sensitive features, and designs a potential boundary constraint activation function to calculate a negative electrode potential estimation value, so that the estimated range of an aged battery is automatically tightened. Meanwhile, the negative electrode potential features are extracted from the negative electrode potential estimation sequence, and the aging state features are generated in combination with the capacity attenuation rate. In the reinforcement learning framework, the aging state features are taken as the conditional input of the strategy network, a comprehensive reward function including a charging efficiency reward function and a lithium extraction risk penalty function is designed, and multi-objective optimization is realized. Finally, the generated charging current curve can be dynamically adjusted according to the actual aging state of the battery and the real-time charging parameters, which not only avoids the limitation of the charging speed of a new battery, but also effectively prevents the lithium extraction risk of an aged battery.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of battery management, and particularly relates to a lithium iron phosphate battery fast charging strategy optimization method and system based on reinforcement learning. BACKGROUND

[0002] Under the background of wide application of lithium ion batteries, especially lithium iron phosphate batteries, fast charging technology has become a key link to improve user experience and equipment operation efficiency. However, as the battery undergoes irreversible aging in the long-term use process, its internal electrochemical characteristics gradually degrade, which leads to serious challenges for the traditional fast charging strategy based on fixed current-voltage curve: on the one hand, the fixed charging curve is difficult to adapt to the dynamic response differences of batteries with different aging degrees, often limits the charging speed on new batteries, and is difficult to fully exert the fast charging capacity, while on the old batteries, the current is too large and induces lithium precipitation on the negative electrode, which seriously threatens the safety and life of the battery; on the other hand, the existing methods mostly rely on experience rules or simplified electrochemical models for charging control, and lack the ability of accurate perception and self-adaptive adjustment to the real-time state of the battery.

[0003] In recent years, deep learning technology has been introduced into the field of battery charging optimization, and intelligent decision-making methods represented by reinforcement learning have shown good potential. Usually, a Markov decision process model is constructed, and the voltage, current, temperature and other observable quantities of the battery are taken as state inputs, and the optimal charging current is output by the strategy network. However, the existing deep learning methods have the following key defects: first, the characterization of the battery aging state mostly stays at the static index level, such as only using cycle number or capacity decay rate as additional features, without fully considering the aging state and real-time charging parameters, which leads to the inability of the strategy to dynamically adjust according to the actual health status of the battery; second, the negative electrode potential, as a core variable for judging the risk of lithium precipitation, is difficult to measure directly, and existing estimation methods mostly rely on simplified models or fixed parameter mapping, which leads to inaccurate measurement of the negative electrode potential and further affects the judgment of the risk of lithium precipitation; third, the design of the reward function often focuses on a single target, such as maximizing the charging speed or minimizing the temperature rise, and lacks modeling of charging efficiency and lithium precipitation risk, which leads to out-of-control safety boundaries.

[0004] The above-mentioned drawbacks jointly restrict the reliability and universality of intelligent charging strategies under real complex working conditions, and it is difficult to achieve the cooperative optimization of safety and efficiency in the charging process of aged batteries. SUMMARY

[0005] In view of the problems existing in the background art, the purpose of the present application is to provide a lithium iron phosphate battery fast charging strategy optimization method and system based on reinforcement learning, which aims to overcome the problems in the prior art that battery life and charging efficiency are difficult to balance, and there is a risk of lithium precipitation in the charging process.

[0006] To achieve the above object, the technical scheme of the present application is as follows:

[0007] A lithium iron phosphate battery fast charging strategy optimization method based on reinforcement learning, comprising the following steps:

[0008] S1: Collect real-time voltage data, real-time current data and real-time temperature data of the lithium iron phosphate battery during the charging process through voltage sensors, current sensors and temperature sensors, and perform preprocessing to obtain preprocessed voltage time series data, current time series data and temperature time series data;

[0009] At the same time, the capacity decay rate is calculated by the capacity calibration method;

[0010] S2: Extract three-factor features according to the preprocessed voltage time series data, current time series data and temperature time series data, and extract capacity decay features according to the capacity decay rate;

[0011] According to the preprocessed voltage time series data, the preprocessed temperature time series data and the capacity decay rate, the potential sensitive feature is calculated;

[0012] According to the three-factor features, the capacity decay features and the potential sensitive features, the gating fusion features are calculated;

[0013] According to the gating fusion features, the potential boundary constraint activation function is designed, and the negative electrode potential estimation value is calculated;

[0014] The potential boundary constraint activation function is based on Sigmoid function and full connection layer to perform nonlinear transformation on the gating fusion features, and is scaled according to the difference between the capacity decay rate and the capacity decay health state threshold, so that the output range of the negative electrode potential estimation value is tightened;

[0015] S3: According to the negative electrode potential estimation value, the negative electrode potential estimation sequence is calculated, and the negative electrode potential feature is extracted; then combined with the capacity decay rate, the aging modulation feature, the risk sensitive feature and the aging state feature are calculated in turn;

[0016] S4: Construct the state space vector and action space of the reinforcement learning model, design the aging strategy network and the comprehensive reward function;

[0017] S5: Train the reinforcement learning model to the convergence state by using the aging strategy network and the comprehensive reward function, and record the real-time charging current action in the action space and arrange them in time sequence according to the converged reinforcement learning model to obtain the charging current curve; then according to the charging current curve, the current regulation instruction is executed through the charging equipment to complete the charging control of the battery.

[0018] Further, the specific process of step S1 is as follows:

[0019] S11: Collect real-time voltage data, real-time current data and real-time temperature data of the lithium iron phosphate battery during the charging process through voltage sensors, current sensors and temperature sensors respectively, and perform digital conversion through an analog-to-digital converter to obtain voltage time series data, current time series data and temperature time series data;

[0020] S12: Eliminate high-frequency noise through moving average filtering and normalize the voltage time series data, current time series data and temperature time series data using the minimum-maximum standardization method to obtain preprocessed voltage time series data, preprocessed current time series data and preprocessed temperature time series data.

[0021] S13: Calculate the capacity attenuation rate of the aged lithium iron phosphate battery through the capacity calibration method based on the historical charge-discharge cycle data stored in the battery management system.

[0022] Further, the lithium iron phosphate battery is a lithium iron phosphate battery with an aging degree reaching a threshold value; the threshold value is that the number of charge-discharge cycles is greater than 500, or the available capacity is less than 80% of the initial nominal capacity, or the direct current internal resistance is greater than 120% of the initial internal resistance.

[0023] Further, the specific process of step S2 is:

[0024] S21: Extract three-factor features from the preprocessed voltage time series data, preprocessed current time series data and preprocessed temperature time series data through a bidirectional long short-term memory network, and the calculation method is:

[0025] ,

[0026] wherein, is the three-factor feature at time t, t is the time index, is the bidirectional long short-term memory network, is the concatenation operation, is the preprocessed voltage time series data at time t, is the preprocessed current time series data at time t, is the preprocessed temperature time series data at time t;

[0027] S22: Extract the capacity attenuation feature through an embedding layer based on the capacity attenuation rate, and the calculation method is:

[0028] ,

[0029] wherein, is the capacity attenuation feature, is the embedding layer, is the capacity attenuation rate;

[0030] S23: Calculate the potential sensitivity feature according to the pre-processed voltage time series data, the pre-processed temperature time series data and the capacity fade rate, and the calculation method is:

[0031] ,

[0032] ,

[0033] ,

[0034] wherein, is the voltage change rate sequence, is the central difference operator, is the pre-processed voltage time series data, is the temperature change rate sequence, is the pre-processed temperature time series data, is the potential sensitivity feature, is the Sigmoid function;

[0035] S24: According to the three-factor feature, the capacity fade feature and the potential sensitivity feature, generate the gated fusion feature through the potential sensitivity gating mechanism, and the calculation method is:

[0036] ,

[0037] ,

[0038] wherein, is the intermediate modulation feature at time t, and is the Hadamard product, is the dynamic feature gain coefficient, is the aging modulation amplitude coefficient, is the hyperbolic tangent function, is the full connection layer, is the gated fusion feature at time t;

[0039] S25: According to the gated fusion feature, design the potential boundary constraint activation function, calculate the negative electrode potential estimation value, and the calculation method is:

[0040] ,

[0041] ,

[0042] wherein, is the negative electrode potential estimation value at time t, is the lower limit of the negative electrode potential safety, is the upper limit of the negative electrode potential, is the potential boundary constraint activation function, is the capacity fade health state threshold value, aging-sensitive gain coefficient, to maximize.

[0043] It needs to be further explained that, in view of the technical difficulties that the negative electrode potential estimation of the aged lithium iron phosphate battery is not accurate during fast charging, and the aged battery is more sensitive to charging conditions, the three-factor feature and the capacity attenuation feature are extracted, and the potential sensitive feature is combined for gated fusion, which can effectively associate the dynamic response of the battery and the aging state, so that the gated fusion feature contains both the electrochemical behavior of the real-time charging process and the battery aging information.

[0044] Specifically, the three-factor feature extracts the dynamic parameter change characteristics of the battery in the charging process from the voltage, current and temperature time series data, such as linear voltage rise in the constant current stage and current decay in the constant voltage stage; the capacity attenuation feature extracts the aging degree information of the battery from the capacity attenuation rate, which reflects the aging phenomena such as increased internal resistance of the battery; the potential sensitive feature models the interaction of voltage change rate, temperature change rate and capacity attenuation rate; in the gated fusion process, the potential sensitive feature is part of the dynamic adjustment weight, which intelligently balances the contribution proportion of the three-factor feature and the capacity attenuation feature, and automatically increases the weight of the capacity attenuation feature when the system detects high-risk working conditions, so that the fusion feature is more inclined to reflect the safety constraints of the aged battery.

[0045] On this basis, the potential boundary constraint activation function can dynamically adjust the negative electrode potential estimation range according to the difference between the capacity attenuation rate and the capacity attenuation health state threshold, so that the battery with higher aging degree can automatically tighten the estimation range to avoid lithium precipitation risk;

[0046] The activation function adds a dynamic scaling term related to the aging state to the standard Sigmoid function When the battery health state is good (for example ), the dynamic scaling term remains at the baseline value, the negative electrode potential estimation range is normal, and the model estimates the negative electrode potential in the normal way; when the battery aging degree is aggravated (for example ), the dynamic scaling term automatically increases, so that the upper limit of the negative electrode potential estimation value is tightened towards the safety boundary, and this mechanism realizes the technical effect of "the more aged the battery, the more conservative the negative electrode potential estimation";

[0047] The design embeds the physical safety prior between the battery aging and the lithium precipitation risk, can automatically adjust the estimation strategy according to the actual aging degree of the battery, and effectively prevents the negative electrode potential from approaching the critical point of lithium precipitation; compared with the prior art, the present application realizes accurate estimation of the negative electrode potential through multi-source information fusion, considers the influence of the battery aging state on the charging safety, and makes the negative electrode potential estimation value self-adaptive to the actual aging degree of the battery; since the core design of the present application is carried out around the lithium precipitation sensitivity of the aged battery, the potential boundary constraint mechanism is used to ensure that a more conservative estimation strategy is automatically adopted when the battery aging intensifies, thereby avoiding the problem of limited charging speed of the new battery and effectively preventing the lithium precipitation risk caused by overcharging of the aged battery, so that the adaptability to the aged lithium iron phosphate battery scene is stronger, and the balance between safe fast charging and battery life protection is realized.

[0048] Further, in the S3 step, the negative electrode potential estimation value at each time is sequentially spliced in time sequence to obtain a negative electrode potential estimation sequence; a negative electrode potential feature is extracted according to the negative electrode potential estimation sequence; an aging modulation feature is generated according to the capacity attenuation rate; the negative electrode potential feature and the aging modulation feature are fused to generate a risk sensitive feature; and an aging state feature is generated through an attention mechanism according to the risk sensitive feature and the negative electrode potential estimation sequence.

[0049] Further, the specific process of step S3 is:

[0050] S31: sequentially splicing the negative electrode potential estimation value at each time in time sequence to obtain a negative electrode potential estimation sequence ;

[0051] S32: extracting a negative electrode potential feature through a one-dimensional convolution layer and a one-dimensional maximum pooling layer according to the negative electrode potential estimation sequence, and the calculation method is:

[0052] ,

[0053] wherein, the negative electrode potential feature is, the one-dimensional convolution layer is, the one-dimensional maximum pooling layer is, the ReLU function is;

[0054] S32: generating an aging modulation feature through an aging sensitive gating mechanism according to the capacity attenuation rate; and the calculation method is:

[0055] ,

[0056] wherein, the aging modulation feature is, the multi-layer perception is, the maximum value is;

[0057] S33: generating a risk-sensitive feature according to the negative electrode potential feature and the aging modulation feature, and the calculation method is as follows:

[0058] ,

[0059] ,

[0060] wherein, is a feature fusion gate weight, is a risk-sensitive feature;

[0061] S34: generating an aging state feature through an attention mechanism according to the risk-sensitive feature and the negative electrode potential estimation sequence, and the calculation method is as follows:

[0062] ,

[0063] wherein, is an aging state feature, is an attention mechanism.

[0064] It needs to be further explained that the aging lithium iron phosphate battery faces the technical difficulty that the static aging state is difficult to drive the dynamic charging decision in the charging process, because the capacity attenuation rate and other aging indicators change slowly and cannot directly reflect the real-time risk change in the charging process, and the traditional method usually inputs the aging state as a fixed parameter or simply splices it with the real-time state, resulting in insufficient adaptability of the charging strategy to the aging battery, which cannot maximize the charging speed under the premise of safety, and it is also difficult to effectively prevent the risk of lithium precipitation;

[0065] The negative electrode potential estimation value at each time is spliced in time sequence to form a sequence to represent the dynamic evolution law of the negative electrode potential in the charging process, and complete time sequence information is provided for subsequent feature extraction; on this basis, the negative electrode potential feature is extracted through a one-dimensional convolution layer and a one-dimensional maximum pooling layer, which can effectively identify the key feature points in the charging curve, such as the width and slope of the voltage platform period, and these features are closely related to the risk of lithium precipitation; at the same time, the aging modulation feature is generated according to the capacity attenuation rate, which quantifies the influence of the battery aging degree on the charging safety, and when At the same time, the aging modulation feature is automatically enhanced, reflecting the physical characteristics of the aging battery that are more sensitive to the same charging conditions; further, the negative electrode potential feature and the aging modulation feature are fused through a feature fusion gate weight to generate a risk-sensitive feature, which realizes dynamic evaluation of the charging process risk - in a safe working condition, the system tends to focus on the negative electrode potential feature to maintain charging efficiency, and in a high-risk working condition, the weight of the aging modulation feature is automatically increased to strengthen safety constraints; finally, the risk-sensitive feature and the negative electrode potential estimation sequence are combined through an attention mechanism to generate an aging state feature, so that the feature not only contains the aging degree information of the battery, but also fuses the real-time risk level under the current charging state, solving the fundamental problem of "static aging state cannot drive real-time decision";

[0066] Compared with the prior art, the capacity decay rate is no longer simply used as a fixed parameter, but through a multi-level feature extraction and fusion mechanism, the static aging information is converted into a dynamic risk evaluation index, so that the charging strategy can be adaptively adjusted according to the actual aging degree of the battery and the current charging state.

[0067] Further, in the S4 step, a state space vector containing the current temperature, the current charging amount, the current voltage change rate and the negative electrode potential estimation value, and an action space containing the minimum charging current to the maximum charging current range are constructed; according to the aging state feature, the hidden state of the strategy network is modulated through an aging gate weight matrix, an aging gate bias vector, an offset weight matrix and an offset bias vector, so that the aging state feature serves as the conditional input of the strategy network.

[0068] A comprehensive reward function including a charging efficiency reward function and a lithium precipitation risk penalty function is designed, wherein the charging efficiency reward function is calculated according to the current charging amount and its first derivative, the lithium precipitation risk penalty function is calculated according to the negative electrode potential estimation value, the current charging amount and the current temperature, and the comprehensive reward function is further calculated by combining the current voltage change rate and the current temperature.

[0069] Further, the specific process of step S4 is:

[0070] S41: Constructing the state space vector and the action space of the reinforcement learning model, the calculation method is:

[0071] ,

[0072] ,

[0073] wherein, is the state space vector, is the current temperature, is the current charging amount, Let A be the current voltage change rate, A be the operating space, and a be the current charging current operating value. Minimum charging current, This is the maximum charging current;

[0074] S42: Based on the characteristics of the aging state, design an aging policy network using a reinforcement learning model to calculate the current charging current. The calculation method is as follows:

[0075] ,

[0076] ,

[0077] ,

[0078] in, This represents the hidden state of the k-th layer of the aging strategy network, where k is the layer index. K represents the total number of layers in the aging strategy network. Let be the optimized weight matrix of the k-th layer of the aging strategy network. For aging strategy network The hidden state of the layer Let be the optimized bias vector of the k-th layer of the aging strategy network. This represents the hidden state of the final layer of the aging strategy network. For aging strategy network The hidden state of the layer For aging gate weight matrix, This is the aging gate bias vector. This is the offset weight matrix. This is the offset vector. This is the final optimized weight matrix for the Kth layer of the aging strategy network. This is the final optimized bias vector for the Kth layer of the aging strategy network;

[0079] S43: Design a comprehensive reward function for the reinforcement learning model based on the state space vector.

[0080] It should be further explained that existing reinforcement learning techniques generally treat the aging state as an external static parameter or simply as an additional dimension of the state space in the optimization of battery fast charging strategies. They fail to effectively integrate it into the decision-making mechanism of the policy network, which makes it difficult for the policy function to dynamically adjust its behavior pattern according to the degree of battery aging and to achieve adaptive control of batteries at different aging stages.

[0081] This invention changes the limitation of the policy function's "passive response" to the aging state in traditional reinforcement learning by deeply embedding the aging state features as conditional inputs to the policy network into the generation process of the network's hidden state.

[0082] Specifically, in the aging strategy network, a dynamic gating mechanism is constructed through the weight matrix and bias vector of each layer, so that the aging state feature can adjust the activation strength of the hidden state of each layer of the strategy network. When the aging state feature reflects that the battery is in a mild aging state, the aging gating weight remains at a high level, the strategy network is dominated by the pursuit of charging efficiency, and a higher charging current action is output. When the aging state feature reflects that the battery enters a serious aging state, the aging gating weight is automatically reduced, and a negative offset is introduced through the offset weight matrix and offset bias vector, so that the overall output of the strategy network is offset to the conservative direction, thereby realizing the continuous evolution of the strategy behavior with the aging degree without changing the network structure. This design makes the strategy function no longer a single fixed mapping relationship, but has the ability to "adaptively deform" according to the battery aging state;

[0083] Compared with the prior art, the present application no longer simply splices the aging state as an isolated input feature, but converts it into a "control signal" that affects the result generation process of the strategy network, so that the strategy network can actively adjust its decision preference according to the aging state, avoiding the problem of strategy rigidity caused by slow changes in the aging state. At the same time, this method does not need to train multiple independent strategy models for different aging stages, significantly reducing the training complexity and deployment cost;

[0084] The present application breaks through the inherent limitations of traditional reinforcement learning in dealing with slow-changing characteristics such as battery aging, and upgrades the aging state from a "observed state variable" to a "conditional variable that regulates strategy behavior", so that the charging strategy can respond to real-time electrochemical state and adapt to long-term aging evolution, achieving the coordinated optimization of charging speed and battery life, and thus having stronger adaptability to the charging scene of aging lithium iron phosphate batteries.

[0085] Further, the specific process of step S43 is:

[0086] S431: According to the current charging capacity, the charging efficiency reward function is calculated, and the calculation method is:

[0087] ,

[0088] Among them, is the charging efficiency reward function, is the first derivative of the current charging capacity, is a natural constant;

[0089] S432: According to the negative electrode potential estimate value, the current charging capacity and the current temperature, the lithium precipitation risk penalty function is calculated, and the calculation method is:

[0090] ,

[0091] ,

[0092] wherein, is a lithium precipitation risk penalty function, is a natural logarithm function, is a risk penalty term;

[0093] S433: According to the charging efficiency reward function, the lithium precipitation risk penalty function, the current voltage rate of change and the current temperature, a comprehensive reward function is calculated in the following manner:

[0094] ,

[0095] wherein, is a comprehensive reward function, is an absolute value.

[0096] It needs to be further explained that, in view of the problem that the charging speed and the battery life are difficult to balance in the fast charging process of the aged lithium iron phosphate battery, the traditional method usually adopts a fixed charging curve or a simple linear weighted reward function, which cannot accurately reflect the lithium precipitation risk under the coupling of multiple physical fields, especially in the high SOC area and low temperature conditions, which is prone to cause the problems of overcharging or insufficient charging speed; The comprehensive reward function designed in the present application realizes the precise regulation and control of the charging process through the synergistic effect of the charging efficiency reward function, the lithium precipitation risk penalty function and the like.

[0097] Specifically, the charging efficiency reward function not only considers the current charging speed, but also matches the electrochemical characteristics of the lithium iron phosphate battery through nonlinear design - fully encourages fast charging in the low SOC area to improve efficiency, and realizes the natural attenuation of the reward through the cooperation of the cubic term and the exponential term in the high SOC area.

[0098] The lithium precipitation risk penalty function takes the negative electrode potential estimation value as a direct evaluation index of the lithium precipitation risk, and when the negative electrode potential approaches the 0.05 volt lithium precipitation critical point, the risk penalty term grows exponentially, while considering the influence of the increased lithium concentration gradient in the high SOC area (more than 75%) and the inhibition of temperature on lithium precipitation, especially when the negative electrode potential drops to 0.03 volts, which is the lithium precipitation acceleration critical point, the reinforcement penalty mechanism is triggered through the Sigmoid function, so that the risk penalty is doubled, effectively preventing the damage of the battery from lithium precipitation;

[0099] The comprehensive reward function realizes the dynamic balance of multiple objectives by subtracting the lithium precipitation risk penalty from the charging efficiency reward and adding the voltage rate of change and the temperature safety constraint term - the voltage rate of change constraint term starts nonlinear penalty when the rate of change exceeds 0.005 volts per second, ensuring system stability, and the temperature constraint term starts linear penalty when it exceeds 45 degrees Celsius, preventing the risk of thermal runaway;

[0100] This design enables the reward function to automatically adjust the reward orientation according to the real-time state, prioritizing charging speed in safe working conditions and automatically switching to safety priority in high-risk working conditions, for example, when the battery is in a low-temperature environment and the negative electrode potential is close to the critical value, the comprehensive reward decreases rapidly, prompting the strategy network to automatically reduce the charging current, and when the temperature is suitable and the negative electrode potential is safe, a higher charging speed is maintained; compared with the prior art, the present application no longer relies on indirect indicators such as terminal voltage to evaluate risk, but directly estimates the risk penalty function based on the negative electrode potential, which is more in line with the physical mechanism of lithium precipitation.

[0101] The application also discloses a lithium iron phosphate battery fast charging strategy optimization system based on reinforcement learning, which comprises:

[0102] The data acquisition module: through voltage, current and temperature sensors, real-time voltage, current and temperature data of the aged lithium iron phosphate battery during the charging process are collected and pretreated to obtain pretreated voltage, current and temperature time series data; the capacity attenuation rate is calculated.

[0103] The negative electrode potential estimation module: according to the pretreated voltage, current and temperature time series data and the capacity attenuation rate, three-factor features and capacity attenuation features are extracted, and then potential sensitive features and gate fusion features are calculated in sequence; according to the gate fusion features, a potential boundary constraint activation function is designed to calculate the negative electrode potential estimation value.

[0104] The aging state feature extraction module: according to the negative electrode potential estimation value, the negative electrode potential estimation sequence is calculated, and the negative electrode potential feature is extracted; then, combined with the capacity attenuation rate, the aging modulation feature, the risk sensitive feature and the aging state feature are calculated in sequence.

[0105] The reinforcement learning module: the state space vector and the action space of the reinforcement learning model are constructed, the aging strategy network and the comprehensive reward function are designed; then the reinforcement learning model is trained to a convergent state by using the aging strategy network and the comprehensive reward function.

[0106] The charging control module: according to the converged reinforcement learning model, the real-time charging current action in the action space is recorded and arranged in time sequence to obtain the charging current curve; then, according to the charging current curve, the current adjustment instruction is executed by the charging equipment to complete the charging control of the battery.

[0107] As described above, due to the adoption of the above technical solutions, the present application has the following advantages:

[0108] (1) The present application proposes a fast charging strategy optimization method based on reinforcement learning to solve the problem that the traditional fast charging strategy of aged lithium iron phosphate battery relies on fixed current curve, which makes it difficult to balance the charging speed and life and easily causes lithium precipitation; by extracting three-factor features and capacity attenuation features, combining potential sensitive features to generate gated fusion features, and innovatively designing potential boundary constraint activation function to calculate negative electrode potential estimation value, the estimated range of the aged battery is automatically tightened; further, the present application extracts negative electrode potential features from the negative electrode potential estimation sequence, and generates aging state features in combination with the capacity attenuation rate; in the reinforcement learning framework, the aging state features are used as the conditional input of the strategy network, a comprehensive reward function including charging efficiency reward function and lithium precipitation risk penalty function is designed, and multi-objective optimization is realized; finally, the generated charging current curve can be dynamically adjusted according to the actual aging state of the battery and the real-time charging parameters, which not only avoids the limitation of charging speed of new battery, but also effectively prevents the risk of lithium precipitation of aged battery, significantly improving the charging safety and battery service life.

[0109] (2) In view of the problem that the negative electrode potential estimation is not accurate in the fast charging of aged lithium iron phosphate battery, and the aged battery is sensitive to charging conditions, the present application innovatively proposes a gated fusion method of three-factor features and capacity attenuation features, which effectively associates the dynamic response of the battery with the aging state; by designing potential sensitive features to intelligently balance the contribution proportion of three-factor features and capacity attenuation features, the safety constraint is automatically enhanced under high-risk working conditions, and the charging efficiency is maintained under safe working conditions; the present application also designs a potential boundary constraint activation function to dynamically adjust the negative electrode potential estimation range according to the difference between the capacity attenuation rate and the capacity attenuation health state threshold, realizing the technical effect that "the older the battery, the more conservative the negative electrode potential estimation": when the battery is in good health, it is estimated in the conventional way; when the battery is aging, the estimation range is automatically tightened.

[0110] (3) In view of the problem that static aging state is difficult to drive dynamic decision in the charging of aged lithium iron phosphate battery, the present application concatenates the negative electrode potential estimation value in time sequence to form a sequence, which completely represents the dynamic evolution law of the charging process; by using one-dimensional convolution layer and one-dimensional maximum pooling layer to extract negative electrode potential features, the feature points closely related to lithium precipitation risk such as voltage platform period are accurately identified; at the same time, aging modulation features are generated according to the capacity attenuation rate, reflecting the sensitivity of aged battery to charging conditions; by intelligently fusing negative electrode potential features and aging modulation features through feature fusion gate weight, risk sensitive features are generated: in safe working conditions, the charging efficiency is given priority to, and in high-risk working conditions, the safety constraint is automatically strengthened; finally, through attention mechanism, aging state features are generated by fusing aging degree and real-time risk level, solving the problem of "static aging state cannot drive real-time decision", so that the charging strategy can be dynamically adjusted according to the actual aging state of the battery and the current charging environment, realizing the optimal balance between safety and efficiency.

[0111] (4) In view of the problem that the existing reinforcement learning technology simply takes the aging state as a static parameter or a state space additional dimension, the aging state feature is innovatively taken as the conditional input of the strategy network and is deeply embedded into the hidden state generation process; when the aging state feature reflects that the battery is in a mild aging state, a higher gating weight is maintained, the strategy network outputs a higher charging current to pursue efficiency; when it reflects that the battery enters a serious aging state, the gating weight is automatically reduced and a negative offset is introduced, so that the output is adjusted in a conservative direction, so that the strategy function has the ability of "self-adaptive deformation", realizing the collaborative optimization of charging speed and battery life, and significantly enhancing the adaptability to the charging scene of the aging lithium iron phosphate battery.

[0112] (5) In view of the problem that the charging speed and the life are difficult to balance in the fast charging of the aging lithium iron phosphate battery, a comprehensive reward function is innovatively designed; wherein, the charging efficiency reward function matches the battery electrochemical characteristics through nonlinear design, encourages fast charging in the low SOC area, and realizes natural attenuation in the high SOC area; the lithium precipitation risk penalty function is constructed based on the negative electrode potential estimation value, the penalty increases exponentially when the negative electrode potential approaches the 0.05 volt lithium precipitation critical point, and the strong punishment mechanism is triggered especially at the 0.03 volt lithium precipitation acceleration critical point, and the influence of high SOC area and temperature is considered; the comprehensive reward function realizes multi-objective dynamic balance by subtracting the lithium precipitation risk penalty from the charging efficiency reward, and adding the voltage change rate and the safety constraint term of temperature, so that the reward function can automatically adjust the direction according to the real-time state, and the speed is given priority in the safe working condition, and the safety is turned to in the high-risk working condition, compared with the traditional method, the physical mechanism of lithium precipitation is more in line with, and the balance ability of charging safety and efficiency is significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0113] Figure 1 It is an algorithm flowchart of the aging state feature of the application;

[0114] Figure 2 It is a system interface schematic diagram provided by the application. DETAILED DESCRIPTION

[0115] In order to make the purpose, technical scheme and advantages of the application clearer, the application will be further described in detail below in combination with embodiments and drawings.

[0116] Embodiment 1

[0117] A lithium iron phosphate battery fast charging strategy optimization method based on reinforcement learning, comprising the following steps:

[0118] S1: Through the voltage, current and temperature sensors, the real-time voltage, current and temperature data of the aging lithium iron phosphate battery in the charging process are collected and preprocessed respectively to obtain the preprocessed voltage, current and temperature time series data; the capacity attenuation rate is calculated; including:

[0119] S11: Collecting real-time voltage data, real-time current data and real-time temperature data of the aged lithium iron phosphate battery during the charging process through voltage sensors, current sensors and temperature sensors respectively, and carrying out digital conversion through an analog-to-digital converter to obtain voltage time series data, current time series data and temperature time series data; the aged lithium iron phosphate battery is a lithium iron phosphate battery with 700 times of charge-discharge cycle times;

[0120] S12: Eliminating high-frequency noise through moving average filtering and normalizing the voltage time series data, the current time series data and the temperature time series data by using the minimum-maximum standardization method to obtain preprocessed voltage time series data, preprocessed current time series data and preprocessed temperature time series data;

[0121] S13: According to the historical charge-discharge cycle data stored in the battery management system, the capacity attenuation rate of the aged lithium iron phosphate battery is calculated through the capacity calibration method.

[0122] S2: According to the preprocessed voltage, current and temperature time series data and the capacity attenuation rate, three-factor features and capacity attenuation features are extracted respectively, and then potential sensitive features and gating fusion features are calculated in turn; according to the gating fusion features, a potential boundary constraint activation function is designed to calculate the negative electrode potential estimation value, which is specifically:

[0123] S21: According to the preprocessed voltage time series data, the preprocessed current time series data and the preprocessed temperature time series data, three-factor features are extracted through a bidirectional long short-term memory network, and the calculation method is:

[0124] ,

[0125] wherein, is the three-factor feature at time t, t is the time index, is the bidirectional long short-term memory network, is the splicing operation, is the preprocessed voltage time series data at time t, is the preprocessed current time series data at time t, is the preprocessed temperature time series data at time t;

[0126] S22: According to the capacity attenuation rate, the capacity attenuation feature is extracted through an embedding layer, and the calculation method is:

[0127] ,

[0128] wherein, is the capacity attenuation feature, is the embedding layer, is the capacity attenuation rate;

[0129] S23: Calculate the potential sensitivity feature according to the pre-processed voltage time series data, the pre-processed temperature time series data and the capacity fade rate, and the calculation method is:

[0130] ,

[0131] ,

[0132] ,

[0133] wherein, is the voltage change rate sequence, is the central difference operator, is the pre-processed voltage time series data, is the temperature change rate sequence, is the pre-processed temperature time series data, is the potential sensitivity feature, is the Sigmoid function;

[0134] S24: According to the three-factor feature, the capacity fade feature and the potential sensitivity feature, generate the gated fusion feature through the potential sensitivity gating mechanism, and the calculation method is:

[0135] ,

[0136] ,

[0137] wherein, is the intermediate modulation feature at time t, and is the Hadamard product, is the dynamic feature gain coefficient, is the aging modulation amplitude coefficient, is the hyperbolic tangent function, is the full connection layer, is the gated fusion feature at time t;

[0138] S25: According to the gated fusion feature, design the potential boundary constraint activation function, calculate the negative electrode potential estimation value, and the calculation method is:

[0139] ,

[0140] ,

[0141] wherein, is the negative electrode potential estimation value at time t, is the lower limit of the negative electrode potential safety, is the upper limit of the negative electrode potential, is the potential boundary constraint activation function, is the capacity fade health state threshold, aging-sensitive gain coefficient;

[0142] S3: According to the negative electrode potential estimation value, the negative electrode potential estimation sequence is calculated, and the negative electrode potential feature is extracted; then combined with the capacity attenuation rate, the aging modulation feature, the risk-sensitive feature and the aging state feature are calculated in turn, wherein the algorithm flow chart of the aging state feature is as shown in Figure 1

[0143] S31: The negative electrode potential estimation value at each time is sequentially spliced according to time sequence to obtain the negative electrode potential estimation sequence

[0144] S32: According to the negative electrode potential estimation sequence, the negative electrode potential feature is extracted through one-dimensional convolution layer and one-dimensional maximum pooling layer, and the calculation method is:

[0145] ,

[0146] wherein, is the negative electrode potential feature, is the one-dimensional convolution layer, is the one-dimensional maximum pooling layer, is the ReLU function;

[0147] S32: According to the capacity attenuation rate, the aging modulation feature is generated through the aging-sensitive gating mechanism; the calculation method is:

[0148] ,

[0149] wherein, is the aging modulation feature, is the multi-layer perception, is the maximum value;

[0150] S33: According to the negative electrode potential feature and the aging modulation feature, the risk-sensitive feature is generated, and the calculation method is:

[0151] ,

[0152] ,

[0153] wherein, is the feature fusion gating weight, is the risk-sensitive feature;

[0154] S34: According to the risk-sensitive feature and the negative electrode potential estimation sequence, the aging state feature is generated through the attention mechanism, and the calculation method is:

[0155] ,

[0156] wherein,​​ This is a characteristic of an aging state. This is an attention mechanism.

[0157] S4: Construct the state space vector and action space of the reinforcement learning model, and design the aging policy network and comprehensive reward function, including:

[0158] S41: Construct the state space vector and action space of the reinforcement learning model, calculated as follows:

[0159] ,

[0160] ,

[0161] in, It is a state-space vector. The current temperature. Current charging level Let A be the current voltage change rate, A be the operating space, and a be the current charging current operating value. Minimum charging current, This is the maximum charging current;

[0162] S42: Based on the characteristics of the aging state, design an aging policy network using a reinforcement learning model to calculate the current charging current. The calculation method is as follows:

[0163] ,

[0164] ,

[0165] ,

[0166] in, This represents the hidden state of the k-th layer of the aging strategy network, where k is the layer index. K represents the total number of layers in the aging strategy network. Let be the optimized weight matrix of the k-th layer of the aging strategy network. For aging strategy network The hidden state of the layer Let be the optimized bias vector of the k-th layer of the aging strategy network. This represents the hidden state of the final layer of the aging strategy network. For aging strategy network The hidden state of the layer This is the aging gate weight matrix. This is the aging gate bias vector. This is the offset weight matrix. This is the offset vector. This is the final optimized weight matrix for the Kth layer of the aging strategy network. is the final optimization bias vector of the aging strategy network Kth layer;

[0167] S43: According to the state space vector, design the comprehensive reward function of the reinforcement learning model, including:

[0168] S431: According to the current charge amount, calculate the charging efficiency reward function, and the calculation method is:

[0169] ,

[0170] wherein, is the charging efficiency reward function, is the first derivative of the current charge amount, is a natural constant;

[0171] S432: According to the negative electrode potential estimate value, the current charge amount and the current temperature, calculate the lithium extraction risk penalty function, and the calculation method is:

[0172] ,

[0173]

[0174] wherein, is the lithium extraction risk penalty function, is a natural logarithm function, is a risk penalty term;

[0175] S433: According to the charging efficiency reward function, the lithium extraction risk penalty function, the current voltage change rate and the current temperature, calculate the comprehensive reward function, and the calculation method is:

[0176] ,

[0177] wherein, is the comprehensive reward function, is the absolute value.

[0178] S5: Use the aging strategy network and the comprehensive reward function to train the reinforcement learning model to the convergence state, and according to the converged reinforcement learning model, record the real-time charging current action in the action space and arrange it in time sequence to obtain the charging current curve; According to the charging current curve, execute the current adjustment instruction through the charging equipment to complete the charging control of the battery.

[0179] Embodiment 2

[0180] A lithium iron phosphate battery fast charging strategy optimization system based on reinforcement learning, comprising:

[0181] Data acquisition module: through voltage, current and temperature sensors, real-time voltage, current and temperature data of the aged lithium iron phosphate battery during the charging process are collected and pretreated respectively to obtain pretreated voltage, current and temperature time series data; the capacity attenuation rate is calculated;

[0182] Negative electrode potential estimation module: according to the pretreated voltage, current and temperature time series data and the capacity attenuation rate, three-factor features and capacity attenuation features are extracted respectively, and then potential sensitive features and gate fusion features are calculated in turn; according to the gate fusion features, a potential boundary constraint activation function is designed to calculate the negative electrode potential estimation value;

[0183] For the aged lithium iron phosphate battery of embodiment 1, if the negative electrode potential estimation value is between 0.05V and 0.1V, it is in the critical safety zone and can be accepted, but needs to be monitored, and long-term operation in this interval may cause slight lithium precipitation; if the negative electrode potential estimation value is between 0.02V and 0.05V, it is in the high-risk zone, and lithium precipitation begins to occur significantly, and should be avoided for long-term operation; if the negative electrode potential estimation value is less than 0.02V, it is in the extremely high-risk zone, and lithium precipitation is severe, and lithium dendrites are easy to form, which may cause internal short circuit and thermal runaway;

[0184] Aging state feature extraction module: according to the negative electrode potential estimation value, the negative electrode potential estimation sequence is calculated, and the negative electrode potential feature is extracted; then combined with the capacity attenuation rate, the aging modulation feature, the risk sensitive feature and the aging state feature are calculated in turn;

[0185] Reinforcement learning module: the state space vector and the action space of the reinforcement learning model are constructed, the aging strategy network and the comprehensive reward function are designed; then the reinforcement learning model is trained to the convergence state by using the aging strategy network and the comprehensive reward function;

[0186] Charging control module: according to the converged reinforcement learning model, the real-time charging current action in the action space is recorded and arranged in time sequence to obtain the charging current curve; then according to the charging current curve, the current adjustment instruction is executed through the charging device to complete the charging control of the battery.

[0187] Figure 2 The system interface schematic diagram provided by the present application is shown in the figure. In the figure, the man-machine interface of the present system only presents the "data acquisition module" and the "charging control module" in real time; the "negative electrode potential estimation module", the "aging state feature extraction module" and the "reinforcement learning module" belong to the intermediate calculation link of deep / reinforcement learning, and their operation process is not readable to the operator, so they are hidden in the background and not visualized.

[0188] The above merely provides the specific implementation of the present application, any feature disclosed in the specification can be replaced by other equivalent or similar purpose alternative features unless specifically described, and all features disclosed or all steps in the method or process can be combined in any manner except for mutually exclusive features and / or steps.

Claims

1. A method for optimizing fast charging strategies for lithium iron phosphate batteries based on reinforcement learning, characterized in that, Includes the following steps: S1: Real-time voltage data, real-time current data, and real-time temperature data of lithium iron phosphate battery during charging are collected by voltage sensor, current sensor, and temperature sensor, respectively, and preprocessed to obtain preprocessed voltage time series data, current time series data, and temperature time series data. At the same time, the capacity decay rate is calculated using the capacity calibration method; S2: Extract three-factor features based on the preprocessed voltage time series data, current time series data, and temperature time series data, and extract capacity decay features based on the capacity decay rate. Based on the preprocessed voltage time series data, preprocessed temperature time series data, and capacity decay rate, the potential sensitivity characteristics are calculated. Based on the three-factor characteristics, capacity decay characteristics, and potential sensitivity characteristics, the gating fusion characteristics are calculated. Based on the gating fusion characteristics, a potential boundary constraint activation function is designed, and the estimated value of the negative electrode potential is calculated. The potential boundary constraint activation function performs a nonlinear transformation on the gated fusion features based on the Sigmoid function and the fully connected layer, and scales the values ​​according to the difference between the capacity decay rate and the capacity decay health state threshold, so that the negative electrode potential estimate dynamically tightens the output range. S3: Based on the estimated negative electrode potential, calculate the estimated negative electrode potential sequence and extract the negative electrode potential features; then, in conjunction with the capacity decay rate, calculate the aging modulation features, risk sensitivity features, and aging state features in sequence. S4: Construct the state space vector and action space of the reinforcement learning model, and design the aging policy network and comprehensive reward function; S5: The reinforcement learning model is trained to convergence using an aging strategy network and a comprehensive reward function. Based on the converged reinforcement learning model, the real-time charging current actions in the action space are recorded and arranged in chronological order to obtain the charging current curve. Then, based on the charging current curve, the charging device executes the current regulation command to complete the charging control of the battery.

2. The method for optimizing fast charging strategies for lithium iron phosphate batteries based on reinforcement learning as described in claim 1, characterized in that, The specific process of step S1 is as follows: S11: Real-time voltage data, real-time current data, and real-time temperature data of lithium iron phosphate battery during charging are collected by voltage sensor, current sensor, and temperature sensor, respectively, and then digitally converted by analog-to-digital converter to obtain voltage timing data, current timing data, and temperature timing data. S12: For voltage time series data, current time series data, and temperature time series data, high-frequency noise is eliminated by moving average filtering and normalization is performed by minimum-maximum normalization method to obtain preprocessed voltage time series data, preprocessed current time series data, and preprocessed temperature time series data. S13: Based on the historical charge-discharge cycle data stored in the battery management system, the capacity decay rate of the aged lithium iron phosphate battery is calculated using the capacity calibration method.

3. The method for optimizing fast charging strategies for lithium iron phosphate batteries based on reinforcement learning as described in claim 2, characterized in that, The lithium iron phosphate battery is a lithium iron phosphate battery whose aging degree has reached a threshold; the threshold is that the number of charge-discharge cycles is greater than 500, or its usable capacity is less than 80% of the initial nominal capacity, or its DC internal resistance is greater than 120% of the initial internal resistance.

4. The method for optimizing fast charging strategies for lithium iron phosphate batteries based on reinforcement learning as described in claim 1, characterized in that, The specific process of step S2 is as follows: S21: Based on the preprocessed voltage time-series data, preprocessed current time-series data, and preprocessed temperature time-series data, three-factor features are extracted using a bidirectional long short-term memory network. The calculation method is as follows: ， in, These are the three features at time t, where t is the time index. It is a bidirectional long short-term memory network. For splicing operations, This represents the preprocessed voltage time series data at time t. This represents the preprocessed current time series data at time t. The temperature time series data at time t is the preprocessed data. S22: Based on the capacity decay rate, the capacity decay feature is extracted through the embedding layer. The calculation method is as follows: ， in, This is a characteristic of capacity decay. For embedding layer, Capacity decay rate; S23: Based on the preprocessed voltage time-series data, preprocessed temperature time-series data, and capacity decay rate, calculate the potential sensitivity characteristics. The calculation method is as follows: ， ， ， in, It is a voltage change rate sequence. For the central difference operator, This is the preprocessed voltage timing data. It is a temperature change rate sequence. This is the preprocessed temperature time series data. It is characterized by its potential sensitivity. For the Sigmoid function; S24: Based on the three-factor characteristics, capacity decay characteristics, and potential sensitivity characteristics, a gated fusion characteristic is generated through a potential-sensitive gating mechanism. The calculation method is as follows: ， ， in, Let represent the intermediate modulation feature at time t, and ⊙ represent the Hadamard product. For dynamic characteristic gain coefficients, The aging modulation amplitude coefficient, It is the hyperbolic tangent function. It is a fully connected layer. Let be the gated fusion feature at time t; S25: Based on the gating fusion characteristics, design a potential boundary constraint activation function and calculate the estimated value of the negative electrode potential. The calculation method is as follows: ， ， in, Let be the estimated value of the negative electrode potential at time t. This is the lower limit of the negative electrode potential safety. This is the upper limit of the negative electrode potential. The activation function is the potential boundary constraint. This is the threshold for healthy capacity decay. This is the aging-sensitive gain coefficient.

5. The method for optimizing fast charging strategies for lithium iron phosphate batteries based on reinforcement learning as described in claim 1, characterized in that, In step S3, the estimated negative electrode potential values ​​at each time point are sequentially spliced ​​together in chronological order to obtain a negative electrode potential estimation sequence; negative electrode potential features are extracted based on the negative electrode potential estimation sequence; aging modulation features are generated based on the capacity decay rate; the negative electrode potential features and aging modulation features are fused to generate risk-sensitive features; and aging state features are generated based on the risk-sensitive features and the negative electrode potential estimation sequence through an attention mechanism.

6. The method for optimizing fast charging strategies for lithium iron phosphate batteries based on reinforcement learning as described in claim 4, characterized in that, The specific process of step S3 is as follows: S31: Concatenate the estimated negative electrode potential values ​​at each time point in chronological order to obtain the negative electrode potential estimation sequence. ; S32: Based on the negative electrode potential estimation sequence, the negative electrode potential features are extracted through a one-dimensional convolutional layer and a one-dimensional max pooling layer. The calculation method is as follows: ， in, It is a negative electrode potential characteristic. It is a one-dimensional convolutional layer. It is a one-dimensional max pooling layer. For ReLU functions; S32: Based on the capacity decay rate, aging modulation characteristics are generated through an aging-sensitive gating mechanism; the calculation method is as follows: ， in, It is an aging modulation characteristic. It is a multilayer perceptron. To obtain the maximum value; S33: Based on the negative electrode potential characteristics and aging modulation characteristics, a risk-sensitive characteristic is generated, calculated as follows: ， ， in, For feature fusion gating weights, It is characterized by risk sensitivity; S34: Based on risk-sensitive characteristics and negative electrode potential estimation sequences, aging state characteristics are generated through an attention mechanism. The calculation method is as follows: ， in, This is a characteristic of an aging state. This is an attention mechanism.

7. The method for optimizing fast charging strategies for lithium iron phosphate batteries based on reinforcement learning as described in claim 1, characterized in that, In step S4, a state space vector containing the current temperature, current charge amount, current voltage change rate, and estimated negative electrode potential is constructed, as well as an action space containing the range from minimum charging current to maximum charging current. Based on the aging state characteristics, the hidden state of the policy network is modulated through the aging gate weight matrix, aging gate bias vector, offset weight matrix, and offset bias vector, so that the aging state characteristics serve as the conditional input of the policy network. The design incorporates a comprehensive reward function that includes a charging efficiency reward function and a lithium plating risk penalty function. The charging efficiency reward function is calculated based on the current charge amount and its first derivative, while the lithium plating risk penalty function is calculated based on the estimated negative electrode potential, the current charge amount, and the current temperature. Furthermore, the comprehensive reward function is calculated by combining the current voltage change rate and the current temperature.

8. The method for optimizing fast charging strategies for lithium iron phosphate batteries based on reinforcement learning as described in claim 6, characterized in that, The specific process of step S4 is as follows: S41: Construct the state space vector and action space of the reinforcement learning model, calculated as follows: ， ， in, It is a state-space vector. The current temperature. Current charging level Let A be the current voltage change rate, A be the operating space, and a be the current charging current operating value. Minimum charging current, This is the maximum charging current; S42: Based on the characteristics of the aging state, design an aging policy network using a reinforcement learning model to calculate the current charging current. The calculation method is as follows: ， ， ， in, This represents the hidden state of the k-th layer of the aging strategy network, where k is the layer index. K represents the total number of layers in the aging strategy network. Let be the optimized weight matrix of the k-th layer of the aging strategy network. For aging strategy network The hidden state of the layer Let be the optimized bias vector of the k-th layer of the aging strategy network. This represents the hidden state of the final layer of the aging strategy network. For aging strategy network The hidden state of the layer For aging gate weight matrix, This is the aging gate bias vector. This is the offset weight matrix. This is the offset vector. This is the final optimized weight matrix for the Kth layer of the aging strategy network. This is the final optimized bias vector for the Kth layer of the aging strategy network; S43: Design a comprehensive reward function for the reinforcement learning model based on the state space vector.

9. The method for optimizing fast charging strategies for lithium iron phosphate batteries based on reinforcement learning as described in claim 8, characterized in that, The specific process of step S43 is as follows: S431: Calculate the charging efficiency reward function based on the current charging amount. The calculation method is as follows: ， in, For charging efficiency reward function, The first derivative of the current charge amount. It is a natural constant; S432: Calculate the lithium plating risk penalty function based on the estimated negative electrode potential, current charge level, and current temperature. The calculation method is as follows: ， ， in, For the lithium plating risk penalty function, It is the natural logarithm function. This is a risk penalty item; S433: Calculate the comprehensive reward function based on the charging efficiency reward function, lithium plating risk penalty function, current voltage change rate, and current temperature. The calculation method is as follows: ， in, For the comprehensive reward function, To take the absolute value.

10. The system used in the reinforcement learning-based fast charging strategy optimization method for lithium iron phosphate batteries as described in any one of claims 1-9, characterized in that, The system includes: Data acquisition module: It collects real-time voltage, current and temperature data of aged lithium iron phosphate batteries during the charging process through voltage, current and temperature sensors, and performs preprocessing to obtain preprocessed voltage, current and temperature time-series data; and calculates the capacity decay rate. Negative electrode potential estimation module: Based on the preprocessed voltage, current, temperature time series data and capacity decay rate, the three-factor features and capacity decay features are extracted respectively, and then the potential sensitivity features and gated fusion features are calculated in sequence; based on the gated fusion features, the potential boundary constraint activation function is designed, and the estimated value of the negative electrode potential is calculated. Aging state feature extraction module: Based on the negative electrode potential estimate, calculate the negative electrode potential estimate sequence and extract the negative electrode potential features; then, combined with the capacity decay rate, calculate the aging modulation features, risk sensitivity features and aging state features in sequence. Reinforcement learning module: Construct the state space vector and action space of the reinforcement learning model, design the aging policy network and the comprehensive reward function; then train the reinforcement learning model to convergence using the aging policy network and the comprehensive reward function. Charging control module: Based on the converged reinforcement learning model, it records the real-time charging current actions in the action space and arranges them in chronological order to obtain the charging current curve; then, based on the charging current curve, it executes current adjustment commands through the charging equipment to complete the charging control of the battery.

Images