Electric scooter intelligent charging management method, device, equipment and storage medium

Abstract

The present application relates to the field of charging management technology, and discloses an intelligent charging management method, device, equipment and storage medium for electric scooters. The method includes: collecting parameters and initializing settings for a single-phase bidirectional converter and a permanent magnet synchronous motor in an electric scooter, obtaining basic system parameters and communication data, and obtaining a network authentication identifier and charging station operation parameters. According to these parameters, dynamic programming of the charging strategy is performed to generate a segmented charging control curve. The input is input into an adaptive predictive controller to achieve real-time adjustment of the charging process, and to obtain charging state data and power compensation parameters. A dual neural network model structure is used to obtain a battery safety feature matrix and a fault probability vector. A three-dimensional charging state space model is constructed to optimize the power compensation parameters, and the deep reinforcement learning network is input for iterative optimization to generate the charging control parameters for the next cycle, thereby improving the charging efficiency and power quality of the electric scooter.

Description

Technical Field

[0001] The present application relates to the field of charging management technology, and in particular to an electric scooter intelligent charging management method, device, equipment and storage medium. Background Art

[0002] Traditional electric scooter charging management systems have problems such as low charging efficiency, short battery life, and poor safety performance. Existing charging management methods mainly use constant current and constant voltage methods, which cannot dynamically adjust the charging strategy according to the battery status and usage environment, resulting in excessive temperature rise and serious polarization during the charging process.

[0003] At the same time, the existing charging system lacks intelligent fault diagnosis and early warning mechanisms, and cannot effectively identify and prevent safety hazards during charging, increasing the risk of battery overcharging, over-discharging and temperature runaway. In addition, unreasonable allocation of charging station resources and poor user charging experience also restrict the promotion and application of electric scooters. Summary of the invention

[0004] The present application provides an electric scooter intelligent charging management method, device, equipment and storage medium, thereby improving the charging efficiency and power quality of the electric scooter.

[0005] The first aspect of the present application provides an electric scooter intelligent charging management method, the electric scooter intelligent charging management method comprising:

[0006] Perform hardware parameter collection and system initialization configuration on the single-phase bidirectional converter and permanent magnet synchronous motor in the electric scooter to obtain basic parameters of the charging system and communication interface data;

[0007] Perform intelligent charging network authentication and encryption channel establishment based on the charging system basic parameters and the communication interface data to obtain a network authentication identifier and charging station parameters;

[0008] Performing charging strategy dynamic programming and interval threshold calculation according to the network authentication identifier and the charging station parameters to obtain a segmented charging control curve;

[0009] The segmented charging control curve is input into an adaptive prediction controller to adjust the charging process in real time, thereby obtaining charging state monitoring data and power compensation parameters;

[0010] Inputting the charging state monitoring data into a first neural network model and a second neural network model respectively, wherein the first neural network model performs multi-dimensional feature extraction of battery temperature, voltage, and current, and the second neural network model performs battery internal resistance and polarization state analysis to obtain a battery safety feature matrix and a fault probability vector;

[0011] A three-dimensional charging state space model is constructed based on the battery safety feature matrix and the fault probability vector, and the power compensation parameters are optimized online by recursive least squares method. The calculation results are input into a deep reinforcement learning network for iterative optimization of the charging strategy to obtain the charging control parameters for the next cycle.

[0012] A second aspect of the present application provides an intelligent charging management device for an electric scooter, the intelligent charging management device for an electric scooter comprising:

[0013] The initialization module is used to collect hardware parameters and perform system initialization configuration on the single-phase bidirectional converter and permanent magnet synchronous motor in the electric scooter, and obtain basic parameters of the charging system and communication interface data;

[0014] An establishment module, used to perform intelligent charging network authentication and encryption channel establishment based on the charging system basic parameters and the communication interface data, and obtain a network authentication identifier and charging station parameters;

[0015] A calculation module, used to perform dynamic programming of charging strategy and interval threshold calculation according to the network authentication identifier and the charging station parameters to obtain a segmented charging control curve;

[0016] An adjustment module is used to input the segmented charging control curve into an adaptive prediction controller to adjust the charging process in real time, so as to obtain charging state monitoring data and power compensation parameters;

[0017] An analysis module, used to input the charging state monitoring data into a first neural network model and a second neural network model respectively, wherein the first neural network model performs multi-dimensional feature extraction of battery temperature, voltage, and current, and the second neural network model performs battery internal resistance and polarization state analysis to obtain a battery safety feature matrix and a fault probability vector;

[0018] The optimization module is used to construct a three-dimensional charging state space model based on the battery safety feature matrix and the fault probability vector, perform online optimization calculation on the power compensation parameters by recursive least squares method, and input the calculation results into a deep reinforcement learning network for iterative optimization of the charging strategy to obtain the charging control parameters for the next cycle.

[0019] The third aspect of the present application provides an electronic device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the electronic device executes the above-mentioned electric scooter intelligent charging management method.

[0020] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned electric scooter intelligent charging management method.

[0021] Compared with the prior art, the present application has the following beneficial effects: through the integrated design of a single-phase bidirectional converter and a permanent magnet synchronous motor, the hardware cost is reduced, and at the same time, the permanent magnet synchronous motor is used as a charging loop filter to improve the charging efficiency and power quality; a dual neural network model is used for battery status monitoring and fault diagnosis, and combined with a three-dimensional charging state space model, real-time evaluation of the battery safety status and accurate early warning of faults are achieved; charging strategy optimization based on a deep reinforcement learning network and dynamic adjustment of charging parameters significantly reduce the battery temperature rise and extend the battery life; through the authentication mechanism and encryption channel of the intelligent charging network, the data security of the charging process is guaranteed and the system reliability is improved; by using the recursive least squares method and adaptive predictive control algorithm, precise control and dynamic optimization of the charging process are achieved, and the charging efficiency is improved; based on dynamic programming, the charging strategy is formulated, combined with electricity price information and user needs, the charging time schedule is optimized and the charging cost is reduced; through segmented charging control and multiple protection mechanisms, the occurrence of overcharging, over-discharging and other faults is effectively prevented, and the charging safety is guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0023] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0024] Figure 1 is a flow chart of an electric scooter intelligent charging management method provided by an embodiment of the present invention;

[0025] Figure 2 is a schematic block diagram of the structure of an intelligent charging management device for an electric scooter provided in an embodiment of the present invention;

[0026] Figure 3It is a schematic block diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0029] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0030] It should be further understood that the term "and / or" used in the specification and appended claims of this application refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations. Figure 1 , an embodiment of the electric scooter intelligent charging management method in the embodiment of the present application includes:

[0031] Step 100: collect hardware parameters and perform system initialization configuration on the single-phase bidirectional converter and the permanent magnet synchronous motor in the electric scooter to obtain basic parameters of the charging system and communication interface data;

[0032] It is understandable that the execution subject of the present application can be an electric scooter intelligent charging management device, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0033] Specifically, the switch tube drive signal configuration and dead time parameter setting of the single-phase bidirectional converter are performed to ensure that the switching operation of the switch tube can be carried out efficiently without overlapping, and obtain reasonable PWM reference waveform and switch timing data. Based on the PWM reference waveform and switch timing data, the parameters of the inductance filter of the permanent magnet synchronous motor charging circuit are set, and the inductance filter coefficient and current ripple ratio are calculated. The design of the filter helps to reduce the current ripple and stabilize the charging current, and the selection of the inductance filter parameters directly affects the current characteristics and overall operating efficiency of the charging circuit. The charging circuit stability analysis and compensation calculation are performed on the inductance filter coefficient and the current ripple ratio to ensure that the entire charging system maintains a stable working state under different load conditions. By calculating the stability compensation vector of the system in a targeted manner, the unstable factors existing in the charging circuit are pre-compensated to ensure that it can show excellent dynamic response under various working conditions. The system stability compensation vector is substituted into the multiple feedback control equation, and the charging circuit is modeled and simulated and analyzed to obtain the response characteristic data of the charging circuit, reflecting the stability and dynamic characteristics of the charging circuit. Based on the response characteristic data of the charging circuit, a multi-level protection control strategy is formulated for the charging system, and the fault judgment criteria are analyzed to create a fault protection control matrix, so that the faults that occur during the charging process can be quickly responded to and processed according to the actual working conditions to ensure the safe operation of the battery and related systems. On the basis of the fault protection control matrix, dynamic threshold division and gradient grading processing are performed to obtain a hierarchical protection threshold group, which effectively manages faults of different degrees in a hierarchical manner, so that corresponding protection measures can be taken according to the severity of the fault. The hierarchical protection mechanism helps to prevent over-protection caused by minor faults, while ensuring that the charging process is quickly cut off when a major fault occurs to ensure safety. On the basis of the hierarchical protection threshold group, a fault diagnosis and processing mechanism is established, and the relevant protection parameters are calibrated in combination with the battery charging safety characteristic curve to obtain standardized protection control data. The battery charging safety characteristic curve includes the changing trends of key parameters such as temperature, voltage, and current. By combining these characteristic curves, the protection parameters are accurately calibrated to adapt to the actual working state of the battery. Improve the response speed of the system, effectively reduce the probability of false alarms and missed alarms, and improve the reliability and safety of the charging system. According to the specifications of the communication protocol, the standardized protection control data is converted and encoded to ensure that it meets the requirements of the communication interface and realizes the digital representation of the basic parameters of the charging system. The basic parameters of the charging system are encrypted and encapsulated in communication frames so that they can be transmitted between the charging management system and the charging station in the form of standardized data frames. The encapsulation of the communication interface data enables the entire charging system to achieve real-time data exchange and monitoring.

[0034] Step 200: Perform intelligent charging network authentication and establish an encrypted channel based on charging system basic parameters and communication interface data to obtain a network authentication identifier and charging station parameters;

[0035] Specifically, the battery identification code and system hardware model in the basic parameters of the charging system are extracted and spliced. By extracting the battery identification code and the system hardware model, they are combined to form a device unique identification code, which can ensure that each electric scooter system has a unique identity in the charging network. The device unique identification code is XORed with the encrypted frame in the communication interface data and grouped to obtain an identity authentication data packet. Through the XOR operation, the preliminary encryption of the device unique identification code is achieved, effectively preventing unauthorized devices from using identities to participate in the smart charging network. The grouping process enables the identity authentication data packet to be efficiently transmitted in the network, and at the same time, the risk of data packet leakage can be reduced during data transmission. Based on the identity authentication data packet, a secure handshake connection is established with the smart charging station, and a two-way identity verification process is performed. The charging station and the device mutually verify each other's identity to ensure that both are legal and trusted devices, and obtain an encrypted session identifier. The validity period is set and the authority is allocated for the encrypted session identifier to ensure that the usage time limit of each session is clear and the authority level is reasonable. This setting helps prevent the session from being maliciously abused or extended, thereby posing a potential security threat to the charging system. Combined with the IoT communication protocol, the data frame is reorganized to obtain the communication authentication key, and the communication authentication key is embedded in the standard data frame header. The protocol handshake and authority verification are performed with the charging station server to obtain the network authentication identifier. The real-time data of the charging station is parsed based on the network authentication identifier. The network authentication identifier enables the device to obtain real-time electricity price data and power capacity data of the charging station from the charging station. By parsing these data, a resource status matrix is ​​generated. The resource status matrix is ​​a comprehensive reflection of the current resource status of the charging station, including key data such as real-time electricity price information and available power capacity. The charging time window is divided and the power limit is calculated based on the resource status matrix. The division of the charging time window is based on the fluctuation of electricity prices and the dynamic changes of the available power of the charging station, ensuring that the charging process can avoid the peak period of electricity prices as much as possible to reduce the charging cost and avoid excessive burden on the power resources of the charging station. In this process, the charging system dynamically adjusts the charging plan according to the changes in real-time electricity price information and power capacity to maximize the economy and sustainability of charging. By calculating the power limit, the maximum charging power allowed for the electric scooter in different charging time periods is clarified, thereby effectively avoiding excessive pressure on the load of the charging station and ensuring that the charging station can maintain normal power supply capacity when multiple devices are charging at the same time. Get the charging station parameters.

[0036] Step 300: Perform dynamic programming of charging strategy and interval threshold calculation according to the network authentication identifier and charging station parameters to obtain a segmented charging control curve;

[0037] It should be noted that the user level data in the network authentication identifier and the power capacity data in the charging station parameters are prioritized and matched. By combining the user level with the current resources of the charging station, a scheduling authorization matrix is ​​generated, which reflects the charging priority enjoyed by different users in different charging station environments. Through this step, when the power resources of the charging station are limited, the charging power is allocated to high-level users first, ensuring that the resources of the entire system are reasonably and effectively utilized. The scheduling authorization matrix is ​​substituted into the dynamic programming model, and the optimal path is searched according to the electricity price period and the charging demand to obtain the timing optimization sequence. By considering the electricity price period, the dynamic programming model can adjust the charging strategy according to the fluctuation of electricity prices in different time periods to minimize the cost. After obtaining the timing optimization sequence, the charging time is estimated and the power is reasonably allocated in combination with the remaining capacity of the battery to generate the initial charging power curve. Based on the initial charging power curve, segmented fitting and threshold feature extraction are performed to divide the boundaries of the three main charging intervals of constant current, constant voltage and trickle flow in the charging process, and obtain the charging partition parameters. The constant current interval is used to quickly replenish the battery capacity, the constant voltage interval is used to stabilize the voltage to prevent the battery from overcharging, and the trickle current interval is used to supplement the current slightly when the battery is close to being fully charged, so as to maximize the safety and life of the battery. The state transfer matrix is ​​constructed according to the charging partition parameters, and the optimal control strategy is solved for each charging interval. The state transfer matrix is ​​used to describe the state changes in each stage of the charging process, and a smooth transition is achieved between different charging intervals to ensure that there is no obvious power fluctuation during the charging process. According to the solution results of the optimal control strategy, the partition control vectors of each charging interval are obtained. These vectors are descriptions of the optimal power and current parameters under different charging states. On this basis, the partition control vectors are dynamically bounded and smoothly transitioned to eliminate mutations between different intervals, making the charging trajectory data more continuous and stable. Based on the charging trajectory data, the charging process is simulated and safety verified to ensure that the generated charging strategy is safe and effective in practical applications. Through the simulation process, potential risks are discovered and adjusted accordingly, and the entire charging process is safety evaluated to ensure that dangerous situations such as overheating and overvoltage will not occur. After simulation and verification, optimized control sequences are obtained, which can maximize charging efficiency and ensure safety. The optimized control sequence is reconstructed in the form of a piecewise function, and appropriate transition constraints are added between each segment to obtain the final piecewise charging control curve.

[0038] Step 400: Input the segmented charging control curve into the adaptive prediction controller to adjust the charging process in real time, and obtain charging state monitoring data and power compensation parameters;

[0039] Specifically, the segmented charging control curve is sampled and data segmented at equal time intervals. Through equal time interval sampling, the charging process is finely discretized in time, so that the control system can monitor and adjust the charging state in a timely manner. At the same time, combined with the dynamic characteristics of the battery during the charging process, a suitable prediction step size is set to ensure that the predictive control can better adapt to the changes in the charging process. After this step, a benchmark sequence for predictive control is obtained. The benchmark sequence is the basis of the entire adaptive predictive control and ensures the controllability and predictability of the charging process in the time domain. The predictive control benchmark sequence is substituted into the linear quadratic optimal control equation, and the control increment matrix in the prediction time domain is obtained by recursive solution. The goal of linear quadratic optimal control is to find the optimal control increment matrix under the premise of minimizing control error and controlling energy consumption, so that the charging process can maintain the best state in the entire prediction time domain. After obtaining the control increment matrix, the battery charging state is predicted according to these increments, and the constraint boundaries in the charging process are calculated to obtain dynamic constraint parameters. The rolling prediction can timely update the deviations that occur during the charging process, and the calculation of dynamic constraint parameters helps to ensure that the entire charging process meets safety constraints in various state variables (such as voltage, current, temperature, etc.). The difference between the dynamic constraint parameters and the data in the actual charging process is calculated to detect whether the current charging process deviates from the predicted control path. In order to adjust the output of the controller in real time, the feedback gain of the controller is updated by the gradient descent method to obtain the adaptive control quantity. The gradient descent method enables the system to converge to the optimal control state faster by continuously adjusting the direction and amplitude of the feedback gain, ensuring that the charging process always remains on the predetermined trajectory. The role of the adaptive control quantity is to adjust the real-time response of the system so that the controller can dynamically respond to changes in the battery state and improve the response speed and stability of the charging process. Based on the adaptive control quantity, the corresponding PWM modulation signal is generated for real-time adjustment of the switch tube drive signal of the single-phase bidirectional converter. The PWM modulation signal drives the switch tube in real time through precise modulation frequency and duty cycle control, so that the current and voltage can change according to the predetermined trajectory, thereby maintaining the stability of the charging process. In this process, the charging state data is continuously collected and characterized according to the three main dimensions of voltage, current and temperature to better understand the current charging state. Through this characteristic decomposition, the parameters that have the greatest impact on safety and efficiency in the charging process can be effectively extracted to obtain charging state monitoring data. The power fluctuation analysis is performed on the charging status monitoring data, and the harmonic compensation value is calculated in combination with the filtering characteristics of the permanent magnet synchronous motor. The compensation value of the harmonic components generated during the charging process is calculated to obtain the power regulation coefficient. Based on the power regulation coefficient, the power of the entire charging process is closed-loop compensated to improve the stability of charging.In order to achieve fine power regulation, a proportional integral controller (PI controller) is used to dynamically adjust the charging output power. The dual regulation mechanism of proportion and integral is used to eliminate the steady-state error in the system and improve the dynamic response speed. Through the closed-loop compensation process, the power compensation parameters are finally obtained.

[0040] Step 500: input charging state monitoring data into a first neural network model and a second neural network model respectively, wherein the first neural network model performs multi-dimensional feature extraction of battery temperature, voltage, and current, and the second neural network model performs battery internal resistance and polarization state analysis to obtain a battery safety feature matrix and a fault probability vector;

[0041] Specifically, the charging state monitoring data is sampled in segments according to the time series, and the data is divided into several relatively independent parts to ensure that the model can capture the state change characteristics in different time periods. The sampled voltage, current and temperature data are normalized to keep each feature in the same numerical range, reduce the scale difference between different features, thereby improving the convergence speed and accuracy of the neural network model during training, and obtaining the feature input matrix for model input. The feature input matrix is ​​input into the temporal convolution layer of the first neural network model for processing. The design of the temporal convolution layer includes three parallel one-dimensional convolution sublayers. Each one-dimensional convolution sublayer uses a different convolution kernel size, and convolution kernels of different scales are used to capture features of different granularities in the charging process. After processing each convolution kernel, the characteristic information of the battery in different time windows can be extracted. This information is processed nonlinearly by the ReLU activation function to enhance the model's ability to capture complex patterns. At the same time, batch normalization is used to standardize the data to reduce the problem of gradient disappearance or explosion, and obtain the initial feature map. Based on the initial feature map, the long short-term memory (LSTM) layer is processed. The LSTM layer contains 64 memory units, each of which consists of an input gate, a forget gate, and an output gate, which are used to selectively memorize time series features. With this structure, the model can effectively learn long-term dependencies in time series, such as the temperature and current change trends of the battery throughout the charging process. Through the processing of LSTM, the time series correlation features are obtained to capture the dynamic changes of the battery at different time stages. The attention mechanism layer is used to weight the time series correlation features. The attention mechanism layer adopts a multi-head self-attention structure, which contains 8 attention heads. Each attention head processes the time series features independently and analyzes the charging data from multiple angles. By weighting the features at different time series positions, the model's attention to the key moments in the charging process is enhanced, and a more representative battery safety feature matrix is ​​generated. The battery safety feature matrix contains multi-dimensional features in the charging process. At the same time, the feature input matrix is ​​input into the fully connected layer of the second neural network model. The fully connected layer contains three hidden layers, with 128, 64, and 32 nodes in each layer, respectively. The feature information is gradually concentrated by reducing the number of nodes layer by layer. The output of each layer is processed by the Sigmoid activation function to ensure that the output is between 0 and 1, so as to effectively capture the changes in the battery's internal resistance. After this process, the internal resistance feature vector is obtained, which contains relevant information about the battery's internal resistance and can reflect the electrochemical changes inside the battery. The internal resistance feature vector is input into the polarization state evaluation subnetwork of the second neural network model. The polarization state evaluation subnetwork contains two layers of graph convolution layers. Each graph convolution layer uses Chebyshev polynomials for feature propagation. This method can efficiently propagate information in the graph structure, allowing the model to learn more features of the battery polarization state from adjacent nodes.Through the operation of graph convolution, the local features of the battery internal resistance are fully propagated and integrated, and a concentrated polarization feature vector is obtained after the maximum pooling operation, which contains the key information of the battery polarization state. The polarization feature vector is input into the fault diagnosis subnetwork of the second neural network model, which contains a three-layer feedforward neural network for predicting the failure probability of the battery. Each layer of the neural network uses Dropout regularization to prevent overfitting during training, and batch normalization is used to accelerate convergence. After being processed by the three-layer neural network, the output is finally normalized by the Softmax function to obtain a fault probability vector, which reflects the probability of failure of the battery under the current charging state.

[0042] Step 600: construct a three-dimensional charging state space model based on the battery safety feature matrix and the fault probability vector, perform online optimization calculation on the power compensation parameters by recursive least squares method, and input the calculation results into the deep reinforcement learning network for iterative optimization of the charging strategy to obtain the charging control parameters for the next cycle.

[0043] Specifically, the battery safety feature matrix is ​​constructed in state space according to the temperature state, voltage state and current state. By dividing these feature data into different state dimensions, a state space containing multi-dimensional information is formed, so that the charging strategy can take into account the various conditions of the battery at every moment. At the same time, the fault probability vector is mapped to the interval of [0,1] as the state weight in the state space to characterize the fault risk of each state and form a state space basic model. The state space basic model is constructed in a hierarchical grid, divided into 5×5 grids in the temperature dimension, 8×8 grids in the voltage dimension, and 6×6 grids in the current dimension. Each possible state is described in detail in different state spaces, and the state transition probability between each grid point is calculated in combination with the power compensation parameter to obtain a Markov decision process model. In this way, the system describes the transition relationship between each state in the charging process through a Markov decision process. The Markov decision process model is input into the recursive least squares algorithm for online optimization calculation. In the application of recursive least squares, the forgetting factor and the initial covariance matrix are set. Based on the state error and control cost, a quadratic objective function is constructed to obtain the online optimization objective function. The quadratic objective function combines the control error and the control cost to minimize the deviation and energy consumption during the charging process. Based on the online optimization objective function, the policy network of the deep reinforcement learning network is constructed. In the policy network, the first layer contains 256 nodes and uses the ReLU activation function to achieve nonlinear mapping of complex features; the second layer contains 128 nodes and uses the PReLU activation function to enhance the expression ability of the model and adapt to the diversity of inputs; the third layer contains 64 nodes and uses the tanh activation function to achieve smooth processing of features at the output. In order to improve the convergence performance of the network, the input data is normalized by Batch Normalization to ensure that the input data distribution of each layer remains stable during the training process, and the distribution of the action space is obtained. In order to evaluate the action space distribution of the policy network, the architecture of the dual Q network is adopted, in which the main Q network and the target Q network both adopt a three-layer structure with 512, 256 and 128 nodes respectively. In order to improve the accuracy of policy evaluation, the Dueling DQN architecture is used to separate the state value and advantage function, so as to better estimate the value of each action in different states and obtain more accurate policy evaluation results. Based on the policy evaluation results, the temporal difference error is calculated, and the proximal policy optimization (PPO) algorithm is used to perform gradient updates on the network parameters to obtain the optimized policy parameters. The PPO algorithm ensures that the policy remains stable during the continuous optimization process by limiting the amplitude of each update, and effectively prevents performance degradation caused by excessive policy updates.Substitute the optimized policy parameters into the target Q network for sliding average update to ensure that the parameters of the target Q network can smoothly follow the update of the main Q network. At the same time, the priority experience replay mechanism is used to sample from the experience pool, learn the best control strategy from the historical charging experience, and obtain the updated amount of the charging strategy. Through the priority experience replay, the key historical experience is used to enhance the adaptability of the model to complex charging states. Based on the obtained charging strategy update, the charging current is controlled in sections. In the constant current charging stage, the current is set to 1C to ensure that the battery can be quickly charged during the initial charging; in the constant voltage charging stage, the voltage is limited to 4.2V to prevent battery damage caused by overcharging; and in the final trickle charging stage, the cutoff current is set to 0.05C to ensure that the battery can be safely and gradually stopped charging when it is close to full. By combining the action noise suppression mechanism, the fluctuations in the control process are effectively reduced, ensuring a smoother charging process, and obtaining the charging control parameters for the next cycle.

[0044] In the embodiment of the present application, the hardware cost is reduced through the integrated design of the single-phase bidirectional converter and the permanent magnet synchronous motor, and the permanent magnet synchronous motor is used as a charging loop filter to improve the charging efficiency and power quality; a dual neural network model is used for battery status monitoring and fault diagnosis, and combined with a three-dimensional charging state space model, real-time evaluation of the battery safety status and accurate warning of faults are achieved; charging strategy optimization based on a deep reinforcement learning network dynamically adjusts charging parameters, significantly reduces battery temperature rise, and extends battery life; through the authentication mechanism and encryption channel of the intelligent charging network, data security of the charging process is guaranteed and system reliability is improved; recursive least squares method and adaptive predictive control algorithm are used to achieve precise control and dynamic optimization of the charging process and improve charging efficiency; charging strategy formulation based on dynamic programming, combined with electricity price information and user needs, optimizes charging time scheduling and reduces charging costs; through segmented charging control and multiple protection mechanisms, the occurrence of overcharging, over-discharging and other faults is effectively prevented, ensuring charging safety.

[0045] In a specific embodiment, the process of executing step 100 may specifically include the following steps:

[0046] The switch drive signal configuration and dead time parameter setting of the single-phase bidirectional converter are performed to obtain the PWM reference waveform and switch timing data. Based on the PWM reference waveform and switch timing data, the inductance filter parameters of the charging circuit of the permanent magnet synchronous motor are set to obtain the inductance filter coefficient and the current ripple ratio.

[0047] Perform charging circuit stability analysis and compensation calculation on the inductance filter coefficient and current ripple ratio to obtain the system stability compensation vector, and substitute the system stability compensation vector into the multiple feedback control equation to perform charging circuit modeling and simulation analysis to obtain the charging circuit response characteristic data;

[0048] Based on the charging circuit response characteristic data, the multi-level protection control strategy and fault judgment criteria are analyzed to create a fault protection control matrix, and the fault protection control matrix is ​​dynamically divided into thresholds and gradient graded to obtain a graded protection threshold group;

[0049] Establish a fault diagnosis and processing mechanism based on the hierarchical protection threshold group, calibrate the protection parameters in combination with the battery charging safety characteristic curve, and obtain standardized protection control data;

[0050] The standardized protection control data is converted and encoded in data format according to the communication protocol specification to obtain the basic parameters of the charging system, and the basic parameters of the charging system are encrypted and encapsulated in communication frames to obtain communication interface data.

[0051] Specifically, the switch tube drive signal is configured and the dead time parameters are set for the single-phase bidirectional converter to generate a PWM reference waveform and switch timing data to ensure safe and efficient operation of the converter. The switch tube of the single-phase bidirectional converter is composed of MOSFET or IGBT, and an accurate drive signal is required to control its switching state. When configuring the switch tube drive signal, the setting of the dead time is crucial. Its main function is to avoid the danger of short circuit caused by the simultaneous conduction of the switch tubes of the upper and lower bridge arms. The dead time is set according to the turn-on and turn-off speed of the switch tube, and the formula is:

[0052] ;

[0053] in, is the dead time, represents the turn-on delay time of the switch tube, and Indicates the shutdown delay time. Set and generate safe and reliable PWM reference waveform and switch timing data for driving single-phase bidirectional converter. Based on the PWM reference waveform and switch timing data, set the inductance filter parameters for the charging circuit of the permanent magnet synchronous motor. The function of the filter inductor is to reduce the current ripple, ensure the current stability when the battery is charged, and reduce the impact of high-frequency noise on the battery life. The setting of the inductance filter parameters is calculated by the following formula:

[0054] ;

[0055] in, Indicates the size of the filter inductor. is the DC bus voltage, is the duty cycle, is the switching frequency of PWM, is the allowable current ripple size. Through this formula, the appropriate inductance value is calculated to achieve the purpose of reducing current ripple. The current ripple ratio is defined as the ratio of ripple current to average current, expressed as:

[0056] ;

[0057] in, is the average current. The reasonable setting of the inductance filter coefficient and the current ripple ratio has an important impact on the efficiency of the charging system and the life of the battery. The charging loop stability analysis is performed on the inductance filter coefficient and the current ripple ratio, and compensation calculations are performed. By using the Bode plot and Nyquist plot in control theory, the gain and phase margin of the charging loop are evaluated to analyze the stability of the system. In order to ensure the stability of the system under different load conditions, the inductance filter coefficient and the current ripple ratio are compensated and calculated to obtain the stability compensation vector of the system, which is recorded as The calculation of the stability compensation vector is achieved through the following quadratic optimization objective function:

[0058] ;

[0059] in, is the objective function, represents the error signal, is the weight factor, is the control signal. By minimizing , and obtain the optimal compensation parameters of the system to ensure that the charging process achieves the best balance between dynamic characteristics and stability. Substitute the system stability compensation vector into the multiple feedback control equation, perform charging circuit modeling and simulation analysis, and obtain the response characteristic data of the charging circuit. The modeling process is described by the state space model, and the state equation is expressed as:

[0060] ;

[0061] ;

[0062] in, is the state vector, is the state matrix, is the input matrix, is the output, and They are the output matrix and the direct transfer matrix respectively. By simulating the model, the response characteristic data of the system under different inputs are obtained, especially the dynamic change characteristics of current, voltage and battery temperature. Based on the charging circuit response characteristic data obtained by simulation, the multi-level protection control strategy and fault judgment criteria are analyzed to create a fault protection control matrix. The core of the multi-level protection control strategy is to ensure that the system can respond to abnormal conditions in a timely manner under different working conditions, such as over-temperature, over-voltage or over-current. Fault protection control matrix It is constructed by setting different thresholds for each protection parameter. In order to improve the accuracy and response speed of protection, the fault protection control matrix is ​​dynamically divided into thresholds and gradient graded to obtain a graded protection threshold group. For example, for battery temperature, the temperature range is divided into three areas: normal, warning and dangerous, and each area corresponds to a different protection level. According to the graded protection threshold group, a fault diagnosis and processing mechanism is established. Combined with the battery charging safety characteristic curve, the relevant protection parameters are finely calibrated to ensure the adaptability and accuracy of the protection mechanism. For example, on the battery temperature characteristic curve, the increase in temperature will cause the internal resistance of the battery to increase, thereby affecting the charging efficiency. Therefore, the protection parameters are dynamically adjusted to adapt to this change to obtain standardized protection control data. In order to ensure the reliable transmission of protection control data, the standardized protection control data is converted and encoded according to the communication protocol specification. For example, the protection data is encapsulated according to a specific protocol header, data part and check part, and converted into a format suitable for transmission on the communication network to form the basic parameters of the charging system. The basic parameters are encrypted to prevent eavesdropping or tampering during transmission. The encryption process uses a symmetric encryption algorithm, such as AES, and its encryption process is described as:

[0063] ;

[0064] in, Represents ciphertext, Represents plaintext data, Represents the encryption key The encryption function under the action of encryption is used to ensure the security of data. The encrypted basic parameters of the charging system are encapsulated in the communication frame to form the communication interface data to achieve safe communication with the charging management system. The communication frame includes a frame header, a data part and a check code, where the frame header is used to identify the data type and length, and the check code is used to ensure the integrity of data transmission. The encapsulation of the communication frame is described by the following expression:

[0065] ;

[0066] in, Indicates the encapsulated communication frame. It is the frame header. is the data part, It is a cyclic redundancy check code. Through this encapsulation method, the communication interface data of the charging system is transmitted safely and reliably between the charging station and the electric scooter.

[0067] In a specific embodiment, the process of executing step 200 may specifically include the following steps:

[0068] The battery identification code in the basic parameters of the charging system and the system hardware model are extracted and spliced ​​to obtain the unique device identification code, and the unique device identification code is XORed and grouped with the encrypted frame in the communication interface data to obtain an identity authentication data packet;

[0069] Establish a secure handshake connection with the smart charging station based on the identity authentication data packet, and perform two-way identity verification to obtain an encrypted session identifier;

[0070] Set validity period and assign permissions for the encrypted session identifier, reorganize the data frame in combination with the IoT communication protocol, obtain the communication authentication key, and embed the communication authentication key into the standard data frame header, perform protocol handshake and permission verification with the charging station server, and obtain the network authentication identifier;

[0071] The real-time electricity price data and power capacity data of the charging station are parsed based on the network authentication identifier to obtain the resource status matrix, and the charging time window and power limit calculation are performed on the resource status matrix to obtain the charging station parameters.

[0072] Specifically, the battery identification code and system hardware model in the basic parameters of the charging system are extracted and spliced ​​to generate a unique device identification code. The battery identification code is a code that uniquely identifies the battery. It consists of a set of specific strings or numbers and is used to uniquely identify the characteristics and production information of the battery. The system hardware model uniquely identifies the hardware used in the scooter charging system. and system hardware model splicing to get the unique device identification code , and its splicing process is expressed as:

[0073] ;

[0074] Among them, || represents the concatenation operation of the string. Through concatenation, the unique identification code of the device is ensured to be unique and can uniquely identify each charging device. The unique identification code of the device is XORed and grouped with the encrypted frame in the communication interface data to generate an authentication data packet. Assume that the encrypted frame in the communication interface data is , the unique device identification code and the encrypted frame are combined by XOR operation to improve the security of authentication data. The process is expressed as:

[0075] ;

[0076] in, Represents an authentication packet. Represents the exclusive OR operator. Through the exclusive OR operation, the privacy of the device's unique identification code is effectively protected, and the authentication data is made highly resistant to attacks. The generated authentication data packet is grouped and divided into data blocks of equal length for easy transmission and authentication operations. A secure handshake connection is established with the smart charging station based on the identity authentication data packet. During the handshake connection process, the charging device and the charging station server exchange information to ensure that the identities of both parties are authentic and trustworthy. This process includes public key exchange and symmetric key negotiation to achieve encrypted data transmission. The main steps of the handshake connection include the transmission of encrypted identity information, verification of the other party's identity, and generation of symmetric keys. Through two-way identity verification, the identities of the device and the charging station server are verified to be authentic and trustworthy, and an encrypted session identifier is generated after the verification is passed. The session identifier is usually generated by a random number and a pre-shared key negotiated by both parties to ensure uniqueness and security. The generation process is expressed as:

[0077] ;

[0078] in, and are random numbers generated by the device and the server respectively. is the pre-shared key, Represents a hash function, which is used to encrypt the generated session identifier. The validity period is set and permissions are assigned to the encrypted session identifier to prevent the session identifier from being abused for a long time or used without authorization. The session validity period is set based on the timestamp. , used to identify the expiration time of the session identifier. The authority allocation is based on the user level and device authority, giving different operation permissions, such as fast charging, slow charging, etc. Combined with the IoT communication protocol, the data frame is reorganized to obtain the communication authentication key The communication authentication key is generated by combining the encrypted session identifier with the device permission information to construct a new key, which is expressed as:

[0079] ;

[0080] in, It is the permission level information of the device. By combining the session identifier and permission information, a unique authentication key can be generated. The communication authentication key is embedded in the header of the standard data frame for subsequent protocol handshake and permission verification with the charging station server. The encapsulation of the data frame is represented as:

[0081] ;

[0082] in, is the frame header information of the data frame, The data payload content. When the protocol handshake is performed with the charging station server through the data frame, the server verifies the identity and authority of the device according to the authentication key, and generates a network authentication identifier after the verification is passed. , to indicate the legality and operability of the equipment. Based on the obtained network authentication mark, the real-time electricity price data and power capacity data are parsed from the charging station. The electricity price data changes over time, while the power capacity depends on the current load of the charging station. Through the interaction between the network authentication mark and the server, these data are obtained in real time to form a resource status matrix The resource status matrix is ​​expressed as:

[0083] ;

[0084] in, Indicates the currently available power capacity, is the current electricity price, Indicates the current load situation. The resource status matrix is ​​used to understand the operation status of the charging station in real time. The resource status matrix is ​​used to divide the charging time window and calculate the power limit. The charging time window is divided based on the real-time electricity price and load situation, and the charging process is divided into peak period and valley period to minimize the charging cost. For example, if the electricity price is low during a certain period, charging can be carried out during this period to reduce expenses. The power limit is calculated based on the current power capacity. and load , ensuring that the power carrying capacity of the charging station is not exceeded while meeting the charging demand. The calculation formula is:

[0085] ;

[0086] in, The maximum allowed charging power of the device is obtained through this calculation. The charging station parameters are obtained based on the above calculations and are used to control each link of the charging process to ensure the safety, efficiency and cost minimization of the charging process.

[0087] In a specific embodiment, the process of executing step 300 may specifically include the following steps:

[0088] Priority calculation and matching are performed on the user level data in the network authentication identifier and the power capacity data in the charging station parameters to obtain a dispatch authorization matrix;

[0089] Substitute the dispatch authorization matrix into the dynamic programming model, search for the optimal path according to the electricity price period and charging demand, obtain the timing optimization sequence, and estimate the charging time and power allocation for the timing optimization sequence and the remaining battery capacity to obtain the initial charging power curve;

[0090] Based on the initial charging power curve, segmented fitting and threshold feature extraction are performed to divide the charging process into constant current interval, constant voltage interval, and trickle interval to obtain charging partition parameters.

[0091] The state transfer matrix is ​​constructed according to the charging partition parameters, and the optimal control strategy is solved for each charging interval to obtain the partition control vector. The partition control vector is subjected to dynamic boundary constraints and smooth transition processing to obtain the charging trajectory data.

[0092] The charging process is simulated and safety verified based on the charging trajectory data to obtain the optimized control sequence. The optimized control sequence is reconstructed in the form of a piecewise function and transition constraints are added to obtain a piecewise charging control curve.

[0093] Specifically, the user level data in the network authentication identifier and the power capacity data in the charging station parameters are prioritized and matched. The user level data reflects the priority of different users, such as ordinary users, member users or VIP users, etc. These levels are used to allocate different charging permissions. The power capacity data indicates the charging power that can be allocated by the current charging station. and the current power capacity of the charging station Construct a priority scoring function, which is defined as:

[0094] ;

[0095] in, represents the priority score, and is the weight coefficient, which is used to adjust the influence of user level and power capacity in priority scoring. , prioritize different users and the power capacity of charging stations to generate a scheduling authorization matrix , which represents the charging priority of each user in different time periods. Substitute the dispatch authorization matrix into the dynamic programming model and search for the optimal path according to the electricity price period and charging demand. The goal of dynamic programming is to find an optimal charging path that minimizes the charging cost while meeting the charging demand. is a dynamically changing function that reflects the electricity price at different times of the day. For in time The cost function of charging, the charging demand is , then construct the objective function:

[0096] ;

[0097] in, Indicates at time The charging power is to minimize the cost of the entire charging process. , in meeting charging needs At the same time, find the optimal charging power distribution. Through the dynamic programming algorithm, the timing optimization sequence is obtained , which describes the charging power that should be allocated in different time periods. The remaining capacity of the battery Combine, estimate charging time and allocate charging power to get the initial charging power curve The charging time is estimated according to the following formula:

[0098] ;

[0099] in, It is charging time. is the remaining battery capacity, is the average charging power. By combining the timing optimization sequence with the current state of the battery, the initial charging power curve is obtained , describes the charging power in each time period during the charging process. Based on the initial charging power curve, segmented fitting and threshold feature extraction are performed to divide the boundaries of the charging process into constant current interval, constant voltage interval and trickle interval to obtain the charging partition parameters. The constant current interval is used to quickly charge the battery in the initial charging stage, and the charging current remains constant; the constant voltage interval is used to keep the voltage constant and gradually reduce the current after the battery is charged to the set voltage; the trickle interval is used to continue charging with a small current when the battery is close to full charge, so as to prevent overcharging and extend the battery life. The boundaries of the constant current interval, constant voltage interval and trickle interval are divided according to the following formula:

[0100] ;

[0101] in, is the power in the constant current range, is a constant charging current, is the charging voltage, is the power in the constant voltage range is a constant voltage, is the trickle charging current. The charging partition parameters obtained by segmented fitting control the charging process at each stage. According to the charging partition parameters, the state transfer matrix is ​​constructed , which is used to describe the state transition relationship between each charging interval. The state transition matrix represents the transition probability between different charging states, such as the probability of transferring from the constant current interval to the constant voltage interval. , and the probability of transition from the constant pressure range to the trickle flow range Based on the state transfer matrix, the optimal control strategy is solved for each charging interval to obtain the partition control vector , which is used to describe the optimal power and current settings at different charging states. Perform dynamic boundary constraints and smooth transition processing to obtain charging trajectory data . In order to make the transition between different charging intervals smoother, dynamic boundary constraint technology is used to ensure that there are no sudden changes in power and current during the charging process. The charging trajectory is smoothly transitioned through the interpolation algorithm to eliminate the drastic fluctuations caused by interval conversion during the charging process, thereby improving the stability and safety of the charging process. Based on the generated charging trajectory data, the charging process is simulated and safety verified. The simulation process uses a system modeling tool to simulate the charging trajectory to verify whether there are any safety hazards in the entire charging process, such as overtemperature, overcurrent and other problems. The response data obtained through simulation is evaluated in combination with the safety verification standards to ensure that the charging trajectory meets safety requirements. The results of simulation and verification are used to generate an optimized control sequence , these optimized control sequences are verified and adjusted to further optimize the charging strategy. The optimized control sequence is reconstructed in the form of piecewise functions, and transition constraints are added between each segment to obtain the final piecewise charging control curve The segmented charging control curve is the core control signal in the entire charging management process, which is used to guide the charging equipment to charge at different currents and voltages in different time periods. By combining the optimized control sequences in each interval and setting transition conditions between intervals, the charging process is ensured to be smooth and efficient, reducing safety hazards caused by sudden changes in current or voltage.

[0102] In a specific embodiment, the process of executing step 400 may specifically include the following steps:

[0103] The segmented charging control curve is sampled and data segmented at equal time intervals, and the prediction step size is set in combination with the dynamic characteristics of the charging process to obtain a prediction control benchmark sequence;

[0104] Substitute the prediction control benchmark sequence into the linear quadratic optimal control equation for recursive solution to obtain the control increment matrix in the prediction time domain, and perform rolling prediction of the battery charging state and calculation of the constraint boundary of the charging process based on the control increment matrix to obtain the dynamic constraint parameters;

[0105] The difference between the dynamic constraint parameters and the actual charging process data is calculated, and the controller feedback gain is updated based on the gradient descent method to obtain the adaptive control quantity;

[0106] Generate PWM modulation signal according to adaptive control quantity, adjust the switch tube drive signal of single-phase bidirectional converter in real time, obtain charging state sampling data, and decompose the charging state sampling data according to three dimensions of voltage, current and temperature to obtain charging state monitoring data;

[0107] The power fluctuation analysis is performed on the charging status monitoring data, and the harmonic compensation value is calculated in combination with the filtering characteristics of the permanent magnet synchronous motor to obtain the power regulation coefficient. At the same time, the charging power is closed-loop compensated based on the power regulation coefficient, and the output power is dynamically adjusted through the proportional-integral controller to obtain the power compensation parameter.

[0108] Specifically, the segmented charging control curve is sampled and data segmented at equal time intervals, and the charging control curve is discretized at equal time intervals to capture the dynamic changes of each stage in the charging process. Suppose the charging control curve is The sampling time interval is , then a series of discrete control data points are obtained by sampling at equal intervals ,in , , Represents the total number of sampling points. By segmenting these data points, the charging process is divided into several sub-intervals, so that more precise control can be performed in each interval. Combined with the dynamic characteristics of the charging process, the prediction step size is set to obtain the prediction control benchmark sequence The prediction step size is set based on the dynamic changes in the battery's current, voltage, temperature, etc. during the charging process to ensure that the controller can respond to these changes in a timely manner. The prediction step size is usually set to , reflecting the system in the future The prediction range is within time steps. By choosing the appropriate , ensuring that the controller can achieve high accuracy and stability in the face of dynamic charging processes, generating a reference sequence for control . The predictive control benchmark sequence Substitute into the linear quadratic optimal control equation and calculate the control increment matrix in the prediction time domain by recursive solution The goal of linear quadratic optimal control is to minimize the control error and energy consumption, and its cost function is expressed as:

[0109] ;

[0110] in, is the cost function, Indicates the system status. To control the increment, and They are the state weighted matrix and the control weighted matrix, respectively, which are used to adjust the trade-off between state deviation and control input. By minimizing the cost function, the control increment matrix is ​​obtained , which describes the optimal control increment at each time step in the prediction time domain. Based on this control increment matrix, the battery charging state is predicted in a rolling manner, and the constraint boundaries in the charging process are calculated in combination with the physical characteristics and safety requirements of the battery to obtain the dynamic constraint parameters These parameters are used to limit the range of charging current and voltage to ensure the safety of the charging process. Real-time data of the actual charging process A difference calculation is performed to evaluate the deviation between the current control and the target control. The result of the difference calculation It is expressed as:

[0111] ;

[0112] The feedback gain of the controller is updated based on the gradient descent method to obtain the adaptive control quantity The gradient descent method adjusts the gradient of the cost function in the negative direction to minimize the error. The update rule is expressed as:

[0113] ;

[0114] in, represents the feedback gain at the previous moment, is the learning rate, which is used to control the step size of the adjustment. , so that the controller can dynamically adapt to the changes in battery status during charging and improve the accuracy of control. Generate PWM modulation signal for real-time adjustment of the switch drive signal of the single-phase bidirectional converter. The generation of PWM modulation signal is based on the comparison between the reference signal and the feedback signal to obtain the duty cycle , and its calculation formula is:

[0115] ;

[0116] in, is the reference voltage, is the input voltage. By adjusting the duty cycle, the on and off time of the switch tube is controlled in real time, thereby adjusting the charging current and voltage to obtain the sampling data of the charging state. The sampling data includes voltage , Current and temperature The charging status sampling data is characterized by voltage, current, and temperature to obtain the charging status monitoring data. The charging status monitoring data is used to analyze the power fluctuation during the charging process. Assume that the charging power is the product of voltage and current, that is:

[0117] ;

[0118] Through Perform Fourier transform to obtain the frequency domain representation of power, which is used to identify the harmonic components in the charging process. Combined with the filtering characteristics of the permanent magnet synchronous motor, the harmonic compensation value is calculated. , in order to reduce the impact of harmonics on the charging system. The harmonic compensation value is calculated by the following formula:

[0119] ;

[0120] in, Indicates The amplitude of the harmonics of each order is calculated by compensating for the harmonics of each order and reducing the total harmonic distortion rate to obtain the power regulation coefficient. The charging power is compensated in a closed loop based on the power regulation coefficient, and the output power is dynamically adjusted through the proportional integral (PI) controller to improve the stability of the charging process. The control law of the PI controller is expressed as:

[0121] ;

[0122] in, is the proportional gain, is the integral gain, The control error is the control error. Proportional control can quickly respond to changes in charging power, while integral control can eliminate steady-state errors to ensure that the charging power reaches the set value. The control output after closed-loop compensation is the power compensation parameter. .

[0123] In a specific embodiment, the process of executing step 500 may specifically include the following steps:

[0124] The charging status monitoring data is sampled in sections according to the time series, and the voltage, current and temperature are normalized to obtain the characteristic input matrix;

[0125] The feature input matrix is ​​input into the temporal convolution layer of the first neural network model. The temporal convolution layer contains three parallel one-dimensional convolution sublayers. Each one-dimensional convolution sublayer uses a different convolution kernel size. The initial feature map is obtained through the ReLU activation function and batch normalization.

[0126] Based on the initial feature map, the long short-term memory layer is processed. The long short-term memory layer contains 64 memory units, each of which contains an input gate, a forget gate, and an output gate. The time series features are selectively memorized to obtain time series correlation features.

[0127] The time-series correlation features are weighted through the attention mechanism layer. The attention mechanism layer adopts a multi-head self-attention structure, which includes 8 attention heads. The features at different time positions are weighted to obtain the battery safety feature matrix.

[0128] The feature input matrix is ​​input into the fully connected layer of the second neural network model. The fully connected layer contains three hidden layers, and the number of nodes in each layer is 128, 64, and 32 respectively. The internal resistance feature vector is obtained through the Sigmoid activation function.

[0129] The internal resistance feature vector is input into the polarization state evaluation subnetwork of the second neural network model. The polarization state evaluation subnetwork includes two graph convolution layers. Each graph convolution layer uses Chebyshev polynomials for feature propagation and obtains the polarization feature vector through maximum pooling.

[0130] The polarized feature vector is input into the fault diagnosis subnetwork of the second neural network model. The fault diagnosis subnetwork contains three layers of feedforward neural network. Dropout regularization and batch normalization are used in each layer, and the fault probability vector is obtained through the Softmax function.

[0131] Specifically, the charging status monitoring data is sampled in segments according to the time series, and the voltage, current, and temperature data during the charging process are discretized in the form of time series for subsequent feature extraction and pattern recognition. Assume that the charging status monitoring data is ,in , , Respectively expressed in time The voltage, current and temperature at the moment. The data sequence is obtained by segmented sampling at equal time intervals. ,in is the number of sampling segments. The sampled data is normalized to eliminate the scale differences between the data. The normalization formula is:

[0132] ;

[0133] in, is the voltage , Current or temperature The sampling value of and They are The normalized data is used to construct the feature input matrix, which It consists of normalized voltage, current and temperature values ​​and is used as the input of the neural network. Input to the temporal convolution layer of the first neural network model. The temporal convolution layer consists of three parallel one-dimensional convolution sub-layers. Each sub-layer uses a different convolution kernel size to extract features at different time scales. Assuming that the convolution kernel sizes are 3, 5, and 7 respectively, the convolution operation is expressed as:

[0134] ;

[0135] in, Indicates that the convolution kernel size is The convolution result of the input matrix is Represents the parameters of the convolution kernel. Each convolution kernel performs a sliding convolution operation on the input matrix to extract local pattern features in the input data. Convolution performs a nonlinear transformation on the convolution result through the ReLU activation function. The ReLU activation function is defined as:

[0136] ;

[0137] The use of ReLU can introduce nonlinear features while reducing the gradient vanishing problem. Use batch normalization to normalize the convolution output to accelerate the convergence of the model and stabilize the training process. The batch normalization formula is:

[0138] ;

[0139] in, and The current batch The mean and variance of It is a small amount to prevent the denominator from being zero. After convolution and batch normalization, the initial feature map is obtained, which contains the timing information and local features in the charging state data. Based on the initial feature map, it is processed by the long short-term memory (LSTM) layer. The LSTM layer contains 64 memory units, and each unit selectively memorizes the timing features through the input gate, forget gate, and output gate. The input gate of LSTM is used to control the impact of the current input on the unit state, the forget gate controls the degree of retention of the previous state, and the output gate determines how the unit state affects the output. The update formula of the memory unit is:

[0140] ;

[0141] ;

[0142] ;

[0143] ;

[0144] ;

[0145] in, , , Represent the activation values ​​of the forget gate, input gate, and output gate respectively, , , , represents the weight matrix, , , , represents the bias vector, represents the Sigmoid activation function, Represents the bitwise multiplication operation of elements. Through the processing of LSTM, the time series features are selectively memorized to obtain the time series correlation features. The time series correlation features are weighted through the attention mechanism layer. The attention mechanism layer adopts a multi-head self-attention structure, which contains 8 attention heads. Each attention head independently calculates the correlation between different time positions. The calculation formula of multi-head attention is:

[0146] ;

[0147] in, , , denote the query matrix, key matrix and value matrix respectively, is the dimension of the key. By weighting the features at different time positions, the battery safety feature matrix is ​​obtained: , which describes the key safety features of the charging process and is used for subsequent battery status assessment. At the same time, the features are input into the matrix The input is sent to the fully connected layer of the second neural network model. The fully connected layer contains three hidden layers, and the number of nodes in each layer is 128, 64, and 32 respectively. The output of each layer is processed nonlinearly by the Sigmoid activation function, which is defined as:

[0148] ;

[0149] Through the Sigmoid function, the output is limited to (0,1) to adapt to the subsequent internal resistance feature modeling. After processing by the fully connected layer, the internal resistance feature vector is obtained. , which is used to describe the change of battery internal resistance and is an important indicator of battery health. The input is sent to the polarization state evaluation subnetwork of the second neural network model. The polarization state evaluation subnetwork contains two graph convolution layers. Each graph convolution layer performs feature propagation by using Chebyshev polynomials. The graph convolution calculation formula of Chebyshev polynomials is:

[0150] ;

[0151] in, Indicates The input feature matrix of the layer, is the Chebyshev polynomial, is the Laplace matrix, is the learned weight matrix. Through the feature propagation of the graph convolution layer, the internal resistance feature is fused with the features of the adjacent nodes to obtain a more accurate representation of the polarization state. Through the maximum pooling operation, the features of each node are aggregated to obtain the polarization feature vector . The polarization eigenvector The input is sent to the fault diagnosis subnetwork of the second neural network model. The fault diagnosis subnetwork contains three layers of feedforward neural network, which is used to evaluate the failure probability of the battery. Each layer of neural network uses Dropout regularization during the training process to prevent overfitting. The Dropout operation is expressed as:

[0152] ;

[0153] in, For the The output of the layer, To retain the probability of the node. The output of each layer is standardized through batch normalization to accelerate the convergence of the model and stabilize the training process. Finally, the fault category is normalized through the Softmax function, which is defined as:

[0154] ;

[0155] in, Indicates that it belongs to The probability of the class, Indicates The score of the class, is the total number of categories. Through the processing of the Softmax function, the fault probability vector is obtained , which is a vector used to describe the probability of battery failure at different states of charge.

[0156] In a specific embodiment, the process of executing step 600 may specifically include the following steps:

[0157] The battery safety feature matrix is ​​constructed in state space according to the temperature state, voltage state, and current state, and the fault probability vector is mapped to the [0,1] interval as the state weight to obtain the state space basic model;

[0158] The state space basic model is constructed by hierarchical grids, with 5×5 grids in the temperature dimension, 8×8 grids in the voltage dimension, and 6×6 grids in the current dimension. The state transition probability between grid points is calculated in combination with the power compensation parameters to obtain the Markov decision process model.

[0159] The Markov decision process model is input into the recursive least squares algorithm, the forgetting factor and the initial covariance matrix are set, and a quadratic objective function is constructed based on the state error and the control cost to obtain the online optimization objective function;

[0160] The policy network of the deep reinforcement learning network is constructed based on the online optimization objective function. The first layer of 256 nodes uses ReLU activation, the second layer of 128 nodes uses PReLU activation, and the third layer of 64 nodes uses tanh activation. The data is normalized through BatchNormalization to obtain the action space distribution;

[0161] A dual Q network is used to evaluate the action space distribution. Both the main Q network and the target Q network adopt a three-layer structure [512, 256, 128]. The Dueling DQN architecture is used to separate the state value and the advantage function to obtain the strategy evaluation result.

[0162] The temporal difference error is calculated based on the strategy evaluation results, and the proximal strategy optimization algorithm is used to perform gradient update on the network parameters to obtain the optimized strategy parameters;

[0163] Substitute the optimized strategy parameters into the target Q network for sliding average update, and use the priority experience playback mechanism to sample from the experience pool to obtain the charging strategy update;

[0164] The charging current is controlled in sections based on the charging strategy update amount. The current in the constant current charging section is set to 1C, the voltage limit in the constant voltage charging section is 4.2V, and the cut-off current is set to 0.05C. Combined with action noise suppression, the charging control parameters for the next cycle are obtained.

[0165] Specifically, the battery safety feature matrix is ​​constructed in the state space according to the temperature state, voltage state and current state, and the fault probability vector is mapped to the [0,1] interval as the state weight. A basic model that can fully reflect the current charging state of the battery is constructed. Battery safety feature matrix Contains temperature, voltage, and current features that are used to define the state vector of the battery at different time points ,in Indicates temperature, Indicates voltage, represents the current. The fault probability vector It represents the possibility of battery failure in the current state, and is mapped to the [0,1] interval as the state weight , which is used to represent the risk level in each state, and the basic model of the state space is obtained as follows:

[0166] ;

[0167] in, represents the state vector of the battery, and Represents the corresponding state weight. The state space basic model is constructed by layered grid. In the temperature dimension, it is divided into The temperature range is divided into 25 small grids; the voltage dimension is divided into The grid divides the voltage range into 64 intervals; the current dimension is divided into The grid divides the current range into 36 states. Through multi-dimensional division, the battery state space is discretized into a grid consisting of A three-dimensional space model consisting of grid points. At each grid point, combined with the power compensation parameter , calculate the state transition probability between each grid point , we get the Markov decision process model The state transition probability describes the possibility of the system transferring from the current state to the target state, which is important information in control strategy planning. Enter the recursive least squares algorithm to update the control parameters of the system online to adapt to the changing battery status. The recursive least squares algorithm sets the forgetting factor and the initial covariance matrix The forgetting factor is set in the range of [0.9, 1] to control the impact of historical data on the current estimate. The value means giving more weight to historical data. The initial covariance matrix Set to a multiple of the identity matrix to ensure the robustness of the initial state. Based on the state error and control costs Construct a quadratic objective function:

[0168] ;

[0169] in, and are weighted matrices of state error and control input, respectively, which are used to adjust the control target to balance between error and control energy. The objective function is minimized by recursive least squares algorithm , and obtain the online optimization objective function, thereby optimizing the performance of the system controller. Based on the online optimization objective function, a policy network of the deep reinforcement learning network is constructed. The structure of the policy network consists of a three-layer neural network. The first layer contains 256 nodes and uses the ReLU activation function to perform nonlinear mapping on the input data. The expression of the ReLU activation function is:

[0170] ;

[0171] The second layer contains 128 nodes and uses the PReLU activation function. PReLU introduces learnable parameters To control the slope of the negative input, the expression is:

[0172] ;

[0173] This activation function can enhance the expressiveness of the model and adapt to the nonlinear characteristics of different data. The third layer contains 64 nodes and uses the tanh activation function. The expression of the tanh function is:

[0174] ;

[0175] The tanh function limits the output to the range of [-1, 1], which facilitates subsequent control operations on the output. The data of each layer is normalized by Batch Normalization to reduce the offset of internal covariates, accelerate model convergence and stabilize the training process, and finally obtain the distribution of the action space . For the action space distribution A dual Q network evaluation is performed to improve the accuracy and stability of the charging strategy. The dual Q network consists of a main Q network and a target Q network, both of which use a three-layer structure with 512, 256, and 128 nodes respectively. In order to enhance the learning ability of state values ​​and advantage functions, the Dueling DQN architecture is used to convert state values ​​into and advantage function Separation, the final Q value calculation formula is:

[0176] ;

[0177] Through this separation, the network can better understand the importance of the current state, rather than just selecting a specific action, and obtain the strategy evaluation result. The temporal difference error is calculated based on the strategy evaluation result, and its calculation formula is:

[0178] ;

[0179] in, It’s an instant reward. is the discount factor, is the evaluation value of the target Q network, is the evaluation value of the main Q network. The time difference error can measure the deviation between the current strategy and the target strategy, and the proximal policy optimization (PPO) algorithm is used to gradient update the network parameters to minimize the error and obtain the optimized policy parameters. PPO adds restrictions when updating the policy to ensure that the amplitude of the policy update is not too large, thereby improving the stability of training. Substitute the optimized policy parameters into the target Q network for sliding average update. The formula for sliding average update is:

[0180] ;

[0181] in, represents the parameters of the target Q network, represents the parameters of the main Q network, is the sliding average coefficient, which is used to control the update speed of the parameters. Through sliding average, the parameter update of the target Q network is ensured to be smoother and the fluctuation during training is reduced. The priority experience replay mechanism is used to sample from the experience pool, and the experience with greater impact on the strategy improvement is prioritized for training to obtain a more effective charging strategy update. Based on the obtained charging strategy update, the charging current is controlled in sections. In the constant current charging stage, the charging current is set to 1C to charge the battery at a faster speed; in the constant voltage charging stage, the voltage is limited to 4.2V to prevent the battery from overcharging and protect the battery life; in the final trickle charging stage, the cutoff current is set to 0.05C, and the charging current is gradually reduced to ensure charging safety and battery health. In order to improve the robustness of the control, combined with the action noise suppression strategy, a noise suppression term is added when outputting the charging control parameters to reduce the impact of external disturbances on the charging process and obtain the charging control parameters for the next cycle.

[0182] The above describes the intelligent charging management method for an electric scooter in the embodiment of the present application. The following describes the intelligent charging management device 10 for an electric scooter in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the electric scooter intelligent charging management device 10 includes:

[0183] Initialization module 11, used to collect hardware parameters and perform system initialization configuration on the single-phase bidirectional converter and permanent magnet synchronous motor in the electric scooter, and obtain basic parameters of the charging system and communication interface data;

[0184] Establishing module 12, used to perform intelligent charging network authentication and establish an encrypted channel based on charging system basic parameters and communication interface data, and obtain network authentication identification and charging station parameters;

[0185] A calculation module 13 is used to perform dynamic programming of charging strategies and interval threshold calculation according to the network authentication identifier and charging station parameters to obtain a segmented charging control curve;

[0186] The adjustment module 14 is used to input the segmented charging control curve into the adaptive prediction controller to adjust the charging process in real time, and obtain the charging state monitoring data and power compensation parameters;

[0187] An analysis module 15 is used to input the charging state monitoring data into the first neural network model and the second neural network model respectively, wherein the first neural network model performs multi-dimensional feature extraction of battery temperature, voltage, and current, and the second neural network model performs battery internal resistance and polarization state analysis to obtain a battery safety feature matrix and a fault probability vector;

[0188] The optimization module 16 is used to construct a three-dimensional charging state space model based on the battery safety feature matrix and the fault probability vector, perform online optimization calculation on the power compensation parameters by recursive least squares method, and input the calculation results into the deep reinforcement learning network for iterative optimization of the charging strategy to obtain the charging control parameters for the next cycle.

[0189] Through the coordinated cooperation of the above-mentioned components, the hardware cost is reduced through the integrated design of the single-phase bidirectional converter and the permanent magnet synchronous motor, and the permanent magnet synchronous motor is used as a charging loop filter to improve the charging efficiency and power quality. The dual neural network model is used for battery status monitoring and fault diagnosis, and combined with the three-dimensional charging state space model, the real-time evaluation of the battery safety status and the accurate early warning of faults are realized. The charging strategy optimization based on the deep reinforcement learning network dynamically adjusts the charging parameters, significantly reduces the battery temperature rise and prolongs the battery life. The authentication mechanism and encryption channel of the intelligent charging network ensure the data security of the charging process and improve the system reliability. The recursive least squares method and adaptive predictive control algorithm are used to achieve precise control and dynamic optimization of the charging process and improve the charging efficiency. The charging strategy formulation based on dynamic programming, combined with electricity price information and user needs, optimizes the charging time schedule and reduces the charging cost. The segmented charging control and multiple protection mechanisms effectively prevent the occurrence of overcharging, over-discharging and other faults, ensuring the charging safety.

[0190] See also Figure 3 , Figure 3 This is a schematic block diagram of the structure of an electronic device 300 provided in an embodiment of the present application. The electronic device 300 includes a processor 301 and a memory 302. The processor 301 and the memory 302 are connected via a device bus 303, wherein the memory 302 may include a non-volatile storage medium and an internal memory.

[0191] The non-volatile storage medium can store a computer program. The computer program includes program instructions, and when the program instructions are executed by the processor 301, the processor 301 can execute any of the above-mentioned electric scooter intelligent charging management methods.

[0192] The processor 301 is used to provide computing and control capabilities to support the operation of the entire electronic device 300 .

[0193] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor 301, the processor 301 can execute any of the above-mentioned electric scooter intelligent charging management methods.

[0194] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the electronic device 300 involved in the scheme of the present application. The specific electronic device 300 may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0195] It should be understood that the processor 301 may be a central processing unit (CPU), and the processor 301 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0196] It should be noted that those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the electronic device 300 described above can refer to the corresponding process of the aforementioned electric scooter intelligent charging management method, which will not be repeated here.

[0197] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the one or more processors implement the electric scooter intelligent charging management method provided in the embodiment of the present application.

[0198] The computer-readable storage medium may be an internal storage unit of the electronic device 300 in the aforementioned embodiment, such as a hard disk or memory of the electronic device 300. The computer-readable storage medium may also be an external storage device of the electronic device 300, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped with the electronic device 300.

[0199] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0200] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable an electronic device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0201] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An intelligent charging management method for an electric scooter, characterized in that: The method comprises: The hardware parameters of the single-phase bidirectional converter and the permanent magnet synchronous motor in the electric scooter are collected and the system is initialized to obtain the basic parameters of the charging system and the communication interface data; specifically, the switch tube drive signal configuration and dead time parameter setting of the single-phase bidirectional converter are performed to obtain the PWM reference waveform and the switch timing data, and the inductance filter parameters of the charging circuit of the permanent magnet synchronous motor are set based on the PWM reference waveform and the switch timing data to obtain the inductance filter coefficient and the current ripple ratio; the charging circuit stability analysis and compensation calculation of the inductance filter coefficient and the current ripple ratio are performed to obtain the system stability compensation vector, and the system stability compensation vector is substituted into the multiple feedback control equation for charging. Modeling and simulation analysis of the electric circuit to obtain charging circuit response characteristic data; based on the charging circuit response characteristic data, a multi-level protection control strategy and fault judgment criterion analysis are performed to create a fault protection control matrix, and the fault protection control matrix is ​​dynamically divided into thresholds and gradient graded to obtain a graded protection threshold group; a fault diagnosis and processing mechanism is established according to the graded protection threshold group, and protection parameter calibration is performed in combination with the battery charging safety characteristic curve to obtain standardized protection control data; data format conversion and encoding are performed on the standardized protection control data according to the communication protocol specification to obtain basic parameters of the charging system, and the basic parameters of the charging system are encrypted and encapsulated in communication frames to obtain communication interface data; Perform intelligent charging network authentication and encryption channel establishment based on the charging system basic parameters and the communication interface data to obtain a network authentication identifier and charging station parameters; Performing dynamic programming of charging strategies and interval threshold calculation according to the network authentication identifier and the charging station parameters to obtain a segmented charging control curve; The segmented charging control curve is input into an adaptive prediction controller to adjust the charging process in real time, thereby obtaining charging state monitoring data and power compensation parameters; Inputting the charging state monitoring data into a first neural network model and a second neural network model respectively, wherein the first neural network model performs multi-dimensional feature extraction of battery temperature, voltage, and current, and the second neural network model performs battery internal resistance and polarization state analysis to obtain a battery safety feature matrix and a fault probability vector; A three-dimensional charging state space model is constructed based on the battery safety feature matrix and the fault probability vector, and the power compensation parameters are optimized online by recursive least squares method. The calculation results are input into a deep reinforcement learning network for iterative optimization of the charging strategy to obtain the charging control parameters for the next cycle.

2. The electric scooter intelligent charging management method according to claim 1, characterized in that: The smart charging network authentication and encryption channel establishment based on the charging system basic parameters and the communication interface data to obtain the network authentication identifier and charging station parameters include: Extract and concatenate the battery identification code and the system hardware model in the basic parameters of the charging system to obtain a unique device identification code, and perform XOR operation and group processing on the unique device identification code and the encrypted frame in the communication interface data to obtain an identity authentication data packet; Establishing a secure handshake connection with the smart charging station based on the identity authentication data packet, and performing a two-way identity verification to obtain an encrypted session identifier; The validity period and authority allocation are set for the encrypted session identifier, and the data frame is reorganized in combination with the Internet of Things communication protocol to obtain a communication authentication key, and the communication authentication key is embedded in the standard data frame header, and a protocol handshake and authority verification are performed with the charging station server to obtain a network authentication identifier; The real-time electricity price data and power capacity data of the charging station are parsed based on the network authentication identifier to obtain a resource state matrix, and the charging time window division and power limit calculation are performed on the resource state matrix to obtain charging station parameters.

3. The electric scooter intelligent charging management method according to claim 2, characterized in that: The performing of dynamic charging strategy programming and interval threshold calculation according to the network authentication identifier and the charging station parameters to obtain a segmented charging control curve includes: Priority calculation and matching are performed on the user level data in the network authentication identifier and the power capacity data in the charging station parameters to obtain a scheduling authorization matrix; Substituting the dispatch authorization matrix into the dynamic programming model, searching for the optimal path according to the electricity price period and the charging demand, obtaining a timing optimization sequence, and performing charging time estimation and power allocation on the timing optimization sequence and the remaining battery capacity to obtain an initial charging power curve; Based on the initial charging power curve, segmented fitting and threshold feature extraction are performed to divide the charging process into constant current interval, constant voltage interval, and trickle current interval to obtain charging partition parameters; A state transfer matrix is ​​constructed according to the charging partition parameters, an optimal control strategy is solved for each charging interval to obtain a partition control vector, and dynamic boundary constraints and smooth transition processing are performed on the partition control vector to obtain charging trajectory data; The charging process is simulated and safety checked based on the charging trajectory data to obtain an optimized control sequence, which is reconstructed in a piecewise function form and transition constraints are added to obtain a piecewise charging control curve.

4. The electric scooter intelligent charging management method according to claim 3, characterized in that: The step of inputting the segmented charging control curve into the adaptive prediction controller to adjust the charging process in real time to obtain charging state monitoring data and power compensation parameters includes: The segmented charging control curve is sampled and data segmented at equal time intervals, and a prediction step is set in combination with the dynamic characteristics of the charging process to obtain a prediction control reference sequence; Substituting the prediction control reference sequence into the linear quadratic optimal control equation for recursive solution, obtaining a control increment matrix in the prediction time domain, and performing rolling prediction of the battery charging state and charging process constraint boundary calculation according to the control increment matrix to obtain dynamic constraint parameters; Calculating the difference between the dynamic constraint parameter and the actual charging process data, and updating the controller feedback gain based on the gradient descent method to obtain an adaptive control amount; Generate a PWM modulation signal according to the adaptive control amount, adjust the switch tube drive signal of the single-phase bidirectional converter in real time, obtain charging state sampling data, and perform feature decomposition of the charging state sampling data according to three dimensions of voltage, current and temperature to obtain charging state monitoring data; A power fluctuation analysis is performed on the charging state monitoring data, and a harmonic compensation value is calculated in combination with the filtering characteristics of the permanent magnet synchronous motor to obtain a power regulation coefficient. At the same time, a closed-loop compensation is performed on the charging power based on the power regulation coefficient, and the output power is dynamically adjusted through a proportional-integral controller to obtain a power compensation parameter.

5. The electric scooter intelligent charging management method according to claim 4, characterized in that: The charging state monitoring data is input into a first neural network model and a second neural network model respectively, wherein the first neural network model performs multi-dimensional feature extraction of battery temperature, voltage, and current, and the second neural network model performs battery internal resistance and polarization state analysis to obtain a battery safety feature matrix and a fault probability vector, including: The charging state monitoring data is sampled in sections according to a time series, and the voltage, current, and temperature are normalized to obtain a characteristic input matrix; Inputting the feature input matrix into the temporal convolution layer of the first neural network model, the temporal convolution layer includes three parallel one-dimensional convolution sublayers, each one-dimensional convolution sublayer adopts a different convolution kernel size, and obtaining an initial feature map through ReLU activation function and batch normalization processing; Based on the initial feature map, long short-term memory layer processing is performed, wherein the long short-term memory layer includes 64 memory units, each unit includes an input gate, a forget gate, and an output gate, and selectively memorizes the time series features to obtain time series correlation features; The time series correlation features are weighted through an attention mechanism layer. The attention mechanism layer adopts a multi-head self-attention structure, including 8 attention heads, and performs weighted processing on features at different time series positions to obtain a battery safety feature matrix; Input the feature input matrix into a fully connected layer of a second neural network model, wherein the fully connected layer includes three hidden layers, and the number of nodes in each layer is 128, 64, and 32 respectively, and an internal resistance feature vector is obtained by a Sigmoid activation function; Inputting the internal resistance feature vector into the polarization state evaluation subnetwork of the second neural network model, wherein the polarization state evaluation subnetwork includes two graph convolution layers, each graph convolution layer uses Chebyshev polynomials for feature propagation, and obtains the polarization feature vector through maximum pooling; The polarized feature vector is input into the fault diagnosis subnetwork of the second neural network model, wherein the fault diagnosis subnetwork comprises three layers of feedforward neural network, each layer uses Dropout regularization and batch normalization, and a fault probability vector is obtained through a Softmax function.

6. The electric scooter intelligent charging management method according to claim 5, characterized in that: The three-dimensional charging state space model is constructed based on the battery safety feature matrix and the fault probability vector, the power compensation parameters are optimized online by recursive least squares method, and the calculation results are input into the deep reinforcement learning network for iterative optimization of the charging strategy to obtain the charging control parameters of the next cycle, including: The battery safety feature matrix is ​​constructed in state space according to the temperature state, voltage state, and current state, and the fault probability vector is mapped to the interval [0, 1] as the state weight to obtain a state space basic model; The state space basic model is constructed by hierarchical gridding, dividing the temperature dimension into 5×5 grids, the voltage dimension into 8×8 grids, and the current dimension into 6×6 grids, and the state transition probability between each grid point is calculated in combination with the power compensation parameter to obtain a Markov decision process model; Inputting the Markov decision process model into a recursive least squares algorithm, setting a forgetting factor and an initial covariance matrix, constructing a quadratic objective function based on a state error and a control cost, and obtaining an online optimization objective function; Based on the online optimization objective function, a policy network of a deep reinforcement learning network is constructed, wherein the first layer of 256 nodes is activated by ReLU, the second layer of 128 nodes is activated by PReLU, the third layer of 64 nodes is activated by tanh, and data is normalized by BatchNormalization to obtain action space distribution; The action space distribution is evaluated by dual Q network. Both the main Q network and the target Q network adopt a three-layer structure [512, 256, 128]. The Dueling DQN architecture is used to separate the state value and the advantage function to obtain the strategy evaluation result. Calculating the temporal difference error according to the strategy evaluation result, and using the proximal strategy optimization algorithm to perform gradient update on the network parameters to obtain optimized strategy parameters; Substituting the optimization strategy parameters into the target Q network for sliding average update, and using the priority experience playback mechanism to sample from the experience pool to obtain the charging strategy update amount; Based on the charging strategy update amount, the charging current is controlled in sections, the current in the constant current charging section is set to 1C, the voltage limit in the constant voltage charging section is 4.2V, the cut-off current is set to 0.05C, and the charging control parameters for the next cycle are obtained in combination with action noise suppression.

7. An intelligent charging management device for an electric scooter, characterized in that: Used to execute the electric scooter intelligent charging management method according to any one of claims 1 to 6, the electric scooter intelligent charging management device comprises: The initialization module is used to collect hardware parameters and perform system initialization configuration on the single-phase bidirectional converter and permanent magnet synchronous motor in the electric scooter, and obtain basic parameters of the charging system and communication interface data; An establishment module, used to perform intelligent charging network authentication and encryption channel establishment based on the charging system basic parameters and the communication interface data, and obtain a network authentication identifier and charging station parameters; A calculation module, used to perform dynamic programming of charging strategy and interval threshold calculation according to the network authentication identifier and the charging station parameters to obtain a segmented charging control curve; An adjustment module is used to input the segmented charging control curve into an adaptive prediction controller to adjust the charging process in real time, so as to obtain charging state monitoring data and power compensation parameters; An analysis module, used to input the charging state monitoring data into a first neural network model and a second neural network model respectively, wherein the first neural network model performs multi-dimensional feature extraction of battery temperature, voltage, and current, and the second neural network model performs battery internal resistance and polarization state analysis to obtain a battery safety feature matrix and a fault probability vector; The optimization module is used to construct a three-dimensional charging state space model based on the battery safety feature matrix and the fault probability vector, perform online optimization calculation on the power compensation parameters by recursive least squares method, and input the calculation results into a deep reinforcement learning network for iterative optimization of the charging strategy to obtain the charging control parameters for the next cycle.

8. An electronic device, characterized in that: The electronic device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the electronic device executes the electric scooter intelligent charging management method according to any one of claims 1-6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the electric scooter intelligent charging management method according to any one of claims 1 to 6 is implemented.

Images