Charging pile adaptive pulse width modulation method, medium and system

Abstract

The invention provides a self-adaptive pulse width modulation method, medium and system for a charging pile, and belongs to the technical field of charging piles. The self-adaptive pulse width modulation method comprises the following steps: acquiring voltage and current values of a battery, calculating a charge state parameter and a tilt change rate of the battery, and calculating a load coefficient and a third-order precision index by combining a load sensing algorithm; and analyzing the current waveform by using Fourier transform to obtain a harmonic distortion index, and calculating a delay compensation function value. The core technology is to call a deep residual neural network model of a double-path parallel architecture, input multi-dimensional characteristic parameters, output optimal pulse width modulation parameters and generate pulse width modulation signals to adjust charging characteristics. Meanwhile, the temperature and the internal resistance value of the battery are monitored in real time, parameter updating is triggered when the preset threshold value is exceeded, a complete closed-loop control cycle is formed, and the technical problem that the pulse width modulation parameter accurately adapts to the change of the charging state of the battery under the complex dynamic load condition is effectively solved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of charging piles, and in particular relates to a charging pile adaptive pulse width modulation method, medium and system. BACKGROUND

[0002] Electric vehicle charging piles are key infrastructure in the new energy vehicle industry chain, and their charging control technology directly affects battery service life and charging efficiency. Traditional charging piles generally use fixed parameter pulse width modulation technology to control the charging process, and achieve charge and discharge control through a pre-set charging curve and a simple voltage and current feedback mechanism. This method can meet the basic charging needs under ideal environmental and stable load conditions, and has been widely used in commercial charging stations and home charging piles. However, the traditional fixed parameter pulse width modulation technology has many limitations in actual operating environments. First, it cannot effectively cope with complex factors such as power grid fluctuations, temperature changes and dynamic changes in battery internal resistance; second, it lacks real-time and accurate assessment of battery state during charging, resulting in low charging efficiency and prolonged charging time; most importantly, it cannot dynamically optimize modulation parameters according to the real-time state of charge of the battery, resulting in poor current waveform quality during charging, generating a large amount of harmonics, which not only reduces energy conversion efficiency, but also can accelerate battery aging. In the face of new generation fast charging and smart grid application scenarios, how to accurately adjust the pulse width modulation parameters under complex and variable load conditions to match the dynamic changes in battery charging state has become a core technical problem that needs to be solved in current charging technology. Solving this problem is of great significance to improving charging efficiency, prolonging battery life and improving grid friendliness. SUMMARY

[0003] Therefore, the present application provides a charging pile adaptive pulse width modulation method, medium and system, which can solve the technical problem that the charging pile cannot accurately adjust the pulse width modulation parameters to adapt to the dynamic changes in the battery charging state under complex dynamic load conditions in the prior art.

[0004] The application is implemented in the following manner: a first aspect of the application provides a charging pile adaptive pulse width modulation method, which comprises the following steps: collecting battery voltage and current values at the output end of the charging pile, calculating battery state of charge parameters and right side tilt change rate; determining a preset charging curve according to the battery state of charge parameters, obtaining modulation parameters required in the current charging stage and left side tilt change rate; monitoring the load fluctuation state of the charging pile, calculating the load coefficient and third-order accuracy index by using a load sensing algorithm; analyzing the battery current waveform based on Fourier transform, obtaining a harmonic distortion index and calculating a delay compensation function value; calling a pulse modulation neural network model, inputting the battery state of charge parameters, load coefficient, right side tilt change rate, left side tilt change rate, third-order accuracy index and delay compensation function value, and outputting optimal pulse width modulation parameters; generating a pulse width modulation signal according to the optimal pulse width modulation parameters, and adjusting the output characteristics of the charging pile; monitoring the battery temperature value and battery internal resistance value in real time, and when the battery temperature value change rate or the battery internal resistance value change rate exceeds a preset threshold, triggering a modulation parameter update cycle and returning to the first step to execute a closed-loop control cycle.

[0005] The battery state of charge parameter is specifically the ratio of the current battery capacity to the rated capacity calculated by the coulomb counting method, the value range is 0 to 1, and it represents the battery charge percentage, which is used to determine the charging stage of the battery.

[0006] The preset charging curve is specifically a charging current and voltage relationship atlas preset according to the characteristics of different types of batteries, which includes three stages of constant current charging, constant voltage charging and trickle charging, each stage corresponding to different modulation strategies.

[0007] The load sensing algorithm specifically calculates the real-time load utilization rate and change trend by measuring the power difference between the input end and the output end of the charging pile, combining historical load data, and using Kalman filtering method to eliminate the influence of random fluctuations.

[0008] The harmonic distortion index is specifically the sum of the ratio of high-order harmonic components to fundamental components in the current waveform calculated by Fourier transform, which is used to measure the quality of the current waveform, and the lower the value, the closer the current waveform is to the ideal state.

[0009] The optimal pulse width modulation parameter is specifically the ratio of pulse width to period, carrier frequency and modulation depth, which directly determines the waveform characteristics and energy conversion efficiency of the charging current.

[0010] The modulation parameter update cycle is specifically the dynamic time interval calculated by the system according to the battery temperature value change rate and the battery internal resistance value change rate, and a new round of parameter optimization calculation is triggered when the change rate exceeds the preset threshold or the fixed time reaches the upper limit value.

[0011] The specific structure of the pulse modulation neural network model is a deep residual network with a two-way parallel architecture, including a feature extraction layer, a residual connection layer, and an output mapping layer. A multi-head attention mechanism is introduced in the middle layer, and the number of attention heads is determined by the battery state of charge parameter and the load coefficient.

[0012] The load coefficient is specifically the ratio of the current actual load of the charging pile to the rated load, which quantifies the working state of the charging pile and provides load information for pulse width modulation parameter optimization.

[0013] The right side slope change rate specifically refers to the slope change trend of the battery voltage value rising curve after the current time during the battery charging process. It is obtained by calculating the voltage difference between consecutive sampling points and performing exponential weighted averaging. This parameter reflects the dynamic change characteristics of the battery charging acceptance ability.

[0014] The left side slope change rate specifically refers to the historical change of the battery voltage value rising curve before the current time during the battery charging process. It is obtained by performing piecewise linear regression analysis on the voltage curve in the historical data window. This parameter represents the historical performance of the battery in responding to the charging current.

[0015] The third-order precision index specifically refers to the residual sum of squares obtained by fitting the battery current value waveform with a third-order polynomial. This index quantifies the degree to which the battery current value waveform deviates from the ideal curve. The smaller the value, the closer the current waveform is to the ideal state.

[0016] The delay compensation function value is used to correct the control error caused by circuit response delay in the charging system. The inputs for calculating the delay compensation function value include the current rise time, current fall time, voltage response delay time, and load coefficient change rate in the current sampling window.

[0017] The steps for establishing the training data set of the pulse modulation neural network model include collecting charging data of various types of batteries under different charging conditions, different environmental temperatures, and different load conditions. Each data contains multiple input features and the corresponding optimal pulse width modulation parameter as a label.

[0018] The steps for training the pulse modulation neural network model include pre-training on the training set, using mean square error as the loss function, using a stochastic gradient descent optimization algorithm with a momentum term for parameter update, setting the initial learning rate to 0.001, and using a cosine annealing strategy for dynamic adjustment.

[0019] The closed-loop control cycle specifically refers to the complete process from battery state acquisition, parameter calculation, model inference, signal generation to parameter update, forming a continuous iteration control loop, and achieving dynamic adjustment of the charging process by continuously optimizing the modulation parameters.

[0020] The second aspect of the present application provides a computer readable storage medium, wherein program instructions are stored in the computer readable storage medium, and the program instructions are used to execute the charging pile adaptive pulse width modulation method when running in a computer.

[0021] The third aspect of the present application provides a charging pile adaptive pulse width modulation system, comprising the computer readable storage medium described above, the system is any one of a computer, a server, and a single-chip microcomputer, the computer readable storage medium is arranged in the system, and the system is provided with a microprocessor for executing the program instructions stored in the computer readable storage medium.

[0022] The present application realizes the comprehensive analysis and processing of multi-dimensional features such as battery state of charge parameters, load coefficients, left and right side tilt change rates, three-order precision indicators and delay compensation function values by constructing a deep residual neural network model with a double-path parallel architecture, thereby dynamically generating optimal pulse width modulation parameters and realizing real-time adaptive control of the charging process. This method effectively solves the limitations of traditional fixed parameter modulation technology. First, by monitoring the battery voltage and current in real time and combining Fourier transform analysis, the small changes in the battery charging state are accurately captured. Second, the load sensing algorithm and Kalman filtering technology are introduced to effectively deal with random fluctuations under complex load conditions. In particular, through the multi-head attention mechanism in the neural network model, intelligent weight distribution of features in different charging stages is realized, so that the pulse width modulation parameters can accurately track the battery state changes. Through the synergistic effect of the above technical means, the present application solves the core technical problem that the pulse width modulation parameters of the charging pile cannot accurately adapt to the dynamic changes of the battery charging state under complex dynamic load conditions, significantly improves the charging efficiency and current waveform quality, reduces harmonic distortion, prolongs the service life of the battery, and enhances the adaptability of the charging pile to power grid fluctuations. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The flowchart of the method of the present application.

[0024] Figure 2 The structural diagram of the pulse modulation neural network model involved in the present application. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application.

[0026] As shown in the flowchart of the charging pile adaptive pulse width modulation method provided by the first aspect of the present application, the method comprises the following steps: Figure 1 ​

[0027] S01, collect the battery voltage value and the battery current value at the output end of the charging pile, calculate the battery state of charge parameter and the right side tilt change rate;

[0028] S02, determine a preset charging curve according to the battery state of charge parameter, and obtain a modulation parameter required in a current charging stage and a left side tilt change rate;

[0029] S03, monitor the load fluctuation state of the charging pile, and calculate a load coefficient and a third-order precision index by using a load sensing algorithm;

[0030] S04, analyze the battery current value waveform based on Fourier transform, obtain a harmonic distortion index, and calculate a delay compensation function value;

[0031] S05, call a pulse modulation neural network model, input the battery state of charge parameter, the load coefficient, the right side tilt change rate, the left side tilt change rate, the third-order precision index, and the delay compensation function value, and output an optimal pulse width modulation parameter;

[0032] S06, generate a pulse width modulation signal according to the optimal pulse width modulation parameter, and adjust the output characteristics of the charging pile;

[0033] S07, monitor the battery temperature value and the battery internal resistance value in real time, when the battery temperature value change rate or the battery internal resistance value change rate exceeds a preset threshold, trigger a modulation parameter update period, and return to step S01 to execute a closed-loop control cycle.

[0034] The battery state of charge parameter specifically refers to the ratio of the current battery capacity to the rated capacity calculated by the coulomb counting method, the value range is 0 to 1, represents the battery charge percentage, and is used to determine the charging stage of the battery.

[0035] The preset charging curve specifically refers to a charging current and voltage relationship graph map preset according to the characteristics of different types of batteries, which includes three stages of constant current charging, constant voltage charging and trickle charging, each stage corresponds to a different modulation strategy.

[0036] The load sensing algorithm specifically calculates the real-time load utilization rate and the change trend by measuring the power difference between the input end and the output end of the charging pile, combining historical load data, and using Kalman filtering method to eliminate the influence of random fluctuations.

[0037] The load coefficient specifically refers to the ratio of the current actual load of the charging pile to the rated load, which is used to quantify the working state of the charging pile and provide load information for pulse width modulation parameter optimization.

[0038] The harmonic distortion index is specifically a ratio sum of high harmonic components and fundamental components in a current waveform calculated by Fourier transform, and is used to measure the quality of the current waveform, and the lower the value is, the closer the current waveform is to an ideal state.

[0039] The optimal pulse width modulation parameter is specifically a ratio of pulse width to period, a carrier frequency, and a modulation depth, which directly determine the waveform characteristics of the charging current and the energy conversion efficiency.

[0040] The modulation parameter update period is specifically a dynamic time interval calculated by the battery temperature value change rate and the battery internal resistance value change rate, and a new round of parameter optimization calculation is triggered when the change rate exceeds a preset threshold or the fixed time reaches an upper limit value.

[0041] The right side slope change rate is specifically a slope change trend of the battery voltage value rising curve after the current time during the battery charging process, and is obtained by calculating the voltage difference between continuous sampling points and performing exponential weighted average, and the parameter reflects the dynamic change characteristics of the battery charging acceptance ability and is an important index for predicting the stability of the charging process.

[0042] The left side slope change rate is specifically a historical change of the slope of the battery voltage value rising curve before the current time during the battery charging process, and is obtained by performing piecewise linear regression analysis on the voltage curve in the historical data window, and the parameter represents the historical performance of the battery in response to the charging current and is used to evaluate the battery charging consistency and predict future charging behavior.

[0043] The third order precision index is specifically a residual sum of squares after third order polynomial fitting of the battery current value waveform, and the index quantifies the degree of deviation of the battery current value waveform from an ideal curve, and the smaller the value is, the closer the current waveform is to an ideal state, and the higher the charging quality is, which is an important parameter for evaluating the effect of pulse width modulation.

[0044] The delay compensation function value is used to correct the control error caused by circuit response delay in the charging system, and the inputs for calculating the delay compensation function value include the current rise time, the current fall time, the voltage response delay time in the current sampling window, and the load coefficient change rate, and the delay compensation function value output is a time compensation amount of the next cycle pulse width modulation signal, and the influence of the inherent delay of the system on the charging performance is eliminated by predictive compensation means, and the charging accuracy and stability are improved.

[0045] The specific structure of the pulse modulation neural network model is a deep residual network with a two-way parallel architecture, including a feature extraction layer, a residual connection layer, and an output mapping layer. The feature extraction layer is composed of eight fully connected layers, which are used to extract feature vectors of battery charging characteristics from input data. The residual connection layer includes six groups of cross-layer connection units, which use a gated recurrent unit mechanism to process time series features. The output mapping layer includes an adaptive normalization layer and three branch prediction heads, which output the pulse width, carrier frequency, and modulation depth as three key parameters. The pulse modulation neural network model introduces a multi-head attention mechanism in the middle layer. The number of attention heads of the multi-head attention mechanism is determined by the battery state of charge parameter and the load coefficient, which realizes the adaptive representation ability of different charging stages.

[0046] The training data set of the pulse modulation neural network model is established by collecting charging data of various types of batteries under different charging states, different environmental temperatures, and different load conditions. Each data includes the battery state of charge parameter, the load coefficient, the right side slope change rate, the left side slope change rate, the third order precision index, and the delay compensation function value as input features, and the optimal pulse width modulation parameter as a label. The above data is collected in a laboratory environment by a test device and normalized. After removing outliers, the data is divided into a training set, a validation set, and a test set in a ratio of seven to two to one.

[0047] The training of the pulse modulation neural network model includes pre-training on the training set, using mean square error as the loss function, using a stochastic gradient descent optimization algorithm with a momentum term to update parameters, setting the initial learning rate to 0.001 and using a cosine annealing strategy to dynamically adjust it, evaluating the model performance on the validation set after completing each training cycle, reducing the learning rate when there is no significant improvement in the validation set performance for five consecutive cycles, introducing batch normalization layers during training to reduce internal covariate shift, applying random dropout technology to prevent overfitting, and finally evaluating the model on the test set after convergence to ensure that the generalization performance meets the requirements of practical applications before deploying it to the charging pile control system.

[0048] The closed-loop control cycle specifically refers to the complete process from battery state acquisition to parameter calculation, model inference, signal generation, and parameter update, forming a continuous iterative control loop. By continuously optimizing the modulation parameters, the charging process is dynamically adjusted to ensure charging efficiency and safety.

[0049] The specific implementation of the above steps is described in detail below.

[0050] The specific implementation of step S01 is to collect the battery voltage value and the battery current value at the output end of the charging pile through a high-precision sampling circuit, and the sampling frequency is set to 10 kHz to ensure the accuracy of signal acquisition. After the collection is completed, the battery state of charge parameter is calculated based on the coulomb counting method, that is, the ratio of the current capacity to the rated capacity, and the calculation formula is where Q current is the current capacity, and Q rated is the rated capacity. At the same time, the differentially calculated battery voltage value is processed by using the exponential weighted moving average algorithm to obtain the right side tilt rate, which represents the slope change trend of the battery voltage rising curve after the current time. The purpose of this step is to obtain the current state parameter of the battery, to provide basic data for subsequent charging strategy optimization, and to ensure accurate control of the charging process according to the actual state of the battery.

[0051] The specific implementation of step S02 is to select a matching charging curve from the pre-set charging curve database according to the battery state of charge parameter calculated in step S01. Specifically, when SOC < 0.2, the constant current charging segment parameters are selected; when 0.2 ≤ SOC < 0.8, the constant voltage charging segment parameters are selected; and when SOC ≥ 0.8, the trickle charging segment parameters are selected. Each stage corresponds to a different set of modulation parameters, including the basic pulse width ratio, the basic carrier frequency and the basic modulation depth. At the same time, the left side tilt rate is calculated by performing piecewise linear regression analysis on the voltage curve in the historical data window, and the window size is set to the sampling data in the past 60 seconds. The purpose of this step is to determine the basic modulation parameters required for the current charging stage and to obtain the battery charging historical response characteristics, which provides a reference basis for subsequent pulse width modulation optimization.

[0052] The specific implementation of step S03 is to measure the power difference between the input end and the output end of the charging pile to monitor the load fluctuation state of the charging pile in real time. The Kalman filter algorithm is used to process the power difference data, and the filter parameters are set as follows: process noise covariance 0.01, measurement noise covariance 0.1, to effectively eliminate the influence of random fluctuations. The real-time load utilization rate is calculated by processing the filtered power data, and the load coefficient is obtained, that is, the ratio of the current actual load to the rated load. At the same time, the battery current value waveform is fitted by a third-order polynomial, and the residual sum of squares is calculated to obtain a third-order accuracy index, and the threshold value of this index is set to 0.05, which means that the current waveform quality is good. The purpose of this step is to accurately obtain the working state of the charging pile and the quality of the current waveform, to provide load information and quality evaluation criteria for pulse width modulation parameter optimization.

[0053] The specific implementation of step S04 is to perform fast Fourier transform on the collected battery current value waveform and analyze its spectral characteristics. By calculating the sum of the ratios of high harmonic components to fundamental components, a harmonic distortion index is obtained, and the qualified threshold value is set to 0.08, which is lower than the value indicating that the current waveform is close to the ideal state. At the same time, the current rise time, current fall time, voltage response delay time and load coefficient change rate are analyzed, a delay characteristic model is established, and a delay compensation function value is calculated. The function uses a second-order Taylor expansion approximation to output the time compensation amount of the next period pulse width modulation signal, and the compensation range is 0-50μs. The purpose of this step is to analyze the current waveform quality and quantify the system delay characteristics, to provide waveform quality evaluation and delay compensation data for realizing high-precision pulse width modulation signal generation.

[0054] The specific implementation of step S05 is to call a pre-trained pulse modulation neural network model, input six feature parameters including battery state of charge parameters, load coefficient, right side tilt change rate, left side tilt change rate, third order accuracy index and delay compensation function value, and output optimal pulse width modulation parameters including pulse width to period ratio, carrier frequency and modulation depth through model inference calculation. The model inference process uses a forward propagation algorithm, sequentially passes through a feature extraction layer, a residual connection layer and an output mapping layer, and finally generates optimized modulation parameters. The purpose of this step is to use the nonlinear mapping capability of the neural network to extract complex relationships from multi-dimensional input features, generate optimal modulation parameters that can adapt to the current charging environment, and improve charging efficiency and safety.

[0055] The specific implementation of step S06 is to generate a pulse width modulation signal using a digital signal processor based on the optimal pulse width modulation parameters output by step S05. The specific implementation is to set the duty cycle according to the pulse width to period ratio, set the signal period according to the carrier frequency, and set the modulation amplitude according to the modulation depth. The generated pulse width modulation signal is transmitted to the power control unit to adjust the on-time of the power tube, thereby accurately controlling the output power and current waveform of the charging pile. The purpose of this step is to convert the optimized modulation parameters into actual control signals to accurately regulate the output characteristics of the charging pile and ensure the efficiency and stability of the charging process.

[0056] The specific implementation of step S07 is to monitor the battery temperature value and the battery internal resistance value in real time through a temperature sensor and a resistance measurement device, and the sampling frequency is set to 1 Hz. The temperature value and the internal resistance value are subjected to sliding window processing, and the change rates thereof are calculated, the temperature change rate threshold is set to 0.5°C / min, and the internal resistance change rate threshold is set to 3% / h. When the temperature change rate or the internal resistance change rate exceeds the preset threshold, or the fixed time reaches the upper limit value of 10 minutes, the modulation parameter update cycle is triggered, and the closed-loop control cycle is returned to step S01 for execution. The purpose of this step is to monitor the battery health status in real time, ensure the safety and reliability of the charging process, and dynamically adjust the charging strategy according to the battery state change to prevent overcharging or overheating phenomenon.

[0057] The detailed structure of the pulse modulation neural network model is a deep residual network with a two-way parallel architecture. The feature extraction layer is composed of eight fully connected layers, and the number of neurons in each layer is 64, 128, 256, 512, 256, 128, 64, and 32, respectively. The activation function uses the GELU (Gaussian Error Linear Unit) function, which can provide smoother gradient information than the traditional ReLU function, which is beneficial to model convergence. The residual connection layer includes six groups of cross-layer connection units, each group including two fully connected layers and a gated recurrent unit. The gated recurrent unit uses a long short-term memory network structure with a hidden state dimension of 128 and a forget gate bias initialization of 1.0, effectively alleviating the gradient vanishing problem. The output mapping layer includes an adaptive normalization layer and three branch prediction heads. The adaptive normalization layer uses group normalization technology with a group size of 8. Each prediction head is composed of two fully connected layers, and the output layer dimensions correspond to pulse width, carrier frequency, and modulation depth, respectively. The model introduces a multi-head attention mechanism in the middle layer, and the number of attention heads is dynamically adjusted according to the battery state of charge parameter and the load coefficient. The specific calculation formula is The number of heads ranges from 1 to 4, and the query, key, and value matrix dimensions of the attention layer are all 64, which realizes adaptive weighting of different charging stage features and enhances the representation ability of the model.

[0058] The detailed steps for establishing the training data set of the pulse modulation neural network model include: first, selecting a plurality of types of battery samples, including lithium ion batteries, lead-acid batteries, nickel-hydrogen batteries, etc., a total of 12 models; second, designing test scenarios with different charging states, environmental temperatures and load conditions, the charging state covers the full range of SOC from 0.05 to 0.95 at intervals of 0.05; the environmental temperature is from -10 DEG C to 45 DEG C at intervals of 5 DEG C; the load condition is from 20% to 100% of the rated load at intervals of 10%; third, collecting data in a laboratory environment using high-precision test equipment, including battery state-of-charge parameters, load coefficients, right side tilt rate, left side tilt rate, third-order accuracy indicators, and delay compensation function values as input features, and corresponding optimal pulse width modulation parameters as labels; then, cleaning and preprocessing the collected data, including removing outliers, filling missing values, and feature standardization; finally, dividing into training set, validation set and test set according to the ratio of seven to two to one, the training set data amount reaches 500,000, ensuring sufficient sample distribution for model learning. The training process uses a small batch gradient descent method with a batch size of 256, the initial learning rate is set to 0.001, the cosine annealing strategy is used to dynamically adjust the learning rate during the training process, and the minimum learning rate is 0.00001; the optimizer selects the Adam algorithm with a momentum term, the momentum parameter is set to 0.9, and the weight decay coefficient is 0.0001; batch normalization layer is introduced in the training process to reduce internal covariate shift, and random dropout technology is applied to prevent overfitting, and the dropout rate is set to 0.3; when the performance of the validation set does not improve significantly for five consecutive training periods, the learning rate is reduced to 0.1 times of the original; training is stopped when it converges or reaches the maximum number of cycles 200, and finally the model is evaluated on the test set to ensure that the generalization ability meets the requirements of practical applications and is deployed to the charging pile control system.

[0059] It should be noted that the first main technical idea of the present application is a multi-dimensional battery state representation mechanism, by simultaneously collecting battery voltage and current values, calculating battery state-of-charge parameters and left and right side tilt rates, a comprehensive representation model of battery charging state is constructed. Unlike the traditional method which only relies on simple voltage or current monitoring, the multi-dimensional representation method used in the present application can accurately capture the microscopic change characteristics of the battery at different charging stages. This comprehensive state perception mechanism enables the system to accurately determine the actual charging acceptance ability of the battery, avoiding the overcharging or undercharging problems caused by state judgment deviation in traditional methods, and at the same time providing a more reliable decision basis for subsequent parameter optimization, effectively improving the safety and efficiency of the charging process.

[0060] The second main technical idea is an adaptive modulation parameter generation technology based on a deep residual neural network. The application constructs a deep residual network model with a two-way parallel architecture, processes time sequence features by introducing residual connection layers and gated recurrent units, and realizes differentiated processing of features in different charging stages by fusing a multi-head attention mechanism. Compared with traditional fixed parameter modulation methods or simple feedback control mechanisms, this deep learning method can establish a nonlinear mapping relationship between the battery state and the modulation parameter, has stronger environmental adaptability and parameter optimization accuracy. In particular, the introduction of the multi-head attention mechanism enables the system to dynamically adjust the feature weight according to the charging stage, optimizes the modulation parameter in a targeted manner, and significantly improves the robustness and accuracy of the model under complex load conditions.

[0061] The third main technical idea is a dynamic parameter updating mechanism based on a closed-loop feedback, which triggers a modulation parameter updating period when the change rate of the battery temperature value and the internal resistance value exceeds a preset threshold, forming a complete closed-loop control loop. Compared with the traditional fixed period parameter updating method, this mechanism can dynamically adjust the control response speed according to the urgency of the battery state change, improve the processing capability of sudden conditions while ensuring the stability of the system. By responding to abnormal battery states in a timely manner, the control deviation accumulation problem caused by parameter updating lag in traditional methods is avoided, effectively reducing the safety risk in the charging process.

[0062] The synergistic effect of the above three main technical ideas forms a complete charging pile adaptive pulse width modulation method system. Multi-dimensional state representation provides comprehensive and accurate battery state information for the system, the deep residual neural network model generates optimal pulse width modulation parameters according to these information, and the closed-loop feedback mechanism ensures the dynamic stability and real-time response capability of the entire system. The organic combination of the three breaks through the adaptability bottleneck of traditional charging control technology in complex environments, realizes precise, efficient and safe control of the charging process. Compared with traditional methods, this synergistic effect not only improves the charging efficiency and current waveform quality, reduces harmonic distortion, but also prolongs the service life of the battery and enhances the adaptability of the charging pile to power grid fluctuations.

[0063] The second aspect of the application provides a computer readable storage medium, the computer readable storage medium stores program instructions, the program instructions are run in the computer, and are used for executing the charging pile adaptive pulse width modulation method.

[0064] The third aspect of the application provides a charging pile adaptive pulse width modulation system, comprising the computer readable storage medium, the system is any one of a computer, a server and a single chip microcomputer, the computer readable storage medium is arranged in the system, and the system is provided with a microprocessor for executing the program instructions stored in the computer readable storage medium.

[0065] Specifically, the principle of the present application is:

[0066] The core principle of the charging pile adaptive pulse width modulation method of the present application is to establish a dynamic mapping relationship between the battery charging state and the pulse width modulation parameters. Through real-time data acquisition, multi-dimensional feature extraction and neural network model inference, intelligent adaptive control of the charging process is realized. The working principle can be described from the following aspects:

[0067] Firstly, the present application uses multi-source data fusion technology to comprehensively perceive the battery charging state. By collecting battery voltage and current values, calculating battery state of charge parameters, and combining left and right side tilt rate calculations, the dynamic characteristics of the battery charging curve are fully characterized. This multi-dimensional state characterization method breaks through the limitations of traditional methods that rely only on single voltage or current parameters, and can more accurately reflect the actual state and acceptance ability of the battery at different charging stages.

[0068] Secondly, the present application introduces a load perception algorithm and Fourier analysis technology to achieve accurate perception of the charging environment. By using Kalman filtering to eliminate random fluctuations, load coefficients and third-order accuracy indicators are calculated; at the same time, based on Fourier transform analysis of current waveform, harmonic distortion index is obtained and delay compensation function value is calculated. The comprehensive application of these technologies enables the system to accurately identify the complex changes of the charging environment, providing a reliable basis for pulse width modulation parameter optimization.

[0069] Thirdly, the core innovation of the present application lies in the use of a deep residual neural network model with a dual-path parallel architecture to process multi-dimensional feature data. This model extracts battery charging characteristics from input data through a feature extraction layer, processes time series features using a residual connection layer and a gated recurrent unit, and finally generates optimal pulse width modulation parameters through an output mapping layer. In particular, the multi-head attention mechanism introduced in the model can dynamically adjust the number of attention heads according to the battery state of charge and load conditions, enabling differentiated processing of features at different charging stages, thereby improving the model's ability to adapt to the complexity of the charging process.

[0070] Finally, the present application ensures system stability through a closed-loop control cycle mechanism. Real-time monitoring of battery temperature and internal resistance values triggers modulation parameter updates when the rate of change exceeds a pre-set threshold, forming a complete feedback loop from state acquisition to parameter optimization. This mechanism enables the system to respond promptly to changes in battery state, avoiding overcharging or undercharging, while preventing efficiency reduction or safety hazards caused by parameter mismatch.

[0071] In conclusion, the application combines four core technologies of multi-dimensional state perception, accurate environment identification, deep learning modeling and closed-loop feedback control to build a complete charging pile adaptive pulse width modulation method, which can effectively solve the technical problem of accurately adapting pulse width modulation parameters to battery charging state changes under complex dynamic load conditions, and realize efficient, safe and intelligent control of the charging process.

[0072] A specific embodiment 1 of the application is provided below, and the specific implementation of each step in embodiment 1 is described in detail as follows.

[0073] The specific implementation of step S01 is to collect the battery voltage value and battery current value at the output end of the charging pile through a high-precision sampling circuit, and the sampling frequency is set to 10 kHz to ensure the accuracy of signal acquisition. After the collection is completed, the battery state of charge parameter is calculated based on the coulomb counting method, that is, the ratio of the current capacity to the rated capacity, and the calculation formula is In the formula, SOC is the battery state of charge parameter, and the value range is 0 to 1; Q current is the current capacity, and the unit is ampere-hour (Ah); Q rated is the rated capacity, and the unit is ampere-hour (Ah). At the same time, the differential calculation is performed on the continuously sampled battery voltage value, and the exponential weighted moving average algorithm is used to process the differential result to obtain the right side tilt change rate R slope , and the calculation formula is In the formula, R slope is the right side tilt change rate, and the unit is volt per second (V / s); α is the smoothing coefficient, and the value range is 0.1 to 0.3; V(t) is the battery voltage value at time t, and the unit is volt (V); Δt is the sampling time interval, and the unit is second (s); is the right side tilt change rate of the last calculation period, and the unit is volt per second (V / s). The purpose of this step is to obtain the current state parameter of the battery, to provide basic data for subsequent charging strategy optimization, and to ensure accurate control of the charging process according to the actual state of the battery.

[0074] The specific implementation of step S02 is to select a matching charging curve from the pre-set charging curve database according to the battery state of charge parameter calculated in step S01. Specifically, when SOC<0.2, the constant current charging segment parameters are selected; when 0.2≤SOC<0.8, the constant voltage charging segment parameters are selected; and when SOC≥0.8, the trickle charging segment parameters are selected. Each stage corresponds to a different set of modulation parameters, including the basic pulse width ratio D base , the basic carrier frequency f base and the basic modulation depth M base . At the same time, the left side tilt change rate L slope is calculated by performing piecewise linear regression analysis on the voltage curve in the historical data window, and the calculation formula is wherein L slope is the left side slope rate, in units of volts per second (V / s); t i is the time of the i-th sample point, in units of seconds (s); is the average of all sample point times within the history window, in units of seconds (s); V i is the voltage value of the i-th sample point, in units of volts (V); is the average of all sample point voltage values within the history window, in units of volts (V); n is the number of sample points within the history window, with the window size set to the past 60 seconds of sample data. The purpose of this step is to determine the basic modulation parameters required for the current charging stage, and to obtain the battery charging history response characteristics, providing a reference basis for subsequent pulse width modulation optimization.

[0075] The specific implementation of step S03 is to monitor the load fluctuation state of the charging pile in real time by measuring the power difference between the input and output ends of the charging pile. The Kalman filtering algorithm is used to process the power difference data, and the filter state equation is x k = Ax k-1 + w k-1 , and the measurement equation is z k = Hx k + v k ; in the formula, x k is the system state vector at time k, including the power difference and its rate of change; A is the state transition matrix; w k-1 is the process noise, which is subject to a normal distribution with a mean of 0 and a covariance of Q, and Q is set to 0.01; z k is the measurement value at time k, i.e., the direct measurement result of the power difference; H is the measurement matrix; v k is the measurement noise, which is subject to a normal distribution with a mean of 0 and a covariance of R, and R is set to 0.1. The prediction step of Kalman filtering is The update step is in the formula, is the prior state estimate; is the prior estimate error covariance; K k is the Kalman gain; is the posterior state estimate; P k is the posterior estimate error covariance; I is the identity matrix. The real-time load utilization rate is calculated by filtering the power data, and the load coefficient L f is obtained, and the calculation formula is in the formula, L f is the load coefficient, dimensionless; P actual is the current actual load power, in units of watts (W); P ratedThe rated load power is in watts (W). Meanwhile, a third-order polynomial is fitted to the battery current value waveform, and the calculation formula is I(t) = a0 + a1t + a2t 2 +a3t 3 ; in the formula, I(t) is the current value at time t, in amperes (A); a0, a1, a2, and a3 are fitting coefficients, which are solved by the least squares method. The residual sum of squares is calculated to obtain a third-order accuracy index P3, and the calculation formula is ; in the formula, P3 is the third-order accuracy index, in amperes squared (A 2 ); I i is the actual current value of the i-th sampling point, in amperes (A); I(t i ) is the fitted current value of the i-th sampling point, in amperes (A); and m is the number of sampling points in the fitting interval. The third-order accuracy index threshold is set to 0.05, and a value lower than this threshold indicates that the current waveform quality is good. The purpose of this step is to accurately obtain the charging pile working state and the current waveform quality, and to provide load information and quality evaluation criteria for pulse width modulation parameter optimization.

[0076] The specific implementation of step S04 is to perform a fast Fourier transform on the collected battery current value waveform to analyze its frequency spectrum characteristics. The Fourier transform calculation formula of the current waveform is ; in the formula, X(k) is the frequency domain representation; x(n) is the time domain current sampling value; N is the number of sampling points; j is the imaginary unit; k is the frequency index, k = 0, 1, 2, …, N-1. The sum of the ratios of high harmonic components to fundamental components is calculated to obtain a total harmonic distortion index THD, and the calculation formula is ; in the formula, THD is the total harmonic distortion index, dimensionless; X(h) is the amplitude of the h-th harmonic component; X(1) is the amplitude of the fundamental component; and H is the highest harmonic number considered, usually taking a value of 20. The index qualified threshold is set to 0.08, and a value lower than this value indicates that the current waveform is close to the ideal state. At the same time, the current rise time t rise , the current fall time t fall , the voltage response delay time t delay , and the load coefficient change rate A delay characteristic model is established to calculate a delay compensation function value D comp , and the calculation formula is ; in the formula, D comp is the delay compensation function value, in microseconds (μs); β1, β2, β3, β4, β5, and β6 are compensation coefficients, which are obtained through experimental calibration; t rise is the current rise time, in microseconds (μs); t fall is the current fall time, in microseconds (μs); and t delayVoltage response delay time, in microseconds (μs); This represents the load coefficient change rate, expressed in seconds (1 / s). The function uses a second-order Taylor expansion approximation to output the time compensation amount for the next cycle of the pulse width modulation signal, with a compensation range of 0–50 μs. The purpose of this step is to analyze the current waveform quality and quantify the system delay characteristics, providing waveform quality assessment and delay compensation data for achieving high-precision pulse width modulation signal generation.

[0077] The specific implementation of step S05 involves calling a pre-trained pulse modulation neural network model, inputting six feature parameters: battery state of charge parameters, load factor, right tilt rate of change, left tilt rate of change, third-order accuracy index, and delay compensation function value. After model inference and calculation, the optimal pulse width modulation parameters are output, including three key parameters: the ratio of pulse width to period, carrier frequency, and modulation depth. The model inference process uses a forward propagation algorithm, sequentially passing through a feature extraction layer, a residual connection layer, and an output mapping layer, ultimately generating the optimized modulation parameters. The model input feature vector is represented as X = [x1, x2, x3, x4, x5, x6]. T In the formula, X is the input feature vector; x1 is the battery state of charge (SOC) parameter; and x2 is the load factor L. f x3 represents the rate of change of the right-side tilt, R. slope x4 represents the rate of change of the leftward tilt, L. slope x5 represents the third-order precision index P3; x6 represents the delay compensation function value D. comp The model output vector is represented as Y = [y1, y2, y3]. T In the formula, Y is the output vector; y1 is the ratio D of pulse width to period, ranging from 0.1 to 0.9; y2 is the carrier frequency f. c The modulation depth M ranges from 10kHz to 100kHz; y3 is the modulation depth M, ranging from 0.2 to 1.0. The purpose of this step is to utilize the nonlinear mapping capability of neural networks to extract complex relationships from multidimensional input features, generate optimal modulation parameters that can adapt to the current charging environment, and improve charging efficiency and safety.

[0078] The specific implementation of step S06 is based on the optimal pulse width modulation parameters output in step S05, using a digital signal processor to generate a pulse width modulation signal. Specifically, it involves setting the duty cycle according to the ratio D of the pulse width to the period, and setting the carrier frequency f... c Set signal period The modulation amplitude is set according to the modulation depth M. The expression for the generated pulse width modulation signal is S. PWM (t)=A·[u(t mod T c )-u(t mod T c -D·T c )];In the formula, SPWM (t) represents the pulse width modulation signal; A is the signal amplitude in volts (V); u(t) is the unit step function; t mod T c Indicates t with respect to T c Modulo operation; D is the duty cycle; T c The carrier period is expressed in seconds (s). In practical applications, the effect of the modulation depth M is considered, and the effective duty cycle D is... eff Calculated as D eff =D·M. The generated pulse width modulation signal is transmitted to the power control unit to adjust the conduction time of the power transistor, thereby precisely controlling the output power and current waveform of the charging pile. The purpose of this step is to convert the optimized modulation parameters into actual control signals, realize precise regulation of the charging pile's output characteristics, and ensure the efficiency and stability of the charging process.

[0079] The specific implementation of step S07 involves real-time monitoring of the battery temperature and internal resistance using a temperature sensor and a resistance measuring device, with a sampling frequency set to 1Hz. A sliding window processing method is applied to the temperature and internal resistance values, and their rate of change is calculated. The rate of temperature change... The calculation formula is: In the formula, Δt represents the rate of temperature change, in degrees Celsius per minute (°C / min); T(t) is the battery temperature at time t, in degrees Celsius (°C); Δt m This represents the time interval for temperature measurements, in minutes (min). Rate of change of internal resistance. The calculation formula is: In the formula, R(t) is the rate of change of internal resistance, expressed as a percentage per hour (% / h); R(t) is the battery internal resistance at time t, expressed in ohms (Ω); Δt r The internal resistance measurement time interval is in hours (h). The temperature change rate threshold is set to 0.5℃ / min, and the internal resistance change rate threshold is set to 3% / h. When the temperature change rate or internal resistance change rate exceeds the preset threshold, or when the fixed time reaches the update upper limit of 10 minutes, the modulation parameter update cycle is triggered, and the process returns to step S01 to execute the closed-loop control cycle. The purpose of this step is to monitor the battery health status in real time, ensure the charging process is safe and reliable, and dynamically adjust the charging strategy according to changes in battery status to prevent overcharging or overheating.

[0080] like Figure 2As shown, the detailed structure of the pulse modulation neural network model is a deep residual network with a two-way parallel architecture. The feature extraction layer is composed of eight fully connected layers, with the number of neurons in each layer being 64, 128, 256, 512, 256, 128, 64, and 32, respectively. The activation function uses the GELU function, which can provide smoother gradient information than the traditional ReLU function, facilitating model convergence. The residual connection layer contains six groups of cross-layer connection units, each containing two fully connected layers and a gated recurrent unit. The gated recurrent unit uses a long short-term memory network structure with a hidden state dimension of 128 and a forget gate bias initialization of 1.0, effectively alleviating the gradient vanishing problem. The output mapping layer includes an adaptive normalization layer and three branch prediction heads. The adaptive normalization layer uses group normalization technology with a group size of 8. Each prediction head is composed of two fully connected layers, with the output layer dimensions corresponding to pulse width, carrier frequency, and modulation depth, respectively. The model introduces a multi-head attention mechanism in the middle layer, with the number of attention heads dynamically adjusted according to the battery state of charge parameter and the load coefficient. The specific calculation formula is where heads is the number of attention heads, ranging from 1 to 4; SOC is the battery state of charge parameter; L f is the load coefficient; represents the floor operation. The query, key, and value matrix dimensions of the attention layer are all 64, realizing adaptive weighting of features in different charging stages and enhancing the representation ability of the model.

[0081] The detailed steps for establishing the training data set of the pulse modulation neural network model include: first, selecting multiple types of battery samples, including lithium-ion batteries, lead-acid batteries, nickel-hydrogen batteries, etc., a total of 12 models; second, designing test scenarios with different charging states, environmental temperatures, and load conditions. The charging state covers the full range of SOC from 0.05 to 0.95 with an interval of 0.05. The environmental temperature ranges from -10°C to 45°C with an interval of 5°C. The load condition ranges from 20% to 100% of the rated load with an interval of 10%. Third, using high-precision test equipment to collect data in a laboratory environment, including battery state of charge parameters, load coefficients, right-side inclination rate, left-side inclination rate, three-order accuracy indicators, and delay compensation function values as input features, and the corresponding optimal pulse width modulation parameters as labels. Then, the collected data is cleaned and preprocessed, including removing outliers, filling missing values, and feature standardization. Finally, the data is divided into training set, validation set, and test set in the ratio of seven to two to one, with the training set data amount reaching 500,000 to ensure sufficient sample distribution for model learning. The standardization processing uses the min-max normalization method, with the calculation formula being where x norm is the normalized feature value; x is the original feature value; x min is the minimum feature value; x maxis the maximum value of the feature. The training process uses the mini-batch gradient descent method with a batch size of 256, the initial learning rate is set to 0.001, and the cosine annealing strategy is used to dynamically adjust the learning rate during the training process. The learning rate adjustment formula is wherein, η t is the learning rate at time t; η min is the minimum learning rate, set to 0.00001; η max is the maximum learning rate, set to 0.001; t is the current training round; T is the total training round. The optimizer uses the Adam algorithm with a momentum term, the momentum parameter is set to 0.9, and the weight decay coefficient is 0.0001; during the training process, the batch normalization layer is introduced to reduce the internal covariate shift phenomenon, and the random dropout technique is applied to prevent overfitting, with a dropout rate of 0.3; when the validation set performance does not improve significantly for five consecutive training periods, the learning rate is reduced to 0.1 times the original; training is stopped when it converges or reaches the maximum number of cycles 200, and finally the model is evaluated on the test set to ensure that the model generalization ability meets the actual application requirements before deploying it to the charging pile control system.

[0082] In order to better understand and implement the present application, the following provides an embodiment 2 of a specific application scenario of the present application: In this embodiment, a 120kWh ternary lithium battery pack is selected for testing, and the charging process is set to standard laboratory conditions at an ambient temperature of 25℃. The initial battery state of charge parameter SOC is 0.15, indicating that the battery is in a low power state and needs to be charged at a constant current. The system uses a sampling frequency of 10kHz to monitor the battery voltage and current in real time.

[0083] In the initial charging stage, the system collects the battery voltage value of 338.5V and the current value of 125.8A through step S01, and calculates the battery state of charge parameter SOC as 0.15. At the same time, the right side slope rate R slope is calculated to be 0.42V / s through continuous sampling, indicating that the battery voltage rising trend is relatively stable. Subsequently, the system determines that it is currently in the constant current charging stage in step S02, selects the corresponding basic pulse width ratio D base is 0.68, the basic carrier frequency f base is 25kHz, and the basic modulation depth M base is 0.85. Through analysis of historical data, the left side slope rate L slope is 0.38V / s.

[0084] The current actual load of the charging pile is 82.5kW, and the load coefficient L fis 0.825. After the battery current waveform is fitted by a third-order polynomial, a third-order precision index P3 of 0.032 is obtained, which is lower than the set threshold 0.05, indicating that the current waveform quality is good. In step S04, by performing a fast Fourier transform on the current waveform, a harmonic distortion index THD of 0.065 is calculated, which is better than the set threshold 0.08, and a delay compensation function value D comp is 28.6 μs.

[0085] As shown in Table 1, the measured parameter data of the system in different charging stages:

[0086] Table 1 Measured parameter data in different charging stages

[0087]

[0088] In step S05, the system inputs the above parameters into the pulse modulation neural network model, and calculates the optimal pulse width modulation parameters by a forward propagation algorithm: a duty cycle D of 0.72, a carrier frequency f c of 32 kHz, and a modulation depth M of 0.91. Subsequently, in step S06, the system generates a pulse width modulation signal according to these parameters and transmits it to the power control unit.

[0089] During the entire charging process, the system continuously monitors the battery temperature and internal resistance value, as shown in Table 2:

[0090] Table 2 Battery temperature and internal resistance monitoring data during charging

[0091]

[0092] When the battery temperature change rate exceeds the preset threshold 0.5℃ / min or the internal resistance change rate exceeds 3% / h, the system triggers the modulation parameter update period and re-executes the closed-loop control cycle. In this embodiment, neither the temperature change rate nor the internal resistance change rate exceeds the threshold during the charging process, and the system updates the parameters at a fixed time interval of 10 minutes.

[0093] After the charging test is completed, the performance comparison data shown in Table 3 is obtained by comparing the traditional fixed parameter pulse width modulation method with the adaptive pulse width modulation method of the present application:

[0094] Table 3 Performance comparison of different charging methods

[0095]

[0096]

[0097] The traditional charging pile pulse width modulation technology mainly adopts a fixed parameter or a simple segmented adjustment method, which cannot dynamically optimize according to the real-time state of the battery and the load condition of the charging pile. This method has poor adaptability under different charging stages and different load conditions, resulting in low charging efficiency and easy generation of large harmonic distortion, which poses a potential risk to the battery life and charging safety.

[0098] The adaptive pulse width modulation method of the present application realizes accurate control of the charging process by introducing a deep learning model and multi-dimensional parameter perception. Compared with the traditional method, the present application improves the charging efficiency by 6.4%, shortens the charging time by 7.7%, reduces the temperature peak by 9.0%, and reduces the average value of harmonic distortion by 15.2%. More importantly, by reducing the temperature fluctuation and current waveform distortion during the charging process, the present application method is expected to extend the battery cycle life by about 8.5%. In practical application, this method also shows stronger environmental adaptability and battery type compatibility, especially in scenes with large temperature fluctuations or different degrees of battery aging, its advantages are more obvious.

[0099] It should be noted that the variables involved in the present application are explained in detail as shown in Tables 4 and 5.

[0100] Table 4 Variable Explanation Table (First Part)

[0101]

[0102]

[0103] Table 5 Variable Explanation Table (Second Part)

[0104]

[0105] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for adaptive pulse width modulation of charging piles, characterized in that, include: Collect the battery voltage and current values ​​at the output end of the charging pile, and calculate the battery state-of-charge parameters and the rate of change of right-side tilt. The system determines a preset charging curve based on battery state-of-charge parameters, obtains the modulation parameters and left tilt rate of change required for the current charging stage, monitors the load fluctuation status of the charging pile, and calculates the load factor and third-order accuracy index using a load sensing algorithm. It analyzes the battery current waveform based on Fourier transform, obtains the harmonic distortion index, and calculates the delay compensation function value. It calls the pulse modulation neural network model, inputting battery state-of-charge parameters, load factor, right tilt rate of change, left tilt rate of change, third-order accuracy index, and delay compensation function value, and outputs the optimal pulse width modulation parameters. It generates a pulse width modulation signal based on the optimal pulse width modulation parameters and adjusts the charging pile's output characteristics. It monitors the battery temperature and internal resistance in real time; when the rate of change of battery temperature or internal resistance exceeds a preset threshold, it triggers a modulation parameter update cycle and returns to the first step to execute the closed-loop control loop.

2. The adaptive pulse width modulation method for charging piles according to claim 1, characterized in that, The battery state-of-charge parameter specifically refers to the ratio of the battery's current charge to its rated charge, calculated using coulometrics. The value ranges from 0 to 1, representing the percentage of battery charge and used to determine the charging stage of the battery.

3. The adaptive pulse width modulation method for charging piles according to claim 2, characterized in that, The preset charging curve refers to the charging current and voltage relationship graph that is pre-set according to the characteristics of different types of batteries. It includes three stages: constant current charging stage, constant voltage charging stage, and trickle charging stage, with each stage corresponding to a different modulation strategy.

4. The adaptive pulse width modulation method for charging piles according to claim 3, characterized in that, The load sensing algorithm specifically measures the power difference between the input and output ends of the charging pile, combines it with historical load data, and uses the Kalman filter method to eliminate the influence of random fluctuations, thereby calculating the real-time load utilization rate and its changing trend.

5. The adaptive pulse width modulation method for charging piles according to claim 4, characterized in that, The harmonic distortion index specifically refers to the sum of the ratios of higher harmonic components to the fundamental component in the current waveform calculated by Fourier transform. It is used to measure the quality of the current waveform; the lower the value, the closer the current waveform is to the ideal state.

6. The adaptive pulse width modulation method for charging piles according to claim 5, characterized in that, The optimal pulse width modulation parameters specifically refer to three key parameters: the ratio of pulse width to period, carrier frequency, and modulation depth. These parameters directly determine the waveform characteristics of the charging current and the energy conversion efficiency.

7. The adaptive pulse width modulation method for charging piles according to claim 6, characterized in that, The modulation parameter update cycle is a dynamic time interval calculated by the system based on the rate of change of battery temperature and the rate of change of battery internal resistance. When the rate of change exceeds the preset threshold or the fixed time reaches the upper limit, a new round of parameter optimization calculation is triggered.

8. The adaptive pulse width modulation method for charging piles according to claim 7, characterized in that, The specific structure of the pulse modulation neural network model is a deep residual network with a dual-path parallel architecture, which includes three main parts: feature extraction layer, residual connection layer and output mapping layer. A multi-head attention mechanism is introduced in the middle layer. The number of attention heads in the multi-head attention mechanism is determined by the battery state of charge parameter and load factor.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions, which, when executed in a computer, are used to perform the adaptive pulse width modulation method for charging piles as described in any one of claims 1-8.

10. A charging pile adaptive pulse width modulation system, characterized in that, The system includes the computer-readable storage medium of claim 9, wherein the system is any one of a computer, a server, or a microcontroller, the computer-readable storage medium is disposed within the system, and the system is provided with a microprocessor that executes the program instructions stored in the computer-readable storage medium.

Images