Charging pile dynamic power allocation method and device based on load prediction

Abstract

The present application relates to the technical field of charging pile power distribution, and discloses a method and device for dynamic power distribution of charging piles based on load forecasting, the method comprising: performing system identification analysis on real-time operation data of charging piles, and constructing a small signal model of the charging pile system; training to obtain a low-sensitivity load forecasting model and a high-sensitivity load forecasting model, and generating charging load forecast data; performing control optimization calculations to obtain dynamic power compensation control parameters; performing genetic optimization and multi-level power distribution to obtain a multi-level power distribution scheme including a conventional load distribution mode and a peak load distribution mode; performing power limitation by reducing the charging power of non-priority users, and performing load transfer by coordinating the power of adjacent charging piles to obtain emergency power regulation control instructions. The present application adopts a closed-loop optimization mechanism, and ensures the stable operation of the charging pile system under various working conditions through online parameter optimization and dynamic adjustment of control strategies.

Description

Technical Field

[0001] The present application relates to the technical field of charging pile power distribution, and in particular to a method and device for dynamic power distribution of charging piles based on load prediction. Background Art

[0002] As charging demand continues to increase, charging pile systems face challenges such as low load forecasting accuracy, poor power allocation efficiency, and poor system stability. These issues stem primarily from the randomness and uncertainty of EV users' charging behavior, making it difficult to accurately predict charging loads, which in turn impacts the power scheduling and allocation of charging piles.

[0003] Current charging pile systems generally employ fixed power allocation strategies, which are unable to dynamically adjust power output based on real-time load fluctuations, easily leading to system overloads during peak load periods. Furthermore, existing power allocation methods fail to consider user charging priorities, making it impossible to rationally allocate charging resources when system capacity is limited, impacting the user charging experience. Furthermore, traditional charging pile control systems struggle to cope with power fluctuations caused by random charging demand under large-scale access conditions due to the lack of effective load forecasting and dynamic power compensation mechanisms. Particularly during peak charging periods, the lack of effective dynamic power allocation measures can easily lead to localized charging pile overloads or grid fluctuations, impacting the safe and stable operation of the entire charging service system. Summary of the Invention

[0004] The present application provides a method and device for dynamic power distribution of charging piles based on load forecasting. The present application adopts a closed-loop optimization mechanism to ensure the stable operation of the charging pile system under various working conditions through online parameter optimization and dynamic adjustment of control strategies.

[0005] In a first aspect, the present application provides a method for dynamic power allocation of a charging pile based on load prediction, the method comprising:

[0006] Conduct system identification and analysis on the real-time operation data of charging piles and build a small signal model of the charging pile system;

[0007] Based on the charging pile system small signal model, a low-sensitivity load prediction model and a high-sensitivity load prediction model are trained to generate charging load prediction data;

[0008] According to the deviation value between the charging load prediction data and the actual load data, a control optimization calculation is performed on the proportional compensation unit and the integral compensation unit to obtain a dynamic power compensation control parameter;

[0009] Based on the dynamic power compensation control parameters, genetic optimization and multi-level power distribution are performed on the power distribution ratio of the charging pile to obtain a multi-level power distribution scheme including a normal load distribution mode and a peak load distribution mode;

[0010] According to the multi-level power allocation scheme, power limitation is performed by reducing the charging power of non-priority users, and load transfer is performed through power coordination of adjacent charging piles to obtain an emergency power adjustment control instruction.

[0011] A second aspect of the present application provides a charging pile dynamic power distribution device based on load prediction, the charging pile dynamic power distribution device based on load prediction comprising:

[0012] Identification module, used to perform system identification analysis on the real-time operation data of charging piles and build a small signal model of the charging pile system;

[0013] A training module, configured to train a low-sensitivity load prediction model and a high-sensitivity load prediction model based on the charging pile system small signal model, and generate charging load prediction data;

[0014] a calculation module, configured to perform control optimization calculation on a proportional compensation unit and an integral compensation unit according to a deviation between the charging load prediction data and the actual load data, and obtain dynamic power compensation control parameters;

[0015] A distribution module is used to perform genetic optimization and multi-level power distribution on the charging pile power distribution ratio based on the dynamic power compensation control parameters to obtain a multi-level power distribution scheme including a normal load distribution mode and a peak load distribution mode;

[0016] The processing module is used to perform power limitation by reducing the charging power of non-priority users according to the multi-level power allocation scheme, and to transfer the load through power coordination of adjacent charging piles to obtain an emergency power adjustment control instruction.

[0017] 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 dynamic power allocation method for charging piles based on load prediction.

[0018] A fourth aspect of the present application provides a computer-readable storage medium, which stores instructions. When the computer-readable storage medium is run on a computer, the computer executes the above-mentioned dynamic power allocation method for charging piles based on load prediction.

[0019] Compared with the existing technology, the present application has the following beneficial effects: by establishing a small signal model of the charging pile system and combining it with the system identification method to accurately describe the operating characteristics of the charging pile, the limitations of the traditional modeling method in large-scale charging pile systems are overcome; a dual-model prediction strategy combining the response surface method and the neural network is adopted to perform load prediction for normal working conditions and peak working conditions respectively, which significantly improves the accuracy of load prediction; a dynamic power compensator is designed based on the pole configuration method, and the influence of load prediction deviation on system stability is effectively suppressed by optimizing the proportional integral control parameters; a genetic algorithm is used to perform multi-objective optimization of the charging pile power distribution, which minimizes the load fluctuation while ensuring the minimum system loss; an emergency power adjustment mechanism based on user priority is designed, which solves the system overload problem during the charging peak period by combining power limitation and load transfer; the present application adopts a closed-loop optimization mechanism, and ensures the stable operation of the charging pile system under various working conditions through online parameter optimization and dynamic adjustment of the control strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. 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 any creative work.

[0021] The structures, proportions, sizes, etc. depicted 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 intended to limit the conditions under which the present invention can be implemented and therefore have no substantive technical significance. Any structural modifications, changes in proportional relationships, or adjustments in size should still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and objectives that can be achieved by the present invention.

[0022] Figure 1 1 is a flow chart of a method for dynamic power allocation of charging piles based on load forecasting provided by an embodiment of the present invention;

[0023] Figure 2 1 is a schematic block diagram of the structure of a dynamic power distribution device for a charging pile based on load prediction provided by an embodiment of the present invention;

[0024] Figure 3 It is a schematic block diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0026] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined, or partially merged, so the actual execution order may vary depending on the actual situation.

[0027] It should also be understood that the terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit the present application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0028] It should be further understood that the term "and / or" used in this specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations. Figure 1 In one embodiment of the present application, a method for dynamic power allocation of charging piles based on load forecasting includes:

[0029] Step 100: Perform system identification analysis on the real-time operation data of the charging pile and construct a small signal model of the charging pile system;

[0030] It is understandable that the execution subject of this application can be a dynamic power distribution device for charging piles based on load forecasting, or a terminal or a server, which is not limited here. The embodiment of this application is described by taking the server as the execution subject as an example.

[0031] Specifically, data is collected on the charging pile's output power, local user load, and charging demand data to determine the real-time operating status of the charging pile. The collected data includes dynamic information such as charging current, voltage, power, charging mode, and user load, covering the diverse operating states of the system under different load conditions. The real-time operating data of the charging pile is preprocessed to ensure the accuracy and effectiveness of subsequent analysis. Data preprocessing includes operations such as outlier removal, noise filtering, data interpolation, and smoothing to eliminate external interference or acquisition errors, generating a preprocessed operating data series with good continuity and consistency. The preprocessed operating data series is input into a recursive least squares algorithm model, and the rated power coefficient and power regulation coefficient of the charging pile are identified through iterative calculations. The recursive least squares algorithm is an online identification tool that can dynamically update parameter estimation results based on time series data, making it suitable for use in real-time systems. This algorithm generates a set of system characteristic parameters that characterize the dynamic characteristics of the charging pile, reflecting its core physical properties and operating status. Based on this set of system characteristic parameters, a state-space model is used to model and analyze the steady-state operating characteristics of the charging pile under rated conditions to construct a steady-state operating model. The state-space model describes the dynamic behavior of the system under the influence of external inputs. Its mathematical expression consists of state equations and output equations, which comprehensively reflect the dynamic relationship between the input and output of the charging pile under steady-state conditions. During the modeling process, attention was paid to stability and sensitivity near the rated power point to ensure that the model accurately describes the steady-state characteristics of the charging pile under normal operating conditions. Simultaneously, the transfer function method was used to model and analyze the response characteristics of power fluctuations caused by random charging demand. The transfer function method is a frequency-domain analysis tool used to describe the dynamic relationship between the input and output of the system. In this stage, a dynamic response model of the charging pile under random charging demand perturbations was constructed using the system's characteristic parameter set, thereby revealing the charging pile's response characteristics to external load perturbations. The dynamic response model is suitable for describing the dynamic characteristics of the charging pile over short periods of time. The steady-state operation model and the dynamic response model are decoupled for calculation, and their independent characteristics in the frequency domain are extracted separately. By mapping the steady-state model and the dynamic model into the frequency domain, the frequency-domain characteristic equations of the charging pile system are obtained. The frequency-domain characteristic equations are converted into time-domain state equations using an inverse Laplace transform to more intuitively describe the system's dynamic behavior, resulting in a small-signal mathematical model of the charging pile system. Eigenvalue analysis and stability criterion calculations are performed on the small-signal mathematical model. By solving the eigenvalues ​​of the state equation, the system's dynamic stability and response characteristics are determined, and model parameters are optimized to better reflect actual operating conditions. The stability criterion calculation comprehensively assesses the system's stability based on the eigenvalue distribution to ensure model reliability. Combining the results of the eigenvalue analysis and stability criterion calculations, a small-signal model of the charging pile system is derived, which includes the charging pile's rated power parameters, load fluctuation parameters, and system response parameters.

[0032] Step 200: Based on the charging pile system small signal model, a low-sensitivity load prediction model and a high-sensitivity load prediction model are trained to generate charging load prediction data;

[0033] Specifically, real-time operating data is classified into operating conditions based on the small-signal model of the charging pile system. By analyzing the amplitude of load fluctuations, the operating data is divided into normal operating condition data and peak operating condition data. This classification is based on the different sensitivities of load changes to the dynamic response of the charging pile system. Normal operating condition data describes stable load fluctuations, while peak operating condition data characterizes sudden and drastic load changes. A response surface polynomial function model is constructed based on the normal operating condition data. By introducing a polynomial expression of linear, square, and cross terms, this model can capture the global trends and interrelationships of the charging load under normal conditions. During the modeling process, the least squares method is used to fit the polynomial coefficients. The least squares method optimizes the model coefficients by minimizing the sum of squared errors between the predicted and actual values, ensuring that the model accurately represents the regularity of load changes under normal operating conditions. This process generates a low-sensitivity load forecasting model suitable for predicting load changes over long timescales. Model evaluation of the low-sensitivity load forecasting model is conducted to verify its reliability and accuracy. Using a cross-validation method, the model's root mean square error (RMS) and coefficient of determination (CDR) are calculated through multiple training and testing cycles on data subsets. The RMS error assesses the model's prediction error range, while the CDR reflects the model's ability to account for data fluctuations. These evaluation metrics ensure that the low-sensitivity model provides robust results for load forecasting under normal operating conditions, yielding reliable low-sensitivity forecasts. Furthermore, for peak operating condition data, a three-layer neural network structure is constructed to establish a high-sensitivity load forecasting model. The neural network consists of an input layer, a hidden layer, and an output layer. The input layer contains nodes for time features, historical loads, and weather factors. These characteristic variables reflect the primary drivers of peak load fluctuations. The hidden layer uses a sigmoid activation function, which introduces nonlinear mapping to capture the complex dynamic behavior of peak loads. The output layer corresponds to the predicted load nodes and directly generates peak load forecasts. During the data processing phase, the peak operating condition data is divided into training, validation, and test sets in a ratio of 7:2:1 to ensure comprehensive and balanced data support for the model during training, validation, and testing. During neural network training, a backpropagation algorithm is used to optimize network weights. This algorithm calculates the error between predicted and actual values, propagates the error back through each layer of the network, and adjusts the weights to minimize the error. During the training optimization process, the Adam optimizer is used to update model parameters. The Adam optimizer, combined with momentum and adaptive learning rate methods, accelerates model convergence and avoids local optima. Using mean squared error as a loss function, it ensures the model better fits actual peak load data, resulting in a highly sensitive load forecasting model. A weighted combination of low- and high-sensitivity forecast results is performed to generate charging load forecast data.The weighted combination assigns different weights to low-sensitivity prediction results and high-sensitivity prediction results based on the prediction characteristics of the two models. These weights are dynamically adjusted through historical prediction performance to achieve comprehensive coverage of load forecasts with different time scales and sensitivity requirements.

[0034] The analysis of charging pile operating data, including variables such as charging pile output power, local user load, and charging demand, was conducted. Correlation analysis was used to assess the strength of the relationship between these variables and the target load. This approach identified the most important set of variables for charging load forecasting, forming a target set of influencing factors. This correlation analysis employed statistical techniques to ensure that the selected variables were both physically meaningful and statistically significant in explaining load variations. Based on the target set of influencing factors, a basic form of a response surface polynomial function was constructed. This function comprehensively describes the direct, nonlinear, and interaction effects between variables through a combination of linear, square, and cross terms. In this process, a linear term coefficient was assigned to each variable to reflect its primary influence; a square term coefficient was set to capture its nonlinear characteristics; and a cross term coefficient was used to describe the interaction between the variables. These terms were combined to form an initial polynomial structure. The coefficients of this initial polynomial structure were then optimized. The least squares method was used to compare the differences between the model predictions and the actual values, minimizing the sum of squares of these differences to obtain the optimal solution for the coefficients. By constructing a sum-of-squares error objective function, the optimization problem was transformed into a problem of solving a system of linear equations. To ensure numerical stability and accuracy during the solution process, matrix decomposition techniques and the QR decomposition method are used to solve the system of linear equations. QR decomposition efficiently decomposes the matrix and extracts the optimal solutions for the coefficients. These optimal solutions directly determine the performance of the polynomial model. After obtaining the optimal solutions for the polynomial model coefficients, the model is tested for significance to assess whether each variable and its interaction terms contribute significantly to the model's forecast results. If certain variables or terms have a small contribution, they are removed based on the test results, simplifying the model structure and retaining the most critical factors for load forecasting. This process effectively reduces model complexity while improving its generalization and computational efficiency. After the significance test, an analysis of variance is performed on the resulting simplified model to verify its goodness of fit. Analysis of variance is a method used to evaluate the predictive ability of a model. It divides the total variation in the data into a regression component and a residual component to measure the model's ability to explain the variation in the variables. The goodness of fit of the model is determined by calculating the regression sum of squares, the residual sum of squares, and the total sum of squares. This metric reflects the overall fit of the model to the data. A high goodness of fit indicates that the model is reliable in its predictive ability. The response surface polynomial function is modified based on the goodness of fit. During the modification process, a penalty term is introduced to limit the weights of some parameters in the polynomial, thus avoiding overfitting caused by excessive model complexity. The modification method includes constraining the weights of higher-order terms or cross terms to simplify the polynomial structure while retaining the model's main predictive capabilities. The modified polynomial model achieves a balance between complexity and accuracy and can more effectively adapt to actual load forecasting needs. Through the above steps, a modified polynomial model is generated, which is a low-sensitivity load forecasting model.

[0035] Step 300: Perform control optimization calculation on the proportional compensation unit and the integral compensation unit according to the deviation between the charging load prediction data and the actual load data to obtain dynamic power compensation control parameters;

[0036] It should be noted that the predicted charging load data and actual load data are acquired in real time, and the difference between them is calculated to generate an input deviation signal for the power compensation unit, reflecting the dynamic difference between the predicted and actual loads. Based on the input deviation signal, mathematical models are developed for the proportional and integral compensation units. For the proportional compensation unit, a first-order transfer function model is established, with the proportional gain coefficient being a key parameter. This model directly generates the control output by multiplying the input deviation signal by the proportional gain coefficient, achieving rapid response to deviations. The primary function of the proportional compensation unit is to promptly correct deviations and provide rapid power adjustment capabilities in dynamic load environments, ensuring that the charging system can adapt to load changes. Simultaneously, a mathematical model is developed for the integral compensation unit, including an integral time constant. This model accumulates the deviation signal to eliminate steady-state errors in the system, thereby maintaining system accuracy during long-term operation. The integral time constant is a key parameter in this model and directly affects the dynamic characteristics of the integral compensation, including the system's response speed and stability. The transfer function models of the integral and proportional compensation units are combined in parallel to construct the overall transfer function of the power compensator. Based on the overall transfer function of the power compensator, the desired closed-loop pole positions of the system are set. Using the pole placement method, appropriate closed-loop pole locations are determined based on system performance requirements (such as response speed, overshoot, and stability). This is used to establish the characteristic equation. The coefficients of the characteristic equation are determined by the proportional gain coefficient and the integral time constant. These parameters are optimized to meet the system's dynamic performance requirements. The characteristic equation is then applied to the Russ-Horwitz stability criterion, and the coefficients are compared to determine stability constraints. These constraints define the range of values ​​for the proportional gain coefficient and the integral time constant, ensuring that the system remains stable during dynamic adjustments. Within the determined parameter range, the proportional gain coefficient is iteratively optimized using the bisection method. This method is an efficient numerical calculation method that gradually narrows the search interval to quickly find the optimal proportional gain coefficient that meets the system's overshoot requirements. Overshoot is a key indicator of system response performance, and its optimization aims to balance system speed and stability. Based on the optimal proportional gain coefficient, the integral time constant is then swept and calculated. By gradually adjusting the value of the integral time constant, comparing the system response time under different parameter combinations, and selecting the parameter combination that can achieve the shortest response time, the dynamic power compensation control parameters are obtained, including the optimal proportional gain coefficient and integral time constant.

[0037] Step 400: Based on the dynamic power compensation control parameters, perform genetic optimization and multi-level power distribution on the charging pile power distribution ratio to obtain a multi-level power distribution scheme including a normal load distribution mode and a peak load distribution mode;

[0038] Specifically, based on the operating characteristics and load demands of the charging pile system, dynamic power compensation control parameters are divided into normal load compensation parameters and peak load compensation parameters, forming a classified compensation parameter set. This classification process is based on the compensation requirements of the charging piles under different operating conditions. Normal load compensation parameters correspond to the system's adjustment requirements under stable load conditions, while peak load compensation parameters address the dynamic response requirements to sudden load changes. An optimization objective function for charging pile power allocation is constructed based on the classified compensation parameter set. The optimization objective function consists of two core components: system power loss and load fluctuation. The power loss term minimizes energy consumption during charging pile operation, while the load fluctuation term aims to reduce the impact of load imbalance during power distribution. By combining these two objectives, the optimization objective function for charging pile power allocation is formed. An encoding design is performed for the charging pile power allocation ratio. The allocation ratio of each charging pile is encoded as a gene fragment, forming a chromosome encoding format. This encoding method treats each charging pile's power allocation as an independent gene, allowing the power allocation status of the entire system to be represented by the chromosome. After the encoding design is completed, key parameters of the genetic algorithm, including population size, crossover probability, and mutation probability, are set according to the chromosome encoding format to form a genetic algorithm optimization model. In the genetic optimization phase, the optimization objective function for charging pile power allocation is input into the genetic algorithm optimization model, and the iterative optimization process begins by generating an initial population. High-quality individuals are selected from the initial population as the parent population. These individuals have high fitness and are better able to meet the objective function requirements. A simulated binary crossover operation is performed on the parent population to achieve genetic recombination and generate a new generation of individuals. Simultaneously, a polynomial mutation operation is incorporated to introduce random variations to increase population diversity and avoid falling into local optima. The crossover and mutation process is continuously iterated, gradually approaching the optimal power allocation solution through multiple generations of evolution. During the iterative optimization process, constraints are applied to the power allocation ratio of each charging pile to ensure that the allocation ratio meets the rated capacity constraints and power balance constraints of the charging piles. The rated capacity constraint ensures that the allocated power of each charging pile does not exceed its physical limitations, while the power balance constraint ensures that the overall power demand and supply of the system are in a dynamic balance. Through constraint processing, individuals that do not meet the requirements are eliminated, forming a set of feasible solutions. This set of feasible solutions is then ranked to select the optimal solution and generate a multi-level power allocation solution. The ranking process evaluates the value of the optimization objective function, prioritizing solutions with minimal power loss and load fluctuation as high-quality solutions. The resulting multi-level power allocation scheme includes both a normal load allocation mode and a peak load allocation mode. The normal load allocation mode is primarily designed for power allocation under steady load conditions, offering stability and efficiency. The peak load allocation mode, on the other hand, addresses sudden load fluctuations and allows for rapid adjustments to power allocation ratios to meet dynamic demands during periods of high load.

[0039] Step 500: According to the multi-level power allocation scheme, power limitation is performed by reducing the charging power of non-priority users, and load transfer is performed through power coordination of adjacent charging piles to obtain an emergency power adjustment control instruction.

[0040] Specifically, based on a multi-level power allocation scheme, a user's charging duration and charging demand are scored. The charging duration score is determined by analyzing the user's cumulative charging time and current charging status, while the charging demand score is calculated based on the user's device's remaining charging capacity and target charge capacity. To integrate the impact of these two scores, a user priority index is generated through a linear weighted combination to reflect the user's priority demand for charging resources. The power allocation scheme is then graded based on the user priority index. Power restrictions are applied to low-priority users in descending order of priority. Power limit thresholds are calculated and the charging power for low-priority users is gradually reduced according to the set levels, generating power limit instructions. The reduction process uses a step-by-step approach, dynamically adjusting power output based on user priority and charging demand to ensure that the charging needs of high-priority users are met first, while preventing the system's power allocation from exceeding the total load capacity. Simultaneously with power restrictions, transfer calculations are performed for restricted loads. A load transfer cost function is established based on the physical distance and remaining capacity between charging piles. This cost function comprehensively considers the transfer distance, load capacity, and user demand matching between adjacent charging piles to generate a load transfer plan. By optimizing this scheme, dynamic load balance within the system is achieved, allowing low-load charging piles to share the pressure of high-load charging piles and improving overall system efficiency. Based on the load transfer plan, a piecewise linear interpolation method is used to calculate the power allocation ratio between adjacent charging piles. This linear interpolation method ensures smooth and continuous load transfer by gradually adjusting the power allocation between adjacent charging piles. Furthermore, to ensure accurate power transfer, a closed-loop iterative algorithm optimizes the transfer power values ​​and generates specific power coordination instructions, including the target power adjustment amount and execution sequence for each charging pile. This ensures efficient system response under dynamic load conditions. After the power coordination instructions are generated, their execution sequence is planned to avoid the impact of large power adjustments on system operation. During the planning process, a buffer time and smoothing coefficient are set for power adjustment. By buffering and gradually controlling the power changes, a smooth transition of charging pile power adjustment is achieved. Furthermore, the planned power adjustment execution sequence is sampled and quantized according to the control cycle. The continuous power limit and transfer power values ​​are converted into digital quantities, generating a control quantity sequence that can be directly used for control execution. After generating the control variable sequence, overload trigger conditions and recovery conditions are set to ensure the effectiveness of the emergency power regulation control instructions. The overload trigger condition determines whether the system needs to initiate emergency power regulation, while the recovery condition determines when the system can exit the emergency state and return to normal operation. These conditions are monitored and determined in real time through a logical judgment mechanism, ultimately generating a clear emergency power regulation control instruction.

[0041] In the embodiment of the present application, a small signal model of the charging pile system is established, and the operating characteristics of the charging pile are accurately described in combination with the system identification method, thereby overcoming the limitations of traditional modeling methods in large-scale charging pile systems; a dual-model prediction strategy combining the response surface method and the neural network is adopted to perform load prediction for normal working conditions and peak working conditions respectively, which significantly improves the accuracy of load prediction; a dynamic power compensator is designed based on the pole configuration method, and the influence of load prediction deviation on system stability is effectively suppressed by optimizing the proportional integral control parameters; a genetic algorithm is used to perform multi-objective optimization of the charging pile power distribution, and load fluctuations are minimized while ensuring minimum system loss; an emergency power adjustment mechanism based on user priority is designed, and the system overload problem during peak charging periods is solved by combining power limitation and load transfer; the present application adopts a closed-loop optimization mechanism, and through online parameter optimization and dynamic adjustment of the control strategy, the stable operation of the charging pile system under various working conditions is ensured.

[0042] In a specific embodiment, the process of executing step 100 may specifically include the following steps:

[0043] Collect data on the charging pile output power, local user load data, and charging demand data to obtain real-time charging pile operation data, and preprocess the real-time charging pile operation data to obtain a preprocessed operation data sequence;

[0044] The pre-processed operating data sequence is input into the recursive least squares algorithm model to identify and calculate the rated power coefficient and power regulation coefficient of the charging pile to obtain the system characteristic parameter set;

[0045] Based on the system characteristic parameter set, the state space model is used to model and analyze the steady-state operation characteristics of the charging pile under rated working conditions to obtain a steady-state operation model;

[0046] Based on the system characteristic parameter set, the transfer function method is used to model and analyze the power fluctuation response characteristics caused by random charging demand, and a dynamic response model is obtained;

[0047] Decouple the steady-state operation model and the dynamic response model to obtain the frequency domain characteristic equation of the charging pile system. This frequency domain characteristic equation is then converted into a time domain state equation through inverse Laplace transform to obtain the small signal mathematical model of the charging pile system.

[0048] The eigenvalue analysis and stability criterion calculation of the small signal mathematical model of the charging pile system are performed to obtain the small signal model of the charging pile system including the rated power parameters of the charging pile, load fluctuation parameters and system response parameters.

[0049] Specifically, the real-time operation data of the charging pile is collected, including the output power of the charging pile , local user load data and users' charging needs , forming the original operation data set of the charging pile system. The real-time operation data of the charging pile is preprocessed, including outlier removal, interpolation and filtering noise reduction. Outlier removal eliminates measurement errors by identifying data points that are out of a reasonable range (such as excessively high instantaneous power values); interpolation is used to repair missing data caused by acquisition delays or transmission interruptions; filtering noise reduction uses a low-pass filter to remove high-frequency noise to obtain a smooth operation data sequence. Assuming the acquisition frequency is , the preprocessed data is represented as a discrete time series ,in is the time index, The pre-processed power data is input into the recursive least squares (RLS) algorithm model for parameter identification. The recursive least squares algorithm is a dynamic parameter estimation method used to identify the system characteristic parameters of the charging pile in real time, including the rated power coefficient. and power regulation coefficient The RLS algorithm updates the model parameters by minimizing the following objective function:

[0050] ;

[0051] in, is the objective function, It is The output power of is the input data vector (such as voltage, current, load, etc.), is the parameter vector to be identified (including and ), Is the forgetting factor, which is used to control the weight of historical data. Through recursive update, the system characteristic parameter set is obtained Based on the identified characteristic parameter set, the state space model is used to model and analyze the rated working conditions of the charging pile. The basic form of the state space model is:

[0052] ;

[0053] ;

[0054] in, is the system state vector, is the input vector (such as load fluctuation), is the output vector (e.g., power response), is the state space matrix. By analyzing the steady-state data of the system, we can determine and The matrix structure of is used to construct a steady-state operation model to describe the stable operation characteristics of the system under rated load. At the same time, based on the characteristic parameter set, the transfer function method is used to model the power fluctuation response characteristics caused by random charging demand. The form of the transfer function is:

[0055] ;

[0056] in, is the transfer function of power fluctuation, and are the Laplace transforms of the output and input, respectively, is a time constant. By constructing a transfer function model, the dynamic response behavior of the system under random load disturbances is revealed, forming a dynamic response model. After obtaining the steady-state operation model and the dynamic response model, the two are decoupled and calculated to obtain the frequency domain characteristic equation of the charging pile system. By mapping the steady-state model and the dynamic model to the frequency domain space respectively, the comprehensive frequency domain characteristic equation is obtained:

[0057] ;

[0058] in, Represents the total frequency domain characteristics of the system. To facilitate time domain analysis, the inverse Laplace transform is used to Convert it into a time domain state equation to form a time domain model that describes the small signal characteristics of the charging pile system. Perform eigenvalue analysis and stability criterion calculation on the small signal mathematical model. The goal of eigenvalue analysis is to solve the state matrix The eigenvalues ​​of the eigenvalues ​​are used to determine whether the system's dynamic behavior is stable. If the real parts of all eigenvalues ​​are less than zero, the system is asymptotically stable; otherwise, there is a risk of oscillation or instability. Stability criteria (such as root locus analysis or gain margin assessment) are used to further verify the model's stability, resulting in a small-signal model of the charging pile system that includes the charging pile's rated power parameters, load fluctuation parameters, and system response parameters.

[0059] In a specific embodiment, the process of executing step 200 may specifically include the following steps:

[0060] Based on the small signal model of the charging pile system, the operating conditions are classified and the operating data is divided into normal operating condition data and peak operating condition data according to the load fluctuation amplitude;

[0061] A response surface polynomial function model is constructed based on conventional operating condition data. The response surface polynomial function model includes linear terms, square terms, and cross terms. The polynomial coefficients are fitted and calculated using the least squares method to obtain a low-sensitivity load prediction model.

[0062] The cross-validation method is used to evaluate the low-sensitivity load forecasting model. The model accuracy is verified by calculating the root mean square error and the coefficient of determination to obtain the low-sensitivity forecasting results.

[0063] A three-layer neural network structure is constructed based on peak operating condition data. The three-layer neural network structure includes an input layer, a hidden layer, and an output layer. The input layer contains time feature nodes, historical load nodes, and weather factor nodes. The hidden layer uses a Sigmoid activation function, and the output layer is a predicted load node.

[0064] The peak operating condition data was divided into training set, validation set and test set in a ratio of 7:2:1. The back propagation algorithm was used to optimize the neural network weights and obtain a high-sensitivity load forecasting model.

[0065] The Adam optimizer is used to update the parameters of the high-sensitivity load forecasting model, and the mean square error function is used as the loss function to obtain the high-sensitivity forecasting results;

[0066] The low-sensitivity prediction results and the high-sensitivity prediction results are weighted and combined to obtain the charging load prediction data.

[0067] Specifically, based on the small signal model of the charging pile system, the operating data is analyzed and the data is divided into normal working condition data and peak working condition data based on the load fluctuation amplitude. The load fluctuation amplitude is calculated by calculating the standard deviation of the power change. To measure, It is The power value at a time point, is the average value of power, is the total number of time points. When the data is below a certain threshold, it is classified as normal operating conditions; when When the threshold is exceeded, the data is classified as peak operating conditions. For normal operating data, a response surface polynomial function model is constructed to predict low-sensitivity load changes. The response surface model takes the following form:

[0068] ;

[0069] in, is the load forecast value, is the input feature variable (such as time, temperature, number of users, etc.), is a constant term, 、 and The coefficients of the linear, square and cross terms are used to fit the coefficients of the model and minimize the actual load. With forecast load The sum of squared errors between:

[0070] ;

[0071] By the objective function The optimal solution of the model parameters is obtained to form a low-sensitivity load forecasting model. After the model is built, the cross-validation method is used to evaluate the low-sensitivity load forecasting model. The data is divided into Each time, one subset is selected as the test set, and the remaining subsets are used as the training set, and the training and testing are repeated. times, and finally calculate the root mean square error :

[0072] ;

[0073] and the coefficient of determination :

[0074] ;

[0075] in, is the average value of the observed values. These indicators verify the prediction accuracy and reliability of the model and generate low-sensitivity prediction results. For peak operating data, a three-layer neural network is used for modeling. The network structure includes input layer, hidden layer and output layer. The input layer contains time feature nodes , historical load nodes and weather factor nodes , these nodes reflect the main factors affecting the peak load; the hidden layer uses the Sigmoid activation function, which is defined as:

[0076] ;

[0077] The output layer contains only one prediction node, which is used to generate load forecast values. Peak operating condition data is divided into training, validation, and test sets in a ratio of 7:2:1. The training set is used to optimize network weights, the validation set is used to prevent overfitting, and the test set is used to evaluate the model's generalization performance. During the training phase, the backpropagation algorithm is used to update the weights and biases of the neural network. Backpropagation adjusts parameters by calculating the gradient of the loss function, using the mean squared error (MSE) as the loss function:

[0078] ;

[0079] To improve training efficiency and stability, the Adam optimizer is used to update model parameters. Combining the momentum method with an adaptive learning rate method, the Adam optimizer can dynamically adjust the parameter update step size to accelerate convergence. Through multiple iterative training cycles, a highly sensitive load forecasting model is ultimately formed, generating forecast results. The low-sensitivity and high-sensitivity forecast results are weighted and combined to obtain the final charging load forecast data. The weighted combination formula is:

[0080] ;

[0081] in, and is the weight coefficient, satisfying .

[0082] In a specific embodiment, the execution step constructs a response surface polynomial function model based on conventional operating condition data, the response surface polynomial function model includes linear terms, square terms, and cross terms, and the polynomial coefficients are fitted and calculated using the least squares method to obtain a low-sensitivity load prediction model. The process can specifically include the following steps:

[0083] Based on normal operating condition data, a correlation analysis is conducted on the output power of charging piles, local user load, and charging demand to obtain the target influencing factor set;

[0084] The basic form of the response surface polynomial function is constructed based on the target influencing factor set, and the linear term coefficient, square term coefficient and cross term coefficient are set for each influencing factor to obtain the initial structure of the polynomial;

[0085] Substitute the initial structure of the polynomial into the least squares calculation formula, construct the error square sum objective function, obtain the coefficient equation group to be solved, and perform matrix decomposition operation on the coefficient equation group. Use QR decomposition method to solve the linear equation group and obtain the optimal solution of the polynomial coefficients;

[0086] According to the optimal solution of the polynomial coefficients, the significance test of the response surface polynomial function is performed to obtain the simplified polynomial model, and the variance analysis of the simplified polynomial model is performed to calculate the regression sum of squares, residual sum of squares and total sum of squares to obtain the goodness of fit of the model;

[0087] The response surface polynomial function is modified based on the goodness of fit, and the complexity of the model is controlled by introducing a penalty term to obtain the modified polynomial function, which is then used as a low-sensitivity load forecasting model.

[0088] Specifically, the normal working condition data of the charging pile system is sorted and analyzed, including the output power of the charging pile , local user load and charging needs These variables directly reflect the characteristics of load changes during the operation of the charging pile. By calculating the correlation coefficients between these variables, their correlation with the target load (i.e. the variables to be predicted) is evaluated, thereby screening out the most influential variables and forming a target influencing factor set. Correlation analysis quantifies the linear correlation between variables by calculating the Pearson correlation coefficient. For two variables and , and the calculation formula of its correlation coefficient is:

[0089] ;

[0090] in, and Respectively The variable value of the sample point, and They are and The mean of is the total number of samples. Correlation coefficient The value range of is [-1, 1], The closer it is to 1 or -1, the stronger the correlation. 、 and The correlation with the target load is used to determine whether these variables have a significant impact on model construction and to screen out the target influencing factor set. Based on the target influencing factor set, the basic form of the response surface polynomial function model is constructed. The response surface model captures the relationship between the input variables and the target load through a polynomial function, which is specifically in the form of:

[0091] ;

[0092] in, represents the target load forecast value, It is Influencing factor variables, is a constant term, is the linear term coefficient, is the square term coefficient, is the cross-term coefficient, is the number of variables. This model includes linear effects, nonlinear effects (square terms), and interaction effects (cross terms), and can fully describe the impact of influencing factors on the target load. Substitute the above polynomial initial structure into the least squares calculation formula to construct the error square sum objective function. The objective function is in the form of:

[0093] ;

[0094] in, is the actual observed value, is the model prediction value, is the total number of samples. In order to minimize , find the optimal solution for the polynomial coefficients. This optimization problem is transformed into the form of a system of linear equations:

[0095] ;

[0096] in, is the design matrix, which contains all input variables and their combinations. is the coefficient vector to be solved, Is the observation vector. The QR decomposition method is used to solve the linear equations, by Decompose into an orthogonal matrix and the upper triangular matrix , calculate , and obtain the optimal solution of the polynomial coefficients. After obtaining the optimal solution, the significance test of the response surface polynomial function is performed to screen out the variables and items that have an important impact on the target load, and at the same time eliminate the redundant items that have little contribution to the model and simplify the model structure. By testing each coefficient -Statistics and -value, determine its significance level. If the value is greater than the preset threshold (such as 0.05), the coefficient is considered insignificant and the corresponding variable or term is removed from the model. ANOVA is performed on the simplified model to assess the goodness of fit of the model. ANOVA calculates the coefficient of determination of the model by decomposing the total sum of squares (SST) into the regression sum of squares (SSR) and the residual sum of squares (SSE):

[0097] ;

[0098] Among them, SST reflects the overall variability of observations, SSR reflects the variability explained by the model, and SSE reflects the variability not explained by the model. The value range of is [0,1]. The closer to 1, the better the model fits the data. Based on the results of variance analysis, the response surface polynomial function is modified and a penalty term is introduced to control the complexity of the model. For example, Lasso regression is used to add -Regularization term:

[0099] ;

[0100] in, is the regularization coefficient, which is used to balance the model complexity and fitting accuracy. , simplifying the model structure and improving the generalization ability, the modified polynomial function is obtained, and the modified polynomial function is used as a low-sensitivity load forecasting model.

[0101] In a specific embodiment, the process of executing step 300 may specifically include the following steps:

[0102] Perform real-time difference calculation on charging load prediction data and actual load data to obtain input deviation signal of power compensation unit;

[0103] Based on the input deviation signal, mathematical modeling is performed on the proportional compensation unit, and a first-order transfer function including a proportional gain coefficient is constructed to obtain a transfer function model of the proportional compensation link;

[0104] According to the input deviation signal, the integral compensation unit is mathematically modeled, a first-order differential equation including the integral time constant is constructed, and the transfer function model of the integral compensation link is obtained;

[0105] The transfer function model of the proportional compensation link and the transfer function model of the integral compensation link are combined in parallel to obtain the overall transfer function of the power compensator;

[0106] Based on the overall transfer function of the power compensator, the desired closed-loop pole position of the system is set, and the characteristic equation is established using the pole placement method to obtain the parameter optimization target equation. The parameter optimization target equation is then substituted into the Russ-Horwitz stability criterion, and the stability constraints are determined by coefficient comparison to obtain the parameter value range.

[0107] According to the parameter value range, the bisection method is used to iteratively optimize the proportional gain coefficient so that the system overshoot is controlled within the set range and the optimal proportional gain coefficient is obtained. Based on the optimal proportional gain coefficient, the integral time constant is scanned and calculated, and the parameter combination with the shortest system response time is selected to obtain the dynamic power compensation control parameters.

[0108] Specifically, by obtaining the charging load prediction data and the actual load data, the real-time difference between the two is calculated to form the input deviation signal of the power compensation unit, which is expressed as follows:

[0109] ;

[0110] in, It's time The deviation signal, is the actual load power, is the predicted load power. Deviation signal Directly reflects the difference between the actual operating status of the charging pile and the prediction model. According to the input deviation signal, the proportional compensation unit is mathematically modeled and a proportional gain coefficient is established. The proportional compensation unit quickly responds to system errors by simply linearly amplifying the deviation signal. Its mathematical expression is:

[0111] ;

[0112] in, is the Laplace transform of the scale-compensated output, is the Laplace transform of the error signal. By introducing the proportional gain coefficient , which can adjust the response strength of the compensation unit. At the same time, the integral compensation unit is modeled and a time constant including the integral The integral compensation unit eliminates the steady-state error by accumulating the deviation signal, and its Laplace domain expression is:

[0113] ;

[0114] in, is the integral compensation output, is the integration time constant, which determines the dynamic characteristics of the integrator. Integral compensation is used to compensate for long-term deviations to ensure that the system can accurately reach the set value. The transfer functions of the proportional compensation unit and the integral compensation unit are combined in parallel to obtain the overall transfer function of the power compensator:

[0115] ;

[0116] This transfer function combines the rapid response of proportional control and the precise compensation of integral control, effectively adjusting the dynamic performance of the charging pile system. After constructing the overall transfer function, the desired closed-loop pole positions of the system are set, and the characteristic equation is established through the pole placement method. The characteristic equation reflects the dynamic characteristics of the system and is in the form of:

[0117] ;

[0118] in, is the feedback path transfer function of the system. Substituting in, we get:

[0119] ;

[0120] By adjusting and , change the position of the closed-loop pole to meet the dynamic performance requirements of the system. In order to ensure the stability of the system, the characteristic equation is substituted into the Russ-Horwitz stability criterion to analyze whether its coefficient distribution meets the stability conditions. This criterion defines the proportional gain coefficient by comparing the coefficients of the characteristic equation. and the integration time constant For example, for a second-order system , need to meet and To ensure the stability of the system. After determining the parameter value range, use the dichotomy method to Perform iterative optimization to control the overshoot of the system within the set range. is the peak deviation in the system response and is calculated as:

[0121] ;

[0122] Continuously adjust through dichotomy , find the The optimal value that meets the requirements. Afterwards, Perform a sweep calculation to determine the parameter combination that can achieve the shortest response time and obtain the dynamic power compensation control parameters. is the time it takes for the system to reach a steady state from its initial state, and its magnitude is related to Directly related.

[0123] In a specific embodiment, the process of executing step 400 may specifically include the following steps:

[0124] Classifying the dynamic power compensation control parameters into conventional load compensation parameters and peak load compensation parameters to obtain a classified compensation parameter set;

[0125] Based on the classified compensation parameter set, the objective function of the system power loss term and the load fluctuation term is constructed to obtain the optimized objective function of the charging pile power distribution;

[0126] The power distribution ratio of the charging piles is coded and designed. The distribution ratio of each charging pile is encoded into a gene fragment to obtain a chromosome coding format. According to the chromosome coding format, the genetic algorithm parameters are set to obtain a genetic algorithm optimization model.

[0127] The optimization objective function of charging pile power distribution is input into the genetic algorithm optimization model, and high-quality individuals are selected to obtain the parent population. The parent population is then simulated with binary crossover and polynomial mutation operations to obtain an iterative optimization sequence.

[0128] According to the iterative optimization sequence, the power allocation ratio of each charging pile is constrained so that the power allocation ratio meets the rated capacity of the charging pile and the power balance constraints. A feasible solution set is obtained, and the feasible solution set is ranked by merit to obtain a multi-level power allocation scheme including conventional load distribution mode and peak load distribution mode.

[0129] Specifically, the dynamic power compensation control parameters are classified and divided into normal load compensation parameters and peak load compensation parameters according to different load conditions. Normal load compensation parameters are used to handle the power distribution of the system under stable load conditions, while peak load compensation parameters are mainly used to deal with the dynamic demand of sudden increase or decrease of load. By classifying the parameters, a classification compensation parameter set is formed, which is recorded as and , representing the normal and peak compensation parameters, respectively. Based on the classified compensation parameter set, an optimization objective function for charging pile power allocation is constructed. The objective function needs to comprehensively consider both system power loss and load fluctuation. The power loss term is used to minimize the transmission and conversion losses of electric energy and is expressed as:

[0130] ;

[0131] in, It is The output power of each charging pile, is the efficiency coefficient of the charging pile, is the total number of charging posts. The load fluctuation term is used to reduce the unbalanced load distribution between charging posts and avoid local overload of the system. Its expression is:

[0132] ;

[0133] in, is the average power output of all charging piles. Combining these two objectives, we get the overall objective function:

[0134] ;

[0135] in, and is the weight coefficient used to balance the effects of power loss and load fluctuation. The power distribution ratio of the charging pile is coded and the power distribution ratio of each charging pile is represented as a gene fragment. Assuming that the total power of the charging pile is , the distribution ratio of each charging column is ,satisfy By encoding the distribution ratio into gene segments, a chromosome encoding format is formed:

[0136] ;

[0137] To achieve optimization, set the key parameters of the genetic algorithm, including the population size , crossover probability and mutation probability . These parameters determine the algorithm's search efficiency and result quality, and combined with chromosome encoding, form a genetic algorithm optimization model. In the optimization stage, the constructed objective function is input into the genetic algorithm optimization model, starting from the randomly generated initial population, and the optimal allocation scheme is found through iterative optimization. According to the value of the objective function, high-quality individuals are selected as the parent population. These individuals have high fitness and can provide excellent genes for subsequent crossover and mutation. Use simulated binary crossover (SBX) to perform genetic recombination on the parent population to generate new individuals. The basic formula of SBX is:

[0138] ;

[0139] in, is the coefficient that controls the crossover range, and are the gene values ​​of the two parent individuals. A crossover operation generates diverse offspring individuals. Polynomial mutation is performed to adjust the gene values ​​of individuals by introducing small random variations to improve population diversity. After each iteration, the power allocation ratio of the charging piles is constrained based on the generated optimization sequence. Constraints include the non-negativity of the power allocation ratio, the rated capacity limit of the charging piles, and the total power balance constraint. Specifically, the following conditions must be met:

[0140] ;

[0141] in, It is The rated power of each charging station is calculated. Constraint processing eliminates solutions that do not meet the requirements, forming a set of feasible solutions. The feasible solutions are ranked, and the solution with the highest fitness is selected as the output based on the objective function value, resulting in a multi-level power allocation scheme. This scheme includes a normal load allocation mode and a peak load allocation mode, suitable for stable and fluctuating load conditions, respectively.

[0142] In a specific embodiment, the process of executing step 500 may specifically include the following steps:

[0143] Based on a multi-level power allocation scheme, the user's charging time and charging demand are scored and calculated, and the charging time score and charging demand score are linearly weighted combined to obtain the user priority index;

[0144] The power allocation scheme is graded based on the user priority index. The charging power of users is restricted from high to low priority to obtain the power limit threshold. Based on the power limit threshold, the charging power of non-priority users is adjusted in different levels. The charging power is restricted by a step-by-step reduction method to obtain the power limit instruction.

[0145] The load transfer calculation involved in the power limit instruction is performed, and a load transfer cost function is established based on the physical distance between charging piles and the remaining capacity to obtain a load transfer plan;

[0146] According to the load transfer plan, the piecewise linear interpolation method is used to calculate the power distribution ratio between adjacent charging piles. The transfer power value is determined through closed-loop iterative calculation to obtain the power coordination instruction;

[0147] Plan the execution timing of power coordination instructions, set the buffer time and smoothing coefficient of power regulation, obtain the power regulation execution sequence, sample and quantize the power regulation execution sequence according to the control period, perform digital conversion on the power limit and transfer power values, and obtain the control quantity sequence;

[0148] Based on the control quantity sequence, the overload trigger condition and the recovery release condition are set, and the effective state of the control instruction is determined through logical judgment to obtain the emergency power regulation control instruction.

[0149] Specifically, the user's charging time and charging needs are scored and calculated. Assuming that the user's actual charging time is , the maximum charging time is , charging time rating The calculation formula is:

[0150] ;

[0151] in, Indicates the proportion of time the user has spent charging. Charging demand score Based on the user's current charging needs and maximum demand Calculation formula is:

[0152] ;

[0153] These two scores reflect the user's time priority and demand priority respectively. The charging time score and charging demand score are linearly weighted to obtain the user priority index. The calculation formula is:

[0154] ;

[0155] in, and is the weight coefficient, satisfying , used to balance the importance of charging time and demand. By setting appropriate weights, long-time charging users or high-demand users are given priority. Priority Index The larger the value of , the higher the priority of the user. After calculating the user priority index, the power allocation scheme is graded according to the priority index. The users are sorted from high to low according to the priority index, and the power limit threshold is set according to the classification standard. Assume that the total system power is , the allocation rule of the limit threshold is:

[0156] ;

[0157] in, is assigned to The power limit threshold for each user, For non-priority users, the charging power is adjusted in stages by a step-by-step reduction method. For example, according to a 10% reduction ratio, the power adjusted each time is , until the power meets the restriction condition, and finally generates a power restriction instruction. After the power restriction instruction is generated, the transfer plan is calculated for the load of the restricted user. The load transfer is optimized based on the physical distance and remaining capacity between the charging piles. The transfer cost function is expressed as:

[0158] ;

[0159] in, It is Users to The physical distance between charging piles, is the power transferred to the pile, It's a charging station The remaining capacity of the charging piles is calculated by optimizing the transfer cost function. Based on the load transfer plan, the power distribution ratio between adjacent charging piles is calculated using the piecewise linear interpolation method. Assume that the power range between two charging piles is , the distribution ratio is expressed as:

[0160] ;

[0161] Through iterative adjustment The value of is combined with the closed-loop control algorithm to optimize the transfer power value and finally generate the power coordination instruction. The execution timing of the power coordination instruction is planned to smooth the power adjustment process and avoid impact on the system. The planning includes setting the buffer time of power adjustment. and smoothing coefficient , the transition is achieved by gradually adjusting the power. The power adjustment sequence is generated by:

[0162] ;

[0163] in, It is The regulation power of time steps, is the target power value. After sampling and quantization of the control cycle, the power limit and transfer power values ​​are converted into digital signals to form a control quantity sequence. Based on the control quantity sequence, the overload trigger condition and recovery release condition are set, and an effective emergency power regulation control instruction is generated through logical judgment. The overload trigger condition is defined as:

[0164] ;

[0165] The conditions for restoration and release are:

[0166] ;

[0167] When the trigger condition is met, the control instruction becomes valid; when the release condition is met, the instruction becomes invalid. This logic enables dynamic adjustment and real-time control.

[0168] The above describes the method for dynamic power distribution of charging piles based on load forecasting in the embodiment of the present application. The following describes the device 10 for dynamic power distribution of charging piles based on load forecasting in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of a dynamic power distribution device 10 for a charging pile based on load prediction includes:

[0169] Identification module 11, used to perform system identification analysis on the real-time operation data of the charging pile and build a small signal model of the charging pile system;

[0170] A training module 12 is used to train a low-sensitivity load prediction model and a high-sensitivity load prediction model based on a small signal model of the charging pile system, and generate charging load prediction data;

[0171] The calculation module 13 is used to perform control optimization calculation on the proportional compensation unit and the integral compensation unit according to the deviation value between the charging load prediction data and the actual load data to obtain the dynamic power compensation control parameters;

[0172] The distribution module 14 is used to perform genetic optimization and multi-level power distribution on the charging pile power distribution ratio based on the dynamic power compensation control parameters to obtain a multi-level power distribution scheme including a normal load distribution mode and a peak load distribution mode;

[0173] The processing module 15 is used to perform power limitation by reducing the charging power of non-priority users according to the multi-level power allocation scheme, and to transfer the load through the power coordination of adjacent charging piles to obtain an emergency power adjustment control instruction.

[0174] Through the collaborative cooperation of the above-mentioned components, by establishing a small signal model of the charging pile system and combining it with the system identification method to accurately describe the operating characteristics of the charging pile, the limitations of traditional modeling methods in large-scale charging pile systems are overcome; a dual-model prediction strategy combining the response surface method and the neural network is adopted to perform load forecasting for normal working conditions and peak working conditions respectively, which significantly improves the accuracy of load forecasting; a dynamic power compensator is designed based on the pole configuration method, and the influence of load forecasting deviation on system stability is effectively suppressed by optimizing the proportional integral control parameters; a genetic algorithm is used to perform multi-objective optimization of the charging pile power distribution, which minimizes the load fluctuation while ensuring the minimum system loss; an emergency power adjustment mechanism based on user priority is designed, which solves the system overload problem during peak charging periods by combining power limitation and load transfer; this application adopts a closed-loop optimization mechanism, which ensures the stable operation of the charging pile system under various working conditions through online parameter optimization and dynamic adjustment of the control strategy.

[0175] 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.

[0176] The non-volatile storage medium may store a computer program. The computer program includes program instructions, which, when executed by the processor 301, may cause the processor 301 to execute any of the above-mentioned methods for dynamic power allocation of charging piles based on load prediction.

[0177] The processor 301 is used to provide computing and control capabilities to support the operation of the entire electronic device 300 .

[0178] 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 charging pile dynamic power allocation methods based on load prediction.

[0179] Those skilled in the art will understand that Figure 3 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the electronic device 300 involved in the solution of the present application. The specific electronic device 300 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0180] It should be understood that the processor 301 may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0181] It should be noted that, those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the electronic device 300 described above can refer to the corresponding process of the aforementioned dynamic power allocation method for charging piles based on load prediction, and will not be repeated here.

[0182] 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 dynamic power allocation method for charging piles based on load prediction as provided in an embodiment of the present application.

[0183] 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 media card (SMC), a secure digital (SD) card, a flash card, etc., equipped with the electronic device 300.

[0184] Those skilled in the art will 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.

[0185] 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, 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 and includes several instructions for enabling an electronic device (which can be a personal computer, server, or network device, etc.) to execute 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, and other media that can store program code.

[0186] 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 above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above 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. A method for dynamic power allocation of charging piles based on load forecasting, characterized in that: The method comprises: Conduct system identification and analysis on the real-time operation data of charging piles and build a small signal model of the charging pile system; Based on the charging pile system small signal model, a low-sensitivity load prediction model and a high-sensitivity load prediction model are trained to generate charging load prediction data; According to the deviation value between the charging load prediction data and the actual load data, a control optimization calculation is performed on the proportional compensation unit and the integral compensation unit to obtain a dynamic power compensation control parameter; Based on the dynamic power compensation control parameters, the power distribution ratio of the charging pile is genetically optimized and multi-level power distribution is performed to obtain a multi-level power distribution scheme including a conventional load distribution mode and a peak load distribution mode; specifically, the scheme includes: classifying the dynamic power compensation control parameters, dividing the dynamic power compensation control parameters into conventional load compensation parameters and peak load compensation parameters, and obtaining a classified compensation parameter set; constructing the objective function of the system power loss term and the load fluctuation term based on the classified compensation parameter set, and obtaining the optimized objective function of the charging pile power distribution; encoding the power distribution ratio of the charging pile, encoding the distribution ratio of each charging pile into a gene fragment, and obtaining a color The genetic algorithm is configured to obtain a genetic algorithm optimization model by using a chromosome encoding format and setting genetic algorithm parameters according to the chromosome encoding format; the optimization objective function of the charging pile power distribution is input into the genetic algorithm optimization model, high-quality individuals are selected to obtain a parent population, and a simulated binary crossover operation and a polynomial mutation operation are performed on the parent population to obtain an iterative optimization sequence; according to the iterative optimization sequence, a constraint processing is performed on the power distribution ratio of each charging pile so that the power distribution ratio satisfies the rated capacity of the charging pile and the power balance constraint, a feasible solution set is obtained, and the feasible solution set is sorted by quality to obtain a multi-level power distribution scheme including a conventional load distribution mode and a peak load distribution mode; According to the multi-level power allocation scheme, power limitation is performed by reducing the charging power of non-priority users, and load transfer is performed through power coordination of adjacent charging piles to obtain an emergency power adjustment control instruction.

2. The method for dynamic power allocation of charging piles based on load forecasting according to claim 1, characterized in that: The system identification and analysis of the real-time operation data of the charging pile and the construction of the small signal model of the charging pile system include: Data collection is performed on the charging pile output power, local user load data, and charging demand data to obtain real-time charging pile operation data, and data preprocessing is performed on the real-time charging pile operation data to obtain a preprocessed operation data sequence; Inputting the pre-processed operating data sequence into a recursive least squares algorithm model, performing identification calculation on the rated power coefficient and power regulation coefficient of the charging pile, and obtaining a system characteristic parameter set; Based on the system characteristic parameter set, a state space model is used to model and analyze the steady-state operation characteristics of the charging pile under rated working conditions to obtain a steady-state operation model; Based on the system characteristic parameter set, a transfer function method is used to model and analyze the power fluctuation response characteristics caused by random charging demand to obtain a dynamic response model; Decoupling the steady-state operation model and the dynamic response model to obtain a frequency domain characteristic equation of the charging pile system, and converting the frequency domain characteristic equation into a time domain state equation through an inverse Laplace transform to obtain a small signal mathematical model of the charging pile system; The small signal mathematical model of the charging pile system is subjected to characteristic value analysis and stability criterion calculation to obtain a small signal model of the charging pile system including charging pile rated power parameters, load fluctuation parameters and system response parameters.

3. The method for dynamic power allocation of charging piles based on load forecasting according to claim 2, characterized in that: The low-sensitivity load prediction model and the high-sensitivity load prediction model are trained based on the small-signal model of the charging pile system, and charging load prediction data are generated, including: Based on the small signal model of the charging pile system, the operating data is classified into normal operating condition data and peak operating condition data according to the load fluctuation amplitude; A response surface polynomial function model is constructed based on the conventional operating condition data, wherein the response surface polynomial function model includes linear terms, square terms, and cross terms, and the polynomial coefficients are fitted and calculated using the least squares method to obtain a low-sensitivity load prediction model; The low-sensitivity load prediction model is evaluated by a cross-validation method, and the model accuracy is verified by calculating the root mean square error and the determination coefficient to obtain a low-sensitivity prediction result; A three-layer neural network structure is constructed based on the peak operating condition data. The three-layer neural network structure includes an input layer, a hidden layer, and an output layer. The input layer includes a time feature node, a historical load node, and a weather factor node. The hidden layer uses a Sigmoid activation function. The output layer is a predicted load node. The peak operating condition data is divided into a training set, a validation set, and a test set in a ratio of 7:2:1, and a back propagation algorithm is used to optimize the neural network weights to obtain a high-sensitivity load prediction model; The Adam optimizer is used to update the parameters of the high-sensitivity load forecasting model, and the mean square error function is selected as the loss function to obtain a high-sensitivity forecasting result; A weighted combination calculation is performed on the low-sensitivity prediction result and the high-sensitivity prediction result to obtain charging load prediction data.

4. The method for dynamic power allocation of charging piles based on load forecasting according to claim 3 is characterized in that: The response surface polynomial function model is constructed based on the conventional operating condition data. The response surface polynomial function model includes linear terms, square terms, and cross terms. The polynomial coefficients are fitted and calculated using the least squares method to obtain a low-sensitivity load prediction model, including: Based on the conventional operating condition data, a correlation analysis is performed on the charging pile output power, local user load, and charging demand to obtain a target influencing factor set; Constructing a basic form of a response surface polynomial function based on the target influencing factor set, setting a linear term coefficient, a square term coefficient, and a cross term coefficient for each influencing factor to obtain an initial polynomial structure; Substituting the initial polynomial structure into the least squares calculation formula, constructing the error square sum objective function, obtaining the coefficient equation group to be solved, and performing matrix decomposition operation on the coefficient equation group, using the QR decomposition method to solve the linear equation group to obtain the optimal solution of the polynomial coefficients; According to the optimal solution of the polynomial coefficients, a significance test is performed on the response surface polynomial function to obtain a simplified polynomial model, and a variance analysis is performed on the simplified polynomial model to calculate the regression sum of squares, the residual sum of squares and the total sum of squares to obtain the goodness of fit of the model; The response surface polynomial function is modified based on the goodness of fit, and the complexity of the model is controlled by introducing a penalty term to obtain a modified polynomial function, which is then used as a low-sensitivity load prediction model.

5. The method for dynamic power allocation of charging piles based on load forecasting according to claim 4 is characterized in that: The control optimization calculation is performed on the proportional compensation unit and the integral compensation unit according to the deviation value between the charging load prediction data and the actual load data to obtain the dynamic power compensation control parameters, including: Performing a real-time difference calculation on the charging load prediction data and the actual load data to obtain an input deviation signal of a power compensation unit; Based on the input deviation signal, mathematical modeling is performed on the proportional compensation unit, a first-order transfer function including a proportional gain coefficient is constructed, and a transfer function model of the proportional compensation link is obtained; According to the input deviation signal, mathematical modeling is performed on the integral compensation unit, a first-order differential equation including an integral time constant is constructed, and a transfer function model of the integral compensation link is obtained; The transfer function model of the proportional compensation link and the transfer function model of the integral compensation link are combined in parallel to obtain an overall transfer function of the power compensator; Based on the overall transfer function of the power compensator, the expected closed-loop pole position of the system is set, and the characteristic equation is established by the pole placement method to obtain the parameter optimization target equation. The parameter optimization target equation is substituted into the Russ-Horwitz stability criterion, and the stability constraint conditions are determined by coefficient comparison to obtain the parameter value range. According to the parameter value range, the proportional gain coefficient is iteratively optimized using the dichotomy method to control the system overshoot within the set range, thereby obtaining the optimal proportional gain coefficient. Based on the optimal proportional gain coefficient, the integral time constant is scanned and calculated, and the parameter combination with the shortest system response time is selected to obtain the dynamic power compensation control parameter.

6. The method for dynamic power allocation of charging piles based on load forecasting according to claim 1, characterized in that: The method of obtaining an emergency power regulation control instruction by reducing the charging power of non-priority users and transferring the load through power coordination of adjacent charging piles according to the multi-level power allocation scheme includes: Based on the multi-level power allocation scheme, the user's charging time and charging demand are scored and calculated, and the charging time score and the charging demand score are linearly weighted combined to obtain a user priority index; The power allocation scheme is graded according to the user priority index, and the charging power of users is limited from high to low priority to obtain a power limit threshold. Based on the power limit threshold, the charging power of non-priority users is adjusted in stages, and the charging power is limited by a step-by-step decrement method to obtain a power limit instruction. Performing a load transfer calculation on the load involved in the power limit instruction, establishing a load transfer cost function based on the physical distance between charging piles and the remaining capacity, and obtaining a load transfer plan; According to the load transfer scheme, the power distribution ratio between adjacent charging piles is calculated using piecewise linear interpolation method, the transfer power value is determined through closed-loop iterative operation, and the power coordination instruction is obtained; Performing execution timing planning on the power coordination instruction, setting a buffer time and smoothing coefficient for power regulation to obtain a power regulation execution sequence, sampling and quantizing the power regulation execution sequence according to a control period, performing digital conversion on the power limit amount and the transfer power value to obtain a control amount sequence; Based on the control quantity sequence, overload triggering conditions and recovery release conditions are set, and the effective state of the control instruction is determined through logical judgment to obtain the emergency power regulation control instruction.

7. A dynamic power distribution device for charging piles based on load prediction, characterized in that: The method for dynamic power allocation of a charging pile based on load prediction according to any one of claims 1 to 6 is configured to include: Identification module, used to perform system identification analysis on the real-time operation data of charging piles and build a small signal model of the charging pile system; A training module, configured to train a low-sensitivity load prediction model and a high-sensitivity load prediction model based on the charging pile system small signal model, and generate charging load prediction data; a calculation module, configured to perform control optimization calculation on a proportional compensation unit and an integral compensation unit according to a deviation between the charging load prediction data and the actual load data, and obtain dynamic power compensation control parameters; A distribution module is used to perform genetic optimization and multi-level power distribution on the charging pile power distribution ratio based on the dynamic power compensation control parameters to obtain a multi-level power distribution scheme including a normal load distribution mode and a peak load distribution mode; The processing module is used to perform power limitation by reducing the charging power of non-priority users according to the multi-level power allocation scheme, and to transfer the load through power coordination of adjacent charging piles to obtain an emergency power adjustment control instruction.

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 to enable the electronic device to execute the dynamic power allocation method for charging piles based on load prediction as described in any one of claims 1 to 6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the dynamic power allocation method for charging piles based on load prediction according to any one of claims 1 to 6 is implemented.

Images