CN115063184A

CN115063184A - Electric vehicle charging demand modeling method, system, medium, equipment and terminal

Info

Publication number: CN115063184A
Application number: CN202210824655.5A
Authority: CN
Inventors: 李恒杰; 梁达明; 陈伟; 周云; 冯冬涵; 朱江皓; 裴喜平; 曾贤强; 刘添一; 安妮; 冯琪; 陈兴旺
Original assignee: Lanzhou University of Technology
Current assignee: Lanzhou University of Technology
Priority date: 2022-07-14
Filing date: 2022-07-14
Publication date: 2022-09-16

Abstract

The invention belongs to the technical field of electric vehicle charging demand modeling, and discloses a method, a system, a medium, equipment and a terminal for modeling electric vehicle charging demand, which are used for analyzing the influence of electric vehicle user difference and electric vehicle performance difference and learning a user driving strategy and a user traveling strategy based on a generated countermeasure model; learning an electric vehicle performance model based on an XGboost machine; the real-time road condition based on the Baidu map proves that the algorithm can extract the driving characteristics and the charging characteristics of the user. According to the regional electric vehicle cluster charging demand model, real-time traffic flow rate of the Baidu map and real user running track data are combined, the obtained regional electric vehicle cluster charging demand model has a wider application prospect, and a prediction result has better generalization capability. The result proves that compared with the current mainstream prediction method, the method has better prediction precision and robustness, and can provide powerful model support for the planning and charging guidance of the charging station in the urban area by combining with the timely traffic data.

Description

Electric vehicle charging demand modeling method, system, medium, equipment and terminal

Technical Field

The invention belongs to the technical field of electric vehicle charging demand modeling, and particularly relates to a method, a system, a medium, equipment and a terminal for electric vehicle charging demand modeling.

Background

In recent years, in order to reduce dependence on petroleum and fossil fuels, many countries and regions have set policies for promoting the development of Electric Vehicles (EVs) and market penetration. In 2021, the global electric automobile sales amount reaches 675 million, which is increased by 108% compared with 2020, and the global share of electric automobiles in global light automobile sales is 8.3%, and 4.2% in 2020. The continuously increasing charging demand of the large-scale electric automobile connected to the power grid inevitably brings challenges to urban road traffic and stable operation of the power grid. At present, research on electric vehicles mainly focuses on charging demand prediction, energy management and charging guidance, which help to reduce negative impact of electric vehicles on a power grid, wherein the electric vehicle charging demand prediction is a basis for conducting impact analysis of electric vehicle access on the power grid, power distribution network planning and control operation, electric vehicle and power grid bidirectional interaction and charging guidance. However, as the permeability of the EV is increased, the charging demand is increased, and the rationality and accuracy of the existing charging prediction method cannot well meet the requirements of power grid dispatching and charging guidance. Therefore, a series of researches on the prediction of the charging requirement of the electric vehicle are urgently needed.

Currently, electric vehicle charging demand prediction research mainly focuses on user behavior analysis, the generation of charging demand is caused by insufficient energy of an electric vehicle, and energy change of the electric vehicle is a result influenced by user behavior, wherein the user behavior comprises charging time, trip mileage, driving strategies and the like, so the user behavior analysis is a difficult point and a key for charging demand prediction. In recent research, real-time travel information (such as 'drip' travel data) is obtained, and data mining and fusion technologies are adopted to obtain regeneration characteristic data so as to analyze the travel distribution rule and the charging behavior characteristics of residents. Recent research approaches have shown that researchers focus on mining individual user charging SOC characteristics and simulating group behaviors and their impact on charging demand prediction based on user charging strategy distributions. Secondly, a Marquardt (LM) training method based on a Rough structure was developed using the feedforward and recursive Artificial Neural Network (ANN) of levenbergu. The method takes into account the correlation between arrival time, departure time and length of travel. The charging demand and the urgency coefficient of the charging behavior of the user are hooked, a mathematical model describing the charging demand behavior is given, but a method for judging the charging demand behavior is lacked. The method is used for predicting the SOC change curve of a single electric automobile in the future 24 hours based on historical SOC data of a user, but the method is only used for predicting user data with strong SOC change regularity, the method lacks of understanding of user behaviors, the user behaviors are deterministic strategies made by the user based on current time, residual SOC and other factors, and due to the fact that the number of influencing factors is large, analysis of mathematical models of user charge and discharge strategies is lacked in the current technical scheme.

Through the above analysis, the problems and defects of the prior art are as follows:

(1) with the increasing permeability of the EV and the increasing charging demand, the rationality and accuracy of the existing charging prediction methods have not been able to meet the demands of grid dispatching and charging guidance well.

(2) The prior art is lack of a method for judging the charging demand behavior and a grasp on the user behavior; the current technical scheme also lacks analysis of a mathematical model of a user charging and discharging strategy.

Disclosure of Invention

The invention provides a method, a system, a medium, equipment and a terminal for modeling the charging requirement of an electric vehicle, and particularly relates to a method, a system, a medium, equipment and a terminal for modeling the charging requirement of the electric vehicle based on generation confrontation imitation learning.

The invention is realized in such a way, and provides a method for modeling the charging requirement of an electric automobile, which comprises the following steps:

analyzing the influence of the differences of the users of the electric vehicles and the differences of the performances of the electric vehicles, and learning a user driving strategy and a user traveling strategy based on a generated countermeasure model; learning an electric vehicle performance model based on an XGboost machine; the real-time road condition based on the Baidu map proves that the algorithm can extract the driving characteristics and the charging characteristics of the user.

Further, the electric vehicle charging demand modeling method comprises the following steps:

the method comprises the steps of firstly, collecting a user operation track data set, cleaning abnormal points of the data set, and dividing a user operation track; constructing a user strategy model training input data set based on MIC, and constructing a 24-hour SOC prediction data set of an electric vehicle user as an input data set;

establishing a generator, a value network and a discriminator neural network based on a linear full-connection network, and initializing the network by using a Bayesian super-parameter optimization method; inputting a training user track data set, optimizing strategy network parameters based on a PPO near-end strategy optimization algorithm, traversing all user tracks, and generating charge and discharge strategies of all users;

training an electric vehicle 24-hour SOC prediction model based on XGboost by adopting a cross validation method, wherein the model is divided into a discharging SOC prediction model and a charging SOC prediction model;

step four, planning the historical driving track passing route, acquiring the real-time traffic flow rate of the route based on a Baidu map, and predicting the SOC change curves of the single vehicles of all users in 24 hours; and (3) forecasting the charging demands and energy demands of all users by combining the charging urgency of the users, and establishing a cluster charging demand forecasting spatio-temporal model of the electric vehicles in the region.

Further, the first step further includes constructing an original data set of the charging requirement of the electric vehicle, and specifically includes:

and screening abnormal points by using an isolated forest, repairing the abnormal points of the data set by using a multiple interpolation method, and extracting user behavior factors strongly related to SOC (state of charge) change of the electric vehicle based on the MIC maximum mutual information coefficient matrix.

The method for constructing the operation original data set of the electric vehicle comprises the following steps:

as user 24 hour true speed data;

as the user 24-hour speed change influencing factor: real-time traffic flow rate;

as the user 24-hour true SOC data;

as a 24-hour SOC variation influencing factor: vehicle speed, single trip mileage, acceleration.

Further, the second step further includes constructing a user charging and discharging strategy learning model based on the generation of confrontation imitation learning, and specifically includes:

(1) trajectory sampling

The main program is a nested loop structure, the first layer of loop is an iterative loop, and all expert track data are traversed into a generation; and the second layer is an expert track cycle, firstly a sampling track is generated through the strategy network, then the corresponding value function, the dominant function and the mixed logarithmic density of the sampling track are calculated, then the expert strategy track and the sampling track are sent to the discriminator to update the parameters of the discriminator, and finally the mixed logarithmic density of the value function, the dominant function and the sampling track is sent to the PPO algorithm to update the strategy network until all the expert tracks are traversed, and the second layer cycle is finished.

(2) PPO strategy optimization

The method is characterized in that a framework is nested in a two-layer loop, the first layer is an iterative loop, the collected sample data (a state set, a return value set, an advantage estimation set, a value estimation set, a feedback estimation set and a state-action mixed logarithmic probability set) is disordered in sequence every time iteration is carried out, and after the sample data are divided into a certain batch, the batch is sent to a PPO algorithm together with a strategy network and a value network for parameter optimization and updating. The second layer is a network parameter updating layer; and traversing all sampling batches to finish the cycle.

Further, the construction of the XGboost-based 24-hour SOC prediction model for the electric vehicle in the third step includes:

according to correlation analysis of the SOC and the user strategy, a single trip mileage curve and a charging time curve are calculated based on a 24-hour user speed curve generated by strategy learning, wherein the speed curve and the mileage curve predict a discharging SOC, and the charging time curve predicts a charging SOC. Therefore, a driving SOC regression prediction model and a charging SOC regression prediction model are respectively established.

The prediction algorithm selects XGboost, wherein the loop is a cross-validation process; the training data set is divided into n _ splits subsets and normalized by mean removal and scaling to unit variance. Assuming that x is raw data, the normalization formula of the raw data x is as follows:

where u is the mean of x, s is the standard deviation of x, and z is the normalized value. One subset is used as a verification set and the rest subsets are used as training sets in each loop until all subsets are traversed. This ensures reliable generalization capability.

Wherein, the construction of the search strategy network based on the Bayesian algorithm and the SOC prediction model hyperparameters comprises the following steps:

in order to make the route planning model and the SOC prediction model converge in a short time to achieve better performance, the Bayesian super-parameter optimization algorithm is used for searching the super-parameters. In the strategy model, KL divergence of a true value and a predicted value is used as a Bayesian optimization objective function, wherein a strategy network and a discriminator network are composed of multiple layers of fully-connected neural networks, the number of expert tracks is set to be 10, in the SOC prediction model, the mean square error of the true value and the predicted value is used as the Bayesian optimization objective function, the search algebra is 50 times, and the hyper-parameters of the XGBoost algorithm are all accurate to 4 decimal numbers.

Further, the step four further comprises the steps of planning the route of the standardized electric automobile in the region and acquiring the real-time traffic flow rate of the route, wherein the steps comprise:

the path planning adopts a real path longitude and latitude coordinate set in a data set, all path longitude and latitude coordinates are subjected to data processing, and then the paths are visualized by using an OSMnx library in python, and the coordinates of road nodes, the distance information of the road nodes and distance initial nodes are extracted (the path nodes are the intersection points of the paths and other roads). Setting each track node data set as:

wherein omega ^j Represents the jth trace data set, j 1,2, 3.,

and

respectively, the longitude and latitude coordinates of the path node and the distance from the starting node.

Considering that a user driving strategy is influenced by real-time traffic flow velocity, so that real-time average speed of each road section needs to be acquired, acquiring running time T of the road section to which the current vehicle coordinate belongs based on a Baidu map real-time information platform, and setting distances between adjacent nodes and an initial node to be l respectively ₁ And l ₂ Then the current link length is L ═ L ₁ -l ₂ Time required to pass through current road sectionIs T _t The real-time traffic average flow velocity of the road section is v _t The real-time traffic average flow velocity can be calculated according to the following expression.

Further, the fourth step includes constructing a regional electric vehicle cluster charging demand prediction model, which specifically includes:

(1) establishing a user charging urgency model

The root cause of the user's charging demand is due to a reduction in battery energy (SOC), i.e., the urgency of the user's charging. The charging urgency of the user is closely related to the habit of the user, the SOC of the battery, the driving purpose and the like, and the higher the charging urgency is, the higher the probability that the user sends a charging demand signal is. Only the charging urgency determined by the battery SOC is considered here. The probability of charging is represented by a dotted line and the urgency of charging is represented by a solid line, and as the depth of discharge (amount of consumed power) is deeper, the amount of remaining power is lower, the probability of charging for the user is higher, and at the same time, the user is more urgent to charge the user is higher.

Considering the general expression of the charging urgency function, let the charging probability function be d (x), d (x) be a function of the depth of discharge DOD, where DOD is 1-SOC, d (x) be a function h ₁ (x) And h ₂ (x) Wherein x is ₁ ，x ₂ And x ₃ Determined by the battery capacity, the larger the battery capacity, x ₁ ，x ₂ And x ₃ The larger.

Charging urgency function C _u (x) Is D (x) the integral from 0 to x, giving the following formula:

here, the

While to the expression of the charging urgency function:

(2) charging demand determination

The user charge initiation SOC profile may reflect the dependence of the user charge rate on SOC. The initial SOC distributions for different users may differ. Consider fitting initial SOC distributions for different users using a normal distribution. When the SOC approaches the historical charging starting SOC of a certain user, namely the SOC is reduced to a charging demand interval of the certain user, the user generates a charging demand, and the charging demand interval is related to the distribution of the charging starting SOC of the user through a user charging urgency coefficient.

Here, it is considered that the charging start SOC distribution of a certain user follows a normal distribution N (μ, σ) in the case where the charging probability of the user is larger as the charging start SOC is smaller ² ) From N (μ, σ) ² ) Extracting an initial SOC value X, and setting a user charging urgency coefficient to be C under the current SOC _u (X), then the user charging requirement interval is [ X, X + 20C ] _u (X)]The smaller X is seen, the smaller C _u The larger (X), the wider the charging demand interval range, and the greater the possibility that the user generates a charging demand.

(3) Regional large-scale electric vehicle energy demand prediction

The method comprises the steps that based on a single electric vehicle SOC prediction result, the regional large-scale electric vehicle charging energy demand prediction is established, the charging energy demand of a user is predicted according to charging urgency and a charging starting SOC of the user, according to the definition of a charging demand interval, when the SOC of the user enters the charging demand interval, the user is considered to have the charging demand, and the charging energy demand of the user is calculated. It is provided that when a user generates a charging request, the difference between the current battery energy and 90 percent of the battery energy is used as the charging energy request, and the charging energy request is E _pc Setting the current battery SOC as SOC _t For use ofCapacity of household battery is C _p At this time, the calculation formula of the charging energy requirement of the single user is as follows:

E _pc ＝(1-SOC _t )×C _p

and completing the serial integration of the basic regressor group by the steps.

Another object of the present invention is to provide an electric vehicle charging demand modeling system for implementing the electric vehicle charging demand modeling method, the electric vehicle charging demand modeling system including:

the input data set construction module is used for constructing an original data set for electric vehicle operation, and dividing the original data set into different weights to be used as an input data set;

the system comprises a discriminator neural network, a strategy neural network and a value neural network construction module, wherein the discriminator neural network, the strategy neural network and the value neural network together form a strategy learning system; the reinforcement learning environment construction module is used for setting a state-action output function and setting an environment state and a strategy network action range; a PPO algorithm building module; the module performs optimization updating on the policy network parameters.

And the single-vehicle 24-hour SOC prediction module is used for dividing a training set based on a cross validation process, training an SOC prediction model under the charging and discharging process 2 on an XGboost algorithm, and meanwhile, carrying out hyperparameter search by utilizing Bayesian optimization.

The regional electric vehicle cluster charging demand prediction building module defines a charging demand interval by using a charging urgency coefficient, judges the charging demand of a single vehicle on the prediction result of the 24-hour single vehicle prediction module, and then integrates a space-time model of the large-scale electric vehicle charging demand.

Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, causes the processor to execute the method for modeling the charging requirement of an electric vehicle.

Another object of the present invention is to provide a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the computer program causes the processor to execute the method for modeling the charging requirement of an electric vehicle.

The invention also aims to provide an information data processing terminal which is used for realizing the electric vehicle charging demand modeling system.

By combining the technical scheme and the technical problem to be solved, the technical scheme to be protected by the invention has the advantages and positive effects that:

first, aiming at the technical problems existing in the prior art and the difficulty in solving the problems, the technical problems to be solved by the technical scheme of the present invention are closely combined with the technical scheme to be protected and the results and data in the research and development process, and some creative technical effects brought after the problems are solved are analyzed in detail and deeply. The specific description is as follows:

the invention provides a charging demand prediction model capable of explaining an electric vehicle user driving strategy and a charging strategy based on Baidu map real-time data based on strategy learning capacity of generation confrontation imitation learning, and the charging demand prediction model is characterized in that firstly, the correlation between strategy factors and SOC in user charging and discharging data is analyzed, then a single vehicle 24-hour SOC prediction model is established, and finally a charging energy demand space-time model in an area is established on the basis.

The modeling method for the charging demand of the electric automobile, provided by the invention, considers the influence of the difference of electric automobile users and the difference of electric automobile performance, and is based on a generation countermeasure model learning (GAIL) user driving strategy and a user traveling strategy; the single electric vehicle SOC prediction model based on the XGboost algorithm; the real-time road condition based on the Baidu map proves that the algorithm can generate the real-time user speed.

The method, the system, the terminal and the medium for predicting the cluster charging demand of the electric automobile provided by the invention are used for establishing a user charging and discharging strategy model. The correlation of the SOC change of the electric vehicle, a driving strategy and a charging strategy is proved based on the MIC maximum mutual information coefficient; and (3) providing a strategy learning model based on generation of a confrontation imitation learning (GAIL) mixed near-end strategy optimization algorithm (PPO) by combining the real-time traffic flow rate of the Baidu map. Compared with the traditional user behavior characteristic extraction, the method establishes an accurate user strategy mathematical model,

the method, the system, the terminal and the medium for predicting the cluster charging demand of the electric automobile provided by the invention establish a 24-hour SOC prediction curve of a single automobile. Based on a strategy learning model, the single electric vehicle SOC prediction method based on the XGboost algorithm is provided, and the prediction method is proved to have good robustness and accuracy.

The method, the system, the terminal and the medium for predicting the cluster charging demand of the electric automobile provided by the invention are used for establishing a space-time model of the cluster charging demand of the electric automobile in an area. Based on a single electric vehicle SOC change curve and in combination with a user charging demand perception model, an electric vehicle cluster charging demand space-time model in a future 24-hour area is established, and the model can explain the space-time characteristics of the charging demand and provide powerful data support for charging guidance and charging station planning problems.

The invention provides a regional electric vehicle cluster charging demand prediction method, a regional electric vehicle cluster charging demand prediction system, a regional electric vehicle cluster charging demand prediction terminal and a regional electric vehicle cluster charging demand prediction medium. The method comprises the steps of firstly analyzing the policy factors and the correlation of SOC in user charging and discharging data, then establishing a 24-hour SOC prediction model of a single vehicle, and finally establishing a charging energy demand space-time model in a region on the basis, so as to obtain a large-scale electric vehicle charging demand space-time distribution map in a future region.

Secondly, considering the technical scheme as a whole or from the perspective of products, the technical effect and advantages of the technical scheme to be protected by the invention are specifically described as follows:

according to the method, the system, the terminal and the medium for predicting the cluster charging demand of the electric automobile, provided by the invention, the obtained regional electric automobile cluster charging demand model has a wider application prospect by combining the real-time traffic flow rate of the Baidu map and the real user driving track data, and meanwhile, the prediction result has better generalization capability.

Compared with the current mainstream prediction method, the electric vehicle charging demand modeling method provided by the invention has better prediction precision and robustness, and can provide powerful model support for urban area charging station planning and charging guidance by combining with timely traffic data.

Third, as inventive supplementary proof of the claims of the present invention, there are several important aspects as follows:

(1) the expected income and commercial value after the technical scheme of the invention is converted are as follows:

and providing accurate charging demand data support for urban charging station planning.

(2) The technical scheme of the invention fills the technical blank in the industry at home and abroad:

establishing a mathematical model with accurate charge and discharge strategies; establishing a 24-hour SOC (state of charge) prediction model of the bicycle based on real-time traffic flow speed; a charging demand determination method based on a charging urgency coefficient is provided.

(3) The technical scheme of the invention solves the technical problems which are always desired to be solved but are not successful: and establishing a mathematical model of the user charging behavior based on the big data.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments of the present invention will be briefly described below, and it is obvious that the drawings described below are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.

FIG. 1 is a flow chart of a method for modeling a charging requirement of an electric vehicle according to an embodiment of the present invention;

FIG. 2 is an electric vehicle charging characteristic MIC matrix thermodynamic diagram provided by an embodiment of the invention;

FIG. 3 is a thermodynamic diagram of an electric vehicle discharge characteristic MIC matrix provided by the embodiment of the invention;

FIG. 4 is a flowchart illustrating an electric vehicle user-charging strategy learning according to an embodiment of the present invention;

FIG. 5 is a flowchart illustrating 24-hour SOC prediction for a single vehicle of an electric vehicle according to an embodiment of the present invention;

FIG. 6 is a flow chart of an XGBoost algorithm provided by an embodiment of the present invention;

FIG. 7 is a graph illustrating a charging urgency function for a user of an electric vehicle according to an embodiment of the present invention;

FIG. 8 is a graph of a bicycle 24 hour speed and range prediction provided by an embodiment of the present invention;

FIG. 9 is a diagram illustrating a 24-hour SOC prediction for a single vehicle, as provided by an embodiment of the present invention;

fig. 10 shows the result of predicting the charging demand of a main urban area for 24 hours according to an embodiment of the present invention;

fig. 11 is a statistical result of the sum of charging demands in each period according to the embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

In order to solve the problems in the prior art, the invention provides a method, a system, a medium, a device and a terminal for modeling the charging requirement of an electric vehicle, and the invention is described in detail with reference to the accompanying drawings.

First, an embodiment is explained. This section is an explanatory embodiment expanding on the claims so as to fully understand how the present invention is embodied by those skilled in the art.

As shown in fig. 1, the method for modeling the charging requirement of the electric vehicle according to the embodiment of the present invention includes the following steps:

s101, collecting a user running track data set, cleaning abnormal points of the data set, and dividing a user running track; constructing a user strategy model training input data set based on MIC, and constructing a 24-hour SOC prediction data set of an electric vehicle user as an input data set;

s102, establishing a generator, a value network and a discriminator neural network based on a linear full-connection network construction, and initializing the network by using a Bayesian super-parameter optimization method; inputting a training user track data set, optimizing strategy network parameters based on a PPO near-end strategy optimization algorithm, traversing all user tracks, and generating charge and discharge strategies of all users;

s103, training an XGboost-based 24-hour SOC prediction model of the electric automobile by adopting a cross validation method, wherein the model is divided into a discharging SOC prediction model and a charging SOC prediction model;

s104, planning a historical driving track passing route, acquiring a real-time traffic flow rate of the route based on a Baidu map, and predicting a 24-hour single-vehicle SOC change curve of all users; and (3) forecasting the charging demands and energy demands of all users by combining the charging urgency of the users, and establishing a cluster charging demand forecasting spatio-temporal model of the electric vehicles in the region.

As a preferred embodiment, the constructing of the raw data set of the charging requirement of the electric vehicle in step S101 provided by the embodiment of the present invention includes the following steps:

(1) MIC matrix analysis of charging characteristics

The charging stage has no human subjective factor, each variable changes along with the interaction process of the electric automobile and the charging pile, the whole process is more regular and can follow than the discharging process, the charging process is divided into fast charging and slow charging, and the characteristic correlation of the two processes is basically uniform. The MIC matrix is shown in fig. 2, and the result proves that the charging SOC is strongly correlated with the user charging period and the charging voltage, and the SOC prediction model is discharged here only in consideration of the user charging period.

(2) MIC matrix analysis of discharge characteristics

The discharging process, i.e. the driving process, is a process in which human factors participate, the SOC of the discharging process is controlled by a plurality of factors, the SOC is not limited to the factors shown in the figure 3, and the SOC also comprises road surface conditions (gradient and concave-convex degree), and the method mainly considers the influence of subjective factors, i.e. the driving strategy of a user, on an SOC curve. The MIC matrix of discharge characteristics is shown in fig. 3, and the results demonstrate: the discharge SOC has the strongest correlation with the running distance and the speed; meanwhile, the correlation between the total voltage of secondary relevant factors of the discharging SOC and the voltage, distance and speed of the battery is strong; the user driving strategy is therefore the main cause of the change in the discharge process SOC.

Example 1

Based on fig. 1, the electric vehicle charging demand modeling method provided by the embodiment of the present invention, further, step S101 provided by the embodiment of the present invention further includes constructing a policy network input feature data set, specifically including:

screening abnormal points by using an isolated forest, repairing the abnormal points of the data set by using a multiple interpolation method, and extracting user behavior factors strongly related to SOC (state of charge) change of the electric vehicle based on an MIC (many integrated core) maximum mutual information coefficient matrix;

the original data set includes:

as the user's 24-hour true speed data;

as the user 24-hour true SOC data;

Example 2

Based on fig. 1, the electric vehicle charging demand modeling method provided by the embodiment of the present invention, further, step S102 provided by the embodiment of the present invention further includes a user policy learning method based on generation confrontation mimic learning, specifically including:

constructing a GAIL algorithm framework; the GAIL utilizes expert data to train a discriminator training generator and confuses judgment of discrimination, and utilizes the discriminator to distinguish data distribution generated by the generator from real data distribution;

initializing the policy function pi ₀ Sampling trajectory

According to the principle of generating a confrontation model, the strategy function performs gradient descending update, and the discriminator performs gradient descending update, so that the probability distribution of real data and generated data is distinguished;

wherein, D (x) ⁽ⁱ⁾ ) Is the probability judgment of the discriminator on the real data; d (G (z) ⁽ⁱ⁾ ) Probability decision for the generated data;

a generator: constructing a loss function that generates a penalty function against the network, and constructing a reward function with the evaluator:

when the discriminator can not distinguish the data generated by the generator from the real data, the generator and the discriminator reach Nash equilibrium, and the generator is successfully matched with the expert strategy;

carrying out parameter optimization on the generator network based on a PPO algorithm, wherein the PPO algorithm has the following targets:

1) generalized dominance estimation

Comparing the return obtained by executing the action a in the state s with the return obtained by executing the current strategy pi (s | a) for evaluating the quality of the action a, so that the strategy updates the probability p (a | s) of the action a to the direction of which the advantage estimation is greater than zero; the generalized dominance estimate is as follows:

wherein,

r _t+l is a reward function of the current time, V(s) _t+l ) Is the function of the state value at the current moment, and gamma is the discount rate of the function of the value at the previous moment;

2) status-action profileRatio of r _t (θ)：

Wherein, pi _θ (a _t |s _t ) Logarithmic probability, π, of new strategy to execute a set of actions of a batch _old (a _t |s _t ) Mean logarithmic probability of action representing execution of the old policy; ε is a hyperparameter, which is the truncation threshold; second item

Modifying the substitute target by clipping the probability ratio, and eliminating the motivation (1-epsilon, 1+ epsilon) moving to the outer side of the interval; taking the minimum value of the clipped target and the uncut target, so that the final target is the lower bound of the uncut target; theta _old Taking the first order of the neighborhood;

wherein,

example 3

Based on fig. 1, the electric vehicle charging demand modeling method provided in the embodiment of the present invention, further, the step S102 of constructing a user behavior strategy learning model based on generation of confrontation-imitation learning includes the following steps:

fig. 4 is a dotted line portion which is a GAIL policy learning portion composed of two upper layer policies and one lower layer policy, and a discriminator; the discriminator and the three strategies form a GAIL strategy learning framework; the three strategy networks use user historical data as learning samples, and the user historical strategy distribution is fitted through a discriminator. The charging and discharging strategy makes action output based on the current time and the current SOC; the charging strategy outputs actions of a single charging time and a charging starting time; the trip strategy outputs a single trip target distance and trip starting time; the lower driving strategy executes the upper strategy to output an action target, and outputs 24-hour acceleration, single mileage and time. The three policy networks adopt the same GAIL structure, and a policy learning flow chart is shown in fig. 5, and the specific flow is as follows:

(1) trajectory sampling

(2) PPO strategy optimization

As a preferred embodiment, the XGboost electrically-constructed electric vehicle SOC prediction model combined with the policy model in step S103 provided in the embodiment of the present invention includes the following steps:

according to the correlation analysis of the SOC and the user strategy, a single trip mileage curve and a charging time curve are calculated based on a 24-hour user speed curve generated by strategy learning, wherein the speed curve and the mileage curve predict the discharging SOC, and the charging time curve predicts the charging SOC. Therefore, a driving SOC regression prediction model and a charging SOC regression prediction model are respectively established.

Where the prediction algorithm selects XGboost, the prediction model flow diagram is shown in fig. 6, where the loop is a cross-validation process; and dividing the training data set into n subsets, taking one subset as a verification set and taking the rest subsets as training sets in each circulation until all the subsets are traversed. This ensures reliable generalization ability.

As shown in fig. 4, the method for predicting the cluster charging demand of the electric vehicle provided by the preferred embodiment includes the following steps:

first, data set construction

The method comprises the steps of constructing an original data set of the charging requirement of the electric automobile, carrying out data interpolation, variable conversion and disturbance increasing on actual requirement data, requirement influence condition data and random noise data to construct the original data set, and dividing different weights of the original data set to serve as the input of the next layer.

Secondly, constructing a strategy learning module

And establishing a generator, a value network and a discriminator neural network based on a linear full-connection network, and initializing the network by using a Bayesian hyper-parameter optimization method. Inputting a training user track data set, optimizing strategy network parameters based on a PPO near-end strategy optimization algorithm, and traversing all user tracks. And generating charge and discharge strategies of all users.

Third, XGBoost prediction module is constructed

The XGboost-based 24-hour SOC prediction model of the electric automobile is divided into a discharging SOC prediction model and a charging SOC prediction model.

Fourthly, outputting all the charging requirements of the electric automobile for 24 hours

The method comprises the steps of planning a path based on historical driving tracks, obtaining real-time traffic flow speed of the path based on a Baidu map, predicting 24-hour single-vehicle SOC change curves of all users, predicting charging demands and energy demands of all users by combining charging urgency of the users, and finally predicting the charging demands of clustered electric vehicles in an area.

In the preferred embodiment, in step one, the raw data set comprises:

as user 24 hour true speed data;

as the user 24-hour true SOC data;

In the preferred embodiment, in step S102, the optimal hyper-parameter of the GAIL model is found based on the bayesian hyper-parameter optimization algorithm, as follows:

in order to make the route planning model and the SOC prediction model converge in a short time to achieve better performance, the Bayesian super-parameter optimization algorithm is used for searching the super-parameters. In the strategy model, KL divergence of a true value and a predicted value is used as an objective function of Bayesian optimization, wherein a strategy network and a discriminator network are composed of a plurality of layers of fully-connected neural networks, and network hyper-parameters are shown in tables 1-4.

TABLE 1 PPO Algorithm parameter settings

Table 2 discriminator network parameter settings

Table 3 policy network parameter settings

TABLE 4 Main program parameter settings

Example 4

Based on fig. 1, the modeling method for the charging requirement of the electric vehicle provided by the embodiment of the present invention, further, the method for predicting the 24-hour SOC of the electric vehicle user based on XGboost in step S103 provided by the embodiment of the present invention includes:

according to input characteristic analysis, calculating a single trip mileage curve and a charging time curve based on a 24-hour user speed curve generated by strategy learning, wherein the speed curve and the mileage curve predict a discharging SOC, and the charging time curve predicts a charging SOC; and respectively establishing a driving SOC regression prediction model and a charging SOC regression prediction model based on XGboost, and searching a road planning model and an SOC prediction model based on a Bayesian hyperparameter optimization algorithm.

In the preferred embodiment, in step S103, the optimal hyper-parameters of the XGboost model are found based on the bayesian hyper-parameter optimization algorithm as follows:

the expert track number is set to be 10, in the SOC prediction model, the mean square error of a true value and a predicted value is used as an objective function of Bayesian optimization, the search algebra is 50 times, XGBoost hyper-parameters are shown in a table 5, and all hyper-parameters are accurate to 4-digit decimal numbers.

TABLE 5 XGBOST parameter settings

Example 5

Based on fig. 1, the method for modeling a charging demand of an electric vehicle according to the embodiment of the present invention further includes a step S104 of obtaining a real-time traffic flow rate and a vehicle route plan based on a Baidu map, which specifically includes:

(1) path planning

The path planning adopts a real path longitude and latitude coordinate set in a data set, after data processing is carried out on all path longitude and latitude coordinates, the paths are visualized by using an OSMnx library in python, and the distance information of road node coordinates, road nodes and distance initial nodes is extracted, wherein the path nodes are intersection points of the paths and other roads;

setting each track node data set as:

wherein omega ^j Represents the jth trace data set, j 1,2, 3.,

and

respectively representing longitude and latitude coordinates of the path node and a distance from the starting node;

(2) in-zone standardized electric vehicle path planning

Analyzing the influence of real-time traffic flow rate on a user driving strategy to obtain the real-time average speed of each road section; acquiring the running time T of a road section to which the current vehicle coordinates belong based on a Baidu map real-time information platform, and setting the distance between each adjacent node and the starting node as l ₁ And l ₂ If the current link length is L ═ L ₁ -l ₂ The time required for passing through the current road section is T _t The real-time traffic average flow velocity of the road section is v _t Calculating the real-time average traffic flow rate according to the following formula:

the regional electric automobile cluster charging demand space-time prediction combined with charging urgency of users comprises the following steps:

(1) urgency of charging for user

The higher the charging urgency is, the higher the probability that the user sends a charging demand signal is;

(2) charging demand determination

Fitting the initial charging SOC distributions of different users by using normal distribution; when the SOC approaches the historical charging initial SOC of a certain user and the SOC is reduced to a charging demand interval of the certain user, the user generates a charging demand, and the charging demand interval is related to the charging initial SOC distribution through a user charging urgency coefficient;

(3) regional large-scale electric vehicle energy demand prediction

Establishing a demand forecast of charging energy of a large-scale electric vehicle in a region based on a single SOC forecast result of the electric vehicle; and predicting the charging energy demand of the user by using the charging urgency of the user and the charging initial SOC, considering that the charging demand exists in the user when the SOC of the user enters the charging demand interval according to the definition of the charging demand interval, calculating the charging energy demand of the user, and completing the serial integration of the basic regressor group.

In the preferred embodiment, in step S104, the method for predicting the charging demand based on the charging urgency of the user and the real path data is as follows:

1) path planning

wherein: omega ^j Represents the jth trace data set, j 1,2, 3.,

and

2) In-zone standardized electric vehicle path planning

Consider usingThe driving strategy of a user is influenced by real-time traffic flow velocity, so that real-time average speed of each road section needs to be acquired, the driving time T of the road section to which the current vehicle coordinate belongs is acquired based on a Baidu map real-time information platform, and the distance between adjacent nodes and an initial node is set to be l ₁ And l ₂ Then the current link length is L ═ L ₁ -l ₂ The time required for passing through the current road section is T _t The real-time traffic average flow velocity of the road section is v _t The real-time traffic average flow velocity can be calculated according to the following expression.

3) User charging urgency and battery capacity parameter selection

According to the expression of charging urgency, three parameters can be selected. The three parameters are determined according to the capacity of the battery of the electric automobile. The larger the battery capacity, the larger these three parameters. In the invention, the charge and discharge processes of four types of electric automobile users are simulated, namely a logistics car, a bus, a taxi and a private car. Where the bus battery capacity is set to 135kWh, the charge urgency parameter is set to 0, 0.7, 0.9, and the remaining three vehicle battery capacities are set to 45kWh, and the charge urgency parameter is set to 0, 0.4, 0.85. And respectively establishing a charging urgency model according to the parameters.

Fig. 7 shows a graph of a charging urgency function for a user of an electric vehicle according to an embodiment of the present invention.

4) Charging demand determination method

Here, it is considered that the charging start SOC distribution of a certain user follows a normal distribution N (μ, σ) in the case where the charging probability of the user is larger as the charging start SOC is smaller ² ) From N (μ, σ) ² ) Extracting an initial SOC value X, and setting a user charging urgency coefficient to be C under the current SOC _u (X), then the user charging demand interval is [ X, X + 20C ] _u (X)]The smaller X is seen, the smaller C _u The larger (X), the wider the charging demand interval range, and the greater the possibility that the user generates a charging demand.

5) Regional large-scale electric vehicle energy demand prediction

The method comprises the step of establishing regional electric automobile cluster charging demand prediction by utilizing a large amount of electric automobile operation data. Firstly, a prediction area is selected, then data of 1000 electric automobiles in one month are extracted to predict 24-hour SOC change of 1000 automobiles, and finally charging requirements and energy requirements of all users are predicted by combining the definition of a charging requirement interval. And on the basis of the above steps, a charging demand space-time distribution diagram in the region is established. It is provided that when a user generates a charging request, the difference between the current battery energy and 90 percent of the battery energy is used as the charging energy request, and the charging energy request is that the current battery SOC is set as SOC _t Let the user battery capacity be C _p At this time, the calculation formula of the charging energy requirement of the single user is as follows:

E _pc ＝(1-SOC _t )C _p

the battery capacities of these four vehicle types were selected as shown in table 6.

TABLE 6 Battery Capacity selection

It should be noted that, the steps in the method provided by the present invention may be implemented by using corresponding modules, devices, units, and the like in the system, and those skilled in the art may implement the composition of the system by referring to the technical solution of the method, that is, the embodiment in the method may be understood as a preferred example for constructing the system, and will not be described herein again.

And II, application embodiment. In order to prove the creativity and the technical value of the technical scheme of the invention, the part is the application example of the technical scheme of the claims on specific products or related technologies.

The experimental data of the invention is the running data of 1000 electric vehicles in one month obtained from a Shanghai new energy electric vehicle monitoring center, wherein the running data contains 20 electric vehicle running parameters such as real-time speed, SOC, running distance and the like, data points are sampled once every ten seconds, wherein the data of private cars, logistics cars, buses and taxis account for 10%, 12%, 35% and 43%, and the data points of single trip tracks are about 2000. The data attributes are shown in table 7.

The model simulation analysis is divided into a plurality of parts: firstly, the robustness and the learning ability of the policy network are evaluated, secondly, the difference of user policies is explained for a speed prediction model based on a policy model, and the speed prediction results of four types of users are compared. And then displaying the 24-hour SOC prediction effect of the single vehicle, comparing the 24-hour SOC prediction effect with a historical SOC-based prediction method, and finally displaying a charging demand space-time diagram at all-day key time in the Shanghai main city area. The algorithm program is implemented entirely by python 3.7.

TABLE 7 data Attribute

An embodiment of the present invention provides a terminal, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor is configured to execute the method in any one of the above embodiments of the present invention or execute the system in any one of the above embodiments of the present invention when executing the program.

Embodiments of the present invention also provide a computer-readable storage medium, on which a computer program is stored, which, when executed by a processor, is operable to perform the method of any one of the above-described embodiments of the present invention, or to run the system of any one of the above-described embodiments of the present invention.

In the above two embodiments, optionally, the memory is used for storing a program; a memory, which may include a volatile memory (english: volatile memory), such as a Random Access Memory (RAM), for example, a Static Random Access Memory (SRAM), a Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), and the like; the memory may also include a non-volatile memory (english) such as a flash memory (english). The memories are used to store computer programs (e.g., applications, functional modules, etc. that implement the above-described methods), computer instructions, etc., which may be stored in partition in the memory or memories. And the computer programs, computer instructions, data, etc. described above may be invoked by a processor.

A processor for executing the computer program stored in the memory to implement the steps of the method according to the above embodiments. Reference may be made in particular to the description relating to the previous method embodiments.

The processor and the memory may be separate structures or may be an integrated structure integrated together. When the processor and the memory are separate structures, the memory, the processor may be coupled by a bus.

And thirdly, evidence of relevant effects of the embodiment. The embodiment of the invention achieves some positive effects in the process of research and development or use, and has great advantages compared with the prior art, and the following contents are described by combining data, diagrams and the like in the test process.

1. Real-time speed prediction based on strategic models

In order to explain the influence of the charge-discharge strategy differences of different users on the charge demand time distribution, data are provided for the subsequent prediction of the 24-hour SOC of the single vehicle. The bicycle 24 hour speed prediction is presented here. The user type composition of the real city electric automobile is considered, and four types of users are considered respectively: private cars, logistics cars, buses and taxis, and learning respective charge and discharge strategies, then predicting a 24-hour speed change curve of the single car based on real-time traffic flow rate, as shown by a broken line in a light gray area in fig. 8, and finally calculating single-time driving mileage based on the speed curve, wherein the initial mileage is obtained by averaging historical data of a user. As indicated by the dark broken lines in fig. 8. The following analyzes the differences between the charging and discharging strategies of the user.

As can be seen from fig. 8, the single driving mileage of the taxi and the bus is the longest, and the charging requirements of the logistics vehicle and the bus are distributed between 21:00PM and 8:00AM in the next day due to the functionality of the logistics vehicle and the bus. While substantially stopping operation after 10:00PM, so both assume slow charging at night. The discharge time intervals of the taxis and private cars are relatively not obvious in regularity. Meanwhile, the taxi has obvious charging behavior at about 13:00PM, and because the taxi needs to obtain more travel orders in the daytime, the taxi is charged quickly, so the charging time of the taxi is about 1 hour, and the charging demand of a private car is basically concentrated between 0:00AM and 8:00AM, so that the private car user charges slowly at night. Secondly, the running speeds of the four types of vehicles are in positive correlation with the real-time traffic flow rate, and the learning result of the driving strategy of the user is considerable. In conclusion, due to different charging and discharging strategies, the difference of the charging demand time distribution of different users is obvious.

2. Bicycle 24 hour SOC change prediction

The prediction result is used as the input characteristic of the XGboost algorithm based on the driving speed and the mileage of the user, so that the 24-hour SOC curve of the bicycle is predicted, and the superiority of the method is proved. As shown in fig. 9, four types of electric vehicles were selected for vehicle use and compared with the historical SOC-based prediction method, where the red line represents the true SOC, the blue line represents the historical SOC-based prediction result, and the sky blue represents the real-time traffic flow rate-based prediction result. The two methods have good prediction results of the charging SOC, the prediction curves are smooth, and the stability of the prediction results is good. However, the SOC prediction results in the discharging process are slightly different, and compared with the prediction results of private car users, the prediction results of the two methods are poor in curve stability and obvious in jitter, which is probably related to the characteristic distribution of data, but the two methods can well grasp the overall change trend of the SOC discharging process; in the prediction result of a taxi user, the method disclosed by the invention shows good stability on the prediction result of the discharging process, but the prediction result of the historical SOC-based method has obvious jitter and abnormal values in the latter half of the discharging process, because the historical data is not obvious in characteristic; the system and the method have better robustness and fitting degree on the SOC prediction result of the logistics vehicle user.

Meanwhile, the method provided by the invention has good prediction precision. According to the table 8, four evaluation indexes of regression prediction are selected, prediction accuracy comparison is respectively carried out on the SOC prediction method and the historical SOC-based prediction method according to the four indexes, training sets with the same size are used for training a network in the two methods, the size of the training sets is 5000 pieces of data, from the prediction results of four types of vehicles SOC, each index obtained by the prediction method is better than that of the historical SOC-based method, the average mean square error is reduced by 13%, the average decision coefficient is improved by 4%, the results prove that the two methods are basically equivalent in terms of the SOC prediction problem of the logistics vehicle, and the prediction accuracy of the method is obviously higher than that of the traditional prediction method in terms of the SOC prediction problems of other three types of electric vehicles.

TABLE 8 prediction index

3. Regional electric vehicle cluster charging demand prediction

The selected area of fig. 10 is a main urban area of a certain city, a charging demand space distribution characteristic diagram of 1000 electric vehicles in one day is displayed, 4 key time points (9:00,12:00,18:00 and 24: 00) in one day are intercepted, the unit is kilowatt-hour, only users with charging demands are displayed in fig. 10, each circle point represents that one vehicle has charging demands, the size and the color depth of the point represent the energy demands of the users, the sum of the charging demands in each hour in one day is counted, and the space-time distribution characteristic of the charging demands is analyzed according to the space-time diagram of the charging demands, so that the effectiveness of the model is proved.

3.1 time distribution analysis of charging demand

As can be seen from fig. 11, the charging request is at 12 pm: 00 and 6 at night: about 00 appears charging peak, which reaches about 150. The charging demand is distributed at 12 noon: 00 and 6 at night: a charging peak occurs around 00, the sum of the charging energy requirements reaches around 4500kW (combined with histogram analysis), and in the morning 3: about 00 charging demands enter the valley. The charging demand time distribution is smoother and the peak period duration is longer on the whole.

3.2 spatial distribution analysis of charging demand

As shown in fig. 10, from the perspective of the spatial distribution of charging demand, night 12:00 to 9 in the morning: the 00 charging demand is in a valley period, the charging demand has no obvious gathering demand, the charging demand is gradually increased after nine points in the morning, the charging demand is radially distributed by taking the main urban area as the center, meanwhile, the charging demand dense areas are formed in the main urban area and the new area, and the charging demand in the suburban area is uniformly distributed.

From the distribution of the charging energy demand all day long, the charging energy demand close to the city center is mostly distributed below 30 kilowatt hours, so that the charging users mainly use taxis and private cars, and the charging energy demand in suburbs is distributed above 60 kilowatt hours, which indicates that the charging users mainly use buses. This is of course related to the ratio of the number of different vehicle types, in the present case the taxis are the majority.

The above description is only for the purpose of illustrating the embodiments of the present invention, and the scope of the present invention should not be limited thereto, and any modifications, equivalents and improvements made by those skilled in the art within the technical scope of the present invention as disclosed in the present invention should be covered by the scope of the present invention.

Claims

1. The electric vehicle charging demand modeling method is characterized by comprising the following steps:

analyzing the influence of the differences of the users of the electric vehicles and the differences of the performances of the electric vehicles, and learning a user driving strategy and a user traveling strategy based on the generated countermeasure model; learning an electric vehicle performance model based on an XGboost machine; the real-time road condition based on the Baidu map proves that the algorithm can extract the driving characteristics and the charging characteristics of the user.

2. The electric vehicle charging demand modeling method of claim 1, wherein the electric vehicle charging demand modeling method comprises the steps of:

training an XGboost-based 24-hour SOC prediction model of the electric automobile by adopting a cross validation method, wherein the model comprises a discharging SOC prediction model and a charging SOC prediction model;

3. The electric vehicle charging demand modeling method of claim 2, wherein the first step further comprises constructing a policy network input feature dataset, specifically comprising:

the original data set includes:

as user 24 hour true speed data;

as the influence factor of the 24-hour speed change of the user: real-time traffic flow rate;

as the user 24-hour true SOC data;

4. The electric vehicle charging demand modeling method according to claim 2, wherein the second step further includes a user policy learning method based on generation confrontation mimic learning, and specifically includes:

initializing a policy function pi ₀ Sampling trajectory

According to the principle of generating the confrontation model, the strategy function performs gradient ascending updating, the discriminator performs gradient descending updating, and then the probability distribution of real data and generated data is distinguished;

a generator: constructing a loss function that generates a challenge network, and constructing a reward function with the authenticator:

performing parameter optimization on the generator network based on a PPO algorithm, wherein the PPO algorithm targets are as follows:

1) generalized dominance estimation

wherein,

2) state-to-action probability ratio r _t (θ)：

Wherein, pi _θ (a _t |s _t ) Logarithmic probability, π, of new strategy to execute a set of actions of a batch _old (a _t |s _t ) Mean logarithmic probability of action representing execution of the old policy; epsilon is a hyperparameter, which is the truncation threshold; second item

Modifying the substitute target by cutting probability ratio to eliminate the motivation (1-epsilon, 1+ epsilon) moving to the outside of the interval; taking the minimum value of the clipped target and the uncut target, so that the final target is the lower bound of the uncut target; theta _old Taking the first order of the neighborhood;

wherein,

5. the modeling method for the charging demand of the electric vehicle as claimed in claim 2, wherein the XGboost-based method for predicting the 24-hour SOC of the electric vehicle user in the third step comprises:

6. The modeling method for electric vehicle charging demand according to claim 2, wherein the fourth step further comprises obtaining real-time traffic flow rate and vehicle path planning based on a Baidu map, and specifically comprises:

(1) path planning

setting each track node data set as:

wherein omega ^j Represents the jth trace data set, j 1,2, 3.,

and

(2) in-zone standardized electric vehicle path planning

(1) urgency of charging for user

(2) charging demand determination

Fitting the charging initial SOC distribution of different users by using normal distribution; when the SOC approaches the historical charging initial SOC of a certain user and the SOC is reduced to a charging demand interval of the certain user, the user generates a charging demand, and the charging demand interval is related to the charging initial SOC distribution through a user charging urgency coefficient;

(3) regional large-scale electric vehicle energy demand prediction

7. An electric vehicle charging demand modeling system for implementing the electric vehicle charging demand modeling method according to any one of claims 1 to 6, the electric vehicle charging demand modeling system comprising:

the system comprises an input data set construction module, a data processing module and a data processing module, wherein the input data set construction module is used for constructing an electric automobile operation original data set, and dividing the original data set into different weights to be used as an input data set;

the system comprises a discriminator neural network, a strategy neural network and a value neural network construction module, wherein the discriminator neural network, the strategy neural network and the value neural network construction module jointly form a strategy learning system; the reinforcement learning environment construction module is used for setting a state-action output function and setting an environment state and a strategy network action range; a PPO algorithm building module; the module carries out optimization updating on the strategy network parameters;

the system comprises a single-vehicle 24-hour SOC prediction module and an electric vehicle 24-hour SOC prediction model based on XGboost, wherein the model is divided into a discharging SOC prediction model and a charging SOC prediction model;

the regional electric vehicle cluster charging demand prediction construction module is used for planning a path based on a historical driving track, acquiring real-time traffic flow rate of the path based on a Baidu map, predicting a 24-hour single vehicle SOC change curve of all users, predicting charging demands and energy demands of all users by combining charging urgency of the users, and finally predicting the charging demands of the regional cluster electric vehicles.

8. A computer device, characterized in that the computer device comprises a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor, so that the processor executes the electric vehicle charging demand modeling method according to any one of claims 1 to 6.

9. A computer-readable storage medium, storing a computer program which, when executed by a processor, causes the processor to execute the method for modeling the charging requirement of an electric vehicle according to any one of claims 1 to 6.

10. An information data processing terminal, characterized in that the information data processing terminal is used for implementing the electric vehicle charging demand modeling system according to claim 7.