Electric vehicle differentiated charging package design method considering user behavior portrait and charging load

Abstract

The invention belongs to the technical field of cooperation of electric vehicles and electric power systems, and discloses an electric vehicle differentiated charging package design method considering user behavior portraits and charging loads, and the method comprises the following steps: constructing a user feature library, and screening key features; user clustering is carried out based on the key features, and user portraits with group distinction degree and feature interpretability are constructed; establishing a user psychological-behavior association model for quantifying a mapping relationship between the psychological latent variable and the charging behavior; the accurate prediction of the charging load is realized by combining a user portrait and a psychological-behavior rule and integrating SOC dynamic constraint and an elastic queuing model; and based on a master-slave game model, designing a differentiated charging package by taking maximization of user utility and profit of an electric vehicle aggregator as a target. The charging package design method provided by the invention solves the problems that the traditional method neglects the intrinsic decision logic of the user, the load prediction deviation is large, the package breaks away from the actual demand and the like, and realizes the win-win situation of the user and the power grid.

Description

Technical Field

[0001] This invention belongs to the field of electric vehicle and power system collaborative operation technology, and specifically relates to a method for designing differentiated charging packages for electric vehicles that takes into account user behavior profiles and charging load. Background Technology

[0002] Vehicle-to-Grid (V2G) interaction is a core technology for the coordinated operation of electric vehicles and power systems. Traditional research often focuses on grid-side load and demand, making it difficult to adapt to the impact of user subjectivity and randomness on electric vehicle (EV) travel habits. The construction and analysis of user profiles can accurately identify users' psychological sensitivities. Introducing user profiles and analyzing the relationship between user psychology and travel behavior can focus research on the user side. Based on this, we can promote user-side-based charging load forecasting and user-differentiated charging package design, systematically making up for the shortcomings of traditional V2G research.

[0003] Existing research has the following shortcomings: 1) It often equates a single user attribute with a complete user profile, unilaterally considering demographic characteristics and vehicle parameters while ignoring key endogenous factors such as psychological preferences; 2) Research on charging load forecasting is relatively simplistic, mostly based on pure data simulation, lacking research on the psychological latent variables driving charging behavior, and focusing on forecast accuracy assessment while ignoring the impact of user profile-behavior correlation on load forecasting and package design; 3) Package design is detached from users' actual needs, failing to fully integrate the peak and valley characteristics of load forecasting with user utility functions, resulting in difficulties in implementation and hindering the achievement of a win-win situation for users and the power grid.

[0004] Therefore, how to combine user profiles to build a psychological-behavioral correlation, improve the accuracy of charging load prediction, and design differentiated charging packages that maximize the benefits for both parties has become an urgent problem to be solved. Summary of the Invention

[0005] Therefore, the purpose of this invention is to provide a method for designing differentiated charging packages for electric vehicles that takes into account user behavior profiles and charging load, so as to solve the problems existing in the prior art.

[0006] The technical solution of the present invention is as follows: A method for designing differentiated charging packages for electric vehicles that considers user behavior profiles and charging load, comprising the following steps:

[0007] Collect relevant data on electric vehicle users, build a user feature database, and use structural equation modeling to screen key features that significantly affect charging behavior;

[0008] Users are clustered based on the key features obtained from the screening, and probability distribution models of key features of different user groups are constructed to form typical user profiles.

[0009] Based on the user profile, user psychology-behavior correlation models are established for different categories of users, wherein the user psychology-behavior correlation models are used to quantify the correlation between user psychological preferences and behavior;

[0010] By combining user profiles and user psychology-behavior correlation models, and introducing SOC dynamic constraints and elastic M / M / c / N queuing models, we can simulate real charging scenarios and achieve accurate prediction of charging load.

[0011] Based on user profile characteristics and load forecast results, several typical packages are designed. A master profit function aimed at maximizing the profit of electric vehicle aggregators and a slave utility function aimed at maximizing user utility are constructed. The Nash-Q reinforcement learning method is used to solve the master-slave game model to obtain the optimal package parameters.

[0012] Preferably, the user feature database includes primary tags, secondary tags, and tag types. The primary tags include basic user attributes, user behavioral characteristics, user attitude preferences, and grid contribution and response. The basic user attributes include two secondary tags: personal attributes and vehicle attributes, both of which are factual. The user behavioral characteristics include three secondary tags: travel characteristics, charging behavior, and time attitude, with tag types of rule, fact, and rule, respectively. The user attitude preferences include three secondary tags: price sensitivity, environmental willingness, and charging satisfaction, with tag types of prediction, forecast, and rule, respectively. The grid contribution and response include two secondary tags: user contribution level and package response behavior, both of which are rule-based.

[0013] Further optimization and selection of key features are achieved through the following methods:

[0014] Construct a structural equation model, which comprises two parts: a measurement model and a structural model. The measurement model is used to observe the relationship between variables and latent variables, and the structural model is used to describe the causal relationship between latent variables.

[0015] The model parameters are solved by the maximum likelihood estimation method, and the goodness of fit of the model is tested by the ratio of chi-square to degrees of freedom, the comparison of fit index, and the root mean square of approximation error. If the goodness of fit of the model does not meet the standard, the model is adjusted until it meets the standard.

[0016] A t-test was performed on the model path coefficients to retain variables with significant influence; the Pearson coefficient was introduced to remove highly redundant features, resulting in a key feature set.

[0017] Further optimization involves the following steps for clustering users based on key features:

[0018] The key features are transformed into p-dimensional feature vectors, and the number of clusters is preset to K. The core information of the user group is compressed into K initial CF units through the clustering features.

[0019] Iterate through all user feature vectors, calculate the distance between each user and the centroids of the K CF unit clusters, and assign the user to the most similar CF unit to complete user classification;

[0020] The silhouette coefficient is used to evaluate the clustering quality, ensuring that the weighted mean of the silhouette coefficient meets the preset requirements.

[0021] Further optimization involves constructing probability distribution models of key features for different user categories to form typical user profiles. The specific methods are as follows:

[0022] Assuming that the key features of each group follow a Gaussian mixture model, the model parameters are iteratively updated until the distribution fit no longer improves. The distribution is then corrected by incorporating behavioral scenario logic to obtain a typical user profile.

[0023] Further optimization involves using the user psychology-behavior correlation model to quantify the relationship between four types of psychological preferences—range anxiety, time tolerance, price sensitivity, and environmental willingness—and the starting state of charge (SOC), charging power selection, off-peak charging ratio, and green electricity charging ratio.

[0024] Further optimization reveals that the quantitative relationships in the user psychology-behavior correlation model include:

[0025] 1) Range anxiety and initial SOC: The higher the anxiety index, the closer the initial SOC is to the maximum safe threshold.

[0026] 2) Time tolerance and charging power selection: The lower the time tolerance, the higher the probability of choosing fast charging;

[0027] 3) Price sensitivity and off-peak charging rate: The lower the price sensitivity, the more positive the response to low prices;

[0028] 4) Environmental awareness and green electricity charging ratio: The stronger the environmental awareness, the higher the green electricity charging ratio.

[0029] Further optimization involves combining user profiles and user psychology-behavior correlation models, introducing SOC dynamic constraints and elastic M / M / c / N queuing models to simulate real charging scenarios and achieve accurate prediction of charging load. Specifically, this includes the following steps:

[0030] By combining user profile proportions with user psychology-behavior correlation models, we define initial parameters for scenarios and quantify the psychology-behavior characteristics of different groups.

[0031] Based on battery energy conservation and travel demand, a dynamic constraint on SOC is constructed to correlate charging behavior and charging characteristics.

[0032] Introducing an elastic M / M / c / N queuing model, considering user stopping behavior and impatience, we calculate the steady-state probability of a charging station containing k electric vehicles.

[0033] Based on the queuing model, the number of charging piles occupied within a specific time period t is statistically analyzed using Monte Carlo simulation, and the charging load is predicted by combining the average charging power.

[0034] Further optimization resulted in four types of packages:

[0035] Peak-hour fast charging + off-peak monthly subscription package: targeting high-frequency, long-distance users with high range anxiety and low tolerance for time, including fast charging subsidies;

[0036] Off-peak tiered plans: designed for commuters with moderate price sensitivity and fixed travel times, offering lower off-peak electricity prices;

[0037] Green electricity off-peak packages: targeting green users with high environmental awareness and low price sensitivity, with a high proportion of green electricity consumption;

[0038] Off-peak pricing packages: Targeting price-sensitive users with low charging frequency and short distances, offering no fixed fees and low prices per charge.

[0039] Further optimization yields the following solution process for the master-slave game model:

[0040] Construct the main party's profit function, where the main party's profit function = basic package revenue + demand response revenue + green electricity revenue - operation and maintenance costs - electricity purchase costs;

[0041] Construct a utility function for the user, where the utility function for the user = cost satisfaction + convenience satisfaction + environmental benefits, and the weighting coefficients are adjusted according to user characteristics;

[0042] The Nash-Q reinforcement learning method is adopted to iteratively adjust the package parameters in response to the user's optimal package selection until a Nash equilibrium is reached, thus obtaining the optimal package parameters.

[0043] The optimal package parameters are verified by load constraints and user acceptance to ensure that the package parameters meet the requirements of power grid regulation and control and the willingness of users to participate.

[0044] This invention provides a method for designing differentiated charging packages for electric vehicles that considers user behavior profiles and charging load. It integrates user behavior profiling, a psychological-behavioral correlation model, and master-slave game optimization. By constructing a multi-dimensional user feature library and filtering key features, it deeply binds users' basic attributes, psychological preferences, and charging behavior, overcoming the limitations of traditional methods that only focus on surface behavior while ignoring the underlying decision-making logic. By establishing a psychological-behavioral correlation model, it transforms static profiles into dynamic behavior predictions, and combines SOC constraints and a flexible queuing model to improve the authenticity and accuracy of charging load predictions. Based on user profiles and psychological needs, it designs differentiated packages and utilizes master-slave game optimization to maximize user utility and grid regulation value, solving the implementation difficulties caused by packages being divorced from user needs.

[0045] The electric vehicle differentiated charging package design method provided by this invention, based on the construction of user profiles, integrates psychological-behavioral models and master-slave game optimization models, realizes the mapping from key user characteristics to package design, and can maximize user utility and grid regulation value. Attached Figure Description

[0046] Figure 1 is a flowchart of the electric vehicle differentiated charging package design method that considers user behavior profiles and charging load provided by the present invention;

[0047] Figure 2 is a flowchart of the key feature selection process in the user feature database;

[0048] Figure 3 is a schematic diagram of the user elastic M / M / c / N queuing model;

[0049] Figure 4 shows the state transition diagram of the elastic M / M / c / N model. Detailed Implementation

[0050] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0051] like Figure 1 As shown, the method for designing differentiated charging packages for electric vehicles that considers user behavior profiles and charging load, provided by the present invention, includes the following steps:

[0052] S1: Collect relevant data on electric vehicle users, build a user feature database, and use structural equation modeling to screen key features that significantly affect charging behavior;

[0053] User-related data can be obtained from channels such as questionnaires and publicly available data from relevant companies;

[0054] The user feature database includes primary tags, secondary tags, and tag types. As shown in Table 1, primary tags include basic user attributes, user behavioral characteristics, user attitude preferences, and grid contribution and response. Basic user attributes include two secondary tags: personal attributes (age, driving experience, income, etc.) and vehicle attributes (range, battery capacity, etc.), both of which are factual. User behavioral characteristics include three secondary tags: travel characteristics (average daily mileage, peak travel times, etc.), charging behavior (start time, power, duration, etc.), and time attitude (queue time threshold, arrival time threshold, etc.), with tag types of rule, fact, and rule, respectively. User attitude preferences include three secondary tags: price sensitivity (acceptance of price increases, willingness to exchange money for time, etc.), environmental willingness (green electricity selection tendency, carbon emission reduction focus, etc.), and charging satisfaction (facility layout satisfaction, predicted mileage accuracy), with tag types of prediction, forecast, and rule, respectively. Grid contribution and response include two secondary tags: user contribution (load transfer volume, off-peak charging ratio) and package response behavior (package selection tendency, renewal rate, etc.), both of which are rule-based.

[0055] surface User Feature Database

[0056]

[0057] Among them, such as Figure 2 As shown, the method for filtering key features is as follows:

[0058] S11: Construct a structural equation model (SEM), which consists of two parts: a measurement model and a structural model. The measurement model is used to observe the relationship between variables and latent variables, while the structural model is used to describe the causal relationship between latent variables.

[0059] The mathematical expression of the measurement model is as follows:

[0060] ;

[0061] ;

[0062] In the formula, For the vector of exogenous observed variables, For the exogenous load matrix, This represents the strength of the influence of the i-th exogenous latent variable on the x-th observed variable. Represents the vector of exogenous latent variables. For the error vector of exogenous observation variables; For the vector of endogenous observed variables, For the endogenous load matrix, This represents the strength of the influence of the i-th endogenous latent variable on the x-th observed variable. Represents the vector of endogenous latent variables. This is the error vector of exogenous observation variables.

[0063] The structural model can be expressed mathematically as follows:

[0064] ;

[0065] in, Let Γ be the relationship matrix among endogenous latent variables, Γ be the influence matrix of exogenous latent variables on endogenous latent variables, and ζ be the structural model error vector. Path coefficient;

[0066] S12: The model parameters are solved by the maximum likelihood estimation method, and the goodness of fit of the model is tested by the ratio of chi-square to degrees of freedom, the comparison fit index (CFI), and the root mean square error of approximation (RMSEA). If the goodness of fit of the model does not meet the standard, the model is adjusted until it meets the standard.

[0067] Among them, the maximum likelihood estimation method is used to estimate the SEM model parameters, with the objective of solving the problem that makes the parameters of the SEM model more likely to be obtained. The largest The objective function is:

[0068] ;

[0069] in, These are the parameters to be estimated (the undetermined coefficients in SEM). The total number of observed variables (determined by the number of labels). For matrix trace operations;

[0070] After completing the maximum likelihood parameter estimation, the core goodness-of-fit index is calculated for the constructed SEM model to verify the fit between the model and the data. The goodness-of-fit test must satisfy formula (5):

[0071] ;

[0072] in, The ratio of chi-square to degrees of freedom is given by , CFI is the comparison fit index, and RMSEA is the root mean square error of approximation. If the goodness of fit is not up to standard, the model is adjusted by modifying the latent variable structure, adding or removing observed variables, and then re-estimating. If the goodness of fit is up to standard, the model proceeds to the subsequent parameter significance test and feature redundancy screening.

[0073] S13: Perform a t-test on the model path coefficients to retain variables with significant influence; introduce the Pearson coefficient to remove highly redundant features and obtain the key feature set;

[0074] For the model Path coefficients in Perform a t-test, null hypothesis Test statistic , To estimate the standard deviation, if ,but Significantly, this exogenous variable has a significant impact on the endogenous latent variable;

[0075] For the observed variables with significant influencing factors identified using the SEM model, the Pearson coefficient is introduced to remove highly redundant features:

[0076] ;

[0077] in, , There are two observed variables; , The mean of the variable; For the sample size, if If the two are highly correlated, then a selection is made, retaining the features with clearer physical meaning and more direct impact on the solution. In this invention, the features with clearer impact on charging behavior are selected, thereby deriving a set of key features of electric vehicle users that meet the target requirements.

[0078] S2: Based on the key features obtained from the screening, users are clustered, and a probability distribution model of the key features of different user groups is constructed to form typical user profiles;

[0079] In this step, users are first aggregated based on the key user features obtained in S1 to form a reasonable group. Then, the distribution pattern of the features of each group is further quantified to finally obtain a user profile that combines "group distinguishability" and "feature interpretability".

[0080] The steps for clustering users based on the key features are as follows:

[0081] S21: Transform key features into p-dimensional feature vectors, preset the number of clusters K=6, and compress the core information of the user group into K initial CF units through cluster features (CF);

[0082] The initial CF cell is represented as follows:

[0083] ;

[0084] in, The number of users represented by a CF unit; It is the linear sum of the feature vectors of all users in this group; This is the sum of squares of the feature vectors of all users in this group;

[0085] S22: Traverse all user feature vectors and calculate the distance between each user and the centroids of the K CF unit clusters. Then, users are assigned to the CF units in the most similar leaf nodes until all users have been classified. The formula is as follows:

[0086] ;

[0087] S23: Use silhouette coefficients to evaluate clustering quality, ensuring that the weighted mean of the silhouette coefficients meets preset requirements. The formula for the silhouette coefficients is as follows:

[0088] ;

[0089] in, For the sample The profile coefficient; For the sample The average distance to all other points within its cluster; For the sample The silhouette coefficient is the average distance to all points in a cluster that does not contain it. Its value ranges from -1 to 1; the closer the value is to 1, the better the clustering. It should satisfy the following:

[0090] ;

[0091] in, The weighted mean of the six cluster profile coefficients. The mean silhouette coefficient of samples in each cluster. The sample size in each cluster, where N is the total number of users;

[0092] At this point, user clustering can be completed, with users with similar characteristics grouped into the same group, resulting in K initial user clusters. This result provides a classification basis for subsequent user profiling analysis.

[0093] The specific method for constructing probability distribution models of key features of different user types to form typical user profiles is as follows:

[0094] S24: Assuming that the key features of each group follow a Gaussian mixture model, iteratively update the model parameters until the distribution fit no longer improves, incorporate behavioral scenario logic to correct the distribution, and obtain a typical user profile.

[0095] 1) Calculate the attribution probability of Gaussian distributions with different characteristics in the population. Use conventional initialization methods to determine the initial parameters of the Gaussian mixture model. Then, update the mean, dispersion, and weight of each Gaussian distribution based on the attribution probability. Iterate repeatedly until the fit between the distribution model and the actual feature data no longer improves. Finally, obtain a clear distribution pattern of each population feature:

[0096] ;

[0097] in, As a characteristic; For the first A cluster; For the first The Gaussian probability density function of each cluster; For the first The weights of each cluster, ;

[0098] Based on Bayesian clustering, the probability of each sample belonging to each Gaussian distribution is calculated as follows:

[0099] ;

[0100] Update the model parameters (weights, mean, variance, etc.) based on the attribution probability, and substitute them into the model expression to obtain the distribution probability of the key feature in the kth cluster.

[0101] 2) In the process of distribution fitting, the behavioral scenario logic of the group is incorporated to ensure that the fitted distribution conforms to the actual behavioral habits. For example, if the charging time distribution of the commuting group shows a "morning peak", the distribution weight is adjusted to correct it to a reasonable distribution of "after the evening travel peak" so that the quantitative results are consistent with the behavioral attributes of the group.

[0102] Therefore, the optimized typical user profile can be obtained based on the fitting results;

[0103] S3: Based on the user profile, establish a user psychology-behavior association model for different categories of users, wherein the user psychology-behavior association model is used to quantify the relationship between user psychological preferences and behavior;

[0104] Among them, user psychological preference tags include four categories of psychological preferences: battery life anxiety, time tolerance, price sensitivity, and environmental willingness.

[0105] Behavioral correlation metrics include initial SOC at charging start, charging power selection, percentage of off-peak charging, and percentage of green electricity charging.

[0106] The user psychology-behavior relationship table is shown in Table 2.

[0107] surface User Psychology-Behavior Correlation Table

[0108]

[0109] Among them, the correlation between quantitative psychological preferences (psychological latent variables) and specific charging behaviors includes range anxiety. With charging at the start of SOC Relevance, Time Tolerance With charging power selection Correlation, price sensitivity Off-peak charging rate Relatedness and environmental consciousness Green electricity charging ratio Association. The relevant calculation method is as follows:

[0110] 1) Range anxiety With charging at the start of SOC Related

[0111] Range anxiety refers to a user's level of concern about insufficient battery power affecting their travel. Higher anxiety levels indicate a greater likelihood of starting to charge when the State of Charge (SOC) is at a higher value. The formula for quantifying the SOC at the start of charging is:

[0112] ;

[0113] in, This indicates the start of SOC (State of Charge) during charging, with a value range of [0.2, 0.8]. The initial SOC (State of Charge) for when there is no anxiety is set at 20%, based on the industry's standard safety threshold. The system initiates SOC (State of Charge) charging when users experience extreme anxiety, which is also the highest safety threshold, set at 80%; (Battery anxiety index) Normalization is performed based on the user data from step 1;

[0114] 2) Time tolerance With charging power selection Related

[0115] Time tolerance This refers to the user's acceptable total time for queuing and charging. The lower the user's tolerance, the more likely they are to choose fast charging. The probability of choosing fast charging is... Then we have the expression:

[0116] ;

[0117] in, For model parameters, , , ; Time tolerance Normalization is performed based on user data in S1;

[0118] 3) Price sensitivity Off-peak charging rate Related

[0119] Price sensitivity This refers to the user's sensitivity to changes in charging electricity prices. If the sensitivity is high... The smaller the value, the more sensitive users are to the incentive of low prices during off-peak hours, the more willing they are to adjust their charging time, and the more actively they respond to the incentive. The expression is as follows:

[0120] ;

[0121] in, For when At that time, the basic low percentage of users with high sensitivity, that is, the lowest percentage without price incentives; For when At that time, the highest and lowest percentage of users with low price sensitivity corresponds to the percentage of ideal users who fully accept price incentives; price sensitivity Normalization is performed based on the user data from step 1;

[0122] 4) Environmental awareness Green electricity charging ratio Related

[0123] Environmental willingness This refers to users' tendency to choose green electricity. The stronger the willingness, the higher the probability that users will pay a higher price for choosing green electricity, and thus the higher the green electricity charging ratio. The higher. Its expression is as follows:

[0124] ;

[0125] Among them, the environmental protection willingness index Normalization is performed based on the user data from step 1; The maximum proportion of green electricity supplied to the power grid, determined based on the current level of renewable energy consumption, is 30%. The green charging pile coverage coefficient should be determined based on the actual charging area conditions of the target user group;

[0126] After quantifying the model and calculating with relevant parameters, the results are judged based on the real-world logic of the scenario to analyze whether they conform to the actual charging scenario. At the same time, the results of groups with the same psychological characteristics are correlated to ensure that the quantification results of the same characteristic groups remain consistent within a specific range, that is, the universality of the rules in the same group is guaranteed. If it is found that it does not conform to the logic, the model parameters should be corrected and reversed until it conforms to the actual rules.

[0127] Therefore, all the obtained psychological-behavioral association results that conform to the actual rules are classified according to the profile to form a user psychological-behavioral association model;

[0128] S4: Combining user profiles and user psychology-behavior correlation models, SOC dynamic constraints and elastic M / M / c / N queuing models are introduced to simulate real charging scenarios and achieve accurate prediction of charging load;

[0129] As electric vehicles gradually become a research hotspot, in order to better promote the precise regulation of the charging behavior of massive electric vehicles on the power grid side and ensure safety and efficiency, this step is based on user profiles and user psychology-behavior correlation models to predict electric vehicle charging load.

[0130] S41: Combining user profile proportions with user psychology-behavior correlation models, define initial parameters for the scenario and quantify the psychology-behavior characteristics of different groups;

[0131] Assuming the total number of users in the scenario is N, based on the obtained user classification and profiling analysis results and the psychological-behavioral correlation model, we utilize users' battery life anxiety. Time tolerance Price sensitivity These indicators are used to quantify the psychological and behavioral characteristics of users in different categories who share the same traits. After classification, K types of user profiles are obtained, where the proportion of the k-th type is denoted as . The number of users in the corresponding category is Thus, the initial parameter definition is completed. Based on the scenario to be studied, the profile can be associated with the specific charging scenario, and the parameters involved in the specific scenario can be further introduced.

[0132] S42: Based on battery energy conservation and travel demand, construct SOC dynamic constraints and associate parameters such as SOC at the start of charging, energy consumption per trip, and charging power;

[0133] SOC constraints help to match charging behavior with charging characteristics. The SOC dynamic constraint calculation formula and related constraints are as follows:

[0134] ;

[0135] ;

[0136] ;

[0137] ;

[0138] in, To determine the target SOC, start the SOC process from charging. Battery life anxiety for users Joint decision, The state of charge of a single user u when arriving at the charging station; This represents the minimum safe state for the battery, typically set at 20%. The SOC reduction for a single trip for users in the kth user profile; The charging energy required for a single user u; This represents the average battery capacity of the vehicle type for user category k. The charging power for a single user u is subject to time tolerance. With the probability of choosing fast charging The impact; The theoretical charging time for a single user u; This indicates the charging efficiency, typically set to [0.85, 0.95].

[0139] S43: Introduce an elastic M / M / c / N queuing model, consider user stopping behavior and impatience behavior, and calculate the steady-state probability of a charging station containing k electric vehicles;

[0140] Among them, the elastic M / M / c / N queuing model (such as...) Figure 3 (As shown) By incorporating user queuing behavior factors (such as stopping and impatience), the system more accurately simulates user charging behavior, making it closer to real-world scenarios. The average service rate of a single charging station in the queuing system is denoted as... It represents the average charging time for users. The reciprocal of, that is ;

[0141] When the number of EVs in the queue is large, users can refuse to enter the queue, even if the number of EVs in the queue has not yet reached the charging station's capacity. Let the probability of a driver refusing to enter the queue increase as the number of electric vehicles in the current queue increases, denoted as . The definition is as follows:

[0142] ;

[0143] In the formula: k represents the EV users in the charging station queue; This is a fixed value, indicating a tendency for EV users to refuse to enter charging stations;

[0144] When the number of EVs in the queue is large, some users in the queue may choose to leave early due to impatience; assuming that the probability of these drivers choosing to leave the queue early increases with the increase of k, this probability is represented here as... And defined as follows:

[0145] ;

[0146] In the formula: δ is a fixed value, which is a parameter representing the user's level of impatience.

[0147] From this, we can obtain the average arrival rate of the elastic M / M / c / N model. and average departure rate Calculation formula:

[0148] ;

[0149] ;

[0150] Under long-term operation, what is the steady-state probability that a charging station contains k electric vehicles? for:

[0151] ;

[0152] In the formula, X(t) represents the number of electric vehicles in the charging station at time t. Figure 4 The state transition diagram of the elastic M / M / c / N model is shown.

[0153] Based on the M / M / c / N queuing theory model, the steady-state probability of a charging station containing k electric vehicles under long-term operation is... for:

[0154] ;

[0155] For steady-state probability constraints;

[0156] The steady-state probability calculation equation and the steady-state probability constraint equation can be used to define... :

[0157] ;

[0158] S44: Based on the queuing model, the number of charging piles occupied within a specific time period t is statistically analyzed using Monte Carlo simulation. Predicting charging load by combining average charging power ;

[0159] in, The calculation formula is as follows:

[0160] ;

[0161] The amount of electricity charged by electric vehicles at the charging station The calculation formula is as follows:

[0162] , The probability of choosing fast charging is based on the average charging power of current charging users and the psychological-behavioral association model. Influence;

[0163] S5: Based on user profile characteristics and load forecast results, design multiple typical packages, construct a master profit function with the goal of maximizing the profit of electric vehicle aggregator (EVA) and a slave utility function with the goal of maximizing user utility, and use the Nash-Q reinforcement learning method to solve the master-slave game model to obtain the optimal package parameters;

[0164] To fully unleash the potential value of electric vehicle users and enhance their contribution to the power grid, this invention proposes charging packages based on user profiles, utilizing master-slave game theory as the core framework to maximize profit and utility. Based on relevant user data, group behavioral characteristics revealed by user profile clustering, and core demands quantified by the psychological-behavioral correlation model (such as the demand for fast charging, low prices, and green electricity), this invention summarizes four typical packages, as shown in Table 3. These package prototypes cover the core value orientations of major user groups, and their specific parameters are optimized and determined through a master-slave game model.

[0165] Four types of packages are designed for K user clusters (K=6 in this embodiment). The number of package types M (M=4 in this embodiment) does not need to be the same as the number of user profile categories K. The packages are standardized products designed for user groups with similar behavioral patterns and value orientations. In the game theory model, the same package can be selected by users from multiple profile categories, and users from the same profile category can also decide between different packages. Based on this step, differentiated charging package design will be carried out. Different types of packages can be designed according to the actual situation, and different tiers can be set for different types to ensure that each type of user corresponds to at least one suitable package.

[0166] surface User types, characteristics, and package design

[0167]

[0168] There are four types of user packages, and the key parameter for the m-th package is the tiered electricity price. Fixed costs Green electricity premium Let the load peak-valley characteristics output in S3 be denoted as the peak period. Low periods Peak-valley load difference is ;

[0169] The solution process for the master-slave game model is as follows:

[0170] 1) Construct the principal profit function based on key user characteristics and calculate the load forecasting results based on S4;

[0171] The goal of an Electric Vehicle Aggregator (EVA) is to maximize total profit. The text is expressed as "the sum of basic package benefits + additional benefits - costs":

[0172] ;

[0173] Basic package benefits The expression obtained by covering different user groups according to package type:

[0174] ;

[0175] in, The average monthly charging amount for user type k in the m-th package; For user of type k, the charging amount ratio at level d under the m-th package; The value is 0 or 1, indicating whether the k-th type of user has selected the m-th package; This is the fixed cost for the m-th package.

[0176] Additional benefits Including demand response revenue and green electricity revenue, therefore:

[0177] ;

[0178] Demand Response Benefits With charging load Related, among which, The unit price for the power grid's demand response to EVA;

[0179] ;

[0180] Green electricity revenue Green electricity consumption of type k users Related to the corresponding green electricity charging ratio ,in, The premium unit price for green electricity packages;

[0181] ;

[0182] Total cost Including operation and maintenance costs and electricity purchase costs, then:

[0183] ;

[0184] 2) Construct utility functions based on user profiles;

[0185] The user-side goal is to maximize the utility of participating in the package. User utility = cost satisfaction + convenience satisfaction + environmental benefits Indicates the weights of different parts:

[0186] ;

[0187] Cost satisfaction The cost is linked to whether or not a package deal is available, and is influenced by price sensitivity. The corresponding off-peak charging rate Adjusting weights:

[0188] ;

[0189] in, The average monthly charging cost for user type k without a data plan; The average monthly charging cost when selecting package m for user type k; The larger the value, the smaller the contribution of cost satisfaction to total utility;

[0190] Convenience satisfaction Related to charging time tolerance and actual charging time, determined by time tolerance. adjust:

[0191] ;

[0192] in, The actual charging time corresponding to the m-th package when the k-th user selects the m-th package, if Then there is That is, exceeding the tolerance level, the convenience offers no benefit;

[0193] Environmental benefits This only applies to users who choose the green electricity package, and is related to their environmental awareness. Related to green electricity consumption:

[0194] ;

[0195] in, The percentage of green electricity charging when the m-th package is selected for user type k; if no green electricity package is selected, then... ;

[0196] 3) Using the Nash-Q reinforcement learning method, the package parameters are iteratively adjusted in response to the user's optimal package selection until a Nash equilibrium is reached, thus obtaining the optimal package parameters;

[0197] The Nash-Q algorithm no longer considers individual actions, but rather multiple agents. Each agent can refer to the actions and rewards of other agents, comprehensively considering the dynamics of the entire system to determine its own action strategy. (Define state) Where k represents the profile type, m represents the package type, t represents the time period, and the action represents the action. initial value Recorded as 0;

[0198] For user class k, calculate the utility of choosing each package m based on the utility function. ,choose The largest package is the best response. ,Right now:

[0199] ;

[0200] Based on the user's optimal response Package parameters , Make adjustments to the profit. Maximize and update the Q value:

[0201] ;

[0202] in, For learning rate , Discount factor and , The next state;

[0203] If the profit difference between the two iterations ( To set the minimum threshold as follows: And the user's optimal package selection If the parameters remain unchanged, a Nash equilibrium is reached, and the corresponding optimal package parameters are output. , ;

[0204] 4) Verify the optimal package parameters through load constraints and user acceptance to ensure that the package parameters meet the requirements of power grid regulation and control and the willingness of users to participate;

[0205] The optimal package output results are verified using load constraints. For the preset threshold, The formula for predicting peak-valley load differences is as follows:

[0206] ;

[0207] Next, user acceptance was used for validation, and the optimal package utility for each user profile was calculated. If you get If the effect is negative, users will be unwilling to participate. The solution is to adjust the package parameters, re-edit the user profile and behavior guidelines, and finally sort out the optimal package parameters, as well as the package selection ratio and load control effect of each group, to form the optimal package for the target user group.

[0208] This invention provides a method for designing differentiated charging packages for electric vehicles that considers user behavior profiles and charging load. It integrates user behavior profiling, a psychological-behavioral correlation model, and master-slave game optimization. By constructing a multi-dimensional user feature library and filtering key features, it deeply binds users' basic attributes, psychological preferences, and charging behavior, overcoming the limitations of traditional methods that only focus on surface behavior while ignoring the underlying decision-making logic. By establishing a psychological-behavioral correlation model, it transforms static profiles into dynamic behavior predictions, and combines SOC constraints and a flexible queuing model to improve the authenticity and accuracy of charging load predictions. Based on user profiles and psychological needs, it designs differentiated packages and utilizes master-slave game optimization to maximize user utility and grid regulation value, solving the implementation difficulties caused by packages being divorced from user needs.

[0209] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0210] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0211] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0212] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0213] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0214] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for designing differentiated charging packages for electric vehicles that considers user behavior profiles and charging load, characterized in that, include: Collect relevant data on electric vehicle users, build a user feature database, and use structural equation modeling to screen key features that significantly affect charging behavior; Users are clustered based on the key features obtained from the screening, and probability distribution models of key features of different user groups are constructed to form typical user profiles. Based on the user profile, user psychology-behavior correlation models are established for different categories of users, wherein the user psychology-behavior correlation models are used to quantify the correlation between user psychological preferences and behavior; By combining user profiles and user psychology-behavior correlation models, and introducing SOC dynamic constraints and elastic M / M / c / N queuing models, we can simulate real charging scenarios and achieve accurate prediction of charging load. Based on user profile characteristics and load forecast results, several typical packages are designed. A master profit function aimed at maximizing the profit of electric vehicle aggregators and a slave utility function aimed at maximizing user utility are constructed. The Nash-Q reinforcement learning method is used to solve the master-slave game model to obtain the optimal package parameters.

2. The method for designing differentiated charging packages for electric vehicles considering user behavior profiles and charging load as described in claim 1, characterized in that, The user feature database includes primary tags, secondary tags, and tag types. Primary tags include basic user attributes, user behavioral characteristics, user attitude preferences, and grid contribution and response. Basic user attributes include two secondary tags: personal attributes and vehicle attributes, both factual. User behavioral characteristics include three secondary tags: travel characteristics, charging behavior, and time attitude, respectively rule-based, factual, and rule-based. User attitude preferences include three secondary tags: price sensitivity, environmental willingness, and charging satisfaction, respectively prediction-predictive, and rule-based. Grid contribution and response include two secondary tags: user contribution level and package response behavior, both rule-based.

3. The method for designing differentiated charging packages for electric vehicles considering user behavior profiles and charging load as described in claim 1, characterized in that, The methods for selecting key features are as follows: Construct a structural equation model, which comprises two parts: a measurement model and a structural model. The measurement model is used to observe the relationship between variables and latent variables, and the structural model is used to describe the causal relationship between latent variables. The model parameters are solved by the maximum likelihood estimation method, and the goodness of fit of the model is tested by the ratio of chi-square to degrees of freedom, the comparison of fit index, and the root mean square of approximation error. If the goodness of fit of the model does not meet the standard, the model is adjusted until it meets the standard. A t-test was performed on the model path coefficients to retain variables with significant influence; the Pearson coefficient was introduced to remove highly redundant features, resulting in a key feature set.

4. The method for designing differentiated charging packages for electric vehicles considering user behavior profiles and charging load as described in claim 1, characterized in that, The steps for clustering users based on key features are as follows: The key features are transformed into p-dimensional feature vectors, and the number of clusters is preset to K. The core information of the user group is compressed into K initial CF units through the clustering features. Iterate through all user feature vectors, calculate the distance between each user and the centroids of the K CF unit clusters, and assign the user to the most similar CF unit to complete user classification; The silhouette coefficient is used to evaluate the clustering quality, ensuring that the weighted mean of the silhouette coefficient meets the preset requirements.

5. The method for designing differentiated charging packages for electric vehicles considering user behavior profiles and charging load as described in claim 1, characterized in that, The specific method for constructing probability distribution models of key features of different user types to form typical user profiles is as follows: Assuming that the key features of each group follow a Gaussian mixture model, the model parameters are iteratively updated until the distribution fit no longer improves. The distribution is then corrected by incorporating behavioral scenario logic to obtain a typical user profile.

6. The method for designing differentiated charging packages for electric vehicles considering user behavior profiles and charging load as described in claim 1, characterized in that, The user psychology-behavior correlation model is used to quantify the relationship between four types of psychological preferences—range anxiety, time tolerance, price sensitivity, and environmental willingness—and the starting state of charge (SOC), charging power selection, off-peak charging ratio, and green electricity charging ratio.

7. The method for designing differentiated charging packages for electric vehicles considering user behavior profiles and charging load as described in claim 1, characterized in that, The quantitative relationships in the user psychology-behavior correlation model include: 1) Range anxiety and initial SOC: The higher the anxiety index, the closer the initial SOC is to the maximum safe threshold. 2) Time tolerance and charging power selection: The lower the time tolerance, the higher the probability of choosing fast charging; 3) Price sensitivity and off-peak charging rate: The lower the price sensitivity, the more positive the response to low prices; 4) Environmental awareness and green electricity charging ratio: The stronger the environmental awareness, the higher the green electricity charging ratio.

8. The method for designing differentiated charging packages for electric vehicles considering user behavior profiles and charging load as described in claim 1, characterized in that, By combining user profiles and user psychology-behavior correlation models, and introducing SOC dynamic constraints and elastic M / M / c / N queuing models, we simulate real charging scenarios and achieve accurate prediction of charging load. The specific steps include: By combining user profile proportions with user psychology-behavior correlation models, we define initial parameters for scenarios and quantify the psychology-behavior characteristics of different groups. Based on battery energy conservation and travel demand, a dynamic constraint on SOC is constructed to correlate charging behavior and charging characteristics. Introducing an elastic M / M / c / N queuing model, considering user stopping behavior and impatience, we calculate the steady-state probability of a charging station containing k electric vehicles. Based on the queuing model, the number of charging piles occupied within a specific time period t is statistically analyzed using Monte Carlo simulation, and the charging load is predicted by combining the average charging power.

9. The method for designing differentiated charging packages for electric vehicles considering user behavior profiles and charging load as described in claim 1, characterized in that, The packages include four types, namely: Peak-hour fast charging + off-peak monthly subscription package: targeting high-frequency, long-distance users with high range anxiety and low tolerance for time, including fast charging subsidies; Off-peak tiered plans: designed for commuters with moderate price sensitivity and fixed travel times, offering lower off-peak electricity prices; Green electricity off-peak packages: targeting green users with high environmental awareness and low price sensitivity, with a high proportion of green electricity consumption; Off-peak pricing packages: Targeting price-sensitive users with low charging frequency and short distances, offering no fixed fees and low prices per charge.

10. The method for designing differentiated charging packages for electric vehicles considering user behavior profiles and charging load as described in claim 1, characterized in that, The solution process for the master-slave game model is as follows: Construct the main party's profit function, where the main party's profit function = basic package revenue + demand response revenue + green electricity revenue - operation and maintenance costs - electricity purchase costs; Construct a utility function for the user, where the utility function for the user = cost satisfaction + convenience satisfaction + environmental benefits, and the weighting coefficients are adjusted according to user characteristics; The Nash-Q reinforcement learning method is adopted to iteratively adjust the package parameters in response to the user's optimal package selection until a Nash equilibrium is reached, thus obtaining the optimal package parameters. The optimal package parameters are verified by load constraints and user acceptance to ensure that the package parameters meet the requirements of power grid regulation and control and the willingness of users to participate.

Images