Mechanism data collaborative driving vehicle-to-grid interaction closed-loop correction method and system and medium

Abstract

This invention relates to the field of vehicle-to-grid power supply and distribution technology, and particularly to a method, system, and medium for closed-loop correction of vehicle-to-grid interaction driven by mechanistic data collaboration. The method includes: acquiring data on electric vehicles, the power grid, and charging facilities within the area to be regulated; constructing an electric vehicle mechanistic scheduling model and generating initial scheduling results; constructing an implicit behavior mining model based on historical charging data and contextual data and generating prediction results; fusing the initial scheduling results and prediction results to generate a fused prediction result; acquiring the actual state of the electric vehicle and identifying its deviation from the predicted state; correcting the fused prediction result based on the deviation and updating the implicit behavior mining model online to generate a corrected state quantity; and performing rolling optimization based on the corrected state quantity and the mechanistic scheduling model to generate a corrected scheduling instruction. This invention effectively solves the problem in existing technologies of how to respond promptly and accurately to the significant differences in behavior among different types of electric vehicles and the dynamic changes in their actual operating states during vehicle-to-grid interaction.

Description

Technical Field

[0001] This invention relates to the field of vehicle-to-grid power supply and distribution technology, and in particular to a method, system and medium for closed-loop correction of vehicle-to-grid interaction driven by mechanism data collaboration. Background Technology

[0002] Against the backdrop of "dual carbon" goals and the construction of a new power system, electric vehicles are gradually transforming from simple energy end-users into flexible resources capable of participating in grid regulation. Vehicle-grid interaction refers to the collaboration between electric vehicles and the power grid through orderly charging and flexible adjustment to enhance system flexibility and the capacity for renewable energy absorption.

[0003] Existing technologies mainly fall into two categories: one is optimization scheduling methods based on operational mechanisms, which have good physical interpretability but often rely on idealized assumptions and struggle to accurately reflect the differences in control capabilities, spatiotemporal distribution, and behavioral characteristics among different types of electric vehicles; the other is behavior prediction methods based on historical data, which, while capable of uncovering certain patterns, lack mechanistic constraints and are less adaptable to changes in scenarios or operational disturbances. Furthermore, existing methods generally lack continuous tracking and closed-loop correction mechanisms for actual operating states, making it difficult to adjust scheduling decisions promptly when deviations occur between predictions and actual behavior. Therefore, there is an urgent need for a method that integrates mechanistic modeling and historical data mining, and dynamically corrects vehicle-to-grid interaction decisions based on real-time state deviations.

[0004] The information disclosed in this background section is intended only to enhance the understanding of the general background of this disclosure and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] This invention provides a method, system, and medium for closed-loop correction of vehicle-to-grid interaction driven by mechanism data collaboration, which can effectively solve the problems in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A mechanism-based data collaboratively driven closed-loop correction method for vehicle-to-grid interaction, the method comprising: The system acquires operational data of electric vehicles, power grids, and charging facilities within the area to be regulated, constructs a mechanistic scheduling model that characterizes the regulation capabilities and operational constraints of various types of electric vehicles, and generates initial scheduling results based on the mechanistic scheduling model. An implicit behavior mining model for electric vehicles is constructed based on historical charging data and contextual data, and behavior prediction results corresponding to the current time period are generated based on the implicit behavior mining model. By combining the initial scheduling result and the behavior prediction result, a vehicle-to-network interaction fusion prediction result is generated; The actual operating status of the electric vehicle in the current time period is obtained, and the deviation between the actual operating status and the predicted status is identified based on the fusion prediction results; Based on the aforementioned deviation, the fusion prediction result is state-corrected, and the implicit behavior mining model is updated online to generate corrected state variables. Rolling optimization is performed based on the corrected state variables and the mechanism scheduling model to generate the corrected scheduling instructions for the current time period.

[0007] Furthermore, a mechanistic scheduling model is constructed, including: The electric vehicles are classified into electric taxis, electric private cars, and electric buses; Adjustable power capacity models for each type of electric vehicle are constructed respectively; Temporal distribution models, spatial distribution models, regulation reliability models, and date correction models for each type of electric vehicle are constructed respectively. The adjustable power capacity model, the time distribution model, the spatial distribution model, the control reliability model, and the date correction model are used to characterize the differences in control capabilities among different types of electric vehicles.

[0008] Further, the initial scheduling results are generated, including: The mechanism scheduling model is constructed with the goal of minimizing the total system cost; The total system cost includes grid operation costs and electric vehicle control costs. In the aforementioned mechanism scheduling model, power balance constraints, power constraints, state of charge constraints, and travel demand energy satisfaction constraints are applied. Solve the aforementioned mechanism scheduling model to generate the initial scheduling result.

[0009] Furthermore, a model for mining the implicit behavior of electric vehicles is constructed, including: Acquire historical charging data, vehicle type data, spatial location data, and contextual data; A multidimensional historical database is constructed based on the historical charging data, the vehicle type data, the spatial location data, and the context data; Outlier and missing value processing is performed on the multidimensional historical database. The implicit behavior mining model is constructed based on the processed multidimensional historical database.

[0010] Furthermore, generate behavioral prediction results corresponding to the current time period, including: The spatiotemporal features of charging behavior in the multidimensional historical database are extracted based on the attention mechanism; The latent distribution of the charging behavior is learned based on a variational autoencoder; The graph neural network is used to model the relationship between charging behavior in different regions; The behavior prediction result is generated based on the spatiotemporal characteristics of the charging behavior, the potential distribution, and the association of the charging behavior.

[0011] Furthermore, the initial scheduling result and the behavior prediction result are fused to generate a vehicle-to-grid interaction fusion prediction result, including: Obtain the mechanism prediction result corresponding to the initial scheduling result; The mechanism prediction results and the behavior prediction results are weighted and fused. The fusion weights are dynamically adjusted based on the historical prediction errors of the respective mechanism prediction results and behavior prediction results. The vehicle-to-network interaction fusion prediction result is generated based on the adjusted fusion weights.

[0012] Furthermore, identifying the deviation between the actual operating state and the predicted state includes: The actual charging power, actual state of charge, actual number of connected vehicles, and actual spatial distribution of the electric vehicle at the current time period are obtained. Based on the vehicle-to-grid interaction fusion prediction results, the deviations between the actual charging power, the actual state of charge, and the actual number of connected vehicles and the predicted state are determined. A comprehensive evaluation index for the deviation is constructed based on the aforementioned deviation. When the deviation comprehensive evaluation index exceeds a preset threshold, a state correction is triggered on the fusion prediction result and the implicit behavior mining model is updated online.

[0013] Furthermore, generate the current time period correction scheduling instruction, including: Extended Kalman filtering is used to dynamically estimate the system state corresponding to the vehicle-to-network interaction fusion prediction results; The parameters of the implicit behavior mining model are updated based on the online learning strategy; Based on the updated system state and the aforementioned mechanism scheduling model, an optimization solution is performed in the rolling time domain; Output the corrected scheduling instruction for the current time period.

[0014] A mechanism-data collaboratively driven vehicle-to-grid interaction closed-loop correction system, the system comprising: The initial scheduling module acquires the operation data of electric vehicles, power grid and charging facilities in the area to be regulated, constructs a mechanism scheduling model that characterizes the regulation capabilities and operation constraints of multiple types of electric vehicles, and generates the initial scheduling results based on the mechanism scheduling model. The behavior prediction module constructs an implicit behavior mining model for electric vehicles based on historical charging data and contextual data, and generates behavior prediction results for the current time period based on the implicit behavior mining model. The fusion prediction module merges the initial scheduling results and behavior prediction results to generate vehicle-to-grid interaction fusion prediction results; The deviation identification module obtains the actual operating status of the electric vehicle in the current time period and identifies the deviation between the actual operating status and the predicted status based on the fusion prediction results. The state correction module corrects the state of the fusion prediction results based on the bias and updates the implicit behavior mining model online to generate corrected state variables. The correction scheduling module performs rolling optimization based on the correction state variables and the mechanism scheduling model to generate the correction scheduling instructions for the current time period.

[0015] A computer-readable storage medium stores a computer program, the computer program including program instructions, which, when executed by a processor, can implement the aforementioned mechanism-data collaboratively driven vehicle-to-grid interactive closed-loop correction method.

[0016] The technical solution of this invention can achieve the following technical effects: A mechanism scheduling model for electric vehicles is constructed and an initial scheduling result is generated. An implicit behavior mining model is constructed by combining historical data and a prediction result is generated. The two are then integrated to identify actual operational deviations, and state correction and online model updates are performed. Corrected scheduling instructions are generated through rolling optimization. This effectively solves the problem that existing technologies cannot respond promptly and accurately to the large differences in behavior of different types of electric vehicles and the dynamic changes in actual operating status in vehicle-to-grid interaction.

[0017] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the closed-loop correction method for vehicle-to-grid interaction driven by mechanism data collaboration; Figure 2A comparative diagram of the power external characteristics of different types of electric vehicles (weekdays); Figure 3 This is a diagram illustrating the comparison of the effects of implicit behavior mining based on historical data. Figure 4 A diagram illustrating the comparison of scheduling effects of real-time correction strategies in emergency scenarios. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0022] Example 1; like Figure 1 As shown, this application provides a mechanism data-driven closed-loop correction method for vehicle-to-grid interaction, the method including: S10: Obtain the operation data of electric vehicles, power grid and charging facilities in the area to be regulated, construct a mechanism scheduling model that characterizes the regulation capability and operation constraints of multiple types of electric vehicles, and generate the initial scheduling results based on the mechanism scheduling model; S20: Construct an implicit behavior mining model for electric vehicles based on historical charging data and contextual data, and generate behavior prediction results for the current time period based on the implicit behavior mining model; S30: Integrate the initial scheduling results and behavior prediction results to generate vehicle-to-network interaction fusion prediction results; S40: Obtain the actual operating status of the electric vehicle in the current time period, and identify the deviation between the actual operating status and the predicted status based on the fusion prediction results; S50: Based on the bias, the fusion prediction results are corrected in state, and the implicit behavior mining model is updated online to generate corrected state variables; S60: Based on the corrected state variables and the mechanism scheduling model, perform rolling optimization to generate the corrected scheduling instructions for the current time period.

[0023] Specifically, the first step is to acquire operational data on electric vehicles, the power grid, and charging facilities within the area to be regulated. Electric vehicles include electric taxis, electric private cars, and electric buses. Operational data includes vehicle type, access status, charging power, state of charge, spatial location, as well as information on power grid load, charging station distribution, and user travel constraints. Based on the operational data, adjustable power capacity models, time distribution models, spatial distribution models, regulation reliability models, and date correction models are established for different types of electric vehicles. On this basis, a mechanistic scheduling model characterizing the regulation capabilities and operational constraints of multiple types of electric vehicles is constructed. By solving the optimization objective function that includes power grid operating costs and electric vehicle regulation costs, and comprehensively considering power balance constraints, power constraints, state of charge constraints, and travel demand energy satisfaction constraints, initial scheduling results are generated. Furthermore, an implicit behavior mining model for electric vehicles is constructed based on historical charging data and contextual data. The historical charging data includes at least timestamps, spatial locations, vehicle types, observed charging power, and observed state of charge. The contextual data includes information such as weather, temperature, and date type. After processing outliers and missing values ​​in the historical data, the spatiotemporal features of charging behavior are extracted using an attention mechanism. The potential distribution of charging behavior is learned using a variational autoencoder, and the correlation of charging behavior between different regions is modeled using a graph neural network. Based on this, the behavior prediction results for the current time period are generated. Furthermore, the mechanism prediction results and behavior prediction results corresponding to the initial scheduling results are fused to obtain the vehicle-to-grid interaction fusion prediction results. During the fusion process, the mechanism model results and implicit behavior model results are weighted and combined, and the fusion weights are dynamically adjusted according to the historical prediction errors of the two types of models, so that the fusion results can simultaneously take into account the interpretability of mechanism constraints and the adaptability of historical behavior patterns. Furthermore, the actual operating status of the electric vehicle in the current time period is obtained, including the actual charging power, actual state of charge, actual number of connected vehicles, and actual spatial distribution. Based on the fusion prediction results, the power deviation, state of charge deviation, and number of vehicles deviation between the actual operating status and the predicted status are calculated respectively, and a comprehensive deviation evaluation index is constructed. When the comprehensive deviation evaluation index exceeds a preset threshold, it is determined that the current operating status has deviated from the prediction result, triggering the subsequent state correction and model update process. Furthermore, state correction is performed on the fusion prediction results based on the deviation, and the implicit behavior mining model is updated online to generate corrected state variables. Among them, the state correction uses extended Kalman filtering to dynamically estimate the system state in order to obtain an estimation result that is more consistent with the current real operating state. The online model update adopts an online learning strategy to update the parameters of the implicit behavior mining model according to real-time deviation information, so as to improve the model's adaptability to new scenarios and sudden disturbances. Finally, rolling optimization is performed based on the corrected state variables and the mechanism scheduling model to generate the corrected scheduling instructions for the current time period. Specifically, the scheduling strategy for the subsequent time domain is re-optimized at the current moment, and only the optimization results corresponding to the current time period are executed. At the next moment, the deviation identification, state correction, model update and rolling optimization are repeated in combination with the latest obtained actual operating status, thereby forming a closed-loop feedback decision-making process oriented to the real-time operating status and realizing the dynamic correction of the vehicle-network interaction scheduling results.

[0024] The technical solution of this invention constructs an electric vehicle mechanism scheduling model and generates initial scheduling results. It combines historical data to construct an implicit behavior mining model and generate prediction results. After fusing the two, it identifies actual operating deviations, performs state correction and online model updates, and generates corrected scheduling instructions through rolling optimization. This effectively solves the problem of how existing technologies can respond promptly and accurately to the large differences in behavior of different types of electric vehicles and the dynamic changes in actual operating status in vehicle-to-grid interaction.

[0025] Furthermore, constructing a mechanism-based scheduling model includes: Electric vehicles are classified into electric taxis, electric private cars, and electric buses; Develop adjustable power capacity models for each type of electric vehicle; Temporal distribution models, spatial distribution models, regulation reliability models, and date correction models for each type of electric vehicle are constructed respectively; The differences in controllability among different types of electric vehicles are characterized by adjustable power capacity model, time distribution model, spatial distribution model, control reliability model and date correction model.

[0026] As a preferred embodiment of the above, the various types of electric vehicles include different models such as electric taxis, electric private cars, and electric buses, and these models have significant differences in power characteristics. 1.1 Modeling of the external power characteristics of electric taxis: Furthermore, electric taxis are characterized by long operating hours, large mileage, and high charging frequency. Their adjustable power capacity model is as follows: ; in, For the first Adjustable power capacity (kW) for time-of-use electric taxis; The number of electric taxis; For charging efficiency; Rated charging power (kW) for a single vehicle; For the first Availability of the time period; Furthermore, the time distribution characteristics of electric taxis take into account shift change times and operating patterns: ; Furthermore, the spatial distribution characteristics of electric taxis are related to the layout of charging stations, and a spatial distribution density function is established: ; in, For spatial location Taxi density at the location; The number of charging stations; For the first The attractiveness coefficient of a charging station; For position to charging station The distance; For spatial diffusion parameters; Represents an exponential function; Furthermore, the reliability model for electric taxis' participation in regulation considers response rate and persistence: ; in, To ensure the reliability of electric taxi control; The response reliability coefficient (values ​​range from 0.8 to 0.9). The continuous reliability coefficient (value ranges from 0.7 to 0.85); Furthermore, considering the differences between holidays and weekdays, the date correction factor for electric taxis is: ; On weekdays, taxi demand is high, and the available capacity is relatively small; on holidays, demand is dispersed, and the available capacity increases. 1.2 Modeling of the external power characteristics of electric private vehicles: Furthermore, electric private vehicles are characterized by dispersed usage time, highly random charging behavior, and relatively low willingness to regulate; therefore, their adjustable power capacity model is as follows: ; in, For the first Adjustable power capacity (kW) of electric private vehicles during specific time periods; The number of electric private cars; For charging efficiency; Rated charging power (kW) for a single vehicle; For the first Availability of the time period; To regulate the willingness coefficient (value range 0.3-0.5); Furthermore, the time distribution characteristics of electric private cars exhibit a distinct "bimodal" pattern: ; Furthermore, the spatial distribution of electric private vehicles is related to residential and office areas, establishing a dual-center distribution model: ; in, For spatiotemporal location The density of private cars; and It comprises residential and office areas respectively; It follows a normal distribution; Centered on; It is the covariance matrix; and They are respectively for nighttime and working hours; Furthermore, the reliability of electric private vehicles participating in regulation is relatively low, as modeled as follows: ; in, To improve the reliability of electric private vehicles; The response reliability coefficient (values ​​range from 0.5 to 0.7). The continuous reliability coefficient (values ​​range from 0.4 to 0.6). Furthermore, there are significant differences between holidays and weekdays for electric private cars: ; Holiday travel patterns for private cars are disrupted, leading to increased time spent at home and a significant increase in available capacity. 1.3 Modeling of the external power characteristics of electric buses: Furthermore, electric buses are characterized by fixed operating routes, predetermined timetables, and centralized charging management. Their adjustable power capacity model is as follows: ; in, For the first Adjustable power capacity (kW) of electric buses during designated time periods; The number of electric buses; For charging efficiency; Rated charging power (kW) for a single vehicle; For the first Availability of the time period; Furthermore, the time distribution characteristics of electric buses are highly regular, exhibiting a clear charging pattern during operational breaks: ; Furthermore, the spatial distribution of electric buses is highly concentrated in bus hubs and parking lots, establishing a discrete location model: ; in, For spatial location Bus density at the location; The number of bus stations; For the first The capacity of each station; It is the Dirac function; For the first Location of each station; Furthermore, electric buses have the highest reliability in participating in regulation: ; in, To ensure the reliability of electric bus control; The response reliability coefficient (values ​​range from 0.9 to 0.95); The continuous reliability coefficient (value ranges from 0.85 to 0.95); Furthermore, the differences between holidays and weekdays for electric buses: ; During holidays, the number of bus services decreases, and the available capacity increases accordingly.

[0027] Furthermore, generating the initial scheduling results includes: Construct a mechanism scheduling model with the goal of minimizing the total system cost; The total system cost includes grid operation costs and electric vehicle control costs. In the mechanistic scheduling model, power balance constraints, power constraints, state of charge constraints, and energy satisfaction constraints for travel demand are imposed. Solve the mechanism scheduling model to generate initial scheduling results.

[0028] As a preferred embodiment of the above, a comprehensive scheduling model for multiple types of electric vehicles is established, with the optimization objective being to minimize the total system cost: ; in, This represents the total number of time periods in the scheduling cycle. For the first Time-limited power grid operating costs; For the first Electric vehicles in the first The cost of regulation during a given period; These respectively represent electric taxis, electric private cars, and electric buses; Furthermore, grid operating costs include generation costs and load deviation costs: ; in, For the first Time-of-use generation cost coefficient; Power generation capacity; This is the load deviation penalty coefficient; This represents the actual load. For reference load; Furthermore, the control costs of electric vehicles are related to reliability and capacity: ; in, For the first The unit control cost of electric vehicles; For the first Electric vehicles in the first Scheduling power during specific time periods; For the reliability of this type of electric vehicle; Furthermore, the power balance constraint is: ; Furthermore, the power constraint for electric vehicles is: ; in, Let be the adjustable power capacity of the i-th type of electric vehicle in time period t; Furthermore, the battery state of charge (SOC) constraint is as follows: ; ; in, For the first Electric vehicles in the first State of charge over a period of time; For the first Average battery capacity (kWh) for electric vehicle type. For time intervals; and These are the lower and upper limits of the state of charge, respectively; Furthermore, consider the energy constraints for meeting user travel demands: in, For the first The expected state of charge of the electric vehicle at the end of the scheduling cycle, where T is the total number of time periods in the scheduling cycle.

[0029] Furthermore, constructing an implicit behavior mining model for electric vehicles includes: Acquire historical charging data, vehicle type data, spatial location data, and contextual data; A multidimensional historical database is constructed based on historical charging data, vehicle type data, spatial location data, and contextual data; Perform outlier and missing value handling on the multidimensional historical database; An implicit behavior mining model is constructed based on the processed multidimensional historical database.

[0030] As a preferred embodiment of the above, an implicit behavior mining model for electric vehicles based on a historical database is established: Furthermore, the actual charging behavior of electric vehicles does not completely follow the mechanism model in step 1. There are individual differences and implicit behavioral patterns, which are contained in historical charging data. 2.1 Historical Database Construction: Furthermore, establish a multi-dimensional historical charging database. ,in For each region, the dataset contains: ; in, For timestamps; Spatial location; Vehicle type; To observe the charging power; To observe the state of charge; This is a contextual feature vector (including weather, temperature, date type, etc.); For the region The number of samples; Furthermore, the historical data is preprocessed to remove outliers and missing values: ; in, The cleaned dataset; and This is within a reasonable power range.

[0031] Furthermore, generating behavioral prediction results corresponding to the current time period includes: Extract spatiotemporal features of charging behavior from a multidimensional historical database based on attention mechanisms; Learning the latent distribution of charging behavior based on variational autoencoders; The graph neural network is used to model the relationship between charging behavior in different regions; Behavior prediction results are generated based on the spatiotemporal characteristics, potential distribution, and correlation of charging behavior.

[0032] As a preferred embodiment of the above embodiments, in a preferred embodiment: 2.2 Spatiotemporal Feature Extraction Based on Attention Mechanism: Furthermore, an attention mechanism is employed to extract the spatiotemporal features of the charging behavior, with the temporal attention weights calculated as follows: ; in, For the first Temporal attention weights for each sample; Let T be the temporal attention parameter matrix, where the superscript T is the transpose of the matrix; For the sample Time feature vector; For the sample Time feature vector; Represents an exponential function; For the region The number of samples; Furthermore, the spatial attention weights are calculated as follows: ; in, For the first Spatial attention weights for each sample; Let T be the spatial attention parameter matrix, where the superscript T is the transpose of the matrix; For the sample Spatial feature vectors; For the sample Spatial feature vectors; Furthermore, by fusing spatiotemporal attention, a comprehensive feature representation is obtained: ; in, For the sample The comprehensive feature representation; This is the context feature weight matrix, where the superscript T is the transpose of the matrix; 2.3 Implicit Behavior Coding Based on Variational Autoencoder: Furthermore, a variational autoencoder (VAE) is employed to learn the latent distribution of implicit charging behavior. The encoder maps the observed data to the latent space: ; in, As latent variables; For input observations, where, To observe the charging power, To observe the state of charge; and These are the mean and standard deviation of the encoder output, respectively. Represents the identity matrix; Represents a normal distribution; Furthermore, the encoder is parameterized as a neural network: ; ; in, This is the weight matrix; It is the bias vector; Furthermore, the decoder reconstructs the observed data from the latent variables: ; ; in, For reconstructing the output; This is the weight matrix; It is the bias vector; To reconstruct the variance; Represents a normal distribution; Furthermore, the loss function of VAE includes reconstruction loss and KL divergence regularization term: ; in, The Kullback-Leibler divergence; It is the prior distribution; The regularization coefficient is used. In distribution Find the expected value of the terms within the parentheses. 2.4 Region Association Modeling Based on Graph Neural Networks: Furthermore, electric vehicle charging behavior in different regions exhibits spatial correlation. A graph neural network (GNN) is used to model these regional correlations and construct a regional correlation graph. ,in For a set of regional nodes, It is an edge set; Furthermore, the edge weights between regions are calculated based on geographical distance and the similarity of charging behavior: ; in, For the region and Edge weights between them; Represents an exponential function; Geographical distance; These are distance normalization parameters; A similarity function for charging behavior; To represent the regions respectively and Charging behavior data representation or charging behavior feature representation; Furthermore, the graph convolutional layer updates the region features: ; in, For the region In the Layer feature representation; For the region The set of neighbors; Indicates the area With the region Edge weights between them; They are nodes and The degree; For the first Layer weight matrix; For activation functions; To represent neighboring areas In the Layer feature representation; Furthermore, after multi-layer graph convolution, an implicit behavioral representation of the fused region association information is obtained: ,in For the region In the The feature representation of a layer is the initial feature representation.

[0033] Furthermore, the initial scheduling results and behavior prediction results are fused to generate vehicle-to-grid interaction fusion prediction results, including: Obtain the mechanism prediction results corresponding to the initial scheduling results; The mechanistic prediction results and behavioral prediction results are weighted and fused. The fusion weights are dynamically adjusted based on the historical prediction errors of the mechanistic prediction results and the behavioral prediction results, respectively. The vehicle-to-network interaction fusion prediction results are generated based on the adjusted fusion weights.

[0034] As a preferred embodiment of the above embodiments, in a preferred embodiment: 2.5 Implicit Behavior Prediction Model: Furthermore, based on the learned latent representation, an implicit charging power prediction model is established: ; in, For the first Electric vehicles in the region No. Implicit charging power prediction for different time periods; For parameters Neural network prediction function; For the region Spatial characteristics; An implicit behavioral representation for integrating regional association information; This is a contextual feature vector (including weather, temperature, date type, etc.); Furthermore, the implicit state of charge prediction model is as follows: ; in, For implicit state of charge prediction; This is an individual difference correction term that follows the learned latent distribution; For the first Average battery capacity (kWh) for electric vehicle type. For time intervals; Furthermore, the individual variability correction term is obtained by sampling from the latent space: ; in, and Output by VAE decoder; Represents a normal distribution; Furthermore, the fusion strategy of the mechanistic model and the implicit behavioral model adopts a weighted combination: ; in, This refers to the combined charging power. For the first Electric vehicles in the first Scheduling power during specific time periods; and The weights for the mechanistic model and the implicit model are respectively, satisfying... ; Furthermore, the fusion weights are dynamically adjusted based on the model prediction error: ; ; in, and These are the historical prediction errors of the mechanistic model and the implicit model, respectively. To adjust the parameters.

[0035] Furthermore, identifying the deviation between the actual operating state and the predicted state includes: Obtain the actual charging power, actual state of charge, actual number of connected vehicles, and actual spatial distribution of electric vehicles in the current time period; Based on the prediction results of vehicle-to-grid interaction and integration, the deviations between the actual charging power, actual state of charge, and actual number of connected vehicles and the predicted state are determined. Construct a comprehensive evaluation index for deviation based on the deviation; When the deviation comprehensive evaluation index exceeds the preset threshold, the state correction of the fusion prediction results is triggered and the implicit behavior mining model is updated online.

[0036] As a preferred embodiment of the above, a charging behavior correction strategy based on the real-time operating status of the electric vehicle is designed. Furthermore, the actual charging behavior of electric vehicles can be affected by factors such as unforeseen events, weather changes, and users' temporary decisions, resulting in deviations from the aforementioned predictions. Therefore, a real-time correction mechanism needs to be established. 3.1 Real-time status monitoring and deviation identification: Furthermore, real-time operating data of electric vehicles is collected through charging piles and on-board terminals to establish a real-time state vector: ; in, For the first Real-time state vector for a given time period; This refers to the actual charging power. This represents the actual state of charge. This represents the actual number of vehicles connected. This represents the actual spatial distribution. Furthermore, the deviation between the real-time state and the fused prediction is calculated: ; ; ; in, These are deviations in power, state of charge, and number of vehicles, respectively. This refers to the combined charging power. The state of charge after fusion; The number of vehicles predicted by fusion; Furthermore, establish a comprehensive evaluation index for deviations: ; in, This is a comprehensive deviation index; and These are the rated power and the rated number of vehicles, respectively, used for normalization; Furthermore, when the overall deviation exceeds a threshold, a correction mechanism is triggered: ; in, This is the preset deviation threshold.

[0037] Furthermore, generating the current time-period correction scheduling instruction includes: Extended Kalman filtering is used to dynamically estimate the system state corresponding to the vehicle-to-network interaction fusion prediction results; The parameters of the implicit behavior mining model are updated based on the online learning strategy; The optimization solution is performed in the rolling time domain based on the updated system state and mechanism scheduling model; Output the corrected scheduling instructions for the current time period.

[0038] As a preferred embodiment of the above embodiments, in a preferred embodiment: 3.2 State estimation correction based on Kalman filter: Furthermore, an extended Kalman filter (EKF) is used to dynamically estimate and correct the charging behavior state. The state equation is as follows: ; in, Let be the state vector, where The first Time zone Inner The actual charging power, actual state of charge, and actual number of vehicles connected to the electric vehicle-like system are represented by the superscript T, which indicates the transpose of the matrix. This is the state transition matrix; To control the input matrix; For control input; For process noise, obey ,in For the first Time zone Inner The actual spatial distribution of electric vehicles; Furthermore, the observation equation is: ; in, For observation vectors; The observation matrix; To observe noise, obey ,in For the first Time zone Inner Rated power of electric vehicles; Further, the EKF prediction steps: ; ; in, For prior state estimation; Let T be the prior error covariance matrix; where the superscript T is the transpose of the matrix. Furthermore, the EKF update steps are as follows: ; ; ; in, Kalman gain; For posterior state estimation; The posterior error covariance matrix; is the identity matrix; where the superscript T is the transpose symbol of the matrix; 3.3 Adaptive Model Parameter Update: Furthermore, based on real-time deviation information, the parameters of the implicit behavior model are adaptively updated; an online learning strategy is adopted, and the loss function is: ; in, The length of the sliding window; Forgetting factor ( This gives higher weight to recent data; for Actual charging power during the time period; For parameters Neural network prediction function; For the region Spatial characteristics; An implicit behavioral representation for integrating regional association information; This is a contextual feature vector (including weather, temperature, date type, etc.); Furthermore, the parameter updates employ stochastic gradient descent: ; in, For the first Model parameters at the next update; The learning rate; ; is the online loss function In parameters Regarding parameters The gradient is used to characterize the descent direction of the loss function under the current parameters; Furthermore, to prevent overfitting, regularization constraints are introduced: ; in, The regularization coefficient is used. 3.4 Rolling optimization scheduling decision update: Furthermore, based on the corrected state estimate, the scheduling decision is updated using Receding Horizon Optimization (RHO) at the 1st... Time, Optimize the Future Scheduling strategy for each time period: ; ; in, To optimize the time domain length; for The power grid operating cost over a given period; for Posterior state estimation for a given time period; For the first Time zone Inner The unit control cost of electric vehicles; For the first Electric vehicles in the first Time zone Internal scheduling power; Furthermore, only the first step is executed. The optimization results for the current time period will be re-optimized in the next time period. ; in, To perform power scheduling; To optimize the optimal solution to the problem; Furthermore, the decision correction quantity is defined as: ; in, This refers to the combined charging power. Furthermore, to ensure the smoothness of scheduling, a correction constraint is introduced: ; ; in, This is the maximum correction amount; To correct the slope constraint; 3.5 Closed-loop feedback mechanism: Furthermore, establish a complete closed-loop feedback process: 1. Real-time monitoring: Collect actual state quantities ; 2. Deviation Calculation: Calculate the comprehensive deviation index. ; 3. State estimation: Run EKF to obtain the posterior state estimate. ; 4. Model Update: Online parameter updates ; 5. Decision Optimization: Solving the rolling optimization problem yields the execution scheduling power. ; 6. Execute the correction: Issue the corrected scheduling instructions; 7. Effect Evaluation: Evaluate the effect of the correction and adjust the parameters accordingly; 8. Iterative loop: Repeat the above process in the next time period; Furthermore, the evaluation indicators for the revised effect are: ; in, To correct the performance indicators, This indicates that the correction is effective; for The execution scheduling power for a given time period; for Charging power during a given period; for Actual charging power during the time period; Furthermore, key parameters are adaptively adjusted based on the correction effect: ; ; in, for Preset deviation threshold for the time period; for Learning rate over a given period; and This is the adaptive adjustment coefficient.

[0039] Example 2; Based on the same inventive concept as the mechanism data collaboratively driven vehicle-to-network interaction closed-loop correction method in the foregoing embodiments, the present invention also provides a mechanism data collaboratively driven vehicle-to-network interaction closed-loop correction system, the system comprising: The initial scheduling module acquires the operation data of electric vehicles, power grid and charging facilities in the area to be regulated, constructs a mechanism scheduling model that characterizes the regulation capabilities and operation constraints of multiple types of electric vehicles, and generates the initial scheduling results based on the mechanism scheduling model. The behavior prediction module constructs an implicit behavior mining model for electric vehicles based on historical charging data and contextual data, and generates behavior prediction results for the current time period based on the implicit behavior mining model. The fusion prediction module merges the initial scheduling results and behavior prediction results to generate vehicle-to-grid interaction fusion prediction results; The deviation identification module obtains the actual operating status of the electric vehicle in the current time period and identifies the deviation between the actual operating status and the predicted status based on the fusion prediction results. The state correction module corrects the state of the fusion prediction results based on the bias and updates the implicit behavior mining model online to generate corrected state variables. The correction scheduling module performs rolling optimization based on the correction state variables and the mechanism scheduling model to generate the correction scheduling instructions for the current time period.

[0040] The system described above in this invention can effectively realize a mechanism-based data collaborative driving method for closed-loop correction of vehicle-to-network interaction, and the technical effects it can achieve are as described in the above embodiments, and will not be repeated here.

[0041] Example 3; Based on the same inventive concept as the mechanism data collaboratively driven vehicle-to-network interaction closed-loop correction method in the foregoing embodiments, the present invention also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, can realize the mechanism data collaboratively driven vehicle-to-network interaction closed-loop correction method.

[0042] Example 4; To verify the effectiveness of this solution, a case study was conducted using the actual operation scenario of a municipal power grid. This city has approximately 6,500 electric taxis, 85,000 electric private cars, and 1,800 electric buses. Data collection spanned from January 1, 2024 to December 31, 2024, with a time resolution of 15 minutes. The city was divided into 12 regions, each with its own independent historical database. The experimental environment consisted of an Intel Xeon Gold 6248R processor, 256GB of RAM, and an NVIDIA A100 GPU, implemented using the PyTorch deep learning framework and the CVXPY optimization library. Figure 2This paper compares the power external characteristics of different types of electric vehicles on a typical weekday. In terms of time distribution, electric buses exhibit a clear dual-peak charging pattern, primarily charging during the 9:30-16:30 and 21:00-5:30 periods, with availability exceeding 0.90 during these charging times, demonstrating strong temporal regularity. Electric taxis show two charging peaks, from 7:00-9:00 and 23:00-7:00 respectively, reflecting shift-based charging and nighttime charging characteristics. Electric private cars show more dispersed charging behavior, mainly concentrated in the evening and nighttime periods from 19:00-7:00, but with greater temporal randomness. Regarding adjustable capacity, although electric private cars are the most numerous, their actual adjustable capacity is only 35%-45% of the theoretical capacity due to a low adjustment willingness coefficient (0.3-0.5). Electric buses and taxis, due to their high reliability, achieve actual adjustable capacities of 85%-92% and 70%-80% of the theoretical capacity, respectively. This figure verifies the accuracy of the established power external characteristic model for multiple types of electric vehicles. Figure 3 This study demonstrates the effectiveness of implicit behavior mining based on historical data. Using electric private car charging data from a certain region as a case study, the performance of three methods—pure mechanistic model prediction, implicit behavior model prediction, and fusion model prediction—was compared. The results show that while the pure mechanistic model can capture the overall trend, it is insufficient in characterizing individual differences and detailed fluctuations, with a mean absolute percentage error (MAPE) of 23.7%. The implicit behavior model, through variational autoencoders, learns the latent behavior distribution and can better fit the actual charging curve, reducing the MAPE to 16.4%, but its generalization ability is slightly insufficient in extreme cases. The fusion model combines the advantages of both mechanistic and data analysis, significantly improving the accuracy of detailed predictions while maintaining the accuracy of the overall trend, further reducing the MAPE to 12.8%, a 46.0% improvement compared to the pure mechanistic model. In addition, the figure also shows the spatiotemporal distribution of attention weights, revealing that high-weight periods are mainly concentrated in the two active user decision-making periods of 19:00-23:00 and 6:00-8:00, verifying the effectiveness of the attention mechanism. Figure 4The performance of vehicle-to-grid (V2G) interaction scheduling with and without a real-time correction strategy was compared. A scenario involving a sudden drop in temperature was selected, leading to decreased electric vehicle battery efficiency, increased user charging demand, and a significant deviation of actual charging behavior from prediction. The static scheduling method without a correction strategy showed a persistent and large deviation between the scheduled power and actual demand after the sudden event (starting at 12:00), with a maximum deviation reaching 38.6%, causing significant fluctuations in grid load. However, with the real-time correction strategy of this invention, the system can identify the deviation within 15 minutes (the comprehensive deviation index exceeds the threshold of 0.15), immediately initiate extended Kalman filtering for state estimation, update implicit behavior model parameters online, and adjust the scheduling plan for the next 4 hours through rolling optimization. The corrected scheduling strategy quickly tracks changes in actual demand, reducing the tracking deviation to 8.3% within 30 minutes and stabilizing below 5% after 1 hour, significantly outperforming the uncorrected method. The closed-loop correction mechanism enables the V2G interaction system to possess "adaptive, self-learning, and self-correcting" capabilities, greatly improving its robustness in dealing with uncertainty.

[0043] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of the application as defined herein, and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Thus, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A mechanism-based data collaborative driving method for closed-loop correction of vehicle-to-grid interaction, characterized in that, The method includes: The system acquires operational data of electric vehicles, power grids, and charging facilities within the area to be regulated, constructs a mechanistic scheduling model that characterizes the regulation capabilities and operational constraints of various types of electric vehicles, and generates initial scheduling results based on the mechanistic scheduling model. An implicit behavior mining model for electric vehicles is constructed based on historical charging data and contextual data, and behavior prediction results corresponding to the current time period are generated based on the implicit behavior mining model. By combining the initial scheduling result and the behavior prediction result, a vehicle-to-network interaction fusion prediction result is generated; The actual operating status of the electric vehicle in the current time period is obtained, and the deviation between the actual operating status and the predicted status is identified based on the fusion prediction results; Based on the aforementioned deviation, the fusion prediction result is state-corrected, and the implicit behavior mining model is updated online to generate corrected state variables. Rolling optimization is performed based on the corrected state variables and the mechanism scheduling model to generate the corrected scheduling instructions for the current time period.

2. The mechanism-based data collaboratively driven vehicle-to-network interaction closed-loop correction method according to claim 1, characterized in that, Constructing a mechanism-based scheduling model, including: The electric vehicles are classified into electric taxis, electric private cars, and electric buses; Adjustable power capacity models for each type of electric vehicle are constructed respectively; Temporal distribution models, spatial distribution models, regulation reliability models, and date correction models for each type of electric vehicle are constructed respectively. The adjustable power capacity model, the time distribution model, the spatial distribution model, the control reliability model, and the date correction model are used to characterize the differences in control capabilities among different types of electric vehicles.

3. The mechanism data collaboratively driven vehicle-to-network interaction closed-loop correction method according to claim 2, characterized in that, Generate initial scheduling results, including: The mechanism scheduling model is constructed with the goal of minimizing the total system cost; The total system cost includes grid operation costs and electric vehicle control costs. In the aforementioned mechanism scheduling model, power balance constraints, power constraints, state of charge constraints, and travel demand energy satisfaction constraints are applied. Solve the aforementioned mechanism scheduling model to generate the initial scheduling result.

4. The mechanism data collaboratively driven vehicle-to-network interaction closed-loop correction method according to claim 1, characterized in that, Constructing an implicit behavior mining model for electric vehicles, including: Acquire historical charging data, vehicle type data, spatial location data, and contextual data; A multidimensional historical database is constructed based on the historical charging data, the vehicle type data, the spatial location data, and the context data; Outlier and missing value processing is performed on the multidimensional historical database. The implicit behavior mining model is constructed based on the processed multidimensional historical database.

5. The mechanism data collaboratively driven vehicle-to-network interaction closed-loop correction method according to claim 4, characterized in that, Generate behavioral prediction results for the current time period, including: The spatiotemporal features of charging behavior in the multidimensional historical database are extracted based on the attention mechanism; The latent distribution of the charging behavior is learned based on a variational autoencoder; The graph neural network is used to model the relationship between charging behavior in different regions; The behavior prediction result is generated based on the spatiotemporal characteristics of the charging behavior, the potential distribution, and the association of the charging behavior.

6. The mechanism data collaboratively driven vehicle-to-network interaction closed-loop correction method according to claim 5, characterized in that, The initial scheduling result and the behavior prediction result are combined to generate a vehicle-to-grid interaction fusion prediction result, including: Obtain the mechanism prediction result corresponding to the initial scheduling result; The mechanism prediction results and the behavior prediction results are weighted and fused. The fusion weights are dynamically adjusted based on the historical prediction errors of the respective mechanism prediction results and behavior prediction results. The vehicle-to-network interaction fusion prediction result is generated based on the adjusted fusion weights.

7. The mechanism data collaboratively driven vehicle-to-network interaction closed-loop correction method according to claim 1, characterized in that, Identifying the deviation between the actual operating state and the predicted state includes: The actual charging power, actual state of charge, actual number of connected vehicles, and actual spatial distribution of the electric vehicle at the current time period are obtained. Based on the vehicle-to-grid interaction fusion prediction results, the deviations between the actual charging power, the actual state of charge, and the actual number of connected vehicles and the predicted state are determined. A comprehensive evaluation index for the deviation is constructed based on the aforementioned deviation. When the deviation comprehensive evaluation index exceeds a preset threshold, a state correction is triggered on the fusion prediction result and the implicit behavior mining model is updated online.

8. The mechanism data collaboratively driven vehicle-to-network interaction closed-loop correction method according to claim 7, characterized in that, Generate the current time period correction scheduling instruction, including: Extended Kalman filtering is used to dynamically estimate the system state corresponding to the vehicle-to-network interaction fusion prediction results; The parameters of the implicit behavior mining model are updated based on the online learning strategy; Based on the updated system state and the aforementioned mechanism scheduling model, an optimization solution is performed in the rolling time domain; Output the corrected scheduling instruction for the current time period.

9. A vehicle-to-grid interactive closed-loop correction system driven by mechanism data collaboration, characterized in that, The system includes: The initial scheduling module acquires the operation data of electric vehicles, power grid and charging facilities in the area to be regulated, constructs a mechanism scheduling model that characterizes the regulation capabilities and operation constraints of multiple types of electric vehicles, and generates the initial scheduling results based on the mechanism scheduling model. The behavior prediction module constructs an implicit behavior mining model for electric vehicles based on historical charging data and contextual data, and generates behavior prediction results for the current time period based on the implicit behavior mining model. The fusion prediction module merges the initial scheduling results and behavior prediction results to generate vehicle-to-grid interaction fusion prediction results; The deviation identification module obtains the actual operating status of the electric vehicle in the current time period and identifies the deviation between the actual operating status and the predicted status based on the fusion prediction results. The state correction module corrects the state of the fusion prediction results based on the bias and updates the implicit behavior mining model online to generate corrected state variables. The correction scheduling module performs rolling optimization based on the correction state variables and the mechanism scheduling model to generate the correction scheduling instructions for the current time period.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which includes program instructions that, when executed by a processor, can implement the mechanism-based data collaborative driving method for closed-loop correction of vehicle-to-grid interaction as described in any one of claims 1-8.

Images