Power distribution network-oriented vehicle-to-grid interaction aggregate intelligent evaluation method and system, and medium

Abstract

The present application relates to the technical field of power system operation control, and particularly relates to a power distribution network-oriented vehicle-network interaction aggregate intelligent evaluation method and system and a medium, the method comprising: acquiring power distribution network operation data, charging facility access data and different types of electric vehicle behavior data, and constructing a vehicle-network interaction data set; extracting multi-dimensional features based on the vehicle-network interaction data set and generating a comprehensive feature vector; constructing a vehicle-network interaction graph model based on the comprehensive feature vector and determining a multi-level vehicle-network interaction aggregate division result; constructing multi-source evaluation evidence based on the comprehensive feature vector and the division result, and obtaining a regulation and control effect evaluation result through evidence fusion; and determining available regulation and control capacity and outputting regulation and control decision information according to the regulation and control effect evaluation result. Through the present application, the problem that the regulation and control effect of a vehicle-network interaction aggregate in a power distribution network cannot be accurately and reliably evaluated in the prior art is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of power system operation and control technology, and in particular to intelligent evaluation methods, systems and media for vehicle-grid interaction aggregation in distribution networks. Background Technology

[0002] With the advancement of new power system construction, electric vehicles, possessing both load and energy storage attributes, are gradually becoming an important component of the adjustable resources on the distribution network side. Vehicle-to-grid (V2G) interaction refers to the energy exchange between electric vehicles and the power grid through charging and discharging facilities, thereby participating in operational regulation such as peak shaving, frequency regulation, and voltage support. Due to the limited capacity of individual electric vehicles and their dispersed and random access behavior, it is usually necessary to aggregate them to form V2G aggregation systems with a certain scale and responsiveness. Therefore, how to scientifically evaluate the regulation effect of such aggregation systems has become a key issue in the coordinated regulation of distribution networks.

[0003] In existing technologies, relevant evaluation methods mostly use a single indicator or a simple combination of indicators to evaluate vehicle-to-grid interaction resources. They often focus on local factors such as capacity or response capability, lacking a comprehensive characterization of the differences in control capability, reliability, and spatiotemporal distribution characteristics among different types of electric vehicles. At the same time, existing aggregation methods mostly make rough divisions based on geographical location or capacity, failing to fully utilize the complementary relationships between different types of electric vehicles in time and space, which can easily lead to unreasonable resource allocation within the aggregation. In addition, existing methods are insufficient in handling uncertainty and information conflicts when dealing with multidimensional heterogeneous information from different vehicle models, regions, and time periods, resulting in poor accuracy and reliability of evaluation results. Therefore, they are difficult to meet the actual needs of accurately and reliably evaluating the control effect of vehicle-to-grid interaction aggregation in distribution network scenarios.

[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 intelligent evaluation of vehicle-grid interaction aggregation in power distribution networks, 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 smart evaluation method for vehicle-to-grid (V2G) interaction aggregates in power distribution networks, the method comprising: Acquire power distribution network operation data, charging facility access data, and behavioral data of different types of electric vehicles to construct a vehicle-to-grid interaction dataset associated with power distribution network nodes; Based on the vehicle-to-grid interaction dataset, multi-dimensional features representing the ability of different types of electric vehicles to participate in power distribution network regulation are extracted, and a comprehensive feature vector of electric vehicle resource units is generated. Based on the comprehensive feature vector, a vehicle-network interaction graph model representing the spatiotemporal correlation between electric vehicle resource units is constructed, and the multi-level vehicle-network interaction aggregate partitioning result for distribution network regulation is determined based on the vehicle-network interaction graph model. Based on the comprehensive feature vector and the segmentation results of the multi-level vehicle-network interaction aggregate, multi-source evaluation evidence is constructed, and the evaluation results of the regulation effect of the vehicle-network interaction aggregate are obtained through evidence fusion. Based on the evaluation results of the control effect, the available control capability of the vehicle-grid interaction aggregator for the distribution network is determined, and the corresponding control decision information is output.

[0007] Furthermore, multidimensional features characterizing the ability of different types of electric vehicles to participate in power distribution network regulation are extracted, including: The different types of electric vehicles are classified into electric taxis, electric private cars, and electric buses; Based on the vehicle-to-grid interaction dataset, the regulation capacity features of different types of electric vehicles are extracted according to battery capacity, maximum charging and discharging power, average access time, charging and discharging efficiency, and availability. Based on the vehicle-to-grid interaction dataset, response reliability features of different types of electric vehicles are extracted according to behavior prediction accuracy, control response success rate, and continuous participation probability.

[0008] Furthermore, a comprehensive feature vector for electric vehicle resource units is generated, including: Based on the access probability and time regularity correction factor for each time period, the temporal distribution characteristics of the different types of electric vehicles are extracted using the improved Shannon entropy. The spatial distribution characteristics of the different types of electric vehicles were extracted based on geographic clustering and mobility indicators, wherein the geographic clustering was determined using the improved Moran index. Based on the differences between the holiday feature set and the weekday feature set, the scene difference features of the different types of electric vehicles are extracted; The comprehensive feature vector is generated by adaptively weighting the regulation capacity feature, the response reliability feature, the temporal distribution feature, the spatial distribution feature, and the scene difference feature using an attention-based neural network.

[0009] Furthermore, a vehicle-to-network interaction graph model is constructed, including: The vehicle-to-grid (V2G) interaction graph model is constructed using the electric vehicle resource units as nodes and the spatiotemporal complementary relationships between the electric vehicle resource units as edges. The combined feature vector and vehicle type code are used as the node features of each node in the vehicle-to-everything (V2X) interactive graph model. The edge weights of each edge in the vehicle-to-network interaction graph model are determined based on the spatial distance, temporal complementary distance, complementary coefficient, and preset weight parameters between the electric vehicle resource units.

[0010] Furthermore, based on the aforementioned vehicle-to-grid interaction graph model, the multi-level vehicle-to-grid interaction aggregation partitioning results for distribution network regulation are determined, including: The node feature matrix and normalized adjacency matrix in the vehicle-to-everything (V2X) interactive graph model are input into a graph convolutional neural network. The graph convolutional neural network is used to perform high-level feature learning on the vehicle-to-network interaction graph model to obtain the node representation of the electric vehicle resource unit; Based on the node representation, an aggregated state is constructed to meet the control objectives of the distribution network.

[0011] Furthermore, determining the results of the multi-level vehicle-to-everything (V2X) interaction aggregation also includes: Construct a set of actions including aggregate merging, aggregate separation, and aggregate adjustment; Construct a reward function consisting of a reward for control effectiveness, a reward for vehicle diversity, and a penalty for constraint violation; The reward for the regulation effect is determined based on the complementarity score between different electric vehicle resource units within each aggregate; A deep Q-network is used to perform action decisions on the aggregated state to obtain the partitioning result of the multi-level vehicle-network interaction aggregate; Under constraints on aggregate size and type diversity, a top-down recursive segmentation method is used to generate the multi-level vehicle-network interaction aggregate partitioning results.

[0012] Furthermore, construct multi-source assessment evidence, including: Establish a framework for identifying regulatory effects, including excellent, good, average, and poor. Based on the comprehensive feature vector and the division results of the multi-level vehicle-to-network interaction aggregate, evidence of regulation capacity, response reliability, temporal distribution, spatial distribution, and aggregation synergy is constructed. The neural network is used to map the regulatory capacity evidence, the response reliability evidence, the temporal distribution evidence, the spatial distribution evidence, and the aggregation synergy evidence to obtain the basic probability assignments corresponding to each evidence source; The basic probability assignment is normalized using an improved Softmax function.

[0013] Furthermore, the evaluation results of the regulation effect of the vehicle-to-everything (V2X) interaction aggregation and the determination of available regulation capabilities are obtained, including: The degree of conflict among the various sources of evidence is detected; The weights of each evidence source are adaptively adjusted based on the uncertainty entropy of each evidence source. The modified Dempster synthesis rule is used to fuse the various evidence sources to obtain the evaluation results of the regulatory effect; The regulation effect level and evaluation score corresponding to the vehicle-to-network interaction aggregate are determined based on the Pignistic probability. The evaluation results of the regulatory effect are dynamically updated by combining confidence interval estimation and forgetting factor. Based on the dynamically updated evaluation results of the control effect, the available control capability of the vehicle-grid interaction aggregate for peak shaving and valley filling, frequency regulation and voltage support of the distribution network in the current period or the predicted period is determined, and the control decision information is output.

[0014] A vehicle-to-grid (V2G) interaction aggregation intelligent evaluation system for power distribution networks, the system comprising: The data acquisition module acquires power distribution network operation data, charging facility access data, and behavioral data of different types of electric vehicles to construct a vehicle-to-grid interaction dataset associated with power distribution network nodes. The feature extraction module extracts multi-dimensional features representing the ability of different types of electric vehicles to participate in power distribution network regulation based on the vehicle-to-grid interaction dataset, and generates a comprehensive feature vector of electric vehicle resource units. The aggregation and partitioning module constructs a vehicle-network interaction graph model representing the spatiotemporal correlation between electric vehicle resource units based on comprehensive feature vectors, and determines the multi-level vehicle-network interaction aggregation partitioning results for distribution network regulation based on the vehicle-network interaction graph model. The regulation and evaluation module constructs multi-source evaluation evidence based on comprehensive feature vectors and multi-level vehicle-network interaction aggregate segmentation results, and obtains the regulation effect evaluation results of the vehicle-network interaction aggregate through evidence fusion. The decision output module determines the available control capability of the vehicle-grid interaction aggregator for the distribution network based on the control effect evaluation results, and outputs the corresponding control decision information.

[0015] A computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, can implement the intelligent evaluation method for vehicle-grid interaction aggregates oriented towards power distribution networks.

[0016] The technical solution of this invention can achieve the following technical effects: By constructing a multi-dimensional feature system for different types of electric vehicles and generating a comprehensive feature vector, a vehicle-network interaction graph model is further established based on spatiotemporal correlation to achieve multi-level aggregation division. Then, the comprehensive feature vector and the aggregation division results are combined to perform multi-source evidence fusion evaluation, thereby obtaining the regulation effect and available regulation capability of the vehicle-network interaction aggregation on the distribution network. This effectively solves the problem that existing technologies are unable to accurately and reliably evaluate the regulation effect of the vehicle-network interaction aggregation in the distribution network.

[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 intelligent evaluation method for vehicle-grid interaction aggregates in power distribution networks; Figure 2 A comparative analysis chart of evaluation indicators for vehicle-to-network interaction effects of different types of electric vehicles; Figure 3 A comparative verification diagram showing the control effect of multi-level aggregate partitioning methods; Figure 4 A convergence characteristic diagram for multi-source information fusion evaluation based on DS evidence theory; Figure 5 This chart compares and analyzes the results of regulatory effectiveness assessments with actual performance across multiple 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 an intelligent evaluation method for vehicle-grid interaction aggregates in power distribution networks, the method including: S10: Acquire power distribution network operation data, charging facility access data, and behavioral data of different types of electric vehicles to construct a vehicle-to-grid interaction dataset associated with power distribution network nodes; S20: Based on the vehicle-to-grid interaction dataset, extract multi-dimensional features representing the ability of different types of electric vehicles to participate in power distribution network regulation and control, and generate a comprehensive feature vector of electric vehicle resource units; S30: Construct a vehicle-network interaction graph model representing the spatiotemporal correlation between electric vehicle resource units based on the comprehensive feature vector, and determine the multi-level vehicle-network interaction aggregate partitioning results for distribution network regulation based on the vehicle-network interaction graph model; S40: Construct multi-source evaluation evidence based on comprehensive feature vectors and multi-level vehicle-network interaction aggregate segmentation results, and obtain the evaluation results of the regulation effect of the vehicle-network interaction aggregate through evidence fusion; S50: Determine the available control capability of the vehicle-grid interaction aggregator for the distribution network based on the control effect evaluation results, and output the corresponding control decision information.

[0023] Specifically, the first step involves data acquisition and dataset construction. This includes collecting power distribution network operation data, charging facility access data, and behavioral data from different types of electric vehicles. The power distribution network operation data includes distribution network node topology information, node load information, voltage information, and operating status information. The charging facility access data includes charging / battery swapping station locations, access nodes, charging / discharging power ranges, and access duration information. The behavioral data from different types of electric vehicles includes vehicle type, charging / discharging records, travel time, parking location, access probability, and response records. The acquired data is then cleaned, aligned, and standardized, and mapped and associated with power distribution network nodes to construct a vehicle-to-grid interaction dataset associated with these nodes. The different types of electric vehicles preferably include electric taxis, electric private cars, and electric buses to reflect the differentiated characteristics of different vehicle types participating in vehicle-to-grid interaction. Secondly, multi-dimensional feature extraction and comprehensive feature vector generation are performed. Based on the vehicle-to-grid interaction dataset, multi-dimensional features representing the ability of different types of electric vehicles to participate in power distribution network regulation are extracted. Among them, regulation capacity features can be determined by combining battery capacity, maximum charging and discharging power, average access time, charging and discharging efficiency, and availability; response reliability features can be determined by combining behavior prediction accuracy, regulation response success rate, and continuous participation probability; temporal distribution features can be quantified by using improved Shannon entropy to quantify the access probability and regularity of vehicles in each time period; spatial distribution features can be represented by combining geographical clustering and mobility, and the geographical clustering can be determined by using the improved Moran index; scene difference features can be obtained by the difference between the feature sets of holidays and weekdays; then, an attention-based neural network is used to adaptively weight the above multi-dimensional features to generate a comprehensive feature vector corresponding to each electric vehicle resource unit. Next, the vehicle-to-grid (V2G) interaction graph model is constructed and multi-level aggregates are partitioned. Based on the comprehensive feature vector, each electric vehicle resource unit is used as a node, and the spatiotemporal relationship between each resource unit is used as an edge to construct a V2G interaction graph model representing the spatiotemporal relationship between electric vehicle resource units. The node features are composed of the comprehensive feature vector and vehicle type code, and the edge weights are determined by the spatial distance, temporal complementary distance, complementary coefficient, and preset weight parameters between resource units. Then, the node feature matrix and normalized adjacency matrix in the V2G interaction graph model are input into a graph neural network, preferably a graph convolutional neural network, to extract high-level feature representations. After obtaining the node representations, an aggregation state oriented towards the distribution network control target is further constructed, and the multi-level V2G interaction aggregate partitioning result is determined based on deep reinforcement learning. Preferably, the reinforcement learning model uses a deep Q-network, and the action space includes aggregate merging, aggregate separation, and aggregate adjustment. The reward function comprehensively considers the control effect, vehicle type diversity, and constraint violation penalty, so that the partitioned aggregates take into account both control capability and resource complementarity. Then, multi-source evaluation evidence construction and evidence fusion evaluation are performed. Based on the comprehensive feature vector and the multi-level vehicle-network interaction aggregate division results, multi-source evaluation evidence is constructed. The multi-source evaluation evidence includes at least regulatory capacity evidence, response reliability evidence, temporal distribution evidence, spatial distribution evidence, and aggregation synergy evidence. On this basis, a regulatory effect recognition framework is established, and each evidence source is mapped to a corresponding basic probability assignment through a neural network. An improved Softmax function is used to implement normalization constraints. Then, the degree of conflict between each evidence source is detected, and adaptive weight correction is performed according to the uncertainty entropy of each evidence source. The improved Dempster synthesis rule is then used to fuse each evidence source to obtain the regulatory effect evaluation result of the vehicle-network interaction aggregate. Furthermore, the regulatory effect level and evaluation score can be determined based on the Pignistic probability, and the evaluation results can be dynamically updated by combining confidence interval estimation and forgetting factor. Finally, the available controllability is determined and control decision is output. Based on the control effect evaluation results, the available controllability of the vehicle-grid interaction aggregate on the distribution network in the current or predicted time period is determined. The available controllability can be reflected in the aggregate's participation level, response level, and applicable scenarios in peak shaving, frequency regulation, and voltage support. Then, the corresponding control decision information is output to the distribution network operation control platform or the vehicle-grid interaction scheduling platform to provide decision-making basis for subsequent charging and discharging scheduling, resource allocation, interaction strategy formulation, and distribution network operation optimization.

[0024] The technical solution of this invention constructs a multi-dimensional feature system for different types of electric vehicles and generates a comprehensive feature vector. Furthermore, a vehicle-network interaction graph model is established based on spatiotemporal correlation to achieve multi-level aggregation division. Then, the comprehensive feature vector and aggregation division results are combined to perform multi-source evidence fusion evaluation, thereby obtaining the regulation effect and available regulation capability of the vehicle-network interaction aggregation on the power distribution network. This effectively solves the problem that existing technologies cannot accurately and reliably evaluate the regulation effect of the vehicle-network interaction aggregation in the power distribution network.

[0025] Furthermore, multidimensional features characterizing the ability of different types of electric vehicles to participate in power distribution network regulation are extracted, including: Different types of electric vehicles are classified into electric taxis, electric private cars, and electric buses; Based on the vehicle-to-grid interaction dataset, the regulation capacity features of different types of electric vehicles are extracted according to battery capacity, maximum charging and discharging power, average access time, charging and discharging efficiency and availability. Based on the vehicle-to-grid interaction dataset, response reliability features of different types of electric vehicles are extracted according to behavior prediction accuracy, control response success rate, and continuous participation probability.

[0026] As a preferred embodiment of the above, an evaluation index system for the effectiveness of different types of electric vehicles participating in vehicle-to-grid interaction is established. Furthermore, various types of electric vehicles include electric taxis (ET), electric private cars (EP), electric buses (EB), and other different models, each with significant differences in vehicle-to-everything (V2X) interaction. Furthermore, establish a vehicle-to-grid (V2G) interaction control capacity assessment index, taking into account factors such as battery capacity, charging and discharging power, and access duration for different vehicle models: ; in, For the regulation capacity index of Class V electric vehicles; This is the maximum charge / discharge power; This refers to the average access duration; For charge and discharge efficiency; Battery capacity; Availability; Furthermore, electric buses have the highest control capacity because their batteries have large capacity and long, relatively fixed access times; electric private cars have relatively low control capacity because their charging power is low and their access times are not fixed; electric taxis have control capacity between that of buses and private cars. Furthermore, establish vehicle-to-grid (V2G) interaction reliability assessment indicators, integrating factors such as behavior prediction accuracy, response success rate, and sustainability: ; in, For the reliability index of Class V electric vehicles; For the accuracy of behavior prediction; To regulate the success rate of response; For the probability of continued participation; , , For the weighting coefficients, satisfying .

[0027] Furthermore, the comprehensive feature vector generated for electric vehicle resource units includes: Based on the access probability and time regularity correction factor for each time period, the temporal distribution characteristics of different types of electric vehicles are extracted using improved Shannon entropy; Spatial distribution characteristics of different types of electric vehicles were extracted based on geographic clustering and mobility indicators, with geographic clustering determined using the improved Moran index. Extract scenario-specific features for different types of electric vehicles based on the differences between the holiday feature set and the weekday feature set; By using an attention-based neural network, adaptive weights are applied to regulation capacity features, response reliability features, temporal distribution features, spatial distribution features, and scene difference features to generate a comprehensive feature vector.

[0028] As a preferred embodiment of the above, based on historical data statistics, electric buses have the highest reliability, followed by electric taxis, while electric private cars have relatively low reliability. Furthermore, an evaluation index for time distribution characteristics was established, and the improved Shannon entropy theory was used to quantify the time distribution patterns of different vehicle models: ; in, Let V be the temporal distribution entropy of the v-th type of electric vehicle; Let be the access probability of vehicle of type v in time period t; As a time-regularity correction factor; Furthermore, electric buses have the most regular time distribution, electric private cars have the most dispersed time distribution, and electric taxis fall in between. Furthermore, establish spatial distribution characteristic assessment indicators, combining geographical clustering and mobility indicators: ; in, This is an indicator of the spatial distribution of electric vehicles of type v; Geographic clustering; Mobility indicators; , These are the weighting coefficients; Furthermore, geographic clustering is calculated using a modified Moran index: ; in, The total number of nodes; This is the spatial weight matrix; Let V be the density of vehicle type v at the nth node; Average density; Furthermore, establish differentiated indicators for the interactive characteristics between holidays and weekdays: ; in, For the difference between holidays and weekdays for Class V electric vehicles; To evaluate the feature set; and These are the characteristic values ​​for holidays and weekdays, respectively; Furthermore, an attention-based neural network algorithm is used to intelligently allocate weights to multi-dimensional indicators: ; in, These are the attention weights for the l-th layer; , , These are the query, key, and value matrices, respectively. Furthermore, the comprehensive evaluation indicator system is constructed as follows: ; in, For the comprehensive evaluation indicators of Class V electric vehicles; For the s-th sub-indicator (regulation capacity, reliability, temporal distribution, spatial distribution, and holiday differences); These are the weights learned through the attention mechanism.

[0029] Furthermore, constructing a vehicle-to-everything (V2X) interaction graph model includes: A vehicle-to-network interaction graph model is constructed using electric vehicle resource units as nodes and the spatiotemporal complementary relationships between electric vehicle resource units as edges. The comprehensive feature vector and vehicle type category code are combined as the node features of each node in the vehicle-to-everything (V2X) interactive graph model. The edge weights in the vehicle-network interaction graph model are determined based on the spatial distance, temporal complementary distance, complementary coefficient, and preset weight parameters between electric vehicle resource units.

[0030] As a preferred embodiment of the above, a method for dividing a multi-level electric vehicle-to-grid interaction aggregate in a power distribution network is established. Furthermore, based on the evaluation index system established in the aforementioned embodiments, graph neural networks and deep reinforcement learning algorithms are used to achieve intelligent partitioning of multi-level aggregates; Furthermore, a power distribution network vehicle-to-grid interaction graph structure model is established, with electric vehicles as nodes and the spatiotemporal complementary relationships between vehicles as edges: ; in, This is an interactive image for the car network; For electric vehicle nodes; Let it be the set of edges; It is an adjacency matrix; Furthermore, the node feature vectors are integrated with the multi-dimensional evaluation metrics described in the aforementioned embodiments: ; in, This is the feature vector of the nth node; For the regulation capacity index of Class V electric vehicles; For the reliability index of Class V electric vehicles; Let V be the temporal distribution entropy of the v-th type of electric vehicle; This is an indicator of the spatial distribution of electric vehicles of type v; For the difference between holidays and weekdays for Class V electric vehicles; Vehicle type codes (EB=1,ET=2,EP=3); Furthermore, the edge weights reflect the degree of spatiotemporal complementarity between different vehicle models: ; in, Let m be the spatial distance between vehicles n and m; For time-complementary distance; The complementarity coefficient; , , These are weight parameters; It is an exponential function.

[0031] Furthermore, based on the vehicle-to-grid interaction graph model, the multi-level vehicle-to-grid interaction aggregation partitioning results for distribution network regulation are determined, including: Input the node feature matrix and normalized adjacency matrix from the vehicle-to-everything (V2X) interactive graph model into the graph convolutional neural network. High-level feature learning is performed on the vehicle-to-network interaction graph model using graph convolutional neural networks to obtain the node representation of electric vehicle resource units; Based on the node representation, an aggregated state is constructed to meet the control objectives of the distribution network.

[0032] As a preferred embodiment of the above, the time complementary distance is further defined as: ; in, This represents the probability of electric vehicles of type n accessing the network during time period t. This represents the probability of electric vehicles of type m accessing the network during time period t. This represents the overlapping portion of the access probabilities of the two types of vehicles in time period t; This represents the coverage portion indicating the probability of two types of vehicles accessing the network in time period t. Furthermore, a graph convolutional neural network (GCN) is used to learn high-level feature representations of nodes and edges: ; in, Let L be the node feature matrix of the l-th layer; The normalized adjacency matrix is ​​denoted by A, and D represents the degree matrix corresponding to A. The weight matrix is ​​trainable. This is the activation function.

[0033] Furthermore, determining the results of the multi-level vehicle-to-everything (V2X) interaction aggregation also includes: Construct a set of actions including aggregate merging, aggregate separation, and aggregate adjustment; Construct a reward function consisting of a reward for control effectiveness, a reward for vehicle diversity, and a penalty for constraint violation; The reward for the regulation effect is determined based on the complementarity scores between different electric vehicle resource units within each aggregate; By using a deep Q-network to make action decisions on the aggregated state, the results of multi-level vehicle-network interaction aggregate partitioning are obtained; Under constraints on aggregate size and type diversity, a top-down recursive segmentation method is adopted to generate multi-level vehicle-network interaction aggregate partitioning results.

[0034] As a preferred embodiment of the above, a reinforcement learning aggregation strategy based on a deep Q-network (DQN) is further designed, and the state space is defined as: ; in, The node features at the current moment; This represents the current aggregation state; As a reward signal; Furthermore, the action space is defined as a set of aggregate operations: ; in, Indicates a merged aggregate and ; Indicates separated polymers ; Indicates adjustment of aggregate ; Furthermore, the reward function comprehensively considers the regulatory effect and diversity balance of the aggregate: ; in, Rewards for effective regulation; Rewards for diversity; To constrain violations and impose penalties; , , These are the weighting coefficients; Furthermore, the incentive for regulatory effects is calculated based on the complementarity of vehicle models within the aggregate: ; in, The number of aggregates; For the k-th aggregate; Size of the aggregate; Score the complementarity of vehicles n and m; Furthermore, the multi-level partitioning strategy employs a top-down recursive partitioning method: ; in, This represents the partitioning result of the l-th layer; The threshold parameter for the l-th layer; Furthermore, the aggregate size constraint and type diversity constraint are as follows: ; ; in, and These are the lower and upper limits for the size of the aggregate; A collection of vehicle model categories; Let be a binary variable, representing whether aggregate k contains model i.

[0035] Furthermore, constructing multi-source assessment evidence includes: Establish a framework for identifying regulatory effects, including excellent, good, average, and poor. Based on the comprehensive feature vector and the multi-level vehicle-to-network interaction aggregation results, we construct evidence for regulation capacity, response reliability, temporal distribution, spatial distribution, and aggregation synergy. By using neural networks to map evidence of regulatory capacity, response reliability, temporal distribution, spatial distribution, and convergent synergy, basic probability values ​​are assigned to each source of evidence. The improved Softmax function is used to normalize the assignment of basic probabilities.

[0036] As a preferred embodiment of the above, a method for evaluating the effect of vehicle-to-grid interactive collaborative regulation based on DS evidence theory is established; Furthermore, a multi-source information fusion-based regulatory effect evaluation framework is constructed based on Dempster-Shafer evidence theory to effectively process evaluation information from different vehicle models and dimensions; Furthermore, a framework for evaluating the effectiveness of regulation is defined: ; in, Indicates "excellent". Indicates "good". Indicates "generally". Indicates "poor"; Furthermore, a multi-source evidence acquisition mechanism based on the aforementioned embodiments is established: ; in, Evidence for regulating capacity; As evidence of reliability; Evidence based on time distribution; Evidence for spatial distribution; Evidence of polymer synergy; Furthermore, the basic probability assignment function (BPA) for each source of evidence is obtained through neural network mapping: ; in, For the e-th source of evidence, the hypothesis is... The basic probability assignment; Let e ​​be the input feature of the e-th evidence source; For neural network parameters; Furthermore, the neural network employs an improved Softmax function to ensure the normalization constraint of BPA: ; in, Let be the scoring function for the e-th source of evidence; Remove the empty set from the power set; Represents the framework for evaluating and identifying the effectiveness of regulation. One of the hypothetical propositions; It is an exponential function.

[0037] Furthermore, obtaining the evaluation results of the regulation effect of the vehicle-to-everything (V2X) interactive aggregation and determining the available regulation capabilities includes: The degree of conflict between various sources of evidence is examined; Adaptive weight adjustment is performed on each evidence source based on the uncertainty entropy of each evidence source; The modified Dempster synthesis rule was used to fuse the various evidence sources to obtain the evaluation results of the regulatory effect; The level of regulation effect and evaluation score corresponding to the vehicle-to-network interaction aggregate are determined based on the Pignistic probability. The evaluation results of the regulatory effect were evaluated by combining confidence interval estimation and dynamic updating of the forgetting factor. Based on the dynamically updated evaluation results of the control effect, determine the available control capabilities of the vehicle-grid interaction aggregator for peak shaving and valley filling, frequency regulation and voltage support of the distribution network in the current or predicted period, and output control decision information.

[0038] As a preferred embodiment of the above, further considering conflict detection between evidence sources, a conflict metric is defined: ; in, The conflict coefficient; Let the basic probability assignment function be the e-th evidence source; The focal element of the e-th evidence source; The total number of evidence sources; Furthermore, an improved Dempster composition rule is used to handle high-collision cases: ; in, The conflict coefficient is the correction factor. , An adaptive weighting factor; Furthermore, the adaptive weighting factor is dynamically adjusted based on the credibility of the evidence source: ; in, The attenuation coefficient; Let be the uncertainty entropy of the e-th evidence source; It is an exponential function; Furthermore, a decision rule based on Pignistic probability is established: ; in, Assumption The Pignistic probability; Assign a basic probability value to focal element A; The size of focal element A; Furthermore, the final evaluation score for the control effect is calculated as follows: ; in, Assumption Corresponding numerical score ( ); Furthermore, confidence interval estimation is introduced to quantify the uncertainty of the assessment results: ; in, The quantiles corresponding to the confidence level; To estimate the standard deviation; Furthermore, establish a dynamic updating mechanism for the evaluation results: ; in, Forgetting factor; The evaluation score for the new moment.

[0039] Example 2; Based on the same inventive concept as the intelligent evaluation method for vehicle-grid interaction aggregation in the aforementioned embodiments, this invention also provides an intelligent evaluation system for vehicle-grid interaction aggregation in the distribution network, the system comprising: The data acquisition module acquires power distribution network operation data, charging facility access data, and behavioral data of different types of electric vehicles to construct a vehicle-to-grid interaction dataset associated with power distribution network nodes. The feature extraction module extracts multi-dimensional features representing the ability of different types of electric vehicles to participate in power distribution network regulation based on the vehicle-to-grid interaction dataset, and generates a comprehensive feature vector of electric vehicle resource units. The aggregation and partitioning module constructs a vehicle-network interaction graph model representing the spatiotemporal correlation between electric vehicle resource units based on comprehensive feature vectors, and determines the multi-level vehicle-network interaction aggregation partitioning results for distribution network regulation based on the vehicle-network interaction graph model. The regulation and evaluation module constructs multi-source evaluation evidence based on comprehensive feature vectors and multi-level vehicle-network interaction aggregate segmentation results, and obtains the regulation effect evaluation results of the vehicle-network interaction aggregate through evidence fusion. The decision output module determines the available control capability of the vehicle-grid interaction aggregator for the distribution network based on the control effect evaluation results, and outputs the corresponding control decision information.

[0040] The adjustment system described above in this invention can effectively realize the intelligent evaluation method for vehicle-grid interaction aggregation in power distribution networks, 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 intelligent evaluation method for vehicle-grid interaction aggregates for power distribution networks 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 intelligent evaluation method for vehicle-grid interaction aggregates for power distribution networks.

[0042] Example 4; To verify the effectiveness of this invention, actual operating data from a municipal power distribution network was selected for case analysis. This network includes three 110kV substations and fifteen 35kV substations, connecting 45,000 electric private cars, 3,200 electric taxis, and 800 electric buses. Data collection spanned from January 1, 2024 to December 31, 2024, with a time resolution of 15 minutes. The experimental environment consisted of an Intel Xeon Gold 6248 processor, 256GB of RAM, and an NVIDIA A100 GPU, implemented using the PyTorch deep learning framework and the NetworkX graph processing library. Figure 2This paper presents a comparative analysis of vehicle-to-grid (V2G) interaction performance evaluation indicators for different types of electric vehicles. The bar chart uses different gray levels and textures to distinguish the three vehicle types: electric buses (black diagonal texture) perform best in terms of control capacity (0.89) and also achieve the highest level in reliability (0.91), mainly due to their large-capacity batteries, long-term access, and highly regular operation characteristics; electric taxis (gray dot texture) have moderate performance across all indicators, with a control capacity of 0.72 and a reliability of 0.78, reflecting the relatively stable participation willingness of operating vehicles; electric private cars (light gray cross texture) have the highest value in terms of time distribution entropy (4.2), indicating strong time flexibility but poor regularity, resulting in a relatively low overall evaluation index. Through the multi-dimensional evaluation system of this invention, the differentiated characteristics of the three vehicle types in different indicators are clearly quantified, and the comprehensive evaluation index achieves a discrimination of 0.65, significantly higher than the 0.32 of the traditional single-indicator evaluation method. Figure 3 The performance differences between the traditional geographic location segmentation method and the multi-level aggregate segmentation method of this invention are compared through two sub-graphs. The bar chart in the left figure shows that in terms of control capacity, the traditional method only reaches 0.65, while the method of this invention improves to 0.87, an improvement of 22.3%. In terms of response reliability, it improves from 0.72 in the traditional method to 0.91 in the method of this invention, an improvement of 18.7%. The line comparison in the right figure shows the changes in the vehicle diversity index of five different aggregates: the diversity index of the traditional method (marked by black dashed squares) fluctuates between 0.32 and 0.45, while the diversity index of the method of this invention (marked by black solid circles) remains above 0.68, with an average of 0.73. This shows that the segmentation strategy of this invention can effectively balance the characteristic differences of different vehicle types and achieve a synergistic effect of "complementing each other's strengths". Figure 4 The diagram illustrates the multi-source information fusion evaluation process and convergence characteristics based on the DS evidence theory. The figure includes three curves: the confidence level (black solid line circle) gradually increases from 0.3 to 0.91, reaching convergence after 150 iterations; the conflict coefficient (black dashed line square) gradually decreases from 0.5 to 0.23, indicating that conflicts between evidence are effectively resolved; and the consistency index (black dotted line triangle) steadily increases from 0.2 to 0.88. A red vertical line marks the convergence point at approximately 150 iterations. Compared to single-source evidence evaluation, the accuracy of multi-source fusion evaluation is improved by 28.5%, and the false positive rate is reduced by 41.2%, providing reliable evaluation results for vehicle-to-everything (V2X) interaction aggregation. Figure 5The bar chart was used to compare and analyze the evaluation results and actual performance of the method of this invention in different scenarios. The chart includes eight different application scenarios, and the evaluation score (black line texture) and actual performance (gray dot texture) show a high degree of consistency. In the peak weekday scenario, the evaluation score is 3.8 and the actual performance is 3.7, with a difference of only 0.1. During special periods such as holidays and the Spring Festival, the method of this invention can still maintain good evaluation accuracy. The upper left corner of the chart shows that the correlation coefficient reaches 0.93, and the evaluation accuracy rate is 89.6%, which fully verifies the reliability and practicality of the evaluation method of this invention and provides strong technical support for distribution network operation decision-making.

[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 smart evaluation method for vehicle-grid interaction aggregation in power distribution networks, characterized in that, The method includes: Acquire power distribution network operation data, charging facility access data, and behavioral data of different types of electric vehicles to construct a vehicle-to-grid interaction dataset associated with power distribution network nodes; Based on the vehicle-to-grid interaction dataset, multi-dimensional features representing the ability of different types of electric vehicles to participate in power distribution network regulation are extracted, and a comprehensive feature vector of electric vehicle resource units is generated. Based on the comprehensive feature vector, a vehicle-network interaction graph model representing the spatiotemporal correlation between electric vehicle resource units is constructed, and the multi-level vehicle-network interaction aggregate partitioning result for distribution network regulation is determined based on the vehicle-network interaction graph model. Based on the comprehensive feature vector and the segmentation results of the multi-level vehicle-network interaction aggregate, multi-source evaluation evidence is constructed, and the evaluation results of the regulation effect of the vehicle-network interaction aggregate are obtained through evidence fusion. Based on the evaluation results of the control effect, the available control capability of the vehicle-grid interaction aggregator for the distribution network is determined, and the corresponding control decision information is output.

2. The intelligent evaluation method for vehicle-to-grid interaction aggregation in power distribution networks according to claim 1, characterized in that, Extracting multidimensional features characterizing the ability of different types of electric vehicles to participate in power distribution network regulation, including: The different types of electric vehicles are classified into electric taxis, electric private cars, and electric buses; Based on the vehicle-to-grid interaction dataset, the regulation capacity features of different types of electric vehicles are extracted according to battery capacity, maximum charging and discharging power, average access time, charging and discharging efficiency, and availability. Based on the vehicle-to-grid interaction dataset, response reliability features of different types of electric vehicles are extracted according to behavior prediction accuracy, control response success rate, and continuous participation probability.

3. The intelligent evaluation method for vehicle-grid interaction aggregation in power distribution networks according to claim 2, characterized in that, Generate a comprehensive feature vector for electric vehicle resource units, including: Based on the access probability and time regularity correction factor for each time period, the temporal distribution characteristics of the different types of electric vehicles are extracted using the improved Shannon entropy. The spatial distribution characteristics of the different types of electric vehicles were extracted based on geographic clustering and mobility indicators, wherein the geographic clustering was determined using the improved Moran index. Based on the differences between the holiday feature set and the weekday feature set, the scene difference features of the different types of electric vehicles are extracted; The comprehensive feature vector is generated by adaptively weighting the regulation capacity feature, the response reliability feature, the temporal distribution feature, the spatial distribution feature, and the scene difference feature using an attention-based neural network.

4. The intelligent evaluation method for vehicle-grid interaction aggregation in power distribution networks according to claim 1, characterized in that, Constructing a vehicle-to-everything (V2X) interaction graph model, including: The vehicle-to-grid (V2G) interaction graph model is constructed using the electric vehicle resource units as nodes and the spatiotemporal complementary relationships between the electric vehicle resource units as edges. The combined feature vector and vehicle type code are used as the node features of each node in the vehicle-to-everything (V2X) interactive graph model. The edge weights of each edge in the vehicle-to-network interaction graph model are determined based on the spatial distance, temporal complementary distance, complementary coefficient, and preset weight parameters between the electric vehicle resource units.

5. The intelligent evaluation method for vehicle-grid interaction aggregation in power distribution networks according to claim 4, characterized in that, Based on the aforementioned vehicle-to-grid interaction graph model, the multi-level vehicle-to-grid interaction aggregation partitioning results for distribution network regulation are determined, including: The node feature matrix and normalized adjacency matrix in the vehicle-to-everything (V2X) interactive graph model are input into a graph convolutional neural network. The graph convolutional neural network is used to perform high-level feature learning on the vehicle-to-network interaction graph model to obtain the node representation of the electric vehicle resource unit; Based on the node representation, an aggregated state is constructed to meet the control objectives of the distribution network.

6. The intelligent evaluation method for vehicle-grid interaction aggregation in power distribution networks according to claim 5, characterized in that, The determination of the multi-level vehicle-to-everything (V2X) interaction aggregation result also includes: Construct a set of actions including aggregate merging, aggregate separation, and aggregate adjustment; Construct a reward function consisting of a reward for control effectiveness, a reward for vehicle diversity, and a penalty for constraint violation; The reward for the regulation effect is determined based on the complementarity score between different electric vehicle resource units within each aggregate; A deep Q-network is used to perform action decisions on the aggregated state to obtain the partitioning result of the multi-level vehicle-network interaction aggregate; Under constraints on aggregate size and type diversity, a top-down recursive segmentation method is used to generate the multi-level vehicle-network interaction aggregate partitioning results.

7. The intelligent evaluation method for vehicle-grid interaction aggregation in power distribution networks according to claim 1, characterized in that, Constructing multi-source assessment evidence, including: Establish a framework for identifying regulatory effects, including excellent, good, average, and poor. Based on the comprehensive feature vector and the division results of the multi-level vehicle-to-network interaction aggregate, evidence of regulation capacity, response reliability, temporal distribution, spatial distribution, and aggregation synergy is constructed. The neural network is used to map the regulatory capacity evidence, the response reliability evidence, the temporal distribution evidence, the spatial distribution evidence, and the aggregation synergy evidence to obtain the basic probability assignments corresponding to each evidence source; The basic probability assignment is normalized using an improved Softmax function.

8. The intelligent evaluation method for vehicle-grid interaction aggregation in power distribution networks according to claim 7, characterized in that, Obtaining the evaluation results of the regulation effect of the vehicle-to-everything (V2X) interaction aggregation and determining the available regulation capabilities includes: The degree of conflict among the various sources of evidence is detected; The weights of each evidence source are adaptively adjusted based on the uncertainty entropy of each evidence source. The modified Dempster synthesis rule is used to fuse the various evidence sources to obtain the evaluation results of the regulatory effect; The regulation effect level and evaluation score corresponding to the vehicle-to-network interaction aggregate are determined based on the Pignistic probability. The evaluation results of the regulatory effect are dynamically updated by combining confidence interval estimation and forgetting factor. Based on the dynamically updated evaluation results of the control effect, the available control capability of the vehicle-grid interaction aggregate for peak shaving and valley filling, frequency regulation and voltage support of the distribution network in the current period or the predicted period is determined, and the control decision information is output.

9. An intelligent evaluation system for vehicle-grid interaction aggregation in power distribution networks, characterized in that: The system includes: The data acquisition module acquires power distribution network operation data, charging facility access data, and behavioral data of different types of electric vehicles to construct a vehicle-to-grid interaction dataset associated with power distribution network nodes. The feature extraction module extracts multi-dimensional features representing the ability of different types of electric vehicles to participate in power distribution network regulation based on the vehicle-to-grid interaction dataset, and generates a comprehensive feature vector of electric vehicle resource units. The aggregation and partitioning module constructs a vehicle-network interaction graph model representing the spatiotemporal correlation between electric vehicle resource units based on comprehensive feature vectors, and determines the multi-level vehicle-network interaction aggregation partitioning results for distribution network regulation based on the vehicle-network interaction graph model. The regulation and evaluation module constructs multi-source evaluation evidence based on comprehensive feature vectors and multi-level vehicle-network interaction aggregate segmentation results, and obtains the regulation effect evaluation results of the vehicle-network interaction aggregate through evidence fusion. The decision output module determines the available control capability of the vehicle-grid interaction aggregator for the distribution network based on the control effect evaluation results, and outputs the corresponding control decision information.

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 intelligent evaluation method for vehicle-grid interaction aggregates for power distribution networks as described in any one of claims 1-8.

Images