Electric vehicle charging load power grid operation real-time scheduling optimization method and system

Abstract

The invention discloses an electric vehicle charging load power grid operation real-time scheduling optimization method and system, belongs to the technical field of electric vehicles, and solves the problems of insufficient coordination of new energy consumption and network congestion regulation and weak real-time dynamic response capability in electric vehicle charging load power grid operation real-time scheduling optimization. Comprising the following steps: constructing a vehicle-road-network coupling three-domain knowledge graph fusion framework, bidirectionally coupling electric vehicle charging and discharging behavior dynamic sensing and charging pile load sensing, outputting a final optimization scheme, and pushing the optimization scheme to a user side, a power grid side and a road network side; according to the method, the coupling relation among the vehicle, the network and the power grid is considered, the power grid blocking peak regulation problem is solved through coupling of the three knowledge maps and the charging and discharging characteristics of the electric vehicle, the collaboration of new energy consumption and network blocking regulation and control is enhanced, the real-time dynamic response capability of the electric vehicle is improved, and the power grid blocking peak regulation and control efficiency is improved. And technical reference is provided for power grid operation management personnel.

Description

Technical Field

[0001] The present invention belongs to the technical field of electric vehicles and relates to a method and system for real-time scheduling optimization of electric vehicle charging load power grid operation. Background Art

[0002] With the rapid adoption of new energy vehicles, electric vehicles are becoming increasingly important in power grid dispatch and operation management. As the impact of electric vehicle (EV) charging loads on the power grid continues to grow, the grid needs to ensure safe, stable, and economical operation through real-time monitoring, optimization, and dispatching. Current real-time grid dispatch of EV charging loads fails to fully consider the quantitative modeling of the bounded rational behavior of EV users. This leads to insufficient coordination between new energy consumption and network congestion control, as well as weak real-time dynamic response capabilities, which are increasingly impacting the safe and stable operation of the power grid.

[0003] Knowledge graph technology has the ability to analyze multi-label big data reasoning knowledge and is currently widely used in the field of power systems. For example, the invention patent with application publication number CN118674175A discloses a method for modeling the characteristics of electric vehicle charging loads based on knowledge graphs. By proposing a method for modeling the characteristics of electric vehicle charging loads based on knowledge graphs, it is possible to refine the characteristics of electric vehicle charging loads and establish a typical load characteristic model of electric vehicle charging loads in a multi-dimensional and autonomous manner. However, the existing technology does not take into account the coupling relationship between vehicles, networks, and power grids, making it difficult to use the charging and discharging characteristics of electric vehicles to solve the problem of grid congestion and peak regulation. Real-time scheduling of electric vehicle charging load grid operations based on the above-mentioned coupling relationship involves multiple knowledge graphs of vehicles, networks, and power grids, and involves path planning for electric vehicles in different scenarios, resulting in a huge amount of calculation and a variety of scenarios. Summary of the Invention

[0004] The technical solution of the present invention is used to solve the problems of insufficient coordination between new energy consumption and network congestion regulation and control, and weak real-time dynamic response capability in real-time scheduling optimization of electric vehicle charging load power grid operation.

[0005] The present invention solves the above technical problems through the following technical solutions: A method for optimizing the real-time dispatching of electric vehicle charging load power grid operation includes the following steps: S1: Build a vehicle-road-network coupled knowledge graph fusion architecture, including vehicle feature graphs, road network resource graphs, and power grid topology graphs. Store knowledge graph data in a graph database and update it in real time. S2 builds a dynamic perception layer based on the vehicle-road-network coupled knowledge graph, including: Build dynamic perception of electric vehicle charging and discharging behavior to predict the spatiotemporal distribution of electric vehicle charging demand; Build charging pile load sensing to predict charging and discharging power of charging piles; Bidirectionally couple the dynamic perception of electric vehicle charging and discharging behavior with the perception of charging pile load to improve the accuracy of dynamic perception of electric vehicle charging behavior; S3: Build a multi-objective optimization layer based on the vehicle-road-network coupling knowledge graph, construct a multi-objective optimization function, and use hybrid adaptive particle swarm optimization to optimize the multi-objective optimization function and output the final optimization solution; S4 builds a strategy generation layer based on the vehicle-road-network coupling knowledge graph, and pushes the optimization plan to the user side, power grid side, and road network side.

[0006] Furthermore, the entity of the vehicle characteristic map described in S1 is the vehicle body, and the categories of the vehicle characteristic map include basic vehicle information, user behavior profile, and on-board BMS real-time data; the attributes of the basic vehicle information include brand, model, fuel type, and rated power; the attributes of the user behavior profile include location preference, time preference, and bounded rationality threshold; the attributes of the on-board BMS real-time data include charging status, health status, charging rate, and location data; The entities of the power grid topology map are electric vehicle charging piles. The categories of the power grid topology map include basic power grid information, power grid information, and real-time status. The attributes of the basic power grid information include user number, capacity, number of parking spaces, charging type, address, and energy consumption category. The attributes of the power grid information include grid-connected lines and grid-connected substations. The attributes of the real-time status include available parking spaces, power margin, and real-time electricity prices. The categories of the road network resource map include traffic flow and dynamic road resistance; the attributes of traffic flow include path time and congestion level, and the attributes of dynamic road resistance include weather conditions and parking space prediction.

[0007] Furthermore, the storage of knowledge graph data in a graph database and real-time updating described in S1 are specifically as follows: The graph database adopts the Neo4j graph database, uses vehicle terminal data to dynamically update the vehicle feature map in real time, uses the power grid SCADA and distribution network OMS data to establish the power grid topology map, and uses the AutoNavi API to build the road network resource map.

[0008] Furthermore, the S2 includes the following steps: S21: Build dynamic perception of electric vehicle charging and discharging behavior. Specifically, build a dynamic perception model of user charging demand, dynamically update the vehicle feature map in real time through the vehicle terminal, obtain the historical SOC and SOH data of electric vehicles based on the vehicle body, use the bidirectional long short-term memory network to predict the charging and discharging power of electric vehicles and judge the charging and discharging behavior of electric vehicles based on the historical data of individual vehicles, and couple it with the user behavior profile of the vehicle body; S22, building a charging pile load sensor, specifically obtaining the charging pile charging load historical data based on the charging pile itself, and using a bidirectional long short-term memory network to predict the charging and discharging power of the charging pile after data preprocessing; S23, at the same time, couples the dynamic perception of electric vehicle charging and discharging behavior with the charging pile load perception in a bidirectional manner, considers the multi-temporal and spatial load characteristics, and improves the accuracy of the dynamic perception of electric vehicle charging behavior.

[0009] Furthermore, the user charging demand dynamic perception model in S21 is specifically an edge computing model based on incremental updates; First, set the incremental update trigger conditions. When the vehicle's state of charge changes by 1% or more, its health changes by 1% or more, and the time interval is greater than or equal to 1 minute, the vehicle's characteristic map is incrementally updated. Only the changed data is updated, and unchanged data is not uploaded for update. Secondly, the edge computing model is adopted to process abnormal jump data only on the vehicle terminal side and filter out instantaneous SOC and SOH jumps.

[0010] Furthermore, the bidirectional long short-term memory network described in S21 is used to predict the charging and discharging power of electric vehicles and judge the charging and discharging behavior of electric vehicles, and the user behavior portrait of the vehicle body is coupled as follows: First, the long short-term memory network outputs the predicted vehicle charge and discharge power based on the historical SOC and SOH data of individual vehicles. It analyzes the vehicle's SOC and considers that the vehicle is charging when the SOC data increases, and discharging when the SOC data decreases. Secondly, based on the predicted vehicle charging and discharging power and the judged charging and discharging behavior, the user's location preference and time preference data in the vehicle characteristic map are coupled to obtain the real-time location and time preference results of electric vehicle charging and discharging; In the S21, the long short-term memory network prediction model is trained by the collaborative method of incremental training and full training, specifically: the prediction model is first incrementally trained, and the historical SOC and SOH data of the previous 7 days are used to dynamically iterate and fine-tune the prediction model, and the prediction results are continuously updated and optimized; the average prediction error of the prediction model is calculated, and when the average prediction error calculated for 3 consecutive days exceeds the preset threshold, the prediction model is fully trained.

[0011] Furthermore, the bidirectional coupling in S23 is specifically: The charging and discharging power of each electric vehicle is aggregated to obtain the aggregated charging and discharging power of the charging pile at this moment:

[0012] Where, express The charging power of the charging piles aggregated at the moment, represents the number of aggregated electric vehicles, express Time The charging power of an electric vehicle, ; Based on the user preference coefficient and the historical feature coefficient, a two-way coupling is calculated and expressed using the following logic:

[0013] Where, Indicates the bidirectional coupling The charging power of the charging pile at this moment, represents the user preference coefficient, represents the historical characteristic coefficient, express The bidirectional long short-term memory network is used to predict the charging and discharging power of the charging pile at the moment; among them, the user preference coefficient and historical characteristic coefficient First, solve the problem based on the data of the previous two historical moments, and then update it dynamically, using the following logic:

[0014] Where, 、 Represents the bidirectional coupling time, The charging power of the charging pile at the moment, 、 Respectively time, Bidirectional long short-term memory network is used to predict the charging and discharging power of charging piles at all times. 、 Respectively time, The charging power of the charging pile aggregated at all times.

[0015] Furthermore, the step S3 includes the following steps: S31, on the grid side, design the objective function of the new energy consumption rate, and add line congestion peak-shaving constraints, power balance constraints, charging pile capacity constraints, charging pile parking space constraints, SOC constraints, and SOH constraints; S32, on the user side, design the user charging cost objective function, add the maximum and minimum constraints of electricity price and user satisfaction constraints; S33, on the road network side, design the optimal objective function for electric vehicle routing; S34, based on the above objective function, construct a multi-objective optimization function, using the following logic representation:

[0016] Where, represents the charging cost target, express Time-of-use electricity prices at different times, express The instantaneous charging power of electric vehicles at this moment, Indicates the total charging cost of the electric vehicle during the charging time. Indicates the time when the electric vehicle starts charging. Indicates the time when the electric vehicle ends charging. Indicates the total charging time of the electric vehicle; Indicates the target of new energy consumption rate, represents the amount of new energy consumed, P EV Represents the predicted power generation of new energy, represents the amount of new energy consumption from the beginning to the end of the multi-objective optimization function, Represents the predicted power generation of renewable energy from the beginning to the end of the multi-objective optimization function, represents the entire optimization cycle of the multi-objective optimization function, represents the new energy consumption rate, , is the optimal path target for electric vehicles, The charging path time for electric vehicles, express 、 and On the basis of satisfying the constraints of S31~S32 respectively, solve 、 and All achieve the smallest global optimal solution set; S35, adopts hybrid adaptive particle swarm optimization for multi-objective optimization function.

[0017] Furthermore, the hybrid adaptive particle swarm optimization multi-objective optimization function described in S35 is specifically: S351, first, generate a particle swarm, each particle representing a charging power allocation scheme; Secondly, dynamic weight adjustment is performed; constraint processing is performed according to the constraint conditions of S31~S32, and the constraint processing is specifically to use a repair strategy to handle out-of-bounds particles; S352, calculate the multi-objective fitness of each particle and screen the non-dominated optimal solution set based on the Pareto frontier; S353: Modify user preferences based on user satisfaction. Specifically, perform satisfaction-weighted sorting on the Pareto solution set and output the final optimization solution.

[0018] Furthermore, the S4 is specifically: S41, on the user side, based on the above final optimization plan, a minute-level rolling real-time optimization mechanism is established to push the user-side path selection to the electric vehicle terminal in real time; S42, on the grid side, forming a real-time response curve of electric vehicle charging electricity prices based on the real-time electricity prices in the grid topology map, and obtaining a line congestion and a new energy consumption hotspot map based on the power flow information and new energy power in the grid topology map; S43, on the road network side, an electric vehicle path flow diagram is formed according to the traffic flow in the road network knowledge map, and a charging pile parking space heat map is obtained according to the real-time status of the charging piles in the power grid topology map.

[0019] The present invention also provides a real-time dispatch optimization system for electric vehicle charging load power grid operation, comprising: The data storage module is used to build a vehicle-road-network coupled three-domain knowledge graph fusion architecture, including vehicle feature graphs, road network resource graphs, and power grid topology graphs. The knowledge graph data is stored in a graph database and updated in real time. The dynamic perception module builds a dynamic perception layer based on the vehicle-road-network coupling knowledge graph, including: Build dynamic perception of electric vehicle charging and discharging behavior to predict the spatiotemporal distribution of electric vehicle charging demand; Build charging pile load sensing to predict charging and discharging power of charging piles; Bidirectionally couple the dynamic perception of electric vehicle charging and discharging behavior with the perception of charging pile load to improve the accuracy of dynamic perception of electric vehicle charging behavior; The multi-objective optimization module builds a multi-objective optimization layer based on the vehicle-road-network coupling knowledge graph, constructs a multi-objective optimization function, and uses a hybrid adaptive particle swarm optimization to optimize the multi-objective optimization function to output the final optimization solution; The strategy generation module builds a strategy generation layer based on the vehicle-road-network coupling knowledge graph, and pushes the optimization plan to the user side, power grid side and road network side.

[0020] The advantages of the present invention are: (1) The present invention takes into account the coupling relationship between vehicles, networks and power grids. By coupling three knowledge graphs, it integrates the multi-dimensional data of vehicles, networks and power grids, and maps entities across domains based on the graph database, thereby enhancing the synergy between new energy consumption and network congestion control.

[0021] (2) The present invention constructs a dynamic perception layer based on the three-domain knowledge graph, and dynamically updates the vehicle feature graph in real time through the vehicle terminal based on the incremental update edge computing model. While improving the quality of the vehicle feature graph data, it effectively reduces the amount of updated data for electric vehicles, alleviates the difficulty of local data processing, and optimizes resource allocation. The present invention uses a bidirectional long short-term memory network to predict the charging and discharging power of each electric vehicle and judge the charging and discharging behavior of the electric vehicle, couples the charging and discharging behavior of the electric vehicle on the user side with the user behavior portrait of the vehicle body, and aggregates the charging and discharging behavior perception results of each electric vehicle on a charging pile basis; on the one hand, the charging pile uses a bidirectional long short-term memory network to predict the charging and discharging power, and on the other hand, uses the user preference coefficient and the historical characteristic coefficient to bidirectionally couple the predicted charging and discharging power with the charging and discharging behavior perception results of the electric vehicle, effectively improving the dynamic perception accuracy of the electric vehicle charging behavior and predicting the spatiotemporal distribution of the charging demand of the electric vehicle.

[0022] (3) Based on the topological analysis capability of the knowledge graph, the present invention constructs a multi-objective optimization function and designs constraints from the three perspectives of the power grid side, the user side, and the road network side, quantitatively models the bounded rational behavior of electric vehicle users, uses HAPSO to solve the multi-objective optimization function, and re-weights and sorts the output optimal solution set according to user satisfaction to generate an optimal charging plan that is more in line with user preferences. It uses the charging and discharging characteristics of electric vehicles to solve the problem of grid congestion and peak regulation, enhances the synergy between new energy consumption and network congestion regulation, improves the real-time dynamic response capability of electric vehicles, and provides technical reference for grid operation managers. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of a method for real-time scheduling optimization of electric vehicle charging load power grid operation according to a first embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the vehicle-road-network coupled three-domain knowledge graph according to the first embodiment of the present invention; Figure 3 Schematic diagram of the structure of the dynamic perception layer of the first embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of the multi-objective optimization layer of the first embodiment of the present invention; Figure 5 It is a structural diagram of the strategy generation in the first embodiment of the present invention. DETAILED DESCRIPTION

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0025] The technical solution of the present invention is further described below with reference to the accompanying drawings and specific embodiments: Example 1 Knowledge graph technology, with its ability to analyze multi-label big data and reason about knowledge, is currently widely used in the power system sector. Real-time scheduling of electric vehicle charging loads and grid operations, involving a knowledge graph of vehicle, network, and grid coupling relationships, involves multiple knowledge graphs and involves route planning for electric vehicles in different scenarios, resulting in a massive amount of computation and a wide variety of scenarios. Knowledge graph technology, however, can leverage graphical models to establish real-time optimized scheduling models for electric vehicles, making it particularly well-suited for the real-time scheduling method for electric vehicle charging loads and grid operations based on a multi-layer vehicle-road-network coupled knowledge graph, as provided in this invention.

[0026] Neo4j is an open-source graph database storage software. Its extensive practical applications both domestically and internationally have led to new developments in both domestic and international markets. Neo4j can also be viewed as a high-performance graph engine with all the features of a mature database, providing a foundation for the real-time scheduling method for electric vehicle charging load power grid operation based on a multi-layer vehicle-road-network coupled knowledge graph provided in this invention.

[0027] In order to provide a further understanding and appreciation of the structural features and effects achieved by the present invention, a detailed description is provided with reference to preferred embodiments and accompanying drawings as follows: like Figure 1 Specifically, a method for optimizing the real-time dispatch of electric vehicle charging load power grid operation based on a multi-layer vehicle-road-network coupling knowledge graph is disclosed, comprising the following steps: S1, builds a vehicle-road-network coupled three-domain knowledge graph fusion architecture, stores knowledge graph data through a graph database and updates it in real time.

[0028] Currently, power grid records contain a large amount of historical data on electric vehicle charging loads and power grid topology data. This data is used to construct a vehicle-road-network coupled three-domain knowledge graph fusion architecture. This architecture primarily comprises three knowledge graph subgraphs: the vehicle characteristic graph (VKG), the road network resource graph (RKG), and the grid topology graph (GKG). The vehicle characteristic graph describes the current operating status of electric vehicles and owner preferences, the road network resource graph describes the vehicle's road traffic status, and the grid topology graph describes the grid's operating status. This architecture enables multi-layered vehicle-road-network data coupling, providing a data foundation for real-time scheduling optimization of electric vehicle charging loads and power grid operations.

[0029] Specifically, such as Figure 2As shown in the figure, the vehicle-road-network coupled three-domain knowledge graph fusion architecture in this embodiment includes three knowledge graph subgraphs: vehicle feature graph (VKG), road network resource graph (RKG), and grid topology graph (GKG). Each knowledge graph subgraph is modeled in the form of a triple: The vehicle feature map utilizes dynamic ontology modeling technology, with the vehicle entity being the vehicle itself. It contains three categories: basic vehicle information, user behavior profiles, and real-time data from the vehicle's battery management system (BMS). Basic vehicle information includes four attributes: brand, model, fuel type, and rated power. User behavior profiles include location preference, time preference, and bounded rationality threshold. Real-time BMS data includes state of charge (SOC), state of health (SOH), charge rate, and location data, and is updated in real time based on the vehicle's current status.

[0030] The grid topology map uses electric vehicle charging stations as entities and includes three categories: basic grid information, grid information, and real-time status. Basic grid information includes six attributes: user number, capacity, number of parking spaces, charging type, address, and energy usage category. Grid information includes two attributes: grid-connected lines and grid-connected substations. Grid-connected lines include four sub-attributes: power flow information, new energy power, thermal stability limit, and energy storage information. Real-time status includes three attributes: available parking spaces, power margin, and real-time electricity price.

[0031] The road network resource map includes two categories: traffic flow and dynamic road resistance. Traffic flow includes two attributes: path time and congestion degree, and dynamic road resistance includes two attributes: weather conditions and parking space prediction.

[0032] In this embodiment, the graph database uses a Neo4j graph database to store vehicle-road-network coupled knowledge graph data, enabling the fusion of the three-domain knowledge graphs. Vehicle feature graphs are dynamically updated in real time using vehicle terminal data, grid SCADA and distribution network OMS data are used to establish grid topology graphs, and the AutoNavi API is used to automatically construct road network resource graphs. By integrating global data from vehicle terminals, roadside units, and the grid SCADA system, a three-domain knowledge graph with spatiotemporal correlation is constructed, capturing electric vehicle movement trajectories, traffic network status, and grid operating conditions in real time.

[0033] S2 builds a dynamic perception layer based on the vehicle-road-network coupled knowledge graph, including: Build dynamic perception of electric vehicle charging and discharging behavior to predict the spatiotemporal distribution of electric vehicle charging demand; Build charging pile load sensing to predict charging and discharging power of charging piles; The dynamic perception of electric vehicle charging and discharging behavior is bidirectionally coupled with the charging pile load perception to improve the accuracy of the dynamic perception of electric vehicle charging behavior.

[0034] like Figure 3 As shown, this embodiment builds a dynamic perception layer based on the vehicle-road-network coupled three-domain knowledge graph fusion architecture, which specifically includes the following steps: S21, build dynamic perception of electric vehicle charging and discharging behavior, specifically build a dynamic perception model of user charging needs, dynamically update the vehicle feature map in real time through the on-board terminal, obtain the historical SOC and SOH data of electric vehicles based on the vehicle body, use the bidirectional long short-term memory network (BiLSTM) to predict the charging and discharging power of electric vehicles and judge the charging and discharging behavior of electric vehicles based on the historical data of individual vehicles, and couple the user behavior portrait of the vehicle body.

[0035] In this embodiment, the real-time dynamic update of the vehicle characteristic map by the vehicle terminal is specifically as follows: the vehicle terminal reads vehicle data at a high frequency of 1Hz and uses the CAN bus through the OBD-II interface, and dynamically updates the real-time data of the electric vehicle by using local caching and compressed uploading.

[0036] Specifically, due to the huge amount of electric vehicle data, this embodiment constructs a dynamic perception model of user charging demand based on an incrementally updated edge computing model, which is used to dynamically update the real-time data of electric vehicles. First, the incremental update trigger conditions are set. When the charging state of the vehicle at the current moment and the previous moment changes by more than or equal to 1%, the health state changes by more than or equal to 1%, and the time interval is more than or equal to 1 minute, the vehicle characteristic map is incrementally updated, and only the changed data is updated. For unchanged data, no more upload and update is performed, thereby effectively reducing the amount of updated data for electric vehicles. Secondly, in order to improve the quality of vehicle characteristic map data and reduce the difficulty of local data processing, this embodiment adopts an edge computing mode, and only processes abnormal jump data on the vehicle terminal side, filters instantaneous SOC and SOH jumps, thereby improving the quality and efficiency of vehicle characteristic map data updates.

[0037] In this embodiment, the following logic is used to set the incremental update trigger condition:

[0038]

[0039]

[0040] Where, Indicates the vehicle charging status data at the previous moment (time k), Indicates the vehicle charging status data at the current time (time k+1), Represents the vehicle health status data at the previous moment (time k), Represents the vehicle health status data at the current time (time k+1), Indicates the time interval between the previous moment and the current moment.

[0041] In this embodiment, the vehicle characteristic map is used to obtain the historical state of charge (SOC) and state of health (SOH) data of individual vehicles. A bidirectional long-short-term memory (BiLSTM) network is used to predict the charging and discharging behavior of electric vehicles. This is then coupled with the vehicle's user behavior profile to achieve EV charging and discharging behavior perception. First, the LSTM outputs a predicted vehicle charge and discharge power based on the individual vehicle's historical SOC and SOH data. This analyzes the vehicle's state of charge (SOC). When the SOC data increases, the vehicle is considered to be charging, and vice versa. Second, based on the predicted vehicle charge and discharge power and the determined charging and discharging behavior, the user's location and time preference data in the vehicle characteristic map is coupled to obtain the real-time location and time preference results for the electric vehicle's charging and discharging.

[0042] In this embodiment, the long short-term memory network prediction model is trained by synergistically using incremental training and full training. Specifically, the prediction model is first incrementally trained, and the historical SOC and SOH data of the previous 7 days are used to dynamically iterate and fine-tune the prediction model, continuously updating and optimizing the prediction results; the average prediction error of the prediction model is calculated, and when the average prediction error calculated for 3 consecutive days exceeds a preset threshold, the prediction model is fully trained; wherein, the calculation of the average prediction error of the prediction model is represented by the following logic:

[0043] Where, represents the average prediction error, Indicates the time involved in calculating the average forecast error, in units of 15 minutes. Divided into multiple moments, Indicates the The prediction error at the moment, that is, to calculate the A 15-minute forecast error.

[0044] Furthermore, through dynamic perception of electric vehicle charging and discharging behavior, regional charging hotspot identification is established in units of 1 minute, providing a reference for electric vehicle charging and discharging behavior for power grid operation managers and electric vehicle owners.

[0045] In this embodiment, through the above-mentioned dynamic perception results of electric vehicle charging and discharging behavior, the charging location preference and time preference of each electric vehicle are aggregated in units of charging piles, and the predicted number of electric vehicle charging at each charging pile can be obtained. Based on the predicted data, regional charging hotspot identification is formed in units of charging piles.

[0046] S22, build charging pile load perception, specifically, take the charging pile itself as the unit, obtain the charging load historical data of the charging pile, after data preprocessing, use the bidirectional long short-term memory network (BiLSTM) to predict the charging and discharging power of the charging pile.

[0047] In this embodiment, the charging pile charging load history data can be obtained through the power flow information in the power grid topology map.

[0048] S23, at the same time, couples the dynamic perception of electric vehicle charging and discharging behavior with the charging pile load perception in a bidirectional manner, considers the multi-temporal and spatial load characteristics, and improves the accuracy of the dynamic perception of electric vehicle charging behavior.

[0049] In this embodiment, the charging and discharging power of each electric vehicle is aggregated to obtain the aggregated charging and discharging power of the charging pile at this moment:

[0050] Where, express The charging power of the charging piles aggregated at the moment, represents the number of aggregated electric vehicles, express Time The charging power of an electric vehicle can be obtained based on the rated power in the vehicle characteristic map. .

[0051] From the user's perspective, the aggregated charging pile charging power can reflect vehicle preference information, such as Compared to other charging piles, electric vehicles prefer to go to the above charging piles for charging. From the perspective of the power grid, the charging and discharging power prediction data of the charging piles reflects the characteristic data of the power grid. This embodiment couples the two information to improve the accuracy of dynamic perception of electric vehicles' charging behavior based on multi-temporal and spatial load characteristics. Specifically, this embodiment calculates bidirectional coupling based on the user preference coefficient and the historical characteristic coefficient, using the following logic:

[0052] Where, Indicates the bidirectional coupling The charging power of the charging pile at this moment, represents the user preference coefficient, represents the historical characteristic coefficient, express The bidirectional long short-term memory network is used to predict the charging and discharging power of the charging pile at the moment. Among them, the user preference coefficient and historical characteristic coefficient First, solve the problem based on the data of the first two historical moments, and then follow the moment, 、 as well as The changes are updated dynamically and expressed using the following logic:

[0053] Where, 、 Represents the bidirectional coupling time, The charging power of the charging pile at the moment, 、 Respectively time, Bidirectional long short-term memory network is used to predict the charging and discharging power of charging piles at all times. 、 Respectively time, The charging power of the charging pile aggregated at all times.

[0054] Currently, grid dispatch models based on electric vehicles often only consider the real-time operating status of the power grid and road network, without considering the spatiotemporal distribution of charging demand due to the mobility characteristics of electric vehicles, or the load transfer of charging stations due to the spread of road network congestion. This embodiment builds a dynamic perception layer based on the vehicle-road-network coupling three-domain knowledge graph. First, it constructs dynamic perception of electric vehicle charging and discharging behavior, predicts the charging and discharging power of a single electric vehicle, and judges the charging and discharging behavior of electric vehicles, coupling with the vehicle-body user behavior profile. Secondly, based on multiple vehicle-body user behavior profiles, it aggregates the charging power of a single charging pile and performs a bidirectional coupling with the predicted charging power output by the charging pile according to the prediction model. This establishes a bidirectional coupling model for electric vehicle charging and discharging power prediction and charging pile charging and discharging power prediction based on the three-domain knowledge graph, realizing dynamic perception of the multi-spatiotemporal load characteristics of electric vehicle charging, and effectively improving the accuracy of dynamic perception of electric vehicle charging behavior.

[0055] S3 builds a multi-objective optimization layer based on the vehicle-road-network coupling three-domain knowledge graph, constructs a multi-objective optimization function and uses a hybrid adaptive particle swarm optimization to optimize the multi-objective optimization function, and outputs the final optimization solution.

[0056] like Figure 4 As shown, this embodiment is based on the above-mentioned vehicle-road-network coupled three-domain knowledge graph, and then constructs a multi-objective optimization layer through the above-mentioned dynamic perception layer based on the vehicle-road-network coupled three-domain knowledge graph. Specifically, it includes the following steps: S31. On the grid side, design the objective function for the new energy consumption rate. Add line congestion peak-shaving constraints, power balance constraints, charging pile capacity constraints, charging pile parking space constraints, SOC constraints, and SOH constraints. Use the following logic to express the objective function for the new energy consumption rate:

[0057] in, Indicates the target of new energy consumption rate, Represents the amount of new energy consumed, specifically the total amount of new energy that can be consumed on the grid side, P EV It indicates the predicted power generation of renewable energy, specifically the power generation of renewable energy predicted by the grid side.

[0058] In this embodiment, to comprehensively consider the coupling between the vehicle, road, and grid, constraints are designed from the perspective of the distribution network. The grid-side constraints include: The line congestion peak-shaving constraint is expressed using the following logic:

[0059] Where, is the line active power, is the minimum load power of the line, is the maximum load power of the line. The above data comes from the power flow information in the power grid topology map.

[0060] The power balance constraint is expressed using the following logic:

[0061] Where, Indicates the bidirectional coupling The charging power of the charging pile at the moment is calculated according to S2 and stored in the power grid topology map and updated dynamically; Indicates other loads, Indicates energy storage power generation. The data comes from the energy storage information in the grid topology map.

[0062] The charging pile capacity constraint is expressed using the following logic:

[0063] Where, 、 They represent the minimum power and maximum power of the charging pile respectively. The data comes from the power margin in the power grid topology map.

[0064] The parking space constraints for charging piles are expressed using the following logic:

[0065] Where, express The current charging pile has occupied the parking space, 、 Respectively The minimum and maximum parking spaces for charging piles at the current moment are calculated from the available parking spaces in the power grid topology map.

[0066] The SOC constraint is expressed using the following logic:

[0067] Where, express The charging status of the electric vehicle at all times, 、 Respectively Minimum time to charge an electric vehicle ,maximum .

[0068] The SOH constraint is expressed using the following logic:

[0069] Where, express The health status of the electric vehicle being charged at all times, 、 Respectively Minimum time to charge an electric vehicle ,maximum .

[0070] S32, on the user side, by adding a charging cost-time game model, a dynamic game equilibrium between individual charging demand and group electricity consumption behavior is achieved. Specifically, the user charging cost objective function is designed, and the maximum and minimum constraints of electricity prices and user satisfaction constraints are added. The user charging cost objective function is expressed using the following logic:

[0071] Where, represents the charging cost target, express The time-of-use electricity price at the moment is derived from the real-time electricity price in the power grid topology map. express The instantaneous charging power of electric vehicles at this moment, Indicates the time when the electric vehicle starts charging. Indicates the time when the electric vehicle ends charging.

[0072] In this embodiment, the user-side constraints include: The maximum and minimum constraints of electricity prices are expressed using the following logic:

[0073] Where, express The charging electricity price of electric vehicles at this moment, 、 Respectively The minimum and maximum charging prices for electric vehicles at the moment are derived from the real-time electricity prices in the power grid topology map.

[0074] User satisfaction constraints are expressed using the following logic:

[0075] Where, express User satisfaction at all times, 、 Respectively The minimum user satisfaction and maximum user satisfaction at a given moment are derived from the user behavior portrait in the vehicle feature map.

[0076] S33, on the road network side, designs the optimal objective function for electric vehicle routing, specifically:

[0077] in, is the optimal path target for electric vehicles, This is the charging path time for electric vehicles. The data comes from the path time in the road network resource map.

[0078] S34, based on the above objective function, construct a multi-objective optimization function, using the following logic representation:

[0079] Where, Indicates the total charging cost of the electric vehicle during the charging time. Indicates the total charging time of the electric vehicle. represents the amount of new energy consumption from the beginning to the end of the multi-objective optimization function, Represents the predicted power generation of renewable energy from the beginning to the end of the multi-objective optimization function, represents the entire optimization cycle of the multi-objective optimization function, represents the new energy consumption rate, , express 、 and On the basis of satisfying the constraints of S31~S32 respectively, solve 、 and All of them reach the smallest global optimal solution set.

[0080] S35, adopts hybrid adaptive particle swarm optimization for multi-objective optimization function.

[0081] This embodiment uses hybrid adaptive particle swarm optimization (HAPSO) to solve the above optimization algorithm, specifically: S351, first, generate a particle swarm, each particle represents a charging power allocation scheme; the charging power allocation scheme is specifically the charging pile and charging time scheme selected by the electric vehicle.

[0082] Secondly, dynamic weight adjustment is performed; constraint processing is performed according to the constraint conditions of S31~S32. The constraint processing specifically adopts a repair strategy to handle out-of-bounds particles, such as adjusting the charging power to the boundary value to ensure that the charging power does not exceed the maximum or minimum charging power, as well as constraining the charging pile parking space, etc.

[0083] S352, calculating the multi-objective fitness of each particle and screening the non-dominated optimal solution set based on the Pareto front. Specifically, using the multi-objective optimization function to calculate the road network side, power grid side, and user side objective fitness of each particle and screening the optimal solution set; The Pareto front is one of the methods used to describe the set of optimal solutions in multi-objective optimization. Since multi-objective optimization problems usually have multiple objectives, which may conflict with each other, the Pareto optimal solution set represents the solution that can be further optimized without sacrificing other objectives.

[0084] In this embodiment, the multi-objective fitness of each particle is calculated. Specifically, a multi-objective optimization function is used to evaluate the target fitness of each particle solution on the road network side, the power grid side, and the user side. By comparing the multi-objective fitness between different particles, the optimal solution set is screened out.

[0085] S353: Modify user preferences based on user satisfaction. Specifically, perform satisfaction-weighted sorting on the Pareto solution set and output the final optimization solution.

[0086] In this embodiment, considering that users may have different preferences for certain goals, such as charging time, real-time electricity prices, or location preferences, this embodiment performs satisfaction-weighted adjustment on the fitness of each solution in the Pareto solution set, so that the priority of different goals is adjusted according to user preferences to generate an optimal charging plan that better meets user preferences.

[0087] In step S3, a multi-objective optimization layer is constructed based on the vehicle-road-network coupling three-domain knowledge graph. On the power grid side, based on the topological analysis capability of the knowledge graph, a new energy consumption rate objective function considering the dynamic thermal stability limit of the line is constructed, and constraints are designed from the distribution network perspective to achieve spatiotemporal matching between peak-shaving resources and fluctuating power sources; on the user side, a charging cost-time game model is designed, and a user charging cost objective function is established to achieve a dynamic game equilibrium between individual charging demand and group electricity consumption behavior; on the road network side, a joint optimization of charging induction and traffic diversion is developed, coupling the capacity constraints of charging stations with traffic flow, and using the path reasoning capability of the knowledge graph to generate the optimal charging navigation strategy. Finally, a multi-objective optimization function from three perspectives is constructed, and HAPSO is used to solve the multi-objective optimization function. The output optimal solution set is re-weighted and sorted according to user satisfaction to generate an optimal charging plan that better meets user preferences.

[0088] S4 builds a strategy generation layer based on the vehicle-road-network coupling knowledge graph, and pushes the optimization plan to the user side, power grid side, and road network side.

[0089] like Figure 5 As shown, this embodiment establishes a strategy generation layer for the vehicle-road-network coupling three-domain knowledge graph based on the above steps S1 to S3, specifically including the following steps: S41: On the user side, based on the above final optimization plan, a minute-level rolling real-time optimization mechanism is established to push the user-side path selection to the electric vehicle terminal in real time. Specifically: Based on the location data in the vehicle feature knowledge graph, the real-time location information of the vehicle body is updated in minutes. Due to the different locations of different vehicles, the paths to the surrounding charging piles are also different. Based on the real-time location information of the vehicle body, the final optimization solution output by S3 (specifically, the path from the current vehicle body position to the charging pile) is pushed to the user side in real time to ensure the effectiveness of the optimization results.

[0090] S42, on the power grid side, a real-time response curve of electric vehicle charging electricity prices is formed according to the real-time electricity prices in the power grid topology map, and a line congestion and a new energy consumption hotspot map are obtained according to the flow information and new energy power in the power grid topology map for reference by power grid operation managers.

[0091] S43, on the road network side, an electric vehicle path flow diagram is formed according to the traffic flow in the road network knowledge graph, and a charging pile parking space heat map is obtained based on the real-time status of the charging piles in the power grid topology graph for reference.

[0092] In step S4, on the user side, this embodiment accurately pushes the user-side path selection to the electric vehicle terminal in real time based on the optimization results output by the optimization layer. A minute-by-minute rolling real-time optimization mechanism is established to push the real-time optimization results to the user side to ensure real-time optimization results. The strategy generation layer, based on the vehicle-road-network coupled knowledge graph, enables real-time scheduling optimization of electric vehicle charging loads and grid operations, thereby improving the quantitative modeling of electric vehicle users' bounded rational behavior, the lack of coordination between new energy consumption and network congestion control, and weak real-time dynamic response capabilities.

[0093] The present invention also provides a real-time dispatch optimization system for electric vehicle charging load power grid operation, comprising: The data storage module is used to build a vehicle-road-network coupled three-domain knowledge graph fusion architecture, including vehicle feature graphs, road network resource graphs, and power grid topology graphs. The knowledge graph data is stored in a graph database and updated in real time.

[0094] In this embodiment, the vehicle-road-network coupled three-domain knowledge graph fusion architecture includes three knowledge graph subgraphs: vehicle feature graph (VKG), road network resource graph (RKG), and grid topology graph (GKG). Each knowledge graph subgraph is modeled in the form of a triple: The vehicle feature map utilizes dynamic ontology modeling technology, with the vehicle entity being the vehicle itself. It contains three categories: basic vehicle information, user behavior profiles, and real-time data from the vehicle's battery management system (BMS). Basic vehicle information includes four attributes: brand, model, fuel type, and rated power. User behavior profiles include location preference, time preference, and bounded rationality threshold. Real-time BMS data includes state of charge (SOC), state of health (SOH), charge rate, and location data, and is updated in real time based on the vehicle's current status.

[0095] The grid topology map uses electric vehicle charging stations as entities and includes three categories: basic grid information, grid information, and real-time status. Basic grid information includes six attributes: user number, capacity, number of parking spaces, charging type, address, and energy usage category. Grid information includes two attributes: grid-connected lines and grid-connected substations. Grid-connected lines include four sub-attributes: power flow information, new energy power, thermal stability limit, and energy storage information. Real-time status includes three attributes: available parking spaces, power margin, and real-time electricity price.

[0096] The road network resource map includes two categories: traffic flow and dynamic road resistance. Traffic flow includes two attributes: path time and congestion degree, and dynamic road resistance includes two attributes: weather conditions and parking space prediction.

[0097] In this embodiment, the graph database uses a Neo4j graph database to store vehicle-road-network coupled knowledge graph data, enabling the fusion of the three-domain knowledge graphs. Vehicle feature graphs are dynamically updated in real time using vehicle terminal data, grid SCADA and distribution network OMS data are used to establish grid topology graphs, and the AutoNavi API is used to automatically construct road network resource graphs. By integrating global data from vehicle terminals, roadside units, and the grid SCADA system, a three-domain knowledge graph with spatiotemporal correlation is constructed, capturing electric vehicle movement trajectories, traffic network status, and grid operating conditions in real time.

[0098] The dynamic perception module builds a dynamic perception layer based on the vehicle-road-network coupling knowledge graph and includes the following units: The dynamic perception unit is used to build a dynamic perception model of user charging needs. It updates the vehicle feature map in real time through the on-board terminal, obtains the historical SOC and SOH data of electric vehicles based on the vehicle body, and uses the bidirectional long short-term memory network (BiLSTM) to predict the charging and discharging power of electric vehicles and judge the charging and discharging behavior of electric vehicles based on the historical data of individual vehicles, and couples the user behavior portrait of the vehicle body.

[0099] In this embodiment, the real-time dynamic update of the vehicle characteristic map by the vehicle terminal is specifically as follows: the vehicle terminal reads vehicle data at a high frequency of 1Hz and uses the CAN bus through the OBD-II interface, and dynamically updates the real-time data of the electric vehicle by using local caching and compressed uploading.

[0100] Specifically, due to the huge amount of electric vehicle data, this embodiment constructs a dynamic perception model of user charging demand based on an incrementally updated edge computing model, which is used to dynamically update the real-time data of electric vehicles. First, the incremental update trigger conditions are set. When the charging state of the vehicle at the current moment and the previous moment changes by more than or equal to 1%, the health state changes by more than or equal to 1%, and the time interval is more than or equal to 1 minute, the vehicle characteristic map is incrementally updated, and only the changed data is updated. For unchanged data, no more upload and update is performed, thereby effectively reducing the amount of updated data for electric vehicles. Secondly, in order to improve the quality of vehicle characteristic map data and reduce the difficulty of local data processing, this embodiment adopts an edge computing mode, and only processes abnormal jump data on the vehicle terminal side, filters instantaneous SOC and SOH jumps, thereby improving the quality and efficiency of vehicle characteristic map data updates.

[0101] In this embodiment, the following logic is used to set the incremental update trigger condition:

[0102]

[0103]

[0104] Where, Indicates the vehicle charging status data at the previous moment (time k), Indicates the vehicle charging status data at the current time (time k+1), Represents the vehicle health status data at the previous moment (time k), Represents the vehicle health status data at the current time (time k+1), Indicates the time interval between the previous moment and the current moment.

[0105] In this embodiment, the vehicle characteristic map is used to obtain the historical state of charge (SOC) and state of health (SOH) data of individual vehicles. A bidirectional long-short-term memory (BiLSTM) network is used to predict the charging and discharging behavior of electric vehicles. This is then coupled with the vehicle's user behavior profile to achieve EV charging and discharging behavior perception. First, the LSTM outputs a predicted vehicle charge and discharge power based on the individual vehicle's historical SOC and SOH data. This analyzes the vehicle's state of charge (SOC). When the SOC data increases, the vehicle is considered to be charging, and vice versa. Second, based on the predicted vehicle charge and discharge power and the determined charging and discharging behavior, the user's location and time preference data in the vehicle characteristic map is coupled to obtain the real-time location and time preference results for the electric vehicle's charging and discharging.

[0106] In this embodiment, the long short-term memory network prediction model is trained by synergistically using incremental training and full training. Specifically, the prediction model is first incrementally trained, and the historical SOC and SOH data of the previous 7 days are used to dynamically iterate and fine-tune the prediction model, continuously updating and optimizing the prediction results; the average prediction error of the prediction model is calculated, and when the average prediction error calculated for 3 consecutive days exceeds a preset threshold, the prediction model is fully trained; wherein, the calculation of the average prediction error of the prediction model is represented by the following logic:

[0107] Where, represents the average prediction error, Indicates the time involved in calculating the average forecast error, in units of 15 minutes. Divided into multiple moments, Indicates the The prediction error at the moment, that is, to calculate the A 15-minute forecast error.

[0108] The load sensing unit is used to obtain the historical charging load data of the charging pile based on the charging pile itself. After data preprocessing, the bidirectional long short-term memory network (BiLSTM) is used to predict the charging and discharging power of the charging pile.

[0109] In this embodiment, the charging pile charging load history data can be obtained through the power flow information in the power grid topology map.

[0110] The bidirectional coupling unit is used to simultaneously couple the dynamic perception of electric vehicle charging and discharging behavior with the charging pile load perception, taking into account the multi-temporal and spatial load characteristics to improve the accuracy of the dynamic perception of electric vehicle charging behavior.

[0111] In this embodiment, the charging and discharging power of each electric vehicle is aggregated to obtain the aggregated charging and discharging power of the charging pile at this moment:

[0112] Where, express The charging power of the charging piles aggregated at the moment, represents the number of aggregated electric vehicles, express Time The charging power of an electric vehicle can be obtained based on the rated power in the vehicle characteristic map. .

[0113] From the user's perspective, the aggregated charging pile charging power can reflect vehicle preference information, such as Compared to other charging piles, electric vehicles prefer to go to the above charging piles for charging. From the perspective of the power grid, the charging and discharging power prediction data of the charging piles reflects the characteristic data of the power grid. This embodiment couples the two information to improve the accuracy of dynamic perception of electric vehicles' charging behavior based on multi-temporal and spatial load characteristics. Specifically, this embodiment calculates bidirectional coupling based on the user preference coefficient and the historical characteristic coefficient, using the following logic:

[0114] Where, Indicates the bidirectional coupling The charging power of the charging pile at this moment, represents the user preference coefficient, represents the historical characteristic coefficient, express The bidirectional long short-term memory network is used to predict the charging and discharging power of the charging pile at the moment. Among them, the user preference coefficient and historical characteristic coefficient First, solve the problem based on the data of the first two historical moments, and then follow the moment, 、 as well as The changes are updated dynamically and expressed using the following logic:

[0115] Where, 、 Represents the bidirectional coupling time, The charging power of the charging pile at the moment, 、 Respectively time, Bidirectional long short-term memory network is used to predict the charging and discharging power of charging piles at all times. 、 Respectively time, The charging power of the charging pile aggregated at all times.

[0116] The multi-objective optimization module builds a multi-objective optimization layer based on the vehicle-road-network coupling knowledge graph, constructs a multi-objective optimization function, and uses a hybrid adaptive particle swarm optimization to optimize the multi-objective optimization function to output the final optimization solution. Specifically, the multi-objective optimization module includes the following units: The grid-side unit is used to design the objective function of the new energy consumption rate. It adds line congestion peak-shaving constraints, power balance constraints, charging pile capacity constraints, charging pile parking space constraints, SOC constraints, and SOH constraints. The objective function of the new energy consumption rate is expressed using the following logic:

[0117] in, Indicates the target of new energy consumption rate, Represents the amount of new energy consumed, specifically the total amount of new energy that can be consumed on the grid side, P EV It indicates the predicted power generation of renewable energy, specifically the power generation of renewable energy predicted by the grid side.

[0118] In this embodiment, to comprehensively consider the coupling between the vehicle, road, and grid, constraints are designed from the perspective of the distribution network. The grid-side constraints include: The line congestion peak-shaving constraint is expressed using the following logic:

[0119] Where, is the line active power, is the minimum load power of the line, is the maximum load power of the line. The above data comes from the power flow information in the power grid topology map.

[0120] The power balance constraint is expressed using the following logic:

[0121] Where, Indicates the bidirectional coupling The charging power of the charging pile at the moment is calculated by the dynamic perception module and stored in the power grid topology map and dynamically updated; Indicates other loads, Indicates energy storage power generation. The data comes from the energy storage information in the grid topology map.

[0122] The charging pile capacity constraint is expressed using the following logic:

[0123] Where, 、 They represent the minimum power and maximum power of the charging pile respectively. The data comes from the power margin in the power grid topology map.

[0124] The parking space constraints for charging piles are expressed using the following logic:

[0125] Where, express The current charging pile has occupied the parking space, 、 Respectively The minimum and maximum parking spaces for charging piles at the current moment are calculated from the available parking spaces in the power grid topology map.

[0126] The SOC constraint is expressed using the following logic:

[0127] Where, express The charging status of the electric vehicle at all times, 、 Respectively Minimum time to charge an electric vehicle ,maximum .

[0128] The SOH constraint is expressed using the following logic:

[0129] Where, express The health status of the electric vehicle being charged at all times, 、 Respectively Minimum time to charge an electric vehicle ,maximum .

[0130] The user-side unit is used to achieve a dynamic game equilibrium between individual charging demand and group electricity consumption behavior by adding a charging cost-time game model. Specifically, the user charging cost objective function is designed, and the maximum and minimum constraints on electricity prices and user satisfaction constraints are added. The user charging cost objective function is expressed using the following logic:

[0131] Where, represents the charging cost target, express The time-of-use electricity price at the moment is derived from the real-time electricity price in the power grid topology map. express The instantaneous charging power of electric vehicles at this moment, Indicates the time when the electric vehicle starts charging. Indicates the time when the electric vehicle ends charging.

[0132] In this embodiment, the user-side constraints include: The maximum and minimum constraints of electricity prices are expressed using the following logic:

[0133] Where, express The charging electricity price of electric vehicles at this moment, 、 Respectively The minimum and maximum charging prices for electric vehicles at the moment are derived from the real-time electricity prices in the power grid topology map.

[0134] User satisfaction constraints are expressed using the following logic:

[0135] Where, express User satisfaction at all times, 、 Respectively The minimum user satisfaction and maximum user satisfaction at a given moment are derived from the user behavior portrait in the vehicle feature map.

[0136] The road network side unit is used to design the optimal objective function of the electric vehicle path, specifically:

[0137] in, is the optimal path target for electric vehicles, This is the charging path time for electric vehicles. The data comes from the path time in the road network resource map.

[0138] The multi-objective optimization unit is used to construct a multi-objective optimization function based on the above objective function, using the following logic:

[0139] Where, Indicates the total charging cost of the electric vehicle during the charging time. Indicates the total charging time of the electric vehicle. represents the amount of new energy consumption from the beginning to the end of the multi-objective optimization function, Represents the predicted power generation of renewable energy from the beginning to the end of the multi-objective optimization function, represents the entire optimization cycle of the multi-objective optimization function, represents the new energy consumption rate, , express 、 and On the basis of satisfying the constraints of S31~S32 respectively, solve 、 and All of them reach the smallest global optimal solution set.

[0140] The particle swarm optimization unit is used to optimize the multi-objective optimization function using hybrid adaptive particle swarm optimization.

[0141] This embodiment uses hybrid adaptive particle swarm optimization (HAPSO) to solve the above optimization algorithm, specifically including: First, a particle swarm is generated, where each particle represents a charging power allocation scheme; the charging power allocation scheme specifically refers to the charging pile and charging time scheme selected by the electric vehicle.

[0142] Secondly, dynamic weight adjustment is performed; constraint processing is performed based on the constraint conditions from the grid-side unit to the user-side unit. The constraint processing specifically adopts a repair strategy to handle out-of-bounds particles, such as adjusting the charging power to the boundary value to ensure that the charging power does not exceed the maximum or minimum charging power, as well as constraints on charging pile parking spaces, etc.

[0143] Calculate the multi-objective fitness of each particle and screen the non-dominated optimal solution set based on the Pareto frontier. Specifically, use the multi-objective optimization function to calculate the road network side, power grid side, and user side target fitness of each particle and screen out the optimal solution set. The Pareto front is one of the methods used to describe the set of optimal solutions in multi-objective optimization. Since multi-objective optimization problems usually have multiple objectives, which may conflict with each other, the Pareto optimal solution set represents the solution that can be further optimized without sacrificing other objectives.

[0144] In this embodiment, the multi-objective fitness of each particle is calculated. Specifically, a multi-objective optimization function is used to evaluate the target fitness of each particle solution on the road network side, the power grid side, and the user side. By comparing the multi-objective fitness between different particles, the optimal solution set is screened out.

[0145] User preferences are modified based on user satisfaction, specifically by performing satisfaction-weighted sorting on the Pareto solution set and outputting the final optimization solution.

[0146] In this embodiment, considering that users may have different preferences for certain goals, such as charging time, real-time electricity prices, or location preferences, this embodiment performs satisfaction-weighted adjustment on the fitness of each solution in the Pareto solution set, so that the priority of different goals is adjusted according to user preferences to generate an optimal charging plan that better meets user preferences.

[0147] The strategy generation module builds a strategy generation layer based on the vehicle-road-network coupling knowledge graph and pushes the optimization plan to the user side, the power grid side, and the road network side. Specifically, the strategy generation module includes the following units: The push unit is used to establish a minute-level rolling real-time optimization mechanism on the user side according to the above final optimization plan, and push the user-side path selection to the electric vehicle terminal in real time. Specifically: Based on the location data in the vehicle feature knowledge graph, the real-time location information of the vehicle body is updated in minutes. Due to the different positions of different vehicles, the paths to the surrounding charging piles are also different. Based on the real-time location information of the vehicle body, the final optimization solution output by the multi-objective optimization module (specifically, the path from the current vehicle body position to the charging pile) is pushed to the user side in real time to ensure the effectiveness of the optimization results.

[0148] The power grid heat map unit is used to form a real-time response curve of electric vehicle charging electricity prices on the power grid side according to the real-time electricity prices in the power grid topology map, and to obtain line congestion and new energy consumption heat maps according to the flow information and new energy power in the power grid topology map for reference by power grid operation managers.

[0149] The road network heat map unit is used to form an electric vehicle path flow map based on the traffic flow in the road network knowledge map on the road network side, and obtain a charging pile parking space heat map for reference based on the real-time status of the charging piles in the power grid topology map.

[0150] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A real-time scheduling optimization method for electric vehicle charging load power grid operation, characterized in that: The following steps are involved: S1: Build a vehicle-road-network coupled knowledge graph fusion architecture, including vehicle feature graphs, road network resource graphs, and power grid topology graphs. Store knowledge graph data in a graph database and update it in real time. S2 builds a dynamic perception layer based on the vehicle-road-network coupled knowledge graph, including: Build dynamic perception of electric vehicle charging and discharging behavior to predict the spatiotemporal distribution of electric vehicle charging demand; Build charging pile load sensing to predict charging and discharging power of charging piles; Bidirectionally couple the dynamic perception of electric vehicle charging and discharging behavior with the perception of charging pile load to improve the accuracy of dynamic perception of electric vehicle charging behavior; S3: Build a multi-objective optimization layer based on the vehicle-road-network coupling knowledge graph, construct a multi-objective optimization function, and use hybrid adaptive particle swarm optimization to optimize the multi-objective optimization function and output the final optimization solution; S4 builds a strategy generation layer based on the vehicle-road-network coupling knowledge graph, and pushes the optimization plan to the user side, power grid side, and road network side.

2. The method for real-time scheduling optimization of electric vehicle charging load power grid operation according to claim 1 is characterized in that: The entity of the vehicle feature map described in S1 is the vehicle itself. The categories of the vehicle feature map include basic vehicle information, user behavior profile, and on-board BMS real-time data. The attributes of the basic vehicle information include brand, model, fuel type, and rated power. The attributes of the user behavior profile include location preference, time preference, and bounded rationality threshold. The attributes of the onboard BMS real-time data include charging status, health status, charging rate, and location data; The entities of the power grid topology map are electric vehicle charging piles. The categories of the power grid topology map include basic power grid information, power grid information, and real-time status. The attributes of the basic power grid information include user number, capacity, number of parking spaces, charging type, address, and energy consumption category. The attributes of the power grid information include grid-connected lines and grid-connected substations. The attributes of the real-time status include available parking spaces, power margin, and real-time electricity prices. The categories of the road network resource map include traffic flow and dynamic road resistance; the attributes of traffic flow include path time and congestion level, and the attributes of dynamic road resistance include weather conditions and parking space prediction.

3. The method for real-time scheduling optimization of electric vehicle charging load power grid operation according to claim 2 is characterized in that: The S2 comprises the following steps: S21: Build dynamic perception of electric vehicle charging and discharging behavior. Specifically, build a dynamic perception model of user charging demand, dynamically update the vehicle feature map in real time through the vehicle terminal, obtain the historical SOC and SOH data of electric vehicles based on the vehicle body, use the bidirectional long short-term memory network to predict the charging and discharging power of electric vehicles and judge the charging and discharging behavior of electric vehicles based on the historical data of individual vehicles, and couple it with the user behavior profile of the vehicle body; S22, building a charging pile load sensor, specifically obtaining the charging pile charging load historical data based on the charging pile itself, and using a bidirectional long short-term memory network to predict the charging and discharging power of the charging pile after data preprocessing; S23, at the same time, couples the dynamic perception of electric vehicle charging and discharging behavior with the charging pile load perception in a bidirectional manner, considers the multi-temporal and spatial load characteristics, and improves the accuracy of the dynamic perception of electric vehicle charging behavior.

4. The method for real-time scheduling optimization of electric vehicle charging load power grid operation according to claim 3 is characterized in that: The user charging demand dynamic perception model in S21 is specifically an edge computing model based on incremental updates; First, set the incremental update trigger conditions. When the vehicle's state of charge changes by 1% or more, its health changes by 1% or more, and the time interval is greater than or equal to 1 minute, the vehicle's characteristic map is incrementally updated. Only the changed data is updated, and unchanged data is not uploaded for update. Secondly, the edge computing model is adopted to process abnormal jump data only on the vehicle terminal side and filter out instantaneous SOC and SOH jumps.

5. The method for real-time scheduling optimization of electric vehicle charging load power grid operation according to claim 4 is characterized in that: S21 describes the use of a bidirectional long short-term memory network to predict the charging and discharging power of electric vehicles and judge the charging and discharging behavior of electric vehicles, coupled with the vehicle body user behavior profile as follows: First, the long short-term memory network outputs the predicted vehicle charge and discharge power based on the historical SOC and SOH data of individual vehicles. It analyzes the vehicle's SOC and considers that the vehicle is charging when the SOC data increases, and discharging when the SOC data decreases. Secondly, based on the predicted vehicle charging and discharging power and the judged charging and discharging behavior, the user's location preference and time preference data in the vehicle characteristic map are coupled to obtain the real-time location and time preference results of electric vehicle charging and discharging; In the S21, the long short-term memory network prediction model is trained by the collaborative method of incremental training and full training, specifically: the prediction model is first incrementally trained, and the historical SOC and SOH data of the previous 7 days are used to dynamically iterate and fine-tune the prediction model, and the prediction results are continuously updated and optimized; the average prediction error of the prediction model is calculated, and when the average prediction error calculated for 3 consecutive days exceeds the preset threshold, the prediction model is fully trained.

6. The method for real-time scheduling optimization of electric vehicle charging load power grid operation according to claim 5, characterized in that: The bidirectional coupling described in S23 is specifically: The charging and discharging power of each electric vehicle is aggregated to obtain the aggregated charging and discharging power of the charging pile at this moment: Where, express The charging power of the charging piles aggregated at the moment, represents the number of aggregated electric vehicles, express Time The charging power of an electric vehicle, ; Based on the user preference coefficient and the historical feature coefficient, a two-way coupling is calculated and expressed using the following logic: Where, Indicates the bidirectional coupling The charging power of the charging pile at this moment, represents the user preference coefficient, represents the historical characteristic coefficient, express The bidirectional long short-term memory network is used to predict the charging and discharging power of the charging pile at the moment; among them, the user preference coefficient and historical characteristic coefficient First, solve the problem based on the data of the previous two historical moments, and then update it dynamically, using the following logic: Where, 、 Represents the bidirectional coupling time, The charging power of the charging pile at the moment, 、 Respectively time, Bidirectional long short-term memory network is used to predict the charging and discharging power of charging piles at all times. 、 Respectively time, The charging power of the charging pile aggregated at all times.

7. The method for real-time scheduling optimization of electric vehicle charging load power grid operation according to claim 1, characterized in that: The S3 includes the following steps: S31, on the grid side, design the objective function of the new energy consumption rate, and add line congestion peak-shaving constraints, power balance constraints, charging pile capacity constraints, charging pile parking space constraints, SOC constraints, and SOH constraints; S32, on the user side, design the user charging cost objective function, add the maximum and minimum constraints of electricity price and user satisfaction constraints; S33, on the road network side, design the optimal objective function for electric vehicle routing; S34, based on the above objective function, construct a multi-objective optimization function, using the following logic representation: Where, represents the charging cost target, express Time-of-use electricity prices at different times, express The instantaneous charging power of electric vehicles at this moment, Indicates the total charging cost of the electric vehicle during the charging time. Indicates the time when the electric vehicle starts charging. Indicates the time when the electric vehicle ends charging. Indicates the total charging time of the electric vehicle; Indicates the target of new energy consumption rate, represents the amount of new energy consumed, P EV Represents the predicted power generation of new energy, represents the amount of new energy consumption from the beginning to the end of the multi-objective optimization function, Represents the predicted power generation of renewable energy from the beginning to the end of the multi-objective optimization function, represents the entire optimization cycle of the multi-objective optimization function, represents the new energy consumption rate, , is the optimal path target for electric vehicles, The charging path time for electric vehicles, express 、 and On the basis of satisfying the constraints of S31~S32 respectively, solve 、 and All achieve the smallest global optimal solution set; S35, adopts hybrid adaptive particle swarm optimization for multi-objective optimization function.

8. The method for real-time scheduling optimization of electric vehicle charging load power grid operation according to claim 7, characterized in that: The multi-objective optimization function using hybrid adaptive particle swarm optimization described in S35 is specifically: S351, first, generate a particle swarm, each particle representing a charging power allocation scheme; Secondly, dynamic weight adjustment is performed; constraint processing is performed according to the constraint conditions of S31~S32, and the constraint processing is specifically to use a repair strategy to handle out-of-bounds particles; S352, calculate the multi-objective fitness of each particle and screen the non-dominated optimal solution set based on the Pareto frontier; S353: Modify user preferences based on user satisfaction. Specifically, perform satisfaction-weighted sorting on the Pareto solution set and output the final optimization solution.

9. The method for real-time scheduling optimization of electric vehicle charging load power grid operation according to claim 1, characterized in that: The S4 is specifically: S41, on the user side, based on the above final optimization plan, a minute-level rolling real-time optimization mechanism is established to push the user-side path selection to the electric vehicle terminal in real time; S42, on the grid side, forming a real-time response curve of electric vehicle charging electricity prices based on the real-time electricity prices in the grid topology map, and obtaining a line congestion and a new energy consumption hotspot map based on the power flow information and new energy power in the grid topology map; S43, on the road network side, an electric vehicle path flow diagram is formed according to the traffic flow in the road network knowledge map, and a charging pile parking space heat map is obtained according to the real-time status of the charging piles in the power grid topology map.

10. A real-time dispatch optimization system for electric vehicle charging load power grid operation, characterized in that: include: The data storage module is used to build a vehicle-road-network coupled three-domain knowledge graph fusion architecture, including vehicle feature graphs, road network resource graphs, and power grid topology graphs. The knowledge graph data is stored in a graph database and updated in real time. The dynamic perception module builds a dynamic perception layer based on the vehicle-road-network coupling knowledge graph, including: Build dynamic perception of electric vehicle charging and discharging behavior to predict the spatiotemporal distribution of electric vehicle charging demand; Build charging pile load sensing to predict charging and discharging power of charging piles; Bidirectionally couple the dynamic perception of electric vehicle charging and discharging behavior with the perception of charging pile load to improve the accuracy of dynamic perception of electric vehicle charging behavior; The multi-objective optimization module builds a multi-objective optimization layer based on the vehicle-road-network coupling knowledge graph, constructs a multi-objective optimization function, and uses a hybrid adaptive particle swarm optimization to optimize the multi-objective optimization function to output the final optimization solution; The strategy generation module builds a strategy generation layer based on the vehicle-road-network coupling knowledge graph, and pushes the optimization plan to the user side, power grid side and road network side.

Images