Charging pile scheduling optimization method and system combined with traffic flow prediction

Abstract

The application provides a charging pile scheduling optimization method and system combined with traffic flow prediction, and relates to the technical field of urban traffic and energy management. First, the historical traffic flow and charging pile usage record set of a target urban area are obtained, an initial dynamic correlation network model is constructed through feature mining and modeling, real-time traffic flow monitoring data is input into the model to predict the traffic flow evolution trend and charging pile occupation state, dynamic scheduling optimization basis information is generated, the charging piles are matched according to the dynamic scheduling optimization basis information and double-dimension joint optimization is performed, a charging scheduling optimization scheme is obtained, and finally the charging scheduling optimization scheme is packaged as a charging guidance instruction set and sent to a user terminal device to trigger navigation and reservation locking operations. The application improves the rationality of charging pile resource allocation and user experience.

Description

Technical Field

[0001] This invention relates to the field of urban traffic and energy management technology, and more specifically, to a method and system for optimizing charging pile scheduling by combining traffic flow prediction. Background Technology

[0002] With the rapid pace of urbanization, urban traffic flow is increasing daily, and the number of electric vehicles is also rising continuously, making the rational scheduling of charging stations a crucial issue in urban traffic and energy management. Currently, traditional charging station scheduling methods have the following shortcomings: Firstly, most methods only consider the current idle status of charging stations, ignoring the impact of traffic flow on their use. Due to significant differences in traffic flow across different time periods and road sections, some charging stations may be idle during traffic congestion, but users may find it difficult to reach them quickly due to traffic conditions; conversely, during smooth traffic, charging stations may be occupied, preventing users from charging in a timely manner. Secondly, existing scheduling methods lack precise analysis of the spatiotemporal distribution of charging demand, making it difficult to predict charging station usage in advance. This hinders dynamic and forward-looking scheduling optimization, leading to unreasonable allocation of charging station resources, long waiting times for users, and reduced efficiency and user experience. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for optimizing charging pile scheduling based on traffic flow prediction, the method comprising: Obtain the historical traffic flow record set and the historical charging pile usage record set of the target city area; The historical traffic flow record set is processed by vehicle flow direction feature mining, and the historical charging pile usage record set is processed by charging demand spatiotemporal distribution modeling. Based on the results of vehicle flow direction feature mining and charging demand spatiotemporal distribution modeling, an initial dynamic correlation network model between urban area traffic flow and charging demand is constructed. The real-time traffic flow monitoring data stream of the target urban area is obtained, and the real-time traffic flow monitoring data stream is input into the initial urban area traffic flow and charging demand dynamic correlation network model for traffic flow evolution trend prediction and charging pile occupancy status prediction. Dynamic scheduling optimization basis information is generated, which includes the predicted traffic flow density distribution of each road section in multiple future prediction time intervals and the predicted idle time window set of each charging pile. Based on the predicted idle time window set in the dynamic scheduling optimization basis information, the charging requests to be allocated in the target city area are matched with charging piles to generate target charging pile identification information and charging time period allocation results for each charging request to be allocated. Based on the charging time period allocation results and the predicted traffic flow density distribution in the dynamic scheduling optimization basis information, the target charging pile identification information and the charging time period allocation results are jointly optimized in two dimensions: charging pile and charging time, to obtain a charging scheduling optimization scheme that includes optimized charging pile identification and optimized charging time interval. The charging scheduling optimization scheme is encapsulated into a set of charging guidance instructions containing the charging pile address coordinates, charging start time, and charging end time. The set of charging guidance instructions is then sent to the corresponding user terminal device to trigger navigation path planning and charging pile reservation locking operations.

[0004] Furthermore, embodiments of the present invention also provide a charging pile scheduling optimization system that incorporates traffic flow prediction, comprising: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the above-described charging pile scheduling optimization method incorporating traffic flow prediction by executing the machine-executable instructions.

[0005] Based on the above, by acquiring historical traffic flow records and historical charging pile usage records for the target urban area, and performing vehicle flow feature mining and charging demand spatiotemporal distribution modeling respectively, an initial dynamic correlation network model of urban area traffic flow and charging demand is constructed. Real-time traffic flow monitoring data is input into this model to predict traffic flow evolution trends and charging pile occupancy status. This generates dynamic scheduling optimization information containing predicted traffic flow density distribution for each road segment within multiple future prediction time intervals and predicted idle time windows for each charging pile. This allows for advance understanding of traffic and charging pile usage dynamics. Based on the dynamic scheduling optimization information, charging pile matching and dual-dimensional joint optimization processing are performed to obtain a charging scheduling optimization scheme that fully considers both traffic flow and charging demand, improving the rationality and efficiency of charging pile resource allocation. Finally, the charging scheduling optimization scheme is encapsulated into a charging guidance instruction set and sent to the user terminal device, triggering navigation route planning and charging pile reservation locking operations, providing users with convenient and efficient charging services, comprehensively improving the user experience and the overall operational efficiency of charging piles. Attached Figure Description

[0006] Figure 1 This is a schematic diagram of the execution flow of the charging pile scheduling optimization method combined with traffic flow prediction provided in an embodiment of the present invention.

[0007] Figure 2 This is a schematic diagram of exemplary hardware and software components of the charging pile scheduling optimization system that combines traffic flow prediction, provided in an embodiment of the present invention. Detailed Implementation

[0008] Figure 1 This is a flowchart illustrating a charging pile scheduling optimization method combining traffic flow prediction, provided in one embodiment of the present invention. A detailed description follows.

[0009] Step S110: Obtain the historical traffic flow record set of the target city area and the historical charging pile usage record set of the target city area.

[0010] In this embodiment, a target urban area is defined, for example, the city's central urban area is designated as a unified management area. First, the historical traffic flow records of all traffic monitoring points within this area over the past year are retrieved from the city's traffic management data center. These monitoring points include geomagnetic induction coils, microwave radar, and checkpoint cameras. The historical traffic flow records contain time-series data on the number of vehicles passing through each monitoring point, recorded minute-by-minute, and time-series data on the average vehicle speed. Each record is associated with a unique device identifier and precise spatial coordinates of the monitoring point. Simultaneously, the historical charging pile usage records of all charging piles within the same time period are obtained from the charging pile operation management platform. This historical charging pile usage record set contains records of each charging event, specifically including the charging pile identifier, charging start time, charging end time, and the charging pile's geographical coordinates. To protect user privacy during the data collection process, all information involving vehicle and user identifiers is anonymized. An irreversible cryptographic hash algorithm is used to convert the original license plate number into a unique, anonymized identifier, ensuring that it cannot be traced back to a specific individual. All data transmission is encrypted using the Secure Sockets Layer protocol, and access control and data encryption technologies are employed during storage to prevent data leakage.

[0011] Step S120: Perform vehicle flow direction feature mining processing on the historical traffic flow record set, and simultaneously perform charging demand spatiotemporal distribution modeling processing on the historical charging pile usage record set. Based on the vehicle flow direction feature mining processing results and the charging demand spatiotemporal distribution modeling processing results, construct an initial urban area traffic flow and charging demand dynamic correlation network model.

[0012] Next, we will conduct in-depth mining and modeling of the two types of historical data obtained above. Specifically, this step is first decomposed into two parallel sub-steps: vehicle flow feature mining and charging demand spatiotemporal distribution modeling, and then a correlation network model is constructed based on the processing results of the two.

[0013] Step S121: Perform vehicle flow direction feature mining processing on the historical traffic flow record set to obtain the vehicle transfer probability distribution matrix and vehicle route selection preference feature set between different road sections in the target city area.

[0014] This step aims to extract macroscopic movement patterns of vehicles from massive amounts of historical traffic data. To achieve this, the following detailed sub-steps are performed: Step S1211: The time-series data of vehicle passage quantity and vehicle average speed in the historical traffic flow record set are processed by spatial grid division according to the spatial distribution of traffic monitoring points. The target urban area is divided into multiple road grid units with uniform area and a unique spatial grid identifier is assigned to each road grid unit. At the same time, a mapping relationship table between the spatial grid identifier and the spatial coordinates of the traffic monitoring points is established.

[0015] Using GIS (Geographic Information System) tools, the target urban area was divided into multiple uniformly sized square grids according to latitude and longitude, with each grid having a side length of 500 meters. Each grid was defined as a road grid unit and assigned a unique code, for example, sequentially numbered from GRID00001 to GRID10000. Subsequently, the spatial coordinates of each traffic monitoring point were matched with its corresponding grid to establish a mapping table. This mapping table records all traffic monitoring point identifiers and their specific coordinates under each spatial grid identifier.

[0016] Step S1212: Extract the time-series data of the number of vehicles with the same vehicle identification information from the historical traffic flow record set, sort the time-series data of the number of vehicles with the same vehicle identification information according to the time sequence, and generate a vehicle trajectory sequence of a single vehicle in a continuous time period. The vehicle trajectory sequence includes multiple consecutive time points and the spatial grid identifier of the road grid unit where the vehicle is located at each time point.

[0017] From historical traffic flow records, using anonymized vehicle identifiers, records of each unique vehicle being captured by all monitoring points within a single day are extracted. These records are then sorted from morning to night according to their timestamps to obtain the vehicle's trajectory sequence. For example, the trajectory sequence of vehicle VID_A can be represented as follows: time point T1 corresponds to spatial grid identifier GRID0012, time point T2 corresponds to spatial grid identifier GRID0025, and time point T3 corresponds to spatial grid identifier GRID0041, where T1, T2, and T3 are consecutive time points.

[0018] Step S1213: Extract transfer pairs from the spatial grid identifiers corresponding to adjacent time points in the vehicle trajectory sequence, count the number of transfers from any first spatial grid identifier to any second spatial grid identifier in all vehicle trajectory sequences, and construct an initial transfer frequency matrix based on the number of transfers. The row index of the initial transfer frequency matrix corresponds to the spatial grid identifier of the starting grid of the transfer, and the column index corresponds to the spatial grid identifier of the reaching grid.

[0019] Iterate through all vehicle trajectory sequences. For each pair of adjacent records in each sequence, such as a transfer from spatial grid identifier GRID_A to spatial grid identifier GRID_B, increment the value of the element with row index GRID_A and column index GRID_B by 1 in the initial transfer frequency matrix. In this way, the vehicle transfer frequency between all grids within the target city area is calculated. Finally, an initial transfer frequency matrix F of dimension N×N is obtained, where N is the total number of road grid cells, and the element F[GRID_A, GRID_B] in the matrix represents the total number of historical transfers from grid A to grid B.

[0020] Step S1214: Perform row normalization on the initial transfer frequency matrix by dividing each element of each row of the initial transfer frequency matrix by the sum of all elements in that row to obtain the normalized vehicle transfer probability distribution matrix. Each element in the vehicle transfer probability distribution matrix represents the probability value of transferring from the starting grid of the corresponding row index to the arriving grid of the corresponding column index.

[0021] For each row in the initial transfer frequency matrix F, such as the row corresponding to the starting grid GRID_A, first calculate the sum of all column elements in that row: SUM_A = Σ_{k=1}^{N}F[GRID_A, GRID_k]. Then, divide each element F[GRID_A, GRID_B] in that row by SUM_A to obtain the probability value P[GRID_A, GRID_B] = F[GRID_A, GRID_B] / SUM_A. After processing all rows, the vehicle transfer probability distribution matrix P is obtained. P[GRID_A, GRID_B] represents the probability that a vehicle starts from grid A and moves to grid B next. This vehicle transfer probability distribution matrix quantitatively describes the macroscopic transfer trend of urban traffic flow.

[0022] Step S1215: Perform sliding window path segment extraction processing on the vehicle trajectory sequence, extracting a set of path segments containing multiple consecutive road grid units from each vehicle trajectory sequence. Each path segment in the set of path segments consists of at least three consecutive spatial grid identifiers arranged in chronological order.

[0023] To uncover more refined path selection preferences of vehicles, a sliding window of length L (L is set to 3 in this embodiment) is used to slide across the trajectory sequence of each vehicle. For example, from the trajectory sequence [GRID0012, GRID0025, GRID0041, GRID0050], path segments PATH1=[GRID0012, GRID0025, GRID0041] and PATH2=[GRID0025, GRID0041, GRID0050] can be extracted. All path segments of length 3 for all vehicles constitute the path segment set P.

[0024] Step S1216: Count the frequency of the same path segment in all vehicle trajectory sequences, calculate the passage frequency ratio parameter of each path segment based on the frequency of the same path segment, and use the passage frequency ratio parameter as the initial path selection weight of the corresponding path segment.

[0025] The total number of occurrences of each unique path segment in the path segment set P is counted. For example, the number of occurrences of path segment PATH_A is denoted as COUNT(PATH_A), and the total number of occurrences of all path segments is TOTAL_COUNT = Σ_{p∈P}COUNT(p). Then, the frequency proportion parameter W_initial(PATH_A) of path segment PATH_A is = COUNT(PATH_A) / TOTAL_COUNT. This frequency proportion parameter W_initial is used as the initial path selection weight for this path segment.

[0026] Step S1217: Perform spatial context weighting on the initial path selection weights. Based on the spatial distance between the starting grid and the destination grid in each path segment and the number of grids passed through in between, perform attenuation adjustment on the initial path selection weights to generate adjusted path selection weights.

[0027] Considering that long-distance paths have a low probability of occurrence but are still important for understanding urban traffic flow, attenuation adjustments are necessary. For a path segment containing a grid sequence [G_start, G_mid1, ..., G_end], its spatial distance D is first calculated, which is the sum of the Euclidean distances between the center points of adjacent grids in the sequence. Simultaneously, the number of grids M it traverses is counted, i.e., the number of grids in the path. The attenuation coefficient α = 1 / (D × M²). The initial path selection weight W_initial is multiplied by the attenuation coefficient α to obtain the adjusted path selection weight W_adjusted = W_initial × α.

[0028] Step S1218: Construct a path selection preference feature set based on the adjusted path selection weights. The path selection preference feature set includes multiple path segment identifiers and a path selection weight value corresponding to each path segment identifier. The path segment identifier is composed of multiple spatial grid identifiers that make up the path segment, connected in chronological order.

[0029] Each path segment is associated with and stored in relation to its adjusted path selection weight. The path segment identifier ID_path is formed by concatenating spatial grid identifiers in the path using a specific delimiter, such as "GRID0012-GRID0025-GRID0041". The aforementioned set of path selection preference features is a series of key-value pairs {ID_path: W_adjusted}.

[0030] Step S1219: Associate and store the vehicle transfer probability distribution matrix and the path segment identifiers and path selection weight values ​​in the path selection preference feature set to establish a mapping relationship database between the transfer probability from the starting grid to the destination grid and the path selection weights through a specific intermediate grid sequence.

[0031] The above processing results are integrated to form a mapping database. This mapping database allows querying the corresponding transition probability P[GRID_start, GRID_end] using the starting grid identifier GRID_start and the ending grid identifier GRID_end. It can also retrieve a list of all possible intermediate path segments containing the start and end point pair and their corresponding path selection weights W_adjusted using GRID_start and GRID_end.

[0032] Step S122: Perform spatiotemporal distribution modeling of charging demand on the historical charging pile usage record set to obtain the charging demand time distribution curve of multiple charging piles in the target city area and the boundary information of the spatial clustering area of ​​charging demand.

[0033] This step aims to reveal the temporal and spatial patterns of charging station usage. To achieve this, the following detailed sub-steps are performed: Step S1221: Arrange the charging start time record and charging end time record of each charging pile in the historical charging pile usage record set according to the time axis to generate a charging event time sequence for a single charging pile. The charging event time sequence includes multiple charging events and the charging start time coordinates and charging end time coordinates corresponding to each charging event.

[0034] For each charging station, for example, with the identifier PILE_A, all its usage records are sorted by time. Each record forms an event EVENT_i, which includes the charging start time START_T_i and the charging end time END_T_i. The entire sequence E_A={EVENT_1, EVENT_2, ..., EVENT_K} is the charging event time series for that charging station.

[0035] Step S1222: Divide the charging event time series into sliding windows according to a preset time window length to obtain multiple charging event subsequences within consecutive time windows. Calculate the cumulative occupancy time of each charging event within each consecutive time window. Calculate the average occupancy rate parameter of a single charging pile within each time window based on the cumulative occupancy time and the time window length.

[0036] The time window length T_win is set to 3600 seconds (1 hour), and the sliding step T_step is set to 900 seconds (15 minutes). For each time window [t, t+T_win] starting at time point t, the charging event time series E_A of charging pile PILE_A is traversed, and the total duration of the charging pile being occupied within the window is calculated as OCC_dur. Specifically, for each event EVENT_i, its occupancy interval is [START_T_i, END_T_i], and the intersection length of this interval with the window [t, t+T_win] is calculated. The intersection lengths of all events and the window are summed to obtain OCC_dur. Then, the average occupancy rate parameter OCC_rate within the time window is calculated as OCC_dur / T_win.

[0037] Step S1223: Perform time series smoothing on the average occupancy rate parameter of a single charging pile in multiple consecutive time windows to eliminate the influence of random fluctuations and generate a charging demand time distribution curve for a single charging pile. The charging demand time distribution curve uses the time window as the horizontal axis and the average occupancy rate parameter as the vertical axis to represent the probability change trend of a single charging pile being occupied by charging vehicles in different time intervals.

[0038] An exponentially weighted moving average algorithm was used to smooth the average occupancy rate parameter sequence {OCC_rate_t}. The smoothing coefficient β was set to 0.2. For the first time window t1, the smoothed value S_t1 = OCC_rate_t1. For each subsequent time window t (t > t1), the smoothed value S_t = β × OCC_rate_t + (1 - β) × S_{t-1}. After processing, a smooth charging demand time distribution curve {S_t} was obtained, which reflects the typical intraday variation pattern of charging pile occupancy rate and eliminates the influence of instantaneous random fluctuations.

[0039] Step S1224: Extract the geographic location coordinate information of each charging pile in the historical charging pile usage record set, project the geographic location coordinate information onto the spatial grid of the target city area, and establish the correspondence between the charging pile and the spatial grid identifier of the road grid unit.

[0040] For each charging station PILE_X, calculate the grid index it falls into based on its latitude and longitude coordinates (LON_X, LAT_X). The grid row index ROW_INDEX = floor((LAT_X - LAT_MIN) / 0.0045), and the column index COL_INDEX = floor((LON_X - LON_MIN) / 0.0045), where LAT_MIN and LON_MIN are the minimum latitude and longitude of the target city area, and 0.0045 degrees approximately corresponds to a distance of 500 meters. A unique spatial grid identifier GRID_ID is generated based on ROW_INDEX and COL_INDEX, and a mapping table M between the charging station identifier PILE_X and GRID_ID is established.

[0041] Step S1225: Based on the correspondence between the charging pile and the spatial grid identifier, associate the charging demand time distribution curve of each charging pile with the corresponding spatial grid identifier to form a set of charging demand time distribution curves with spatial location labels.

[0042] Using the correspondence table M, the charging demand time distribution curve {S_t}_X of each charging pile PILE_X is associated with GRID_ID. Each grid GRID_ID may be associated with multiple curves (if there are multiple charging piles in that grid). This forms a set C={GRID_ID: [curve_PILE_1, curve_PILE_2, ...]}.

[0043] Step S1226: Perform spatial clustering analysis on the set of charging demand time distribution curves with spatial location labels, and calculate the similarity between charging demand time distribution curves corresponding to different spatial grid identifiers. The similarity is calculated based on the differences in curve shape, peak occurrence time, and peak amplitude.

[0044] For any two spatial grids GRID_A and GRID_B, firstly, the representative curves within each grid are calculated. If a grid is associated with multiple curves, the occupancy rates of these curves at the same time point are averaged to form the comprehensive demand curves AVG_A(t) and AVG_B(t) for that grid. Then, the similarity SIM_AB between the two curves is calculated. The similarity is obtained by weighting three parts: curve shape similarity is calculated using the Pearson correlation coefficient to determine the correlation r_AB between AVG_A(t) and AVG_B(t); peak time difference is calculated as the reciprocal of the absolute difference between the peak times T_peak_A and T_peak_B of the two curves, 1 / (|T_peak_A-T_peak_B|+ε); peak amplitude difference is calculated as the reciprocal of the absolute difference between the peak amplitudes P_A and P_B of the two curves, 1 / (|P_A-P_B|+ε). The final similarity SIM_AB = w1×r_AB + w2×(1 / (|T_peak_A-T_peak_B|+ε)) + w3×(1 / (|P_A-P_B|+ε)), where w1, w2, and w3 are preset weights, and ε is a very small positive number to prevent division by zero.

[0045] Step S1227: Based on the similarity, aggregate adjacent spatial grid identifiers with similar charging demand time distribution curves into the same spatial cluster. Each spatial cluster corresponds to a charging demand spatial aggregation region. The spatial cluster contains multiple spatial grid identifiers and the charging demand time distribution curve corresponding to each spatial grid identifier.

[0046] A density-based spatial clustering algorithm is used to merge grids GRID_A and GRID_B that have a similarity SIM_AB higher than a preset threshold θ (θ is 0.75 in this embodiment) and are spatially adjacent (i.e., share edges) into the same cluster CLUSTER_k. This process is repeated iteratively until no new grid can be added to any existing cluster. Ultimately, K spatial clusters are obtained, each representing a spatially contiguous region with highly similar charging demands.

[0047] Step S1228: Extract the boundary grid coordinates of the spatial grid identifier contained in each spatial cluster, draw the circumscribed polygon outline of each spatial cluster according to the boundary grid coordinates, and generate the boundary information of the charging demand spatial cluster area. The boundary information of the charging demand spatial cluster area includes multiple circumscribed polygon outlines and the spatial cluster identifier corresponding to each circumscribed polygon outline.

[0048] For each cluster CLUSTER_k, obtain the boundary grids of all spatial grids contained within it (i.e., grids with at least one neighboring grid not within the cluster). Extract the coordinates of the four vertices of these boundary grids to form a point set. Calculate the convex hull polygon of this point set, which serves as the circumscribed polygon contour POLY_k of the cluster. Associate and store each contour POLY_k with the cluster identifier CLUSTER_ID_k to form the boundary information of the charging demand spatial clustering region.

[0049] Step S1229: Perform a weighted average on the charging demand time distribution curves corresponding to all spatial grid identifiers within each spatial cluster, using the number of charging piles within each spatial grid identifier as the weight, to generate a representative charging demand time distribution curve for that spatial cluster.

[0050] For each cluster CLUSTER_k, iterate through each grid GRID_i it contains, obtaining the number of charging piles NUM_i within that grid and the comprehensive demand curve AVG_i(t) for that grid. Calculate the total number of charging piles in the cluster TOTAL_NUM_k = ΣNUM_i. Then, the representative charging demand time distribution curve REP_k(t) for cluster CLUSTER_k is = Σ(NUM_i × AVG_i(t)) / TOTAL_NUM_k. This representative charging demand time distribution curve represents the typical pattern of charging demand variation over time for the entire cluster area.

[0051] Step S123: Construct an initial dynamic correlation network model between urban area traffic flow and charging demand based on the results of vehicle flow direction feature mining and the results of charging demand spatiotemporal distribution modeling.

[0052] This step structures and correlates the vehicle flow patterns discovered above with the spatiotemporal distribution of charging demand, constructing a unified network model. To achieve this, the following detailed sub-steps are performed: Step S1231: Take all road grid units within the target city area as traffic nodes in the network model, assign a unique traffic node identifier to each traffic node, and establish a one-to-one correspondence between the traffic node identifier and the spatial grid identifier of the road grid unit.

[0053] The 10,000 road grid cells divided in step S1211 are directly mapped to traffic nodes in the network model. Each traffic node identifier NODE_T_i is exactly the same as its corresponding spatial grid identifier GRID_i, for example, NODE_T_0012 corresponds to GRID0012, ensuring conceptual consistency and direct mapping.

[0054] Step S1232: Take all charging piles in the target city area as charging pile nodes in the network model, assign a unique charging pile node identifier to each charging pile node, and establish a mapping table of the affiliation relationship between the charging pile node identifier and the traffic node identifier of the traffic node to which it is located.

[0055] Each physical charging pile is abstracted as an independent charging pile node and assigned a unique identifier NODE_P_j. Using the correspondence table M established in step S1224, the grid GRID_i where each charging pile PILE_j belongs can be determined. Therefore, an affiliation mapping table R is established to record each affiliation relationship: NODE_P_j belongs to NODE_T_i.

[0056] Step S1233: Based on the vehicle transfer probability distribution matrix in the vehicle flow direction feature mining processing result, construct a directed connection edge from the first traffic node corresponding to any first traffic node identifier to the second traffic node corresponding to any second traffic node identifier, and use the transfer probability value corresponding to the vehicle transfer probability distribution matrix as the initial edge weight of the directed connection edge.

[0057] Based on the vehicle transfer probability distribution matrix P obtained in step S1214, for any non-zero element P[GRID_A, GRID_B] in the matrix, in the network model, a directed connection edge EDGE_AB is constructed from traffic node NODE_T_A (corresponding to GRID_A) to traffic node NODE_T_B (corresponding to GRID_B). The initial weight W_edge_initial_AB of this edge is set to P[GRID_A, GRID_B].

[0058] Step S1234: Traverse each path segment identifier in the path selection preference feature set in the vehicle flow direction feature mining processing result, parse the sequence of multiple consecutive traffic node identifiers contained in each path segment identifier, and perform path preference correction processing on the initial edge weight of the directed connection edge according to the path selection weight value corresponding to the path segment identifier to generate the corrected edge weight containing path preference information.

[0059] From the path selection preference feature set generated in step S1218, traverse each path segment ID_path and its corresponding weight W_adjusted. Parse ID_path, for example, "GRID0012-GRID0025-GRID0041", to obtain the traffic node sequence [NODE_T_0012, NODE_T_0025, NODE_T_0041]. For each pair of adjacent nodes in the sequence, such as (NODE_T_0012, NODE_T_0025), find the corresponding directed connection edge EDGE_0012_0025. Distribute the path selection weight W_adjusted proportionally to this edge, with the distribution ratio being the edge's contribution to the path. Since the path length is 3, containing 2 edges, the contribution of each edge is 1 / 2. Therefore, the initial weight W_edge_initial of edge EDGE_0012_0025 is updated by superposition: W_edge_updated = W_edge_initial + λ × W_adjusted × (1 / 2), where λ is the path preference influence factor, which is set to 0.3 in this embodiment. After this correction is performed on all edges of all paths, the corrected edge weights are obtained.

[0060] Step S1235: According to the attribution mapping table, establish an attribution connection edge between each charging pile node and the traffic node to which it belongs, and assign an initial attribution weight to the attribution connection edge. The initial attribution weight is calculated based on the number of charging piles in the traffic node.

[0061] For each record in the attribution mapping table R, an undirected attribution connection edge EDGE_Pj_Ti is established between the charging pile node NODE_P_j and its affiliated traffic node NODE_T_i. The weight W_belonging of this attribution connection edge is set to the reciprocal of the total number of charging piles NUM_i within the traffic node NODE_T_i, i.e., W_belonging = 1 / NUM_i. The higher the weight, the greater the scarcity and importance of the charging pile node within its affiliated traffic node.

[0062] Step S1236: Obtain the charging demand time distribution curve of each charging pile node, and attach the characteristic parameters of the charging demand time distribution curve as the node attribute information of the charging pile node to the charging pile node. The characteristic parameters include the peak occurrence time, peak amplitude and curve fluctuation frequency.

[0063] For each charging pile node NODE_P_j, associate it with the charging demand time distribution curve {S_t}_j generated in step S1223. Extract key feature parameters from this curve: peak occurrence time T_peak_j, i.e., the time corresponding to the maximum value of curve {S_t}_j; peak amplitude A_peak_j, i.e., the maximum value of the curve; curve fluctuation frequency F_j, which is obtained by performing a Fast Fourier Transform on the curve and taking the frequency component with the highest energy as the fluctuation frequency. Attach {T_peak_j, A_peak_j, F_j} as attribute information to NODE_P_j.

[0064] Step S1237: Integrate and encapsulate the traffic nodes, charging pile nodes, directed connection edges, belonging connection edges, corrected edge weights, initial belonging weights, and node attribute information to generate complete structural data of the initial urban area traffic flow and charging demand dynamic correlation network model. The complete structural data is stored in the form of a graph data structure, where traffic nodes and charging pile nodes are the vertices of the graph, directed connection edges and belonging connection edges are the edges of the graph, and the edge weight parameters and node attribute information are the additional features of the graph.

[0065] Thus, a complete heterogeneous graph network model G has been constructed. The vertex set V of this heterogeneous graph network model contains all traffic nodes and charging pile nodes. The edge set E contains two types of edges: directed edges representing traffic flow (with corrected weights W_edge_updated) and undirected edges representing affiliation (with affiliation weights W_belonging). Each vertex carries its own attribute information; the attribute of a traffic node is its current state (which will be used in subsequent steps), and the attributes of a charging pile node are {T_peak_j, A_peak_j, F_j}. The entire model G is stored in a distributed database or in-memory database in the form of an adjacency list, supporting efficient graph traversal and update operations.

[0066] Step S130: Obtain the real-time traffic flow monitoring data stream of the target urban area, input the real-time traffic flow monitoring data stream into the initial urban area traffic flow and charging demand dynamic correlation network model for traffic flow evolution trend prediction processing and charging pile occupancy status prediction processing, and generate dynamic scheduling optimization basis information containing the predicted traffic flow density distribution of each road section in multiple future prediction time intervals and the predicted idle time window set of each charging pile.

[0067] Based on the established dynamic correlation network model and combined with real-time data, future states are predicted. To achieve this, the following detailed sub-steps are performed: Step S131: Continuously receive real-time data on the number of vehicles passing through and the average speed of vehicles at the current moment from multiple traffic monitoring points within the target city area. Classify and organize the data on the number of vehicles passing through and the average speed of vehicles at the current moment according to spatial grid identifiers to generate a real-time traffic flow monitoring data stream. The real-time traffic flow monitoring data stream includes the real-time traffic flow density parameter and real-time vehicle speed parameter of each road grid unit at the current moment.

[0068] The system uses a message queue to access real-time data streams pushed by the city traffic management center. Each record in the data stream contains a monitoring point ID, a timestamp, the number of vehicles passing through (NUM_cur), and the average speed (SPD_cur). Using the mapping table established in step S1211, each record is mapped to its corresponding spatial grid identifier (GRID_i). For each grid (GRID_i), within a time window Δt (e.g., 5 minutes), the NUM_cur of all monitoring points within it is summed to obtain the total number of vehicles passing through that grid. This total number is then divided by the grid area to obtain the real-time traffic flow density parameter (DENS_i). Simultaneously, the average SPD_cur of the aforementioned monitoring points is taken to obtain the real-time vehicle speed parameter (V_i) for that grid.

[0069] Step S132: Load the real-time traffic flow monitoring data stream into the initial urban area traffic flow and charging demand dynamic correlation network model, use the real-time traffic flow density parameter of each road grid unit as the current state value of the corresponding traffic node, and use the real-time vehicle speed parameter of each road grid unit as the current speed attribute value of the traffic node.

[0070] The real-time traffic density parameter DENS_i of each grid GRID_i calculated in the previous step is assigned to the corresponding traffic node NODE_T_i in model G as its current state value STATE_i. At the same time, the real-time vehicle speed parameter V_i is stored as the speed attribute value SPD_i of this node.

[0071] Step S133: Call the traffic flow evolution trend prediction module in the initial urban area traffic flow and charging demand dynamic correlation network model, and calculate the predicted traffic flow density parameters of each traffic node in multiple future prediction time intervals through a multi-step iterative propagation algorithm based on the vehicle transfer probability distribution matrix and the current state value of each traffic node at the current time.

[0072] In this embodiment, model G incorporates a traffic flow evolution trend prediction module, which makes predictions based on the graph propagation concept.

[0073] Step S1331: The calculation process of the multi-step iterative propagation algorithm includes: for the first future prediction time interval, the first predicted traffic flow density parameter of each traffic node is calculated based on the current state value of each traffic node at the current time and the weighted sum of the transfer probabilities from other traffic nodes to this traffic node.

[0074] For the first future prediction time interval T_pred1, such as 15 minutes after the current time, the formula for calculating the predicted traffic flow density parameter PRED_i_1 of traffic node NODE_T_i is: PRED_i_1=Σ_{j∈N_in(i)}(STATE_j×P[GRID_j,GRID_i]). Here, N_in(i) is the set of all traffic nodes with directed edges pointing to node i, STATE_j is the current state value (i.e., current traffic flow density) of these nodes, and P[GRID_j,GRID_i] is the transition probability from grid j to grid i in the vehicle transition probability distribution matrix.

[0075] Step S1332: For each subsequent future prediction time interval, take the predicted traffic flow density parameter of the previous prediction time interval as input, repeat the weighted sum calculation process, and generate the predicted traffic flow density parameters of the second, third and up to the Mth future prediction time intervals in sequence, so as to obtain the predicted traffic flow density distribution of each road grid cell in the multiple future prediction time intervals.

[0076] For the second prediction time interval T_pred2 (the next 30 minutes), the prediction result of the first step is used as input: PRED_i_2=Σ_{j∈N_in(i)}(PRED_j_1×P[GRID_j,GRID_i]). This process is repeated for M iterations to obtain the predicted traffic flow density distribution {PRED_i_1,PRED_i_2,...,PRED_i_M} for the next M time intervals (in this embodiment, M is 12, meaning predicting the next 3 hours, with each interval lasting 15 minutes).

[0077] After loading real-time traffic flow monitoring data streams into the dynamic association network model, a standardized input data format needs to be constructed for the graph attention network model. For each prediction time step, the input data contains two core components: the node feature matrix X and the adjacency matrix A.

[0078] The construction process of the node feature matrix X is as follows: For each traffic node NODE_T_i in the dynamic network model, its multi-dimensional features at the current time t are extracted. These features include: the traffic flow density parameter STATE_i (i.e., DENS_i) at the current time, the vehicle speed parameter SPD_i at the current time, and the historical average traffic flow density HIST_AVG_i and historical average vehicle speed HIST_SPD_i of the node in the same time period (e.g., 9:00 AM to 10:00 AM) obtained from historical data. In addition, time-encoded features are introduced, such as converting the current time t into a sinusoidal position encoding vector PE(t) to help the model understand the time periodicity. The above features of all traffic nodes are concatenated to form a node feature matrix X_t with dimension N×F, where N is the total number of traffic nodes and F is the feature dimension of each node. In this embodiment, F is 64 (including traffic flow density, vehicle speed, historical average, historical vehicle speed, and 60-dimensional time-encoded features).

[0079] The adjacency matrix A is constructed based on the vehicle transition probability distribution matrix P. The original transition probability matrix P describes the transition probabilities between nodes, but graph attention networks require an adjacency structure that reflects the connectivity and importance between nodes. Therefore, matrix P is binarized: for any two traffic nodes NODE_T_i and NODE_T_j, if the transition probability P[GRID_i, GRID_j] is greater than a preset threshold TH_P (TH_P is 0.01 in this embodiment), then in the adjacency matrix A, element A_ij is set to 1, indicating that there is a directed edge from node i to node j; otherwise, it is set to 0. Simultaneously, to consider spatial proximity, if two nodes are adjacent on the spatial grid (i.e., share an edge), but their transition probabilities are below the threshold, their corresponding A_ij is also set to 1 to ensure local spatial connectivity. Finally, an N×N sparse adjacency matrix A is obtained, serving as the structural input to the graph.

[0080] The core of the traffic flow evolution trend prediction module is a stacked graph attention network, which consists of an input layer, three hidden graph attention layers, and an output layer. The core mechanism of each graph attention layer is multi-head self-attention, which can adaptively learn the importance weights of different neighbor nodes to the center node.

[0081] The data first enters the input layer. The input layer receives the node feature matrix X_t and adjacency matrix A constructed in step S133-1. In the input layer, X_t undergoes a linear transformation, mapping the original features to a higher-dimensional hidden space. Specifically, X_t is transformed into an initial hidden state H_0 = X_t × W_in using a trainable weight matrix W_in, where the dimension of W_in is F×H, and H is the hidden layer dimension; in this embodiment, H is set to 128. The dimension of H_0 becomes N×H.

[0082] Subsequently, H_0 is passed to the first graph attention layer, GAT_Layer_1. In this layer, for each node i, the model first calculates the attention coefficients e_ij between it and all its neighboring nodes j (determined by the adjacency matrix A). The attention coefficients are calculated by concatenating the hidden state vector h_i of node i and the hidden state vector h_j of node j, then performing a dot product with a learnable attention weight vector a, and finally applying the LeakyReLU activation function. The specific formula is: e_ij = LeakyReLU(a^T × [W_att × h_i || W_att × h_j]), where W_att is the shared linear transformation matrix of dimension H×H, || denotes the vector concatenation operation, and a^T is the transpose of the attention parameter vector.

[0083] After obtaining the original attention coefficients e_ij for node i and all its neighbors j, the softmax function is used to normalize these coefficients to obtain the final attention weights α_ij. The normalization process is performed on all neighbors of the node: α_ij = exp(e_ij) / Σ_{k∈N(i)} exp(e_ik), where N(i) is the set of all neighbor nodes of node i.

[0084] After obtaining the attention weight α_ij, the new feature representation h'_i of node i is obtained by weighted summation of the transformed features of all neighboring nodes: h'i = σ( Σ{j∈N(i)} α_ij × (W_att × h_j) ), where σ is the ELU activation function.

[0085] To stabilize the learning process of the self-attention mechanism, the first graph attention layer employs a multi-head attention mechanism, which involves running K independent attention mechanisms in parallel (K is set to 4 in this embodiment). The results calculated by the K attention mechanisms are then concatenated as the final output of this layer. Therefore, after the first graph attention layer, the feature representation dimension of each node becomes K×H, i.e., 4×128=512 dimensions. The output of this layer is denoted as H_1, with a dimension of N×512.

[0086] H_1 is then passed to the second graph attention layer, GAT_Layer_2. The structure and operation of this layer are basically the same as the first layer, but the input dimension becomes 512, while the output dimension remains unchanged at 512 (achieved by concatenating multi-head results). In the second layer, the model further aggregates information within higher-order neighbor ranges, learning more complex spatial dependencies. The output of this layer is denoted as H_2, with a dimension of N×512.

[0087] H_2 is then passed to the third graph attention layer, GAT_Layer_3. This layer also employs a multi-head attention mechanism, but instead of concatenating the outputs, it averages the outputs of the K attention heads to obtain the final embedding representation for each node. The output of this layer is denoted as H_3, with dimensions N×H (N×128). At this point, the embedding vector h_i_final of each node NODE_T_i has incorporated its own features as well as the structural and feature information of its multi-level neighbors, effectively representing the traffic state at the current moment.

[0088] The final hidden state H_3 is passed to the output layer. The output layer consists of a fully connected feedforward neural network. This network contains two fully connected sub-layers. The first fully connected sub-layer maps the input N×128-dimensional features to an intermediate representation of dimension N×64 and applies the ReLU activation function. The second fully connected sub-layer then maps the N×64-dimensional features to the final output dimension, i.e., N×M, where M is the number of future prediction time intervals; in this embodiment, M is set to 12. Therefore, the original output tensor O generated by the output layer has a dimension of N×12.

[0089] Each element O_i_k in tensor O represents the original predicted value of traffic flow density for traffic node NODE_T_i in the k-th prediction time interval in the future. However, these original predicted values ​​may not meet physical constraints (e.g., they cannot be negative). Therefore, after the output layer, a post-processing module is applied to O_i_k using the Softplus activation function to ensure that all predicted values ​​are positive: PRED_i_k = ln(1 + exp(O_i_k)). The final PRED_i_k is the predicted traffic flow density parameter for each road grid cell in the multiple prediction time intervals in the future, which is required in step S1332.

[0090] The graph attention network model described above requires extensive offline training before it can be applied to real-time prediction. The training process is performed after constructing the initial dynamic correlation network model between urban area traffic flow and charging demand, but before real-time prediction is performed.

[0091] First, a training dataset is constructed. Data for consecutive time periods is extracted from the historical traffic flow record set. For each time point t, a node feature matrix X_t is constructed according to step S133-1, and the actual traffic flow density sequence Y_t for the subsequent M time intervals is obtained as the label. A large number of training sample pairs (X_t, Y_t) are generated by using a sliding time window. In this embodiment, the training dataset contains 20 million sample pairs randomly selected from the past three years.

[0092] Secondly, the loss function is defined. The weighted sum of the mean absolute error and the root mean square error is used as the model's loss function L. For each sample, the difference between the predicted value PRED_i_k and the true value TRUE_i_k for all nodes across all future time intervals is calculated. The loss function is calculated as follows: L = λ1 × (1 / (N×M)) Σ|PRED_i_k - TRUE_i_k| + λ2 × √((1 / (N×M)) Σ(PRED_i_k - TRUE_i_k)²), where λ1 and λ2 are weighting coefficients balancing the two errors. In this embodiment, λ1 is set to 0.4 and λ2 to 0.6.

[0093] Next, an optimizer was selected and hyperparameters were set. The Adam optimizer was used to update the model parameters. The initial learning rate was set to 0.001, the batch size to 128, and the number of training epochs to 200. To prevent overfitting, a Dropout mechanism was introduced after each graph attention layer, with a dropout rate of 0.1. Simultaneously, an early stopping strategy was employed: training was stopped when the loss function value on the validation set no longer decreased for 10 consecutive epochs, and the parameters of the best-performing model on the validation set were saved.

[0094] Finally, model training is performed. The training dataset is input into the model in batches, and the gradient is calculated and the model parameters are updated using the backpropagation algorithm to minimize the loss function L. After training, the weight parameters of the trained graph attention network model are fixed and loaded into the traffic flow evolution trend prediction module for subsequent real-time prediction and inference.

[0095] During real-time operation, once the latest real-time traffic flow monitoring data stream is acquired and the current state values ​​of traffic nodes are updated, the inference process of the traffic flow evolution trend prediction module is triggered. First, following step S133-1, the node feature matrix X_t and adjacency matrix A for the current time t are constructed (adjacency matrix A is fixed after model training unless significant changes occur in the urban road network requiring reconstruction). Then, X_t and A are input into the pre-trained graph attention network model. The model performs one forward propagation calculation, sequentially passing through the input layer, three graph attention layers, and the output layer, ultimately generating the original output tensor O. Next, the Softplus function of the post-processing module converts O into the final predicted traffic flow density distribution {PRED_i_1, PRED_i_2, ..., PRED_i_M} for subsequent steps. The entire inference process is completed in milliseconds, meeting the requirements of real-time scheduling.

[0096] Step S134: Based on the predicted traffic flow density distribution of each road grid unit in the multiple future prediction time intervals, and combined with the mapping table of the affiliation relationship between charging pile nodes and traffic nodes, determine the predicted traffic flow density parameters of the traffic node where each charging pile node is located in the corresponding future prediction time interval.

[0097] For each charging station node NODE_P_j, its associated traffic node NODE_T_i is found according to the affiliation mapping table R. Then, for the k-th predicted time interval, the predicted traffic flow density of the area where the charging station is located is PRED_i_k.

[0098] Step S135: For each charging pile node, obtain its charging demand time distribution curve, and extract the baseline occupancy probability of the charging pile node in the corresponding future prediction time interval from the charging demand time distribution curve according to the time point corresponding to the future prediction time interval.

[0099] For a charging pile node NODE_P_j, obtain its charging demand time distribution curve {S_t}_j. For the k-th predicted time interval (which corresponds to a specific point in time or time period), extract the baseline occupancy probability BASE_OCC_j_k for that time from the curve. For example, if T_pred_k corresponds to 9 AM, then BASE_OCC_j_k is the smoothed occupancy rate value of the curve at 9 AM.

[0100] Step S136: Dynamically correct the baseline occupancy probability based on the predicted traffic flow density parameter of the traffic node where the charging pile node is located, and generate the corrected occupancy probability. The dynamic correction process is based on the positive correlation between the predicted traffic flow density parameter and the baseline occupancy probability. The predicted traffic flow density parameter is converted into a correction coefficient through a preset correction function. The correction coefficient is multiplied by the baseline occupancy probability to obtain the corrected occupancy probability.

[0101] Considering that increased traffic flow in a region usually implies increased potential charging demand, the baseline occupancy probability needs to be corrected. For charging pile node NODE_P_j and the k-th future interval, the correction coefficient δ_j_k is calculated as follows: δ_j_k = 1 + γ × (PRED_i_k - BASELINE_DENS) / BASELINE_DENS, where PRED_i_k is the predicted traffic flow density of the area where the node is located, BASELINE_DENS is the historical average traffic flow density of the area (which can be statistically derived from historical data), and γ is the sensitivity coefficient, which is set to 0.5 in this embodiment. The corrected occupancy probability ADJ_OCC_j_k = BASE_OCC_j_k × δ_j_k. This correction ensures that the predicted occupancy probability reflects the dynamic changes in real-time traffic conditions.

[0102] Step S137: Calculate the predicted idle probability of each charging pile node in multiple predicted time intervals in the future based on the corrected occupancy probability. The predicted idle probability is equal to one minus the corrected occupancy probability.

[0103] For charging pile node NODE_P_j and the k-th future interval, the predicted idle probability FREE_j_k=1-ADJ_OCC_j_k.

[0104] Step S138: For each charging pile node, arrange the multiple future predicted time intervals in chronological order, identify the continuous time intervals in which the predicted idle probability exceeds the preset idle threshold, extract the above continuous time intervals as the predicted idle time window of the charging pile node, and mark the start time point and end time point of each predicted idle time window to generate a set containing multiple predicted idle time windows.

[0105] The idle threshold TH_FREE is set to 0.3. For charging pile node NODE_P_j, the FREE_j_k values ​​for its future M time intervals are checked sequentially. If all FREE_j_k values ​​are greater than TH_FREE in a continuous interval from interval k_start to interval k_end, then these continuous intervals are merged into a single predicted idle time window WINDOW_j_m, with its start time being the start time of interval k_start and its end time being the end time of interval k_end. All of the above windows constitute the predicted idle time window set WINDOWS_j={WINDOW_j_1, WINDOW_j_2, ...} for this charging pile.

[0106] Step S139: Integrate the predicted traffic flow density distribution of each road grid unit within the multiple predicted time intervals in the future and the predicted idle time window set of all charging pile nodes to generate dynamic scheduling optimization basis information.

[0107] The predicted traffic flow density distribution {PRED_i_k} of each grid generated in step S1332 and the predicted idle time window set WINDOWS_j of all charging pile nodes generated in step S138 are encapsulated to form a data structure DYN_INFO.

[0108] Step S140: Based on the predicted idle time window set in the dynamic scheduling optimization basis information, perform charging pile matching processing on the charging requests to be allocated in the target city area, generate target charging pile identification information and charging time period allocation results corresponding to each charging request to be allocated, and perform joint optimization processing on the target charging pile identification information and the charging time period allocation results based on the charging time period allocation results and the predicted traffic flow density distribution in the dynamic scheduling optimization basis information, and obtain a charging scheduling optimization scheme that includes optimized charging pile identification and optimized charging time interval.

[0109] Based on dynamic scheduling optimization information, user charging requests are matched and optimized globally in real time. First, preliminary matching is performed, followed by joint optimization.

[0110] Step S141: Based on the predicted idle time window set in the dynamic scheduling optimization basis information, perform charging pile matching processing on the charging requests to be allocated in the target city area, and generate target charging pile identification information and charging time period allocation results corresponding to each charging request to be allocated.

[0111] This step aims to quickly find a suitable charging station and time slot for each newly arrived charging request. To achieve this, the following detailed sub-steps are performed: Step S1411: Receive charging requests from user terminal devices in real time. The charging requests include the request initiation time, the current location coordinates of the requesting user, the preset target departure time of the requesting user, and the charging duration parameter.

[0112] The user initiates a charging request REQ_u through a mobile application. This charging request includes: the initiation time T_req_u, the user's current GPS coordinates LOC_u, the desired time when the vehicle will be available after charging is complete T_leave_u (i.e., the time the target leaves), and the estimated charging duration DUR_u.

[0113] Step S1412: Determine the spatial grid identifier of the road grid cell to which the current location of the requesting user belongs based on the current location coordinates of the requesting user, and use it as the request initiation location identifier.

[0114] Based on the user coordinates LOC_u, the road grid cell in which it is located is calculated using the same method as in step S1224, and the spatial grid identifier GRID_req_u is obtained.

[0115] Step S1413: Extract the predicted idle time window set of all charging pile nodes from the dynamic scheduling optimization basis information, and calculate the spatial reachability parameter of each charging pile node based on the spatial distance between the spatial grid identifier of the traffic node where each charging pile node is located and the request initiation location identifier. The spatial reachability parameter is negatively correlated with the spatial distance.

[0116] Obtain the predicted idle time window set for all charging piles from DYN_INFO. For each charging pile node NODE_P_j, whose grid is GRID_j, calculate its Euclidean distance DIST_uj to GRID_req_u. Then, the spatial reachability parameter ACC_uj = 1 / (DIST_uj + ε), where ε is a very small positive number used to prevent division by zero errors.

[0117] Step S1414: For each charging pile node, traverse each predicted idle time window in its predicted idle time window set, determine whether the start time of the predicted idle time window is later than the request initiation time, and whether the duration of the predicted idle time window is greater than or equal to the requested charging duration parameter. If so, mark the corresponding predicted idle time window as a candidate time window.

[0118] For each predicted idle time window WINDOW_j_m of charging pile node NODE_P_j, its start time is T_start_j_m and its end time is T_end_j_m. It is determined whether T_start_j_m ≥ T_req_u + δ (where δ is the minimum preparation time, e.g., 10 minutes) and (T_end_j_m - T_start_j_m) ≥ DUR_u. If both conditions are met, then the window WINDOW_j_m is marked as a candidate time window CAND_j_m for request REQ_u.

[0119] Step S1415: For each charging pile node, if it has at least one candidate time window, mark the charging pile node as a candidate charging pile and record the start and end times of all its candidate time windows.

[0120] Add all charging pile nodes NODE_P_j with at least one candidate time window to the candidate charging pile list CAND_PILES. For each candidate charging pile in the list, record the start and end times of all its candidate windows.

[0121] Step S1416: Calculate the comprehensive matching score of each candidate charging pile and each candidate time window based on the spatial accessibility parameter of each candidate charging pile and the time interval between the start time of its candidate time window and the time point of the request initiation. The comprehensive matching score is positively correlated with the spatial accessibility parameter and negatively correlated with the time interval.

[0122] For each candidate time window CAND_j_m of the candidate charging station NODE_P_j, its starting time point is T_start_j_m. The calculation time interval GAP_j_m = T_start_j_m - T_req_u. The formula for calculating the comprehensive matching score SCORE_ujm is: SCORE_ujm = α × ACC_uj - β × GAP_j_m, where α and β are weighting coefficients balancing the importance of space and time, and in this embodiment, they are taken as 0.7 and 0.3, respectively. The higher the score, the more user-friendly the charging station and the time window are.

[0123] Step S1417: Select the candidate time window with the highest comprehensive matching score from all candidate time windows of all candidate charging piles, determine the charging pile node corresponding to the candidate time window as the target charging pile, and determine the start time point and end time point of the candidate time window as the charging time period allocation result.

[0124] Find the (j*, m*) corresponding to the maximum value among all SCORE_ujm. Then, determine the charging pile node NODE_P_j* as the target charging pile for this request, and determine the start time point T_start_j*_m* and end time point (T_start_j*_m*+DUR_u) of CAND_j*_m* as the charging time period allocation result.

[0125] Step S1418: Output the charging pile node identifier of the target charging pile and the charging time period allocation result as the target charging pile identifier information and charging time period allocation result corresponding to the charging request to be allocated.

[0126] The output result RESULT_u={PILE_ID:NODE_P_j*,START:T_start_j*_m*,END:T_start_j*_m*+DUR_u}.

[0127] Step S1419: For multiple charging requests to be assigned received within the preset response time window, perform matching processing sequentially according to the order in which the requests were initiated, and update the predicted idle time window set of the selected target charging pile after each matching process, and remove the assigned candidate time window from the predicted idle time window set of the charging pile.

[0128] If multiple requests REQ_u1, REQ_u2, ... are received within a very short time (e.g., within 1 second), they are processed in the order of their initiation time T_req. After REQ_u1 is allocated, the corresponding window WINDOW_j*_m* in DYN_INFO is split or removed (if it is completely occupied). The updated DYN_INFO is used as the input for the next request REQ_u2 to ensure that no resource conflicts occur.

[0129] Step S142: Based on the charging time period allocation result and the predicted traffic flow density distribution in the dynamic scheduling optimization basis information, perform joint optimization processing on the target charging pile identification information and the charging time period allocation result in two dimensions: charging pile and charging time, to obtain a charging scheduling optimization scheme that includes the optimized charging pile identification and the optimized charging time interval.

[0130] Step S141's rapid matching ensures real-time performance, but may not be globally optimal. This step, within an optimization cycle (e.g., every 5 minutes), performs joint optimization on a batch of accumulated requests to be assigned, balancing user satisfaction with driving and charging time. To achieve this, the following detailed sub-steps are implemented: Step S1421: Extract the current location coordinates of the requesting user, the geographical coordinates of the charging pile corresponding to the target charging pile identification information, and the charging start time point in the charging time period allocation result for each charging request to be assigned.

[0131] For a set of requests U to be optimized, for each request REQ_u, extract its LOC_u, the geographical coordinates LOC_j* of the target charging pile NODE_P_j* allocated in step S1417, and the charging start time point START_u (i.e. T_start_j*_m*) of the allocation result.

[0132] Step S1422: Based on the current location coordinates of the requesting user and the geographical location coordinates of the charging pile, and combined with the predicted traffic flow density distribution in multiple predicted time intervals in the dynamic scheduling optimization information, generate multiple candidate driving paths for each charging request to be allocated. Each candidate driving path consists of multiple consecutive road grid unit sequences, and calculate the predicted driving time of each candidate driving path.

[0133] For a request REQ_u, it is necessary to travel from the starting grid GRID_req_u (derived from LOC_u) to the ending grid GRID_j* (derived from LOC_j*). Based on the mapping relationship database established in step S1219, all possible path segments from GRID_req_u to GRID_j* (i.e., paths contained in the path selection preference feature set) are queried and used as the candidate driving path set PATH_u={PATH_u1, PATH_u2, ...}. For each candidate path PATH_uq, it consists of the grid sequence [G_start, G_mid1, ..., G_end].

[0134] Step S1422-1: The method for calculating the predicted travel time includes: based on the predicted traffic flow density parameters of each road grid unit on the candidate travel path within the corresponding predicted time interval, converting the predicted traffic flow density parameters into the predicted traffic speed of the road grid unit through a preset congestion degree conversion function, calculating the time required to pass through the corresponding grid based on the spatial distance and predicted traffic speed of each road grid unit, and summing up the time required for all grids to obtain the predicted travel time of the candidate travel path.

[0135] Calculate the predicted travel time T_travel_uq. First, based on the predicted traffic density PRED_m_k for each grid G_m on the path (k depends on the expected arrival time interval for that grid), calculate the predicted speed V_pred_m using the congestion level conversion function V_pred=V_max / (1+(PRED_m_k / DENS_ref)^η), where V_max is the road speed limit, DENS_ref is the reference density, and η is the nonlinear coefficient. Then, divide the spatial distance D_m of grid G_m (i.e., the distance from the center point of the grid to the center point of the next grid, or the road length within the grid) by V_pred_m to obtain the estimated time T_m for passing through that grid. Finally, T_travel_uq=ΣT_m.

[0136] Step S1423: For each charging request to be assigned, calculate the upper limit of the allowed departure time window based on the request initiation time and the charging start time, and filter out candidate driving paths whose predicted driving time is less than or equal to the upper limit of the allowed departure time window as a set of feasible driving paths.

[0137] Users must arrive at the charging station before the charging start time START_u. The latest allowed departure time is START_u. The upper limit of the departure time window is W_upper = START_u - T_req_u. Candidate paths that satisfy T_travel_uq ≤ W_upper are selected as the set of feasible driving paths FEASIBLE_PATH_u.

[0138] Step S1424: Construct a collaborative optimization objective function for electric vehicle driving path and charging plan. The independent variables of the collaborative optimization objective function are a path selected from the set of feasible driving paths for each charging request to be assigned, and the charging start time point in the charging time period allocation result for each charging request to be assigned.

[0139] The objective function F of the collaborative optimization is to minimize the total cost to the user while satisfying all constraints. The independent variables X include: for each request u, selecting a path PATH_uq∈FEASIBLE_PATH_u, and assigning it a new charging start time point START'_u (which can be fine-tuned near the original assignment).

[0140] Step S1425: The collaborative optimization objective function includes a first sub-objective term and a second sub-objective term. The first sub-objective term is used to minimize the sum of the predicted driving times of the selected paths of all charging requests to be assigned. The first sub-objective term is obtained by summing the predicted driving times of the selected paths of each charging request to be assigned.

[0141] The first sub-target item F_travel=Σ_{u∈U}T_travel_uq, where T_travel_uq is the predicted travel time calculated based on the selected path PATH_uq.

[0142] Step S1426: The second sub-target item is used to minimize the sum of the time intervals between the charging start time of all pending charging requests and the user-preset target departure time. The second sub-target item is obtained by summing the absolute values ​​of the differences between the charging start time of each pending charging request and the user-preset target departure time.

[0143] The second sub-objective, F_satisfaction = Σ_{u∈U}|START'_u+DUR_u-T_leave_u|, measures the deviation between the charging completion time and the user's expected time; the smaller the deviation, the higher the user satisfaction.

[0144] Step S1427: Assign a first weight coefficient and a second weight coefficient to the first sub-objective item and the second sub-objective item respectively, and add the product of the first sub-objective item and the first weight coefficient to the product of the second sub-objective item and the second weight coefficient to obtain the overall expression of the collaborative optimization objective function.

[0145] The overall objective function is F_total = ω1 × F_travel + ω2 × F_satisfaction. In this embodiment, the weighting coefficients ω1 and ω2 are both set to 0.5, indicating that equal importance is given to travel time and user time satisfaction.

[0146] Step S1428: Using the currently generated target charging pile identification information and charging time period allocation results as the initial solution space, the collaborative optimization objective function is solved through an iterative optimization algorithm. In each iteration, the charging pile identification or charging time period in the current solution space is finely perturbed to generate new candidate solutions. The collaborative optimization objective function value corresponding to the new candidate solution is calculated, and the new candidate solution is accepted as the current solution based on the quality of the objective function value. The iteration is repeated until the stopping condition is met, and the final optimal solution is taken as the charging scheduling optimization scheme containing the optimized charging pile identification and the optimized charging time interval.

[0147] The simulated annealing algorithm is used for solving. The output of step S141 is used as the initial solution X0. Set the initial temperature T_high and the cooling rate cool. In each iteration, perturb the current solution: randomly select a request u, and randomly replace the allocated charging pile with another feasible candidate pile, or randomly adjust the charging start time START_u of it by one time step (e.g., an integer multiple of 15 minutes). Check the feasibility of the new solution X_new (whether the charging pile is idle in the new interval and whether the selected path is feasible), and calculate its objective function value F_total(X_new). If F_total(X_new) < F_total(X_current), then accept X_new as the new current solution. Otherwise, accept the inferior solution with probability P = exp((F_total(X_current) - F_total(X_new)) / T_current) to jump out of the local optimum. As the temperature T_current gradually decreases, the algorithm converges. When the temperature is lower than T_low or there is no improvement after consecutive iterations for multiple times, stop the iteration and output the optimal solution X_best. X_best is the charging schedule optimization scheme including the optimized charging pile identifier and the optimized charging time interval.

[0148] Step S150: Package the charging schedule optimization scheme into a charging guidance instruction set including the charging pile address coordinates, the charging start time, and the charging end time, and send the charging guidance instruction set to the corresponding user terminal device to trigger the navigation path planning operation and the charging pile reservation and locking operation.

[0149] Finally, implement the optimized scheme, guide the user, and lock the resources. To achieve this, the following refined sub-steps are carried out: Step S151: Analyze the charging schedule optimization scheme, and extract the optimized charging pile identifier and the optimized charging time interval corresponding to each charging request to be allocated. The optimized charging time interval includes the optimized charging start time point and the optimized charging end time point.

[0150] From the optimal solution X_best obtained in step S1428, parse out the optimized charging pile identifier PILE_ID_opt_u (i.e., NODE_P_j*_opt) and the optimized charging time interval [T_start_opt_u, T_end_opt_u] finally determined for each request REQ_u, where T_end_opt_u = T_start_opt_u + DUR_u.

[0151] Step S152: Query the corresponding charging pile geographical location coordinates, the specific address description information of the charging pile, and the charging pile interface type parameters in the preset charging pile information database according to the optimized charging pile identifier.

[0152] Connect to the charging pile information database and query its detailed information based on PILE_ID_opt_u, including: geographical coordinates LON_opt_u, LAT_opt_u; specific address description ADDR_opt_u, such as "XX City XX District XX Road XX No. XX Parking Lot"; and interface type TYPE_opt_u, such as "GB / T20234.3 DC Fast Charging".

[0153] Step S153: Based on the request user identifier information corresponding to each charging request to be assigned, combine and encapsulate the optimized charging pile identifier, the charging pile geographical coordinates, the charging pile specific address description information, the charging pile interface type parameters, the optimized charging start time point, and the optimized charging end time point to generate a charging guidance instruction for a single user.

[0154] For each request REQ_u, its corresponding user identifier USER_ID_u is combined with the above information and encapsulated into a structured charging boot instruction CMD_u. CMD_u={USER_ID: USER_ID_u, PILE_ID: PILE_ID_opt_u, DEST_COORD: (LON_opt_u, LAT_opt_u), DEST_ADDR: ADDR_opt_u, CONN_TYPE: TYPE_opt_u, START_TIME: T_start_opt_u, END_TIME: T_end_opt_u}.

[0155] Step S154: All generated charging guidance instructions are classified and organized according to the requesting user identification information, and a corresponding personalized charging guidance instruction subset is generated for each requesting user. The personalized charging guidance instruction subset contains the charging guidance instruction corresponding to the current charging request to be allocated by the requesting user.

[0156] The CMD_u instruction is grouped by USER_ID_u. Since each user has only one request in this batch, each user's personalized charging guidance instruction subset contains its own CMD_u.

[0157] Step S155: Send a subset of personalized charging guidance instructions for each requesting user to the user terminal device corresponding to the requesting user via a wireless communication network. The user terminal device is a mobile device with navigation and display functions.

[0158] The CMD_u command is pushed to the user's mobile application via cellular network or Wi-Fi. End-to-end encryption is used for data transmission to ensure the security of the command content.

[0159] Step S156: While sending the personalized charging guidance instruction subset, a charging pile reservation lock request is sent to the charging pile management system. The charging pile reservation lock request includes the optimized charging pile identifier, the optimized charging start time, and the optimized charging end time. It is used to trigger the charging pile management system to mark the charging pile status corresponding to the optimized charging pile identifier as reserved within the corresponding time period, and to prohibit other users from using the charging pile between the optimized charging start time and the optimized charging end time.

[0160] While sending instructions to the user, the backend service sends a reservation lock request (LOCK_REQ_u) to the interface of the charging pile operation management system. This request includes PILE_ID_opt_u, T_start_opt_u, and T_end_opt_u. After receiving the reservation lock request, the charging pile management system updates the status of the charging pile for the corresponding time period from "idle" or "unknown" to "reserved," ensuring that the resource is exclusively locked and preventing over-selling.

[0161] Step S157: After receiving the personalized charging guidance instruction subset, the user terminal device automatically extracts the geographical coordinates of the charging pile as the navigation destination, and starts the navigation path planning application in combination with the current positioning coordinates of the user terminal device to generate the optimal navigation path from the current positioning coordinates to the geographical coordinates of the charging pile and displays it on the screen of the user terminal device.

[0162] After receiving CMD_u, the user's mobile application parses the destination coordinates (LON_opt_u, LAT_opt_u). The application automatically calls the phone's built-in map navigation function (such as Google Maps or Apple Maps API), uses the user's current real-time location as the starting point and the parsed destination as the destination, plans the optimal driving route, displays it on the screen, and provides voice guidance throughout the journey.

[0163] Step S158: The user terminal device simultaneously displays the optimized charging start time and the optimized charging end time on the screen, and sends a departure reminder notification to the user at a preset reminder time before the optimized charging start time, prompting the user to go to the designated charging station for charging.

[0164] The application clearly displays the reserved charging time slot on the interface: "Please arrive before [T_start_opt_u], your charging period is from [T_start_opt_u] to [T_end_opt_u]". Simultaneously, a departure reminder alarm can be set; for example, 30 minutes before T_start_opt_u, the application will push a notification: "Your reserved charging station will start in 30 minutes, please head to [ADDR_opt_u] on time," ensuring users arrive on time.

[0165] Step S210: Extract curve feature parameters from the charging demand time distribution curve to obtain the peak charging demand period parameters, the trough charging demand period parameters, and the charging demand fluctuation range parameters for each charging pile.

[0166] After generating the charging demand time distribution curve in step S1223, further refined analysis can be performed. For the curve {S_t}_j of each charging pile node NODE_P_j, the local maximum point of the curve is found by differentiation, and the corresponding time is the peak charging demand period parameter T_peak_j. The local minimum point of the curve is found, and the corresponding time is the trough charging demand period parameter T_valley_j. The standard deviation STD_j of the curve {S_t}_j is calculated as a parameter of charging demand fluctuation amplitude. The larger this value, the more drastic the change in demand of the charging pile over time.

[0167] Step S220: Based on the peak charging demand period parameters and the trough charging demand period parameters, classify all charging piles in the target city area into charging demand period types to generate a set of charging piles with morning peak demand, a set of charging piles with evening peak demand, and a set of charging piles with balanced demand.

[0168] Based on the time period in which T_peak_j falls, all charging stations are categorized. If T_peak_j falls between 7:00 and 9:00, the charging station is assigned to the MORNING_SET (morning peak demand type). If T_peak_j falls between 17:00 and 19:00, it is assigned to the EVENING_SET (evening peak demand type). If T_peak_j falls neither during the morning nor evening peak hours, and the fluctuation range STD_j is less than the preset threshold STD_TH, it is assigned to the BALANCE_SET (balanced demand type).

[0169] Step S230: Based on the boundary information of the charging demand spatial clustering area, identify the spatial adjacency relationship between different charging demand spatial clustering areas, and construct a charging demand spatial clustering area adjacency graph. The vertices of the charging demand spatial clustering area adjacency graph correspond to each charging demand spatial clustering area, and the edges between the vertices indicate that two charging demand spatial clustering areas are spatially adjacent.

[0170] Using the boundary information of the charging demand spatial clustering regions generated in step S1228 (i.e., the circumscribed polygon outline POLY_k of each cluster), for any two different clusters CLUSTER_k and CLUSTER_l, determine whether their circumscribed polygons POLY_k and POLY_l intersect or share a boundary. If so, construct an undirected edge between them. This constructs a graph G_REGION, with CLUSTER_ID as the vertex and edges representing spatial adjacency of regions.

[0171] Step S240: Perform graph cut optimization on the adjacency graph of the charging demand spatial clustering region, and merge the charging demand spatial clustering regions with similar charging demand time distribution curve characteristics and spatial adjacency into a charging demand collaborative scheduling region. Each charging demand collaborative scheduling region contains multiple original charging demand spatial clustering regions.

[0172] A graph cut algorithm is used to optimize G_REGION. The merging cost function is defined as COST(k, l) = 1 - SIM_kl, where SIM_kl is the similarity between the representative curves REP_k(t) and REP_l(t) of the two clusters calculated in step S1226. The algorithm iteratively merges adjacent regions with high similarity (low COST) until a stopping condition is met (e.g., the total number of charging piles in the merged region exceeds a threshold). Finally, multiple charging demand collaborative scheduling regions CO_ZONE are obtained, each containing multiple original clusters.

[0173] Step S250: Establish an intra-regional charging pile resource sharing pool for all charging piles within the collaborative scheduling area for each charging demand. The intra-regional charging pile resource sharing pool records the charging pile identification information, geographical location coordinate information, and charging demand time distribution curve information of all charging piles within the area where the charging pile is established.

[0174] For each coordinated scheduling zone CO_ZONE_z, a resource pool POOL_z is constructed. This resource pool is a data structure that records the identifier list PILE_LIST_z of all charging piles in the zone, the geographical coordinates of each pile, and its charging demand time distribution curve, used to support querying and scheduling within the zone.

[0175] Step S260: Construct cross-regional charging pile collaborative scheduling priority rules. The cross-regional charging pile collaborative scheduling priority rules are determined based on the normalized spatial distance between charging demand collaborative scheduling areas and the complementarity between the charging demand time distribution curves of each area. The closer the spatial distance and the stronger the complementarity of the charging demand time distribution curves, the higher the collaborative scheduling priority between two charging demand collaborative scheduling areas.

[0176] Define the cross-regional collaborative scheduling priority PRIO_zw. First, calculate the normalized spatial distance DIST_norm_zw between regions CO_ZONE_z and CO_ZONE_w, which is the Euclidean distance between the center points of the two regions divided by the maximum distance between cities. Then, calculate the complementarity COMP_zw of the representative demand curves REP_z(t) and REP_w(t) of the two regions, defined as the degree to which one curve is at a peak and the other is at a trough on the time axis. Specifically, this can be quantified by calculating the degree of overlap between the peak period of one curve and the trough period of the other curve. The collaborative scheduling priority PRIO_zw = μ × (1 - DIST_norm_zw) + ν × COMP_zw, where μ and ν are weighting coefficients.

[0177] Step S270: The charging demand collaborative scheduling area, the charging pile resource sharing pool within the area, and the cross-regional charging pile collaborative scheduling priority rules are used as auxiliary scheduling information, and together with the vehicle transfer probability distribution matrix and the vehicle path selection preference feature set, are used to subsequently construct the initial urban area traffic flow and charging demand dynamic correlation network model.

[0178] The auxiliary scheduling information generated in steps S210 to S260 can be injected as additional prior knowledge when constructing the initial network model. For example, in step S1236, an attribute "Cooperative Scheduling Region ID" can be added to the charging pile node, or in the graph structure of step S1237, virtual connection edges between cooperative scheduling regions can be added, with the weight of the edges determined by the cooperative scheduling priority, so that the model can take into account the possibility of cross-regional resource coordination when making predictions or optimizations.

[0179] For example, step S310: perform real-time abnormal fluctuation detection processing on the real-time traffic flow monitoring data stream to identify sudden surges and sudden drops in traffic flow in the real-time traffic flow monitoring data stream.

[0180] During the continuous reception of real-time traffic flow monitoring data streams in step S131, an anomaly detection module runs synchronously. This module maintains a sliding window (e.g., data from the past hour) of a real-time traffic flow density sequence {DENS_i_t} for each grid GRID_i. It calculates the mean μ_i and standard deviation σ_i of the data within this window. When a newly arrived DENS_i_new satisfies DENS_i_new > μ_i + 3σ_i, it is determined to be a sudden surge in traffic flow. When DENS_i_new < μ_i - 3σ_i, it is determined to be a sudden decrease in traffic flow.

[0181] Step S320: When a sudden surge in traffic flow is identified, based on the location coordinates and impact range of the sudden surge in traffic flow, the predicted traffic flow density distribution of the affected road grid units is extracted from the dynamic scheduling optimization basis information. The predicted traffic flow density parameters of the affected road grid units in multiple future prediction time intervals are adjusted upwards to generate the corrected predicted traffic flow density distribution.

[0182] Suppose a surge event is detected in grid GRID_a, with an influence radius of R_imp. Determine the set of all affected grids AFFECTED_GRIDS within the radius R_imp. Extract the predicted values ​​PRED_i_k for all future time intervals of these grids from DYN_INFO. For each affected grid GRID_i, its corrected predicted value PRED'_i_k = PRED_i_k × (1 + θ × exp(-dist_ai / R_imp)), where θ is the surge influence intensity coefficient (e.g., 0.5), and dist_ai is the distance from grid i to the event center grid a.

[0183] Step S330: When a sudden decrease in traffic flow is detected, based on the location coordinates and impact range of the sudden decrease in traffic flow, the predicted traffic flow density distribution of the affected road grid units is extracted from the dynamic scheduling optimization basis information. The predicted traffic flow density parameters of the affected road grid units in multiple future prediction time intervals are adjusted downward to generate the corrected predicted traffic flow density distribution.

[0184] Similar to the surge event handling logic, but with the correction direction reversed. PRED'_i_k=PRED_i_k×(1-θ×exp(-dist_ai / R_imp)).

[0185] Step S340: Based on the corrected predicted traffic flow density distribution, the predicted idle time window set of each charging pile node in the dynamic scheduling optimization basis information is synchronously updated, the predicted traffic flow density parameters of the traffic nodes where the affected charging pile nodes are located are recalculated, and the baseline occupancy probability of the corresponding charging pile node is dynamically corrected based on the recalculated predicted traffic flow density parameters to generate an updated predicted idle time window set.

[0186] The corrected predicted traffic flow density distribution PRED'_i_k is then updated back to DYN_INFO. Subsequently, steps S135 to S138 are repeated to recalculate the predicted idle time windows for all charging pile nodes within the affected area (i.e., those belonging to AFFECTED_GRIDS). For charging piles outside the area, their prediction windows remain unchanged.

[0187] Step S350: Merge the updated predicted idle time window set with the original predicted idle time window set of the unaffected charging pile nodes to generate the corrected dynamic scheduling optimization basis information.

[0188] The recalculated updated predicted idle time window set for affected charging piles is merged with the original predicted idle time window set for unaffected charging piles to generate a new dynamic scheduling optimization information DYN_INFO_UPDATED that reflects the latest traffic anomaly. Subsequent steps S140 will be based on this updated information, thereby ensuring the scheduling scheme's rapid response capability to emergencies.

[0189] In one exemplary embodiment, a charging pile scheduling optimization system incorporating traffic flow prediction is provided. This system can be a terminal, server, etc., and its internal structure diagram can be as follows: Figure 2As shown, the charging pile scheduling optimization system incorporating traffic flow prediction includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, near-field communication, or other technologies. When the computer program is executed by the processor, it implements a charging pile scheduling optimization method incorporating traffic flow prediction. The display unit is used to generate a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, or a button, trackball, or touchpad set on the casing of the charging pile scheduling optimization system that combines traffic flow prediction, or an external keyboard, touchpad, or mouse, etc.

[0190] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for optimizing charging pile scheduling based on traffic flow prediction, characterized in that, The method includes: Obtain the historical traffic flow record set and the historical charging pile usage record set of the target city area; The historical traffic flow record set is processed by vehicle flow direction feature mining, and the historical charging pile usage record set is processed by charging demand spatiotemporal distribution modeling. Based on the results of vehicle flow direction feature mining and charging demand spatiotemporal distribution modeling, an initial dynamic correlation network model between urban area traffic flow and charging demand is constructed. The real-time traffic flow monitoring data stream of the target urban area is obtained, and the real-time traffic flow monitoring data stream is input into the initial urban area traffic flow and charging demand dynamic correlation network model for traffic flow evolution trend prediction and charging pile occupancy status prediction. Dynamic scheduling optimization basis information is generated, which includes the predicted traffic flow density distribution of each road section in multiple future prediction time intervals and the predicted idle time window set of each charging pile. Based on the predicted idle time window set in the dynamic scheduling optimization basis information, the charging requests to be allocated in the target city area are matched with charging piles to generate target charging pile identification information and charging time period allocation results for each charging request to be allocated. Based on the charging time period allocation results and the predicted traffic flow density distribution in the dynamic scheduling optimization basis information, the target charging pile identification information and the charging time period allocation results are jointly optimized in two dimensions: charging pile and charging time, to obtain a charging scheduling optimization scheme that includes optimized charging pile identification and optimized charging time interval. The charging scheduling optimization scheme is encapsulated into a set of charging guidance instructions containing the charging pile address coordinates, charging start time, and charging end time. The set of charging guidance instructions is then sent to the corresponding user terminal device to trigger navigation path planning and charging pile reservation locking operations.

2. The charging pile scheduling optimization method combining traffic flow prediction according to claim 1, characterized in that, The process of mining vehicle flow direction features from the historical traffic flow record set yields a vehicle transfer probability distribution matrix and a vehicle route selection preference feature set between different road sections within the target city area, including: The time-series data of vehicle passage quantity and vehicle average speed in the historical traffic flow record set are spatially gridded according to the spatial distribution of traffic monitoring points. The target urban area is divided into multiple road grid units with uniform area and a unique spatial grid identifier is assigned to each road grid unit. At the same time, a mapping table between the spatial grid identifier and the spatial coordinates of the traffic monitoring points is established. Extract the time-series data of the number of vehicles with the same vehicle identification information from the historical traffic flow record set, sort the time-series data of the number of vehicles with the same vehicle identification information in chronological order, and generate a vehicle trajectory sequence of a single vehicle in a continuous time period. The vehicle trajectory sequence contains multiple consecutive time points and the spatial grid identifier of the road grid unit where the vehicle is located at each time point. The spatial grid identifiers corresponding to adjacent time points in the vehicle trajectory sequence are processed by transfer pair extraction. The number of transfers from any first spatial grid identifier to any second spatial grid identifier in all vehicle trajectory sequences is counted. An initial transfer frequency matrix is ​​constructed based on the number of transfers. The row index of the initial transfer frequency matrix corresponds to the spatial grid identifier of the starting grid of the transfer, and the column index corresponds to the spatial grid identifier of the reaching grid. The initial transfer frequency matrix is ​​row normalized by dividing each element of each row of the initial transfer frequency matrix by the sum of all elements in that row to obtain the normalized vehicle transfer probability distribution matrix. Each element in the vehicle transfer probability distribution matrix represents the probability value of transferring from the starting grid of the corresponding row index to the arriving grid of the corresponding column index. The vehicle trajectory sequence is subjected to sliding window path segment extraction processing to extract a set of path segments containing multiple consecutive road grid units from each vehicle trajectory sequence. Each path segment in the set of path segments consists of at least three consecutive spatial grid identifiers arranged in chronological order. The frequency of the same path segment in all vehicle trajectory sequences is counted. The passage frequency ratio parameter of each path segment is calculated based on the frequency of the same path segment. The passage frequency ratio parameter is used as the initial path selection weight of the corresponding path segment. The initial path selection weights are subjected to spatial context weighting. The initial path selection weights are then attenuated and adjusted based on the spatial distance between the starting grid and the destination grid in each path segment and the number of grids passed through in between, to generate adjusted path selection weights. A path selection preference feature set is constructed based on the adjusted path selection weights. The path selection preference feature set includes multiple path segment identifiers and a path selection weight value corresponding to each path segment identifier. The path segment identifier is composed of multiple spatial grid identifiers that make up the path segment, which are concatenated in chronological order. The vehicle transfer probability distribution matrix and the path segment identifiers and path selection weight values ​​in the path selection preference feature set are associated and stored to establish a mapping relationship database between the transfer probability from the starting grid to the destination grid and the path selection weights through a specific intermediate grid sequence.

3. The charging pile scheduling optimization method combining traffic flow prediction according to claim 1, characterized in that, The process of modeling the spatiotemporal distribution of charging demand on the historical charging pile usage record set yields the charging demand time distribution curves of multiple charging piles within the target city area and the boundary information of spatial clustering areas of charging demand, including: The charging start time and charging end time records of each charging pile in the historical charging pile usage record set are arranged according to the time axis to generate a charging event time sequence for a single charging pile. The charging event time sequence includes multiple charging events and the charging start time coordinates and charging end time coordinates corresponding to each charging event. The charging event time series is divided into sliding windows according to a preset time window length to obtain multiple charging event subsequences within consecutive time windows. The cumulative occupancy time of the charging events within each consecutive time window is counted. Based on the cumulative occupancy time and the time window length, the average occupancy rate parameter of a single charging pile within each time window is calculated. The average occupancy rate parameter of a single charging pile in multiple consecutive time windows is smoothed over time to eliminate the influence of random fluctuations and generate a charging demand time distribution curve for a single charging pile. The charging demand time distribution curve uses the time window as the horizontal axis and the average occupancy rate parameter as the vertical axis to represent the probability change trend of a single charging pile being occupied by charging vehicles in different time intervals. Extract the geographic location coordinates of each charging pile from the historical charging pile usage record set, project the geographic location coordinates onto the spatial grid of the target city area, and establish the correspondence between the charging pile and the spatial grid identifier of the road grid unit where it is located; Based on the correspondence between the charging piles and the spatial grid identifiers, the charging demand time distribution curve of each charging pile is associated with the corresponding spatial grid identifier, forming a set of charging demand time distribution curves with spatial location labels. Spatial clustering analysis is performed on the set of charging demand time distribution curves with spatial location labels to calculate the similarity between charging demand time distribution curves corresponding to different spatial grid identifiers. The similarity is calculated based on the differences in curve shape, peak occurrence time, and peak amplitude. Based on the similarity, adjacent spatial grid identifiers with similar charging demand time distribution curves are aggregated into the same spatial cluster. Each spatial cluster corresponds to a charging demand spatial aggregation region. The spatial cluster contains multiple spatial grid identifiers and the charging demand time distribution curve corresponding to each spatial grid identifier. Extract the boundary grid coordinates of the spatial grid identifier contained in each spatial cluster, draw the outer polygon outline of each spatial cluster according to the boundary grid coordinates, and generate the boundary information of the charging demand spatial cluster area. The boundary information of the charging demand spatial cluster area includes multiple outer polygon outlines and the spatial cluster identifier corresponding to each outer polygon outline. The charging demand time distribution curves corresponding to all spatial grid identifiers within each spatial cluster are weighted and averaged, with the number of charging piles within each spatial grid identifier as the weight, to generate a representative charging demand time distribution curve for that spatial cluster.

4. The charging pile scheduling optimization method combining traffic flow prediction according to claim 1, characterized in that, The initial dynamic correlation network model between urban area traffic flow and charging demand is constructed based on the results of vehicle flow direction feature mining and the results of charging demand spatiotemporal distribution modeling, including: All road grid units within the target city area are used as traffic nodes in the network model. A unique traffic node identifier is assigned to each traffic node, and a one-to-one correspondence is established between the traffic node identifier and the spatial grid identifier of the road grid unit. All charging piles within the target city area are taken as charging pile nodes in the network model. A unique charging pile node identifier is assigned to each charging pile node, and a mapping table of the affiliation relationship between the charging pile node identifier and the traffic node identifier of the traffic node to which it is located is established. Based on the vehicle transfer probability distribution matrix in the vehicle flow direction feature mining processing result, a directed connection edge is constructed from the first traffic node corresponding to any first traffic node identifier to the second traffic node corresponding to any second traffic node identifier, and the transfer probability value corresponding to the vehicle transfer probability distribution matrix is ​​used as the initial edge weight of the directed connection edge. Traverse each path segment identifier in the path selection preference feature set in the vehicle flow direction feature mining processing result, parse the sequence of multiple consecutive traffic node identifiers contained in each path segment identifier, and perform path preference correction processing on the initial edge weight of the directed connection edge according to the path selection weight value corresponding to the path segment identifier to generate the corrected edge weight containing path preference information. The path preference correction process includes: identifying the directed connection edges between every two adjacent traffic node identifiers in the multiple consecutive traffic node identifier sequences, distributing the path selection weight values ​​proportionally to these directed connection edges, and updating the initial edge weights of the directed connection edges by superposition. According to the attribution mapping table, an attribution connection edge is established between each charging pile node and the traffic node to which it belongs, and an initial attribution weight is assigned to the attribution connection edge. The initial attribution weight is calculated based on the number of charging piles in the traffic node. Obtain the charging demand time distribution curve for each charging pile node, and attach the characteristic parameters of the charging demand time distribution curve as the node attribute information of the charging pile node. The characteristic parameters include the peak occurrence time, peak amplitude, and curve fluctuation frequency. The traffic nodes, charging pile nodes, directed edges, belonging edges, corrected edge weights, initial belonging weights, and node attribute information are integrated and encapsulated to generate complete structural data for an initial dynamic correlation network model of urban area traffic flow and charging demand. The complete structural data is stored in the form of a graph data structure, where traffic nodes and charging pile nodes are the vertices of the graph, directed edges and belonging edges are the edges of the graph, and edge weight parameters and node attribute information are additional features of the graph.

5. The charging pile scheduling optimization method combining traffic flow prediction according to claim 1, characterized in that, The process involves acquiring real-time traffic flow monitoring data streams for the target urban area, inputting these data streams into an initial dynamic correlation network model of urban area traffic flow and charging demand for traffic flow evolution trend prediction and charging pile occupancy status prediction. This generates dynamic scheduling optimization information containing predicted traffic flow density distributions for each road segment within multiple future prediction time intervals and predicted idle time windows for each charging pile. The system continuously receives real-time data on the number of vehicles passing through and the average speed of vehicles at the current moment from multiple traffic monitoring points within the target city area. The system then categorizes and organizes the data according to spatial grid identifiers to generate a real-time traffic flow monitoring data stream. The real-time traffic flow monitoring data stream includes the real-time traffic density parameters and real-time vehicle speed parameters of each road grid unit at the current moment. The real-time traffic flow monitoring data stream is loaded into the initial urban area traffic flow and charging demand dynamic correlation network model. The real-time traffic flow density parameter of each road grid unit is used as the current state value of the corresponding traffic node, and the real-time vehicle speed parameter of each road grid unit is used as the current speed attribute value of the traffic node. The traffic flow evolution trend prediction module in the initial urban area traffic flow and charging demand dynamic correlation network model is invoked. Based on the vehicle transfer probability distribution matrix and the current state value of each traffic node at the current moment, the predicted traffic flow density parameter of each traffic node in multiple future prediction time intervals is calculated through a multi-step iterative propagation algorithm. The calculation process of the multi-step iterative propagation algorithm includes: for the first future prediction time interval, the first predicted traffic flow density parameter of each traffic node is calculated based on the current state value of each traffic node at the current time and the weighted sum of the transfer probabilities from other traffic nodes to this traffic node; For each subsequent future prediction time interval, the predicted traffic flow density parameter of the previous prediction time interval is used as input, and the weighted sum calculation process is repeated to generate the predicted traffic flow density parameters of the second, third, and up to the Nth future prediction time intervals in turn, so as to obtain the predicted traffic flow density distribution of each road grid cell in multiple future prediction time intervals. Based on the predicted traffic flow density distribution of each road grid unit within the multiple future prediction time intervals, and combined with the mapping table of the affiliation relationship between charging pile nodes and traffic nodes, the predicted traffic flow density parameters of each traffic node where the charging pile node is located are determined within the corresponding future prediction time interval. For each charging pile node, obtain its charging demand time distribution curve, and extract the baseline occupancy probability of the charging pile node in the corresponding future prediction time interval from the charging demand time distribution curve based on the time point corresponding to the future prediction time interval. The baseline occupancy probability is dynamically corrected based on the predicted traffic flow density parameter of the traffic node where the charging pile node is located, and a corrected occupancy probability is generated. The dynamic correction process is based on the positive correlation between the predicted traffic flow density parameter and the baseline occupancy probability. The predicted traffic flow density parameter is converted into a correction coefficient through a preset correction function. The correction coefficient is multiplied by the baseline occupancy probability to obtain the corrected occupancy probability. The predicted idle probability of each charging pile node in multiple predicted time intervals in the future is calculated based on the corrected occupancy probability. The predicted idle probability is equal to one minus the corrected occupancy probability. For each charging pile node, the multiple predicted time intervals are arranged in chronological order, and the continuous time intervals in which the predicted idle probability exceeds the preset idle threshold are identified. The above continuous time intervals are extracted as the predicted idle time window of the charging pile node, and the start time point and end time point are marked for each predicted idle time window to generate a set containing multiple predicted idle time windows. The predicted traffic flow density distribution of each road grid unit and the predicted idle time window set of all charging pile nodes within the multiple predicted time intervals in the future are integrated to generate dynamic scheduling optimization information.

6. The charging pile scheduling optimization method combining traffic flow prediction according to claim 1, characterized in that, The process of matching charging requests to be allocated within the target city area with charging piles based on the predicted idle time window set in the dynamic scheduling optimization information, and generating target charging pile identification information and charging time period allocation results for each charging request to be allocated, includes: The system receives charging requests from user terminal devices in real time. The charging requests include the request initiation time, the current location coordinates of the requesting user, the preset target departure time of the requesting user, and the charging duration parameter. Based on the current location coordinates of the requesting user, determine the spatial grid identifier of the road grid unit to which the current location of the requesting user belongs, and use it as the request initiation location identifier; Extract the predicted idle time window set of all charging pile nodes from the dynamic scheduling optimization basis information, and calculate the spatial reachability parameter of each charging pile node based on the spatial grid identifier of the traffic node where each charging pile node is located and the request initiation location identifier. The spatial reachability parameter is negatively correlated with the spatial distance. For each charging pile node, iterate through each predicted idle time window in its predicted idle time window set, determine whether the start time of the predicted idle time window is later than the request initiation time, and whether the duration of the predicted idle time window is greater than or equal to the requested charging duration parameter. If so, mark the corresponding predicted idle time window as a candidate time window. For each charging pile node, if it has at least one candidate time window, then the charging pile node is marked as a candidate charging pile, and the start and end times of all its candidate time windows are recorded. Based on the spatial accessibility parameters of each candidate charging station and the time interval between the start time of its candidate time window and the time point of the request initiation, a comprehensive matching score is calculated for each candidate charging station and its candidate time window. The comprehensive matching score is positively correlated with the spatial accessibility parameters and negatively correlated with the time interval. Select the candidate time window with the highest comprehensive matching score from all candidate time windows of all candidate charging piles, determine the charging pile node corresponding to the candidate time window as the target charging pile, and determine the start time point and end time point of the candidate time window as the charging time period allocation result. The charging pile node identifier of the target charging pile and the charging time period allocation result are output as the target charging pile identifier information and charging time period allocation result corresponding to the charging request to be allocated; For multiple charging requests to be assigned received within the preset response time window, matching processing is performed sequentially according to the order in which the requests were initiated. After each matching process, the predicted idle time window set of the selected target charging pile is updated, and the candidate time window that has been assigned is removed from the predicted idle time window set of the charging pile.

7. The charging pile scheduling optimization method combining traffic flow prediction according to claim 1, characterized in that, The charging time period allocation result and the predicted traffic flow density distribution in the dynamic scheduling optimization basis information are used to jointly optimize the target charging pile identification information and the charging time period allocation result in two dimensions: charging pile and charging time. This yields a charging scheduling optimization scheme that includes optimized charging pile identification and optimized charging time intervals, including: Extract the current location coordinates of the requesting user, the geographical location coordinates of the charging pile corresponding to the target charging pile identification information, and the charging start time point in the charging time period allocation result for each charging request to be assigned. Based on the current location coordinates of the requesting user and the geographical location coordinates of the charging pile, and combined with the predicted traffic flow density distribution in multiple predicted time intervals in the dynamic scheduling optimization information, multiple candidate driving paths are generated for each charging request to be allocated. Each candidate driving path consists of multiple consecutive road grid unit sequences, and the predicted driving time of each candidate driving path is calculated. The method for calculating the predicted travel time includes: based on the predicted traffic flow density parameters of each road grid unit on the candidate travel path within the corresponding predicted time interval, converting the predicted traffic flow density parameters into the predicted traffic speed of the road grid unit through a preset congestion degree conversion function, calculating the time required to pass through the corresponding grid based on the spatial distance and predicted traffic speed of each road grid unit, and summing up the time required for all grids to obtain the predicted travel time of the candidate travel path. For each charging request to be assigned, the upper limit of the allowed departure time window is calculated based on the request initiation time and the charging start time. Candidate driving paths with predicted driving time less than or equal to the upper limit of the allowed departure time window are selected as a set of feasible driving paths. Construct a collaborative optimization objective function for electric vehicle driving routes and charging plans. The independent variables of the collaborative optimization objective function are a path selected from the set of feasible driving routes for each charging request to be assigned, and the charging start time point in the charging time period allocation result for each charging request to be assigned. The collaborative optimization objective function includes a first sub-objective term and a second sub-objective term. The first sub-objective term is used to minimize the sum of the predicted driving times of the selected paths of all charging requests to be assigned. The first sub-objective term is obtained by summing the predicted driving times of the selected paths of each charging request to be assigned. The second sub-objective is used to minimize the sum of the time intervals between the charging start time of all pending charging requests and the user-preset target departure time. The second sub-objective is obtained by summing the absolute values ​​of the differences between the charging start time of each pending charging request and the user-preset target departure time. Assign a first weight coefficient and a second weight coefficient to the first sub-objective item and the second sub-objective item respectively. Add the product of the first sub-objective item and the first weight coefficient to the product of the second sub-objective item and the second weight coefficient to obtain the overall expression of the collaborative optimization objective function. The currently generated target charging pile identification information and charging time period allocation results are used as the initial solution space. The collaborative optimization objective function is solved by an iterative optimization algorithm. In each iteration, the charging pile identification or charging time period in the current solution space is finely perturbed to generate new candidate solutions. The collaborative optimization objective function value corresponding to the new candidate solution is calculated. The new candidate solution is then accepted as the current solution based on its quality. The iteration is repeated until the stopping condition is met. The final optimal solution is used as the charging scheduling optimization scheme that includes the optimized charging pile identification and the optimized charging time interval.

8. The charging pile scheduling optimization method combining traffic flow prediction according to claim 1, characterized in that, The step of encapsulating the charging scheduling optimization scheme into a set of charging guidance instructions containing charging pile address coordinates, charging start time, and charging end time, and sending the set of charging guidance instructions to the corresponding user terminal device to trigger navigation path planning and charging pile reservation locking operations, includes: The charging scheduling optimization scheme is analyzed, and the optimized charging pile identifier and optimized charging time interval corresponding to each charging request to be allocated are extracted. The optimized charging time interval includes the optimized charging start time point and the optimized charging end time point. Based on the optimized charging pile identifier, the corresponding charging pile geographical location coordinates, specific address description information, and charging pile interface type parameters are queried in the preset charging pile information database; Based on the request user identifier information corresponding to each charging request to be assigned, the optimized charging pile identifier, the charging pile geographical coordinates, the charging pile specific address description information, the charging pile interface type parameters, the optimized charging start time point and the optimized charging end time point are combined and encapsulated to generate a charging guidance instruction for a single user. All generated charging guidance instructions are categorized and organized according to the requesting user identification information, and a corresponding personalized charging guidance instruction subset is generated for each requesting user. The personalized charging guidance instruction subset contains the charging guidance instruction corresponding to the current charging request to be allocated by the requesting user. Each requesting user's personalized charging guidance instruction subset is sent to the user terminal device corresponding to the requesting user via a wireless communication network. The user terminal device is a mobile device with navigation and display functions. While sending the personalized charging guidance instruction subset, a charging pile reservation lock request is sent to the charging pile management system. The charging pile reservation lock request includes the optimized charging pile identifier, the optimized charging start time, and the optimized charging end time. It is used to trigger the charging pile management system to mark the charging pile status corresponding to the optimized charging pile identifier as reserved within the corresponding time period, and to prohibit other users from using the charging pile between the optimized charging start time and the optimized charging end time. After receiving the personalized charging guidance instruction subset, the user terminal device automatically extracts the geographical coordinates of the charging pile as the navigation destination, and launches a navigation path planning application in combination with the current positioning coordinates of the user terminal device to generate the optimal navigation path from the current positioning coordinates to the geographical coordinates of the charging pile and display it on the screen of the user terminal device. The user terminal device simultaneously displays the optimized charging start time and optimized charging end time on the screen, and sends a departure reminder notification to the user at a preset reminder time before the optimized charging start time, prompting the user to go to the designated charging station for charging.

9. The charging pile scheduling optimization method based on traffic flow prediction according to claim 1, characterized in that, After performing spatiotemporal distribution modeling of charging demand on the historical charging pile usage record set to obtain the charging demand time distribution curves of multiple charging piles within the target city area and the boundary information of the spatial clustering area of ​​charging demand, the method further includes: The charging demand time distribution curve is processed by extracting curve feature parameters to obtain the peak charging demand period parameters, the trough charging demand period parameters, and the charging demand fluctuation amplitude parameters for each charging pile. Based on the peak charging demand time period parameters and the trough charging demand time period parameters, all charging piles in the target city area are classified by charging demand time period type to generate a set of charging piles with morning peak demand, a set of charging piles with evening peak demand, and a set of charging piles with balanced demand. Based on the boundary information of the charging demand spatial clustering area, the spatial adjacency relationship between different charging demand spatial clustering areas is identified, and a charging demand spatial clustering area adjacency graph is constructed. The vertices of the charging demand spatial clustering area adjacency graph correspond to each charging demand spatial clustering area, and the edges between the vertices indicate that two charging demand spatial clustering areas are spatially adjacent. The adjacency graph of the charging demand spatial clustering region is subjected to graph cut optimization processing. The charging demand spatial clustering regions with similar charging demand time distribution curve characteristics and spatial adjacency are merged into a charging demand collaborative scheduling region. Each charging demand collaborative scheduling region contains multiple original charging demand spatial clustering regions. For each charging demand, a regional charging pile resource sharing pool is established for all charging piles within the coordinated scheduling area. The regional charging pile resource sharing pool records the charging pile identification information, geographical location coordinate information, and charging demand time distribution curve information of all charging piles within the area where the charging pile is established. A cross-regional charging pile collaborative scheduling priority rule is constructed. The cross-regional charging pile collaborative scheduling priority rule is determined based on the normalized spatial distance between charging demand collaborative scheduling areas and the complementarity between the charging demand time distribution curves of each area. The closer the spatial distance and the stronger the complementarity of the charging demand time distribution curves, the higher the collaborative scheduling priority between two charging demand collaborative scheduling areas. The charging demand collaborative scheduling area, the charging pile resource sharing pool within the area, and the cross-regional charging pile collaborative scheduling priority rules are used as auxiliary scheduling information, which, together with the vehicle transfer probability distribution matrix and the vehicle path selection preference feature set, are used to subsequently construct an initial urban area traffic flow and charging demand dynamic correlation network model.

10. A charging pile scheduling optimization system combining traffic flow prediction, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the charging pile scheduling optimization method incorporating traffic flow prediction as described in any one of claims 1 to 9 by executing the machine-executable instructions.

Images