Method and device for simulating spatial demand of automobile energy supplement based on multi-source spatio-temporal data

Abstract

The application discloses a vehicle energy supplement demand space simulation method and device based on multi-source space-time data, and relates to the field of urban planning and traffic energy systems. An unsupervised learning method is used to find patterns in the global grid unit, automatically identify energy supplement demand pattern areas with similar energy supplement demand behaviors and environmental characteristics; a causal analysis dataset is constructed based on each energy supplement demand pattern area, and a causal discovery algorithm is used to infer the core driving factor system; based on the spatial partitioning of the energy supplement demand pattern area, the energy supplement demand characteristics and the core driving factor system of each partition are used to adaptively select the target energy supplement demand prediction model for each partition. The application automatically identifies the core driving factors that have a direct causal effect on energy supplement demand, uses an adaptive modeling method constrained by causal structure in different pattern areas, effectively avoids the failure of a unified model in areas with significant heterogeneity, and improves the stability, interpretability and repeatability of the energy supplement demand space distribution results.

Description

Technical Field

[0001] This invention relates to the field of urban planning and transportation energy systems, and more specifically, to a method and apparatus for spatial simulation of vehicle refueling demand based on multi-source spatiotemporal data. Background Technology

[0002] With the continuous growth in the number of new energy vehicles, the demand for various types of energy replenishment for urban transportation energy infrastructure is constantly increasing. Analyzing the energy replenishment demand of new energy vehicles has become a crucial technical factor influencing the layout of urban energy facilities and traffic operations. Accurately identifying the intensity of new energy vehicle energy replenishment demand in different urban areas and at different times, thereby supporting the planning and layout of replenishment facilities and the upgrading of existing facilities, is a key technical challenge currently facing urban transportation and energy systems.

[0003] Current technologies for predicting and simulating urban new energy vehicle refueling demand primarily rely on macro-level data such as statistical yearbooks, vehicle ownership, and administrative divisions. Analysis is conducted at the district / county or traffic zone scale. While this reflects the overall urban demand level, it fails to depict micro-level spatial differences in roads, blocks, and functional units, thus failing to meet the practical needs of refined planning and the upgrading of existing infrastructure. Some technologies incorporate spatiotemporal data such as traffic flow and population activity intensity, but these still have significant limitations: First, they rely on manually preset indicator systems and driving factors, primarily based on empirical analysis, lacking the ability to automatically mine potential refueling demand patterns from high-dimensional data. Second, they often use a single model covering the entire city, failing to fully consider the high spatial heterogeneity caused by urban functions, traffic, and population distribution, resulting in insufficient model adaptability and prediction accuracy. Third, their ability to analyze the spatial formation mechanism and causal driving relationship of refueling demand is weak, leading to poor stability and interpretability of the results. Summary of the Invention

[0004] In view of this, the present invention provides a method and apparatus for spatial simulation of vehicle refueling demand based on multi-source spatiotemporal data, which is used to solve the problems of existing refueling demand prediction methods, such as coarse prediction scale and insufficient integration of multi-source urban operation elements.

[0005] To achieve the above objectives, the following solution is proposed: A method for spatial simulation of vehicle refueling demand based on multi-source spatiotemporal data includes: Collect multi-source spatiotemporal data related to vehicle refueling demand within the urban area; The study area is divided into grids, and multi-source spatiotemporal data are uniformly mapped to the grid system; Based on multi-source spatiotemporal data, a high-dimensional feature vector is constructed for each grid cell. An unsupervised learning method is used to perform pattern discovery on the grid cells across the entire domain, and the energy demand pattern regions with similar energy demand behaviors and environmental characteristics are automatically identified. Causal analysis datasets were constructed for each energy replenishment demand pattern region, and causal discovery algorithms were used to infer the core driving factor system. The research scope is spatially partitioned based on the energy demand pattern region, and the target energy demand prediction model is adaptively selected for each energy demand region based on the energy demand characteristics and core driving factor system of each energy demand region. The energy demand indexes output by the target energy demand prediction models corresponding to each energy demand zone are integrated to generate the spatial distribution results of energy demand and visualize them.

[0006] Preferably, before mapping the multi-source spatiotemporal data to the grid system, the method further includes: Data cleaning and preprocessing of the collected multi-source spatiotemporal data were performed, and a unified spatial reference coordinate system was established.

[0007] Preferably, the process of using unsupervised learning methods to discover patterns in global grid cells includes: The high-dimensional feature vectors of grid cells are clustered by a clustering algorithm, and grid cells with similar energy replenishment requirements are grouped into the same category to obtain grid pattern labels. Map high-dimensional feature vectors to low-dimensional latent representations; Clustering algorithms are used to cluster low-dimensional latent representations in the latent space to identify latent pattern types; Based on grid pattern labels and potential pattern types, all grid cells are divided into several energy replenishment demand pattern zones.

[0008] Preferably, the process of constructing causal analysis datasets based on each energy replenishment demand pattern region and using causal discovery algorithms to infer the core driving factor system includes: Within each energy replenishment demand pattern region, grid cells are used as analysis samples to construct a set of target variables and a set of candidate driving variables; Based on the spatial adjacency relationship of grid cells, spatial constraints are constructed between the target variable set and the candidate driving variable set; Under spatial constraints, a causal relationship structure between variables is inferred from the target variable set and the candidate driving variable set through a causal discovery algorithm, forming a causal network diagram of the energy replenishment demand driving mechanism. Identify variables in the causal network diagram that have a direct relationship with the energy replenishment demand representation, and use them as the core driving factor set of the current energy replenishment demand pattern region; By clarifying the order of action among the core driving factors and their influence path on the formation of energy replenishment demand, a core driving factor system is obtained.

[0009] Preferably, the causal discovery algorithm is one or more of the following: a constraint-based causal discovery algorithm, a causal structure learning algorithm based on a scoring function, and a causal inference method based on deep learning.

[0010] Preferably, the process of spatially partitioning the research area based on the energy replenishment demand pattern region includes: Based on the energy replenishment demand pattern area and urban spatial structure characteristics, the urban area is divided into zones, resulting in energy replenishment demand zones with clear spatial boundaries and relatively consistent energy replenishment demand formation mechanisms.

[0011] Preferably, the process of adaptively selecting a target energy demand prediction model for each energy demand zone based on its energy demand characteristics and core driving factor system includes: Using the core driving factor system as candidate input variables, and combining it with the high-dimensional feature vectors of grid units within the energy replenishment demand partition, a partition-level energy replenishment demand modeling dataset is constructed. Using the regional energy demand modeling dataset as input for each candidate model, the predictive performance, stability, and interpretability of different candidate models are evaluated through cross-validation or information criteria. The model with the best overall performance within the energy demand region is automatically selected as the target energy demand prediction model.

[0012] Preferably, the process of integrating the energy demand indices output by the target energy demand prediction models for each energy demand zone includes: Using the high-dimensional feature vectors of grid cells and the core driving factor system of the energy replenishment demand pattern region as inputs, the energy replenishment demand index is calculated for each grid cell in the energy replenishment demand partition through the target energy replenishment demand prediction model. The energy replenishment demand index of each grid cell is uniformly mapped to the grid system; The energy demand indices of different energy demand zones are integrated to generate spatial distribution results of energy demand.

[0013] Preferably, the process of visualizing the spatial distribution of energy replenishment demand includes: Based on the integrated spatial distribution results of energy replenishment demand, a spatial simulation map of energy replenishment demand is generated through a geographic information system platform. The spatial simulation map of energy replenishment demand includes at least one of a demand index distribution map, a hotspot distribution map, and a zoning comparison map.

[0014] A spatial simulation device for vehicle refueling demand based on multi-source spatiotemporal data, comprising: The data acquisition module collects multi-source spatiotemporal data related to vehicle refueling demand within the city. The data mapping module divides the study area into grids and maps multi-source spatiotemporal data to the grid system in a unified manner. The energy replenishment demand spatial pattern discovery module constructs a high-dimensional feature vector for each grid cell based on multi-source spatiotemporal data, and uses an unsupervised learning method to discover patterns in the entire grid cell, automatically identifying energy replenishment demand pattern areas with similar energy replenishment demand behaviors and environmental characteristics. The causal inference module of the driving mechanism constructs causal analysis datasets based on each energy replenishment demand pattern region and uses causal discovery algorithms to infer the core driving factor system. The spatial partitioning and adaptive modeling module divides the research scope into spatial partitions based on the energy demand pattern region, and adaptively selects the target energy demand prediction model for each energy demand partition based on the energy demand characteristics and core driving factor system of each energy demand partition. The energy replenishment demand spatial simulation output module integrates the energy replenishment demand index output by the target energy replenishment demand prediction model corresponding to each energy replenishment demand partition, generates the energy replenishment demand spatial distribution results, and outputs them in a visual format.

[0015] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: (1) The spatial simulation method for vehicle refueling demand based on multi-source spatiotemporal data provided by this invention constructs a high-dimensional feature vector for each grid cell based on multi-source spatiotemporal data, and uses an unsupervised learning method to perform pattern discovery on the grid cells across the entire domain, automatically identifying refueling demand pattern areas with similar refueling demand behaviors and environmental characteristics; constructs causal analysis datasets based on each refueling demand pattern area, and uses a causal discovery algorithm to infer the core driving factor system; spatially partitions the research scope based on the refueling demand pattern areas, and adaptively selects target refueling demand prediction models for each refueling demand partition based on the refueling demand characteristics and core driving factor system of each refueling demand partition. This invention characterizes the spatial heterogeneity of the refueling demand formation mechanism within a city through the technical chain of "refueling demand spatial pattern discovery - causal inference of driving mechanism - partition adaptive modeling". By automatically identifying spatial pattern areas with similar refueling demand formation mechanisms and identifying core driving factors that have a direct causal effect on refueling demand, it avoids the problem of the model being highly sensitive to the selection of indicators caused by relying solely on manually preset indicator systems or correlation analysis. Meanwhile, by adopting an adaptive modeling approach constrained by causal structure in different model regions, the failure of the unified model in regions with significant heterogeneity is effectively avoided, thereby improving the stability, interpretability, and repeatability of the simulation results of energy demand space from a mechanism perspective.

[0016] (2) This invention divides the study area into a grid and maps multi-source spatiotemporal data to the grid system. By constructing a unified micro-spatial analysis unit, this invention achieves a refined expression of energy replenishment demand at the street block scale, effectively avoiding the spatial scale bias caused by traditional analysis based on administrative districts or macro-regions. By constructing a unified micro-spatial analysis unit, a refined expression of energy replenishment demand at the street block scale is achieved, outputting high-resolution spatial distribution results of energy replenishment demand.

[0017] (3) This invention integrates the energy demand index output by the target energy demand prediction model corresponding to each energy demand zone, generates the spatial distribution results of energy demand, and visualizes them. This invention can output high-resolution, continuous, and mechanism-consistent spatial distribution results of energy demand, providing a reliable quantitative basis for the planning of new energy vehicle charging facilities, the layout of roadside energy replenishment stations, and the transformation of existing gas stations into integrated energy stations. Attached Figure Description

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

[0019] Figure 1 A flowchart of a method for spatial simulation of vehicle refueling demand based on multi-source spatiotemporal data provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a spatial simulation device for vehicle refueling demand based on multi-source spatiotemporal data, provided in an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] First, combined Figure 1 The present invention introduces a spatial simulation method for vehicle refueling demand based on multi-source spatiotemporal data, as provided in the embodiments of the present invention. Figure 1 As shown, the simulation methods include: Step S01: Collect multi-source spatiotemporal data related to vehicle refueling demand within the urban area.

[0022] Specifically, multi-source spatiotemporal data includes geographic location data, road traffic data, population activity data, supporting infrastructure data, and energy and logistics data.

[0023] Geographic location data: city centers, rural settlements, suburban counties, etc.

[0024] Road traffic data: OSM road network, Gaode Map traffic conditions during peak commuting hours, etc.

[0025] Population activity data: distribution of active population via mobile phone signaling, population heat map from Baidu Smart Eye, and nighttime light imagery.

[0026] Supporting facilities data: Points of interest (POI) data, including residential, corporate, commercial and public service facilities.

[0027] Energy and logistics data: Information on existing gas stations, logistics channels, and logistics nodes.

[0028] Step S02: Divide the study area into a grid and map the multi-source spatiotemporal data to the grid system.

[0029] Specifically, the study area is divided into a 250m×250m standard grid. This grid is used as the basic unit for spatial analysis. Through GIS platform technologies such as resampling, spatial connection, and raster-to-vector conversion, multi-source spatiotemporal data of different resolutions, such as road network and population heat data, are uniformly mapped to this standard grid system, thereby obtaining grid-level information on road traffic, population activities, and facility attributes.

[0030] Step S03: Construct a high-dimensional feature vector for each grid cell based on multi-source spatiotemporal data, and use an unsupervised learning method to perform pattern discovery on the grid cells across the entire domain, automatically identifying energy replenishment demand pattern regions with similar energy replenishment demand behaviors and environmental characteristics.

[0031] Specifically, high-dimensional urban operation characteristics at the grid scale are constructed from multi-source heterogeneous spatiotemporal data, and potential spatial patterns of energy replenishment demand in the city are automatically identified through unsupervised learning, providing a basis for subsequent causal structure inference and adaptive zoning modeling.

[0032] (1) Construction of high-dimensional feature vectors At a standard grid scale of 250m×250m, based on multi-source spatiotemporal data (including raw data on geographical location, road traffic, population activity, urban infrastructure, logistics and transportation, and their derived feature variables), each grid cell is... g Constructing high-dimensional feature vectors:

[0033] in Let g be the high-dimensional eigenvector of the grid. For the first i One characteristic, i= 1, 2, 3... k . k This refers to the number of feature dimensions. The organization and combination of features within a high-dimensional feature vector are not pre-defined as a fixed-dimensional system, but rather utilize a data-driven approach for pattern discovery and factor selection. The types of features can be expanded according to data availability and the target task, and must include at least one of the following types or a combination thereof: a. Spatial location and accessibility features: distance from the grid center to the city center, distance from suburban counties and towns, distance from logistics nodes, distance from major logistics channels, distance from major roads, etc.

[0034] b. Road traffic flow characteristics: peak / off-peak traffic congestion index and change rate, road accessibility, distance from grid center point to expressway entrances and exits, road network length density, intersection density, proportion of roads of different grades, etc.

[0035] c. Characteristics of population activity intensity: mobile phone signaling activity, heat map distribution during peak commuting hours, nighttime light intensity, daytime / nighttime population density, population job-to-residence ratio, distribution of different age groups / occupations, social media check-in activity, etc.

[0036] d. Characteristics of urban supporting facilities: density and distribution diversity index of various POIs (residential communities, companies, commercial complexes, restaurants, leisure and entertainment, education, medical care, public services, logistics stations, industrial parks, etc.).

[0037] Population activity dimension: active population distribution density within the grid, heat distribution during peak commuting hours, and nighttime light intensity.

[0038] Surrounding amenities: Total density of POIs within the grid, density of residential POIs, density of corporate POIs, and density of commercial market POIs.

[0039] Logistics and transportation dimension: the distance from the grid center point to logistics channels, main roads, and logistics nodes.

[0040] The aforementioned features can be calculated using known spatial analysis methods or customized algorithms. For example, distance-related features can be calculated and standardized using Euclidean distance or shortest path distance; traffic congestion indicators can be calculated using a weighted average road congestion index. This invention uses distance-related features, traffic congestion indicators, and road accessibility indicators as examples for illustration. Specific calculation methods may include, but are not limited to, the following forms: 1. Distance-related indicators Based on location data of the city's central urban area, rural residential areas, suburban county towns, all expressway and highway entrances / exits, logistics channels, major roads, and logistics nodes, using the center point of a 250m × 250m grid as a reference, the Euclidean distance between each grid center point and the target is calculated. The distance values ​​are then standardized to obtain a node proximity index. The formula for calculating the node proximity index is as follows:

[0041] Among them, P g D represents the node proximity of grid g (range 0–1, larger values ​​indicate that the grid center is closer to the target); g D represents the Euclidean distance from the center point of grid g to the target being measured; min With D max These represent the minimum and maximum distance values ​​within the city limits. The measurement targets include the central urban area, rural residential areas, suburban county towns, highway and expressway entrances and exits, logistics channels, and major roads or logistics nodes.

[0042] 2. Traffic congestion indicators Real-time traffic data from typical weekday evening peak hours (17:00–19:00) and off-peak hours (14:00–16:00) were used to record the maximum congestion value for each road during these periods. Spatial segments of each road within different grid cells were identified, and the length of each segment within the grid cell was calculated. The length-weighted average method is used to aggregate the road-level congestion index into a grid-level traffic flow index:

[0043] in, This represents the traffic flow index during peak commuting hours for grid g. This represents the congestion index of road segment r; This represents the length of road segment r within grid g; This represents the set of all road segments that fall within grid g.

[0044] 3. Road accessibility indicators Based on urban road centerline data, spatial syntactic analysis is performed to calculate the global integration degree of all road axes in the city. The integration degree value of each road segment is mapped to the grid cells it covers. The integration degree values ​​of road segments are aggregated into a grid-level road accessibility index using a length-weighted average method.

[0045] in, Indicates the road reachability of grid g; This represents the integration degree value of road segment r; This represents the length of road segment r within grid g; This represents the set of all road segments that fall within grid g.

[0046] (2) Discovery of spatial patterns of energy replenishment demand First, based on the high-dimensional feature vector F g In this embodiment of the invention, an unsupervised learning method is used to discover patterns in global grid cells to automatically identify spatial pattern types with similar energy replenishment demand behaviors and environmental characteristics.

[0047] Clustering algorithms are used to cluster the high-dimensional feature vectors of grid cells, grouping grid cells with similar energy replenishment requirements into the same category to obtain grid pattern labels:

[0048] in ，f cluster This represents the clustering mapping, where Θ is the clustering parameter. It is a high-dimensional feature vector.

[0049] Clustering algorithms can be one or a combination of density-based clustering algorithms (DBSCAN, ST-DBSCAN), hierarchical clustering algorithms (agglomerative, divisive), K-means, spectral clustering, etc. The clustering results can form several energy demand pattern zones, such as residential, commercial, commuter, and logistics hub zones.

[0050] Then, to extract more expressive and stable latent structures from high-dimensional features, this invention can also employ representation learning methods to transform the high-dimensional feature vectors... Mapping to low-dimensional latent representation :

[0051] in, For representation learning models, These are the model parameters. Representation learning methods include, but are not limited to: a. Autoencoder: Learns low-dimensional latent variables by reconstructing loss; b. Graph Neural Networks (GNNs): Construct graph structures using grid adjacency relationships, and learn latent representations that integrate spatial neighborhood information on the graph; c. Other deep representation models: contrastive learning, variational autoencoders, etc.

[0052] In obtaining low-dimensional latent representations Subsequently, the present invention further performs clustering or density estimation in the latent space to identify latent pattern types. In this latent space, grids with similar energy replenishment needs will be close to each other, thereby revealing different energy replenishment demand patterns through the analysis of latent variables.

[0053] High-dimensional feature vectors, constructed from multi-source spatiotemporal data, directly characterize the energy replenishment demand-related features of grid cells and serve as the primary basis for pattern segmentation. Low-dimensional latent representations extract latent structures from high-dimensional features using representation learning methods, reducing noise interference and enhancing the stability of pattern recognition. Finally, based on grid pattern labels and latent pattern types, all grid cells are divided into several energy replenishment demand pattern zones. Each of these zones represents a similar energy replenishment demand formation mechanism with similar traffic behavior, population activity, and infrastructure environment characteristics. This can generate: a. Grid Pattern Label Map: Outputs a pattern category for each grid cell. .

[0054] b. Pattern region boundaries and statistical characteristics: Output the spatial range, core feature distribution and typical time period behavior characteristics of each pattern region.

[0055] c. Pattern region boundaries and statistical characteristics: Output the spatial range, core feature distribution and typical time period behavior characteristics of each pattern region.

[0056] Step S04: Construct causal analysis datasets based on each energy replenishment demand pattern region, and use causal discovery algorithms to infer the core driving factor system.

[0057] Specifically, based on the findings of the spatial pattern discovery of energy replenishment demand, the key driving factors influencing the formation of different energy replenishment demand patterns and their causal interaction structures are automatically inferred from multi-source high-dimensional spatiotemporal data. Through causal discovery algorithms, core driving factors with direct causal impact on energy replenishment demand are identified, and their action paths are clarified, providing structural constraints and a variable basis for subsequent adaptive modeling of different regions.

[0058] (1) Causal analysis objects and modeling data construction Based on the energy replenishment demand spatial pattern regions identified in the pattern discovery module, this invention constructs a causal analysis dataset for each type of energy replenishment demand pattern region.

[0059] Within each pattern region, using grid cells as analysis samples, the following set of variables is constructed: a. Set of target variables: Energy demand characteristics used to characterize the intensity of energy demand, including but not limited to grid-level total energy demand, charging facility utilization rate, or calibrated energy demand behavior intensity indicators.

[0060] b. Candidate driving variable set: Traffic operation characteristics, population activity characteristics, facility and environment characteristics and their time-series derived characteristics from the high-dimensional spatiotemporal feature set.

[0061] This invention decouples "pattern discovery" from "causal inference," enabling causal analysis to be performed within a more homogeneous spatial subset, thus avoiding interference from the overall spatial heterogeneity of the city on the causal identification results.

[0062] (2) Causal discovery algorithm and causal structure inference For the constructed pattern-level analysis dataset, a causal discovery algorithm is used to automatically infer the causal relationship structure between variables in two categories, forming a causal network diagram of the energy replenishment demand-driven mechanism.

[0063] Before inferring causal structure, constraints are constructed between variables based on the spatial adjacency relationships of urban spatial analysis units to limit the search space for causal inference and avoid generating causal connections without actual physical meaning in high-dimensional feature data.

[0064] Under the aforementioned spatial constraints, causal discovery algorithms include, but are not limited to, any one or a combination of the following: a. Constraint-Based Causal Discovery Algorithm: This algorithm employs a causal discovery method based on conditional independence testing. By examining the independence relationships of variables under given conditions of other variables, a causal graph structure is gradually constructed. The algorithm includes, but is not limited to, the PC algorithm and the FCI algorithm. This type of algorithm identifies driving factors that have a direct causal effect on the energy replenishment demand, and distinguishes between direct influence paths and indirect influence paths generated through mediating variables. This provides a basis for variable selection and structural constraints in subsequent energy replenishment demand prediction models.

[0065] b. Causal Structure Learning Algorithm Based on Scoring Function: This method employs a causal discovery approach based on scoring function optimization. It determines the causal relationships between variables by searching for the structure that optimizes the overall score of the causal graph. Algorithms include, but are not limited to, the GES algorithm and the NOTEARS algorithm. This type of method can identify stable causal structures in urban high-dimensional spatiotemporal feature data scenarios, preventing variables with no causal contribution from entering the energy replenishment demand prediction model, and improving the stability and interpretability of the inference results of the driving mechanism.

[0066] c. Deep Learning-Based Causal Inference Method: For high-dimensional time-series feature data containing significant time dependencies, this invention can also employ a deep learning-based causal inference model under the aforementioned spatial constraints to model the dynamic causal relationships between variables. The method includes, but is not limited to, a neural network-based causal structure learning model to characterize the role of the energy replenishment demand driving mechanism in the spatial dimension and to provide structural constraints for the partitioned adaptive prediction model.

[0067] Using a causal discovery algorithm, a causal network diagram describing the causal relationships between variables is obtained within each energy replenishment demand pattern region. Further, based on the causal network diagram, a core driving factor system is constructed. The core driving factor system processing steps are as follows: First, identify variables that have a direct relationship with the energy replenishment demand representation in the causal network, and use them as the core driving factor set for this energy replenishment demand pattern region. Then, redundant variables that are only superficially related to energy replenishment demand but do not have a direct or indirect causal path can be eliminated. Finally, clarify the order of action among the core driving factors and their influence path on the formation of energy replenishment demand.

[0068] Through the above process, a system of driving factors based on causal structure can be formed that can be dynamically adjusted as data is updated, rather than a fixed set of artificial indicators.

[0069] Step S05: Spatial partitioning of the research scope based on the energy replenishment demand pattern region, and adaptive selection of target energy replenishment demand prediction model for each energy replenishment demand region based on the energy replenishment demand characteristics and core driving factor system of each energy replenishment demand region.

[0070] Specifically, based on the high heterogeneity of urban space in terms of energy replenishment demand formation mechanisms, an adaptive energy replenishment demand prediction modeling method based on spatial partitioning is constructed. By dividing the study area into several sub-regions with relatively homogeneous energy replenishment demand characteristics and driving mechanisms, and selecting and training the optimal prediction model in each sub-region, a refined simulation of the spatial distribution of energy replenishment demand is achieved.

[0071] (1) Urban spatial zoning Based on the identified energy replenishment demand pattern areas, and combined with urban spatial structure characteristics (such as administrative divisions, urban planning functional zones, or natural zones obtained through geographical feature clustering), the study area is spatially partitioned to form several sub-regions with relatively homogeneous energy replenishment demands. Through spatial partitioning, a set of energy replenishment demand zones with clear spatial boundaries and relatively consistent energy replenishment demand formation mechanisms is formed.

[0072] Spatial partitioning can be achieved using any of the following methods or a combination thereof: a. Partitioning based on energy demand patterns: The clustering results in the energy demand pattern area are directly used as the initial partition, so that the grid cells in the same partition have high similarity in energy demand behavior characteristics and environmental characteristics.

[0073] b. Zoning based on spatial heterogeneity identification: Introduce spatial analysis methods to identify the spatial non-stationarity characteristics of energy replenishment demand or its driving factors. For example, use the local parameter estimation of the geographically weighted regression (GWR) model to identify areas with significant spatial heterogeneity, so as to form spatial sub-regions with relatively consistent energy replenishment demand mechanisms.

[0074] c. Zoning combined with administrative or functional zoning: Combine the zoning results of energy demand patterns obtained through data-driven approaches with existing administrative or functional zoning, so as to improve the real-world interpretability and planning usability of the zoning results while maintaining data homogeneity.

[0075] (2) Selection and training of regional adaptive prediction model For each energy replenishment demand zone, based on the energy replenishment demand characteristics within that zone and the core driving factors identified in Module 3, the optimal energy replenishment demand prediction model is adaptively selected and trained, achieving an adaptive modeling mechanism of "different models for the same city but different zones". Specific steps include: First, using the core driving factor system as candidate input variables, and combining it with the high-dimensional feature vectors of the corresponding grid units within the energy replenishment demand partition, a partition-level energy replenishment demand modeling dataset is constructed.

[0076] The partition-level energy replenishment demand modeling dataset is used as input for each candidate model, for performance evaluation, model comparison, and model training. A candidate model set is constructed based on the feature complexity of different energy replenishment demand partitions. The candidate model set includes, but is not limited to: a. Linear or generalized linear regression model: used for partitions with relatively clear driving relationships and relatively simple structures.

[0077] b. Tree-based ensemble models (such as random forests and XGBoost): used to characterize nonlinear relationships and variable interactions.

[0078] c. Deep learning-based models (such as LSTM, CNN-LSTM, and Graph Neural Networks GNN): used to handle partitions with significant temporal dependencies, spatial correlations, or complex nonlinear structures.

[0079] d. Spatial regression or geographic weighted model: used to further characterize the local spatial heterogeneity within a partition.

[0080] Within each energy demand partition, the predictive performance, stability, and interpretability of different candidate models are evaluated using methods such as cross-validation or information criteria. The model with the best overall performance within that partition is automatically selected as the target energy demand prediction model. The predictive performance evaluation metrics may include error metrics, goodness-of-fit metrics, or a combination thereof, to ensure that the selected model has stable predictive capabilities within that partition.

[0081] Step S06: Integrate the energy demand index output by the target energy demand prediction model corresponding to each energy demand zone to generate the spatial distribution result of energy demand and visualize it.

[0082] Specifically, based on the target energy demand prediction model trained within each energy demand zone, energy demand calculations are performed on all grid units within the study area, and the results are integrated to generate simulation results of the spatial distribution of energy demand at the city scale.

[0083] (1) Calculation of Energy Demand Index For each grid cell, based on the target energy demand prediction model of its corresponding energy demand partition, and using the high-dimensional feature vector of that grid cell and the core driving factor system identified within the corresponding energy demand partition, the grid-level energy demand index is calculated:

[0084] in, The energy replenishment demand index for grid g. Let g be the high-dimensional eigenvector of the grid. Zoned for energy replenishment demand Z The target energy replenishment demand prediction model.

[0085] This approach avoids using uniform weights or models to estimate the city's grid, thus effectively reflecting the differences in the formation mechanisms of energy replenishment demand in different spatial areas.

[0086] (2) Integration of Energy Demand Index The grid-level energy replenishment demand index output by each zoning model is uniformly mapped to a 250m×250m standard grid system, and the prediction results of different zoning are integrated to form continuous and comparable spatial distribution results of urban energy replenishment demand. To ensure the consistency and interpretability of the results, this invention can normalize or scale-calibrate the output results of different zoning models, so that the demand intensity results of different zoning are under a unified analytical framework.

[0087] (3) Spatial visualization and decision support output Based on the integrated spatial distribution of energy replenishment demand, a spatial simulation map of energy replenishment demand is generated through a geographic information system platform. This simulation map includes, but is not limited to, demand index distribution maps, hotspot distribution maps, or zonal comparison maps. It aims to identify areas with high energy replenishment demand, areas with potential demand growth, and areas with weak existing infrastructure services within the city, providing quantitative decision support for the planning of new energy vehicle charging facilities, roadside energy replenishment layout, and the transformation of existing energy facilities into integrated energy service stations.

[0088] This invention proposes a technical solution for fine-grained simulation of the spatial distribution of vehicle energy replenishment demand at the urban micro-scale. By uniformly modeling and analyzing multi-source heterogeneous spatiotemporal data such as urban road traffic, population activity, supporting facilities, and logistics transportation, a high-dimensional spatial feature vector is constructed under a unified micro-spatial unit. Unsupervised learning methods are used to automatically identify spatial patterns with similar energy replenishment demand formation mechanisms. Furthermore, causal structure inference identifies the core driving factors influencing different energy replenishment demand patterns and their relationships. Addressing the high heterogeneity of urban space in terms of energy replenishment demand formation mechanisms, this invention employs a strategy of first partitioning and then adaptive modeling. Within different spatial partitions, energy replenishment demand simulation models matching their driving mechanisms are constructed, thereby achieving stable modeling and quantitative expression of the spatial distribution of urban energy replenishment demand.

[0089] This invention can improve the accuracy and modeling stability of spatial simulation of energy replenishment demand at the microscale, and provide quantifiable and implementable technical support for the planning of new energy vehicle charging facilities, the transformation analysis of existing energy facilities, and the optimization of the spatial layout of urban transportation energy infrastructure.

[0090] To avoid patterns that are "similar in features but spatially discrete," this embodiment of the invention, based on the aforementioned embodiments, may introduce at least one constraint mechanism during the clustering process: Spatial adjacency constraint: Spatial adjacent grids are only allowed to be grouped into the same cluster if their similarity meets a threshold. Spatiotemporal joint distance metric: Considers both feature distance and spatial / temporal proximity in similarity metrics; Post-cluster processing: Regularize isolated small patches (e.g., merge patches based on minimum patch area threshold) to ensure that the pattern region has spatial connectivity available for planning.

[0091] In addition, to better simulate demand space, before uniformly mapping multi-source spatiotemporal data of different resolutions to the grid system, the following steps are also included: Data cleaning and preprocessing of the collected multi-source spatiotemporal data were performed, and a unified spatial reference coordinate system was established.

[0092] Specifically, the collected data will undergo data cleaning (removing invalid and duplicate data) and unified spatial reference system conversion (identifying the source coordinate system and using ArcGIS projection tools or Python library coordinate conversion functions to uniformly convert the collected multi-source data into the WGS1984 coordinate system) to build a unified spatial analysis database.

[0093] The data processing and spatial analysis steps of this invention can be implemented by any geographic information system or data analysis tool with corresponding functions, and are not limited to specific software or platforms.

[0094] The following describes the vehicle energy demand spatial simulation device based on multi-source spatiotemporal data provided in the embodiments of the present invention. The vehicle energy demand spatial simulation device based on multi-source spatiotemporal data described below can be referred to in correspondence with the vehicle energy demand spatial simulation method based on multi-source spatiotemporal data described above.

[0095] Combination Figure 2 This paper introduces a spatial simulation device for vehicle refueling demand based on multi-source spatiotemporal data, such as... Figure 2 As shown, the vehicle refueling demand spatial simulation device based on multi-source spatiotemporal data may include: Data acquisition module 100 collects multi-source spatiotemporal data related to vehicle refueling demand within the urban area; The data mapping module 200 divides the study area into grids and maps multi-source spatiotemporal data to the grid system in a unified manner. The energy replenishment demand spatial pattern discovery module 300 constructs a high-dimensional feature vector for each grid cell based on multi-source spatiotemporal data, and uses an unsupervised learning method to discover patterns in the entire grid cell, automatically identifying energy replenishment demand pattern areas with similar energy replenishment demand behaviors and environmental characteristics. The causal inference module 400 of the driving mechanism constructs causal analysis datasets based on each energy replenishment demand mode region and uses causal discovery algorithms to infer the core driving factor system. The Spatial Partitioning and Adaptive Modeling Module 500 spatially partitions the research scope based on the energy replenishment demand pattern region, and adaptively selects the target energy replenishment demand prediction model for each energy replenishment demand region based on the energy replenishment demand characteristics and core driving factor system of each energy replenishment demand region. The energy replenishment demand spatial simulation output module 600 integrates the energy replenishment demand index output by the target energy replenishment demand prediction model corresponding to each energy replenishment demand partition, generates the energy replenishment demand spatial distribution results, and outputs them in a visual format.

[0096] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0097] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0098] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for spatial simulation of vehicle refueling demand based on multi-source spatiotemporal data, characterized in that, include: Collect multi-source spatiotemporal data related to vehicle refueling demand within the urban area; The study area is divided into grids, and multi-source spatiotemporal data are uniformly mapped to the grid system; Based on multi-source spatiotemporal data, a high-dimensional feature vector is constructed for each grid cell. An unsupervised learning method is used to perform pattern discovery on the grid cells across the entire domain, and the energy demand pattern regions with similar energy demand behaviors and environmental characteristics are automatically identified. Causal analysis datasets were constructed for each energy replenishment demand pattern region, and causal discovery algorithms were used to infer the core driving factor system. The research scope is spatially partitioned based on the energy demand pattern region, and the target energy demand prediction model is adaptively selected for each energy demand region based on the energy demand characteristics and core driving factor system of each energy demand region. The energy demand index output by the target energy demand prediction model corresponding to each energy demand zone is integrated to generate the spatial distribution results of energy demand and visualize them. The process of constructing causal analysis datasets based on each energy replenishment demand pattern region and using causal discovery algorithms to infer the core driving factor system includes: Within each energy replenishment demand pattern region, grid cells are used as analysis samples to construct a set of target variables and a set of candidate driving variables; Based on the spatial adjacency relationship of grid cells, spatial constraints are constructed between the target variable set and the candidate driving variable set; Under spatial constraints, a causal relationship structure between variables is inferred from the target variable set and the candidate driving variable set through a causal discovery algorithm, forming a causal network diagram of the energy replenishment demand driving mechanism. Identify variables in the causal network diagram that have a direct relationship with the energy replenishment demand representation, and use them as the core driving factor set of the current energy replenishment demand pattern region; Eliminate redundant variables; By clarifying the order of action among the core driving factors and their influence path on the formation of energy replenishment demand, a core driving factor system is obtained.

2. The method for spatial simulation of vehicle refueling demand based on multi-source spatiotemporal data according to claim 1, characterized in that, Before mapping multi-source spatiotemporal data to a grid system, the process also includes: Data cleaning and preprocessing of the collected multi-source spatiotemporal data were performed, and a unified spatial reference coordinate system was established.

3. The method for spatial simulation of vehicle refueling demand based on multi-source spatiotemporal data according to claim 1, characterized in that, The process of pattern discovery in global grid cells using unsupervised learning methods includes: The high-dimensional feature vectors of grid cells are clustered by a clustering algorithm, and grid cells with similar energy replenishment requirements are grouped into the same category to obtain grid pattern labels. Map high-dimensional feature vectors to low-dimensional latent representations; Clustering algorithms are used to cluster low-dimensional latent representations in the latent space to identify latent pattern types; Based on grid pattern labels and potential pattern types, all grid cells are divided into several energy replenishment demand pattern zones.

4. The method for spatial simulation of vehicle refueling demand based on multi-source spatiotemporal data according to claim 1, characterized in that, The causal discovery algorithm is one or more of the following: constraint-based causal discovery algorithm, causal structure learning algorithm based on scoring function, and causal inference method based on deep learning.

5. The method for spatial simulation of vehicle refueling demand based on multi-source spatiotemporal data according to claim 1, characterized in that, The process of spatially partitioning the research scope based on the energy replenishment demand pattern region includes: Based on the energy replenishment demand pattern area and urban spatial structure characteristics, the urban area is divided into zones, resulting in a set of energy replenishment demand zones with clear spatial boundaries and relatively consistent energy replenishment demand formation mechanisms.

6. The method for spatial simulation of vehicle refueling demand based on multi-source spatiotemporal data according to claim 5, characterized in that, The process of adaptively selecting a target energy demand prediction model for each energy demand region based on its energy demand characteristics and core driving factor system includes: Using the core driving factor system as candidate input variables, and combining it with the high-dimensional feature vectors of grid units within the energy replenishment demand partition, a partition-level energy replenishment demand modeling dataset is constructed. Using the regional energy demand modeling dataset as input for each candidate model, the predictive performance, stability, and interpretability of different candidate models are evaluated through cross-validation or information criteria. The model with the best overall performance within the energy demand region is automatically selected as the target energy demand prediction model.

7. The method for spatial simulation of vehicle refueling demand based on multi-source spatiotemporal data according to claim 1, characterized in that, The process of integrating the energy demand indices output by the target energy demand prediction models for each energy demand zone includes: Using the high-dimensional feature vectors of grid cells and the core driving factor system of the energy replenishment demand pattern region as inputs, the energy replenishment demand index is calculated for each grid cell in the energy replenishment demand partition through the target energy replenishment demand prediction model. The energy replenishment demand index of each grid cell is uniformly mapped to the grid system; The energy demand indices of different energy demand zones are integrated to generate spatial distribution results of energy demand.

8. The method for spatial simulation of vehicle refueling demand based on multi-source spatiotemporal data according to claim 1, characterized in that, The process of visualizing the spatial distribution of energy replenishment demand includes: Based on the integrated spatial distribution results of energy replenishment demand, a spatial simulation map of energy replenishment demand is generated through a geographic information system platform. The spatial simulation map of energy replenishment demand includes at least one of a demand index distribution map, a hotspot distribution map, and a zoning comparison map.

9. A spatial simulation device for vehicle refueling demand based on multi-source spatiotemporal data, characterized in that, include: The data acquisition module collects multi-source spatiotemporal data related to vehicle refueling demand within the city. The data mapping module divides the study area into grids and maps multi-source spatiotemporal data to the grid system in a unified manner. The energy replenishment demand spatial pattern discovery module constructs a high-dimensional feature vector for each grid cell based on multi-source spatiotemporal data, and uses an unsupervised learning method to discover patterns in the entire grid cell, automatically identifying energy replenishment demand pattern regions with similar energy replenishment demand behaviors and environmental characteristics. The causal inference module of the driving mechanism constructs causal analysis datasets based on each energy replenishment demand pattern region and uses causal discovery algorithms to infer the core driving factor system. The spatial partitioning and adaptive modeling module divides the research scope into spatial partitions based on the energy demand pattern region, and adaptively selects the target energy demand prediction model for each energy demand partition based on the energy demand characteristics and core driving factor system of each energy demand partition. The energy replenishment demand spatial simulation output module integrates the energy replenishment demand index output by the target energy replenishment demand prediction model corresponding to each energy replenishment demand partition, generates the energy replenishment demand spatial distribution results, and visualizes them. The process of constructing causal analysis datasets based on each energy replenishment demand pattern region and using causal discovery algorithms to infer the core driving factor system includes: Within each energy replenishment demand pattern region, grid cells are used as analysis samples to construct a set of target variables and a set of candidate driving variables; Based on the spatial adjacency relationship of grid cells, spatial constraints are constructed between the target variable set and the candidate driving variable set; Under spatial constraints, a causal relationship structure between variables is inferred from the target variable set and the candidate driving variable set through a causal discovery algorithm, forming a causal network diagram of the energy replenishment demand driving mechanism. Identify variables in the causal network diagram that have a direct relationship with the energy replenishment demand representation, and use them as the core driving factor set of the current energy replenishment demand pattern region; Eliminate redundant variables; By clarifying the order of action among the core driving factors and their influence path on the formation of energy replenishment demand, a core driving factor system is obtained.

Images