Energy supplement facility positioning method and system

Abstract

The application discloses a kind of energy supplement facilities positioning method and system. Obtain multi-modal multi-source vector data and carry out pretreatment;The spatial constraint system of demarcating energy supplement facilities and energy supplement demand distribution boundary;According to the topological structure of road network, select initial candidate site and extract characteristic attribute based on spatial constraint system, generate corresponding multidimensional space characteristic vector;Construct simulation positioning scene to extract the spatial characteristic vector of simulation demand point set, combine heuristic optimization result to generate corresponding construction value score label, obtain labeled dataset;Neural network is trained using labeled dataset, the mapping relationship of space characteristic and construction value is established, and the model parameter trained is stored as offline value evaluator;Using offline value evaluator, the initial candidate site is carried out quickly value retrieval screening, obtain high-value candidate site pool and carry out heuristic optimization, output the final energy supplement facility layout scheme.

Description

Technical Field

[0001] This invention relates to the field of urban resource scheduling and energy replenishment planning technology, and more specifically to a method and system for locating energy replenishment facilities. Background Technology

[0002] With the accelerated construction of urban energy infrastructure, how to scientifically locate energy facilities to balance urban planning constraints and user needs has become an important issue in the field of energy internet. Traditional energy facility layout planning mainly relies on analytical modeling or simple heuristic algorithms. Analytical modeling methods construct mathematical models by integrating factors such as administrative boundaries, geographic information, and population distribution. However, when dealing with large-scale urban areas, the coupling relationships of actual geographical constraints (such as complex water systems and restrictions on non-construction land) are extremely complex, making it difficult for the models to accurately reproduce the complexity of the physical environment, thus limiting the engineering feasibility of the planning results.

[0003] For solving challenging problems under complex constraints, existing heuristic algorithms (such as particle swarm optimization or genetic algorithms) have demonstrated strong nonlinear processing capabilities. However, the problem of locating energy replenishment facilities is a large-scale combinatorial optimization problem, and heuristic algorithms are highly dependent on the number of iterations during the search process. When the number of candidate points surges, the algorithms are prone to getting trapped in local optima or facing a "computational explosion" problem, and they struggle to demonstrate good generalization performance under different urban constraint configurations.

[0004] Furthermore, although some improved solutions attempt to use neural networks for demand forecasting or constraint boundary correction, in existing fusion strategies, deep learning is often only used as an auxiliary means, failing to deeply participate in the complete decision-making chain from "spatial value assessment" to "global layout optimization." Especially when dealing with refined deployment across the entire city, existing methods often reduce the research area or simplify influencing factors (such as ignoring the interaction between dynamic population distribution and multimodal spatial data), resulting in a lack of flexibility in the assessment system and making it difficult to meet the high precision and adaptability requirements of modern smart cities for energy replenishment network deployment.

[0005] Therefore, how to provide a method for locating energy replenishment facilities that can integrate multi-dimensional spatial constraints and take into account both offline value assessment efficiency and online global search capabilities is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a method and system for locating energy replenishment facilities. First, a spatial constraint system for energy replenishment facilities is constructed. By integrating multimodal data on administrative boundaries and water system distribution, multidimensional evaluation features reflecting the potential for facility deployment are extracted. Subsequently, in the offline evaluation phase, a training set is constructed using a sample generation mechanism driven by the Dung Beetle Optimization (DBO) algorithm. A Deep Neural Network (DNN) is used to learn the mapping relationship between the construction value and spatial features of each candidate site, thereby quantifying the site construction score across the entire city. In the online optimization phase, the system calls the trained evaluation model for rapid screening based on a preset initial scheme and iteratively searches for the globally optimal solution among candidate sites using a heuristic search algorithm. This invention decouples evaluation accuracy from search efficiency, significantly improving the scientific rigor and generalization adaptability of energy replenishment facility layout schemes while satisfying complex spatial constraints.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A method for locating energy replenishment facilities, comprising: Acquire vector data of urban administrative boundaries and vector data of water system distribution, and perform preprocessing; A spatial constraint system is defined based on the preprocessed data to delineate the distribution boundaries of energy replenishment facilities and energy replenishment demand. Initial candidate sites are selected based on the road network topology, and feature attributes reflecting the site service efficiency and competitiveness are extracted based on the spatial constraint system to generate corresponding multi-dimensional spatial feature vectors. A simulated positioning scenario is constructed to extract spatial feature vectors of a set of simulated demand points. These vectors are then combined with a heuristic optimization algorithm to generate corresponding construction value scoring labels, resulting in a labeled dataset. A neural network is trained using labeled datasets to establish a mapping relationship between spatial features and construction value, and the trained model parameters are stored as an offline value evaluator. An offline value evaluator was used to quickly retrieve and screen the initial candidate sites to obtain a pool of high-value candidate sites. Heuristic optimization is performed on the pool of high-value candidate sites to output the final energy replenishment facility layout scheme.

[0008] Preferably, the city's administrative boundary vector data and water system distribution vector element data are obtained separately, and the geometric attributes and attribute statistics of each modality data are read using the corresponding geographic information engine; Using the city's administrative boundary vector data as a reference system, coordinate projection transformation is performed on the water system distribution vector element data. Perform topology checks on urban administrative boundary vector data and water system distribution vector element data, correct topology errors, obtain standardized spatial constraint metadata, and form a spatial constraint system that defines the permitted construction scope of facilities through spatial analysis operations.

[0009] Preferably, generating the corresponding multidimensional spatial feature vector includes: Based on the urban road network topology of the target area, each road intersection node is extracted, the betweenness centrality feature of each road intersection node is calculated, and probability sampling is performed according to the betweenness centrality from high to low to select several nodes as initial candidate sites. The set of urban points of interest within the target area is obtained as the original demand carrier. Population density spatial distribution data is introduced as a weight gain factor. The set of urban points of interest is weighted and randomly sampled according to the regional population density intensity of each urban point of interest location to determine the set of energy replenishment demand points. Based on the aforementioned spatial constraint system, the access cost and service coordination efficiency of each candidate energy replenishment facility site are quantified; through spatial overlay and buffer analysis, the coverage contribution of each initial candidate site to surrounding demand points is calculated, and the coordination coverage efficiency features reflecting the complementarity of spatial layout are extracted. Based on the access cost ranking and corresponding collaborative coverage efficiency components within the collaborative service cluster to which each initial candidate site belongs, a function mapping relationship reflecting the competitive advantages and disadvantages among sites is constructed. The relative attractiveness of the initial candidate sites within the shared service area is quantified through the mapping relationship, and a multi-dimensional spatial feature vector is extracted.

[0010] Preferably, the process of obtaining the labeled dataset includes: By repeatedly executing random sampling procedures for urban interest point sets and road node sets independently, several sets of simulated demand point sets and simulated candidate site sets with spatial distribution differences are generated. Each set of demand points and candidate sites constitutes an independent simulated positioning scenario. In various scenarios, the dung beetle optimization algorithm is invoked to perform multi-objective optimization, resulting in a set of optimal site selection schemes that reflect the overall global benefits. By statistically analyzing the frequency of occurrence and fitness ranking of each simulated candidate facility site in the set of optimal site selection schemes, a weighted evaluation score representing the construction potential of the site is calculated. The spatial feature vectors of the simulated candidate facility sites are paired with the evaluation scores to form a labeled dataset containing feature-label pairs.

[0011] Preferably, the process of training the neural network includes: A deep neural network architecture is constructed, with the multidimensional spatial feature vector of the initial candidate sites as the input layer and the construction value score as the output layer; The deep neural network was iteratively trained using the labeled dataset. The trained network weights and bias parameters are stored to form an offline value evaluator.

[0012] Preferably, the process of obtaining the high-value candidate site pool includes: The receiving planning side receives the initial candidate point set pre-selected according to actual needs, and uses the spatial feature vector extracted for each initial candidate point as input to the offline value evaluator that has been trained, outputting the corresponding continuous value, and using the continuous value as the construction value evaluation score that characterizes the construction potential. A threshold for assessing construction value was set, and all initial candidate sites were sorted in descending order based on their assessment scores. A quantile screening mechanism was used to identify and remove low-potential sites with scores below the threshold, and the remaining sites were used to obtain a high-value candidate pool.

[0013] Preferably, a multi-objective optimization model is constructed that comprehensively considers planning benefits and user experience; the multi-objective optimization model takes maximizing demand coverage and minimizing the average access path distance as the joint objective function; a target search space is constructed based on the high-value candidate pool, and the dung beetle optimization algorithm is executed for iterative optimization; The optimal combination of energy replenishment facility sites is determined from the high-value candidate pool and mapped to a layout decision scheme with geospatial attributes, outputting the optimal network configuration that meets urban planning constraints.

[0014] Preferably, a power replenishment facility positioning system includes: Data preprocessing module: used to construct a spatial constraint system for the target region and extract multidimensional spatial feature vectors of initial candidate sites; Simulation training and annotation module: used to construct simulated localization scenarios and generate an annotated dataset containing value rating labels through heuristic optimization; Neural network evaluation module: used to build and train a deep neural network evaluator to achieve offline mapping from spatial features to construction value; Space reduction and filtering module: used to call the evaluator to quickly retrieve the value of real candidate sites and remove low-potential sites based on a preset threshold; The refined optimization module is used to perform secondary heuristic iterations in the high-value candidate pool and output the final energy replenishment facility layout scheme.

[0015] As can be seen from the above technical solution, compared with the prior art, this invention discloses a method and system for locating energy replenishment facilities. This invention first constructs a multimodal spatial constraint system, using data such as administrative boundaries, water system distribution, and population density to accurately delineate the legal distribution space of candidate sites and energy replenishment demand points. Based on this, by quantifying the spatial coverage increment of candidate sites to surrounding demand points and their relative cost ranking within the collaborative cluster, multidimensional evaluation features reflecting the construction value of the sites are extracted. This approach effectively solves the problem that traditional site selection models struggle to balance complex geographical constraints with refined service value quantification, ensuring the planning compliance and scientific layout of the solution.

[0016] During the evaluation and optimization process, this invention successfully transformed knowledge from "static spatial features" to "dynamic construction value" by introducing a dynamic annotation mechanism based on DBO optimization frequency. By leveraging the offline learning of this nonlinear mapping relationship using deep neural networks, the system can achieve second-level value retrieval and rapid solution space reduction during the online optimization phase. This "offline training-online evaluation" architecture significantly reduces the computational complexity of large-scale candidate site screening, overcoming the bottleneck of low search efficiency and susceptibility to local optima in heuristic algorithms at the city-wide scale.

[0017] This invention employs a multi-objective collaborative optimization mechanism in the final decision-making stage, comprehensively considering the competitive needs of both planners and users. By performing a secondary refined search within the high-value candidate space, this invention can output an optimal layout scheme that balances economy and universality. Furthermore, due to the model's deep generalization ability to spatial features, this method can be flexibly transferred to urban areas of different sizes and geographical characteristics, demonstrating strong engineering application value and universality. 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 The overall flowchart of the method provided by this invention; Figure 2 The structural block diagram of the energy replenishment facility layout optimization system provided by 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] like Figure 1 As shown in the figure, an embodiment of the present invention discloses a method for locating energy replenishment facilities, including: Acquire vector data of urban administrative boundaries and vector data of water system distribution, and perform preprocessing; A spatial constraint system is defined based on the preprocessed data to delineate the distribution boundaries of energy replenishment facilities and energy replenishment demand. Initial candidate sites are selected based on the road network topology, and feature attributes reflecting the site service efficiency and competitiveness are extracted based on the spatial constraint system to generate corresponding multi-dimensional spatial feature vectors. A spatial feature vector of a set of simulated demand points is extracted by constructing a simulated positioning scenario. The corresponding construction value scoring labels are generated by combining the heuristic optimization results to obtain a labeled dataset. A neural network is trained using labeled datasets to establish a mapping relationship between spatial features and construction value, and the trained model parameters are stored as an offline value evaluator. An offline value evaluator was used to quickly retrieve and screen the initial candidate sites to obtain a pool of high-value candidate sites. Heuristic optimization is performed on the pool of high-value candidate sites to output the final energy replenishment facility layout scheme.

[0022] Specifically, the city's administrative boundary vector data and water system distribution vector element data are obtained separately, and the geometric attributes and attribute statistics of each modality data are read using the corresponding geographic information engine; Using the city's administrative boundary vector data as a reference system, coordinate projection transformation is performed on the water system distribution vector element data. Perform topology checks on urban administrative boundary vector data and water system distribution vector element data, correct topology errors, obtain standardized spatial constraint metadata, and form a spatial constraint system that defines the permitted construction scope of facilities through spatial analysis operations (such as buffer generation and overlay clipping).

[0023] In a specific embodiment of the present invention, urban administrative boundary vector data in .shp format and water system distribution vector features in .geojson format are acquired, and the geometric attributes and attribute statistics of each modality data are read using the corresponding geographic information engine. Unified alignment of the multi-source data coordinate system is achieved through resampling and spatial transformation, and abnormal pixels and topological errors are cleaned. Using the administrative boundary data as a reference system, coordinate projection transformation is performed on the water system distribution data to ensure that each modal data precisely overlaps in geographic space.

[0024] Furthermore, the physical areas where energy replenishment facilities can be constructed are delineated. Using the administrative boundary vector data as a reference layer, spatial overlay analysis is used to identify the areas occupied by the water system distribution vector elements and designate them as non-statutory construction restricted areas. These restricted areas are then removed from the administrative boundaries to extract candidate physical spaces for the deployment of energy replenishment facilities. These spaces constitute the specific implementation range of the aforementioned spatial constraint system at the physical site selection level.

[0025] Specifically, generating the corresponding multidimensional spatial feature vector includes: Based on the urban road network topology of the target area, each road intersection node is extracted, the betweenness centrality feature of each road intersection node is calculated, and probability sampling is performed according to the betweenness centrality from high to low to select several nodes as initial candidate sites. The set of urban points of interest within the target area is obtained as the original demand carrier. Population density spatial distribution data is introduced as a weight gain factor. The set of urban points of interest is weighted and randomly sampled according to the regional population density intensity of each urban point of interest location to determine the set of energy replenishment demand points. Based on the aforementioned spatial constraint system, the access cost and service coordination efficiency of each candidate energy replenishment facility site are quantified; through spatial overlay and buffer analysis, the coverage contribution of each initial candidate site to surrounding demand points is calculated, and the coordination coverage efficiency features reflecting the complementarity of spatial layout are extracted. Based on the access cost ranking and corresponding collaborative coverage efficiency components within the collaborative service cluster to which each initial candidate site belongs, a function mapping relationship reflecting the competitive advantages and disadvantages among sites is constructed. The relative attractiveness of the initial candidate sites within the shared service area is quantified through the mapping relationship, and a multi-dimensional spatial feature vector is extracted.

[0026] In a specific embodiment of the present invention, based on the urban road network topology of the target area, each road intersection node is extracted; the betweenness centrality feature of each node is calculated, and probability sampling is performed according to the betweenness centrality from high to low, thereby selecting several nodes as candidate sites; the specific process includes: First, the urban road network is simplified by eliminating redundant connections of branch roads and extracting road centerlines. Then, geographic information tools are used to map the road network into a graph theory model. ,in It represents the set of nodes consisting of road intersections. The set of road segments connecting intersections; the final traversal graph. Extract all the original road nodes to be calculated.

[0027] For road network topology map For each node v in the bounded space, calculate its Euclidean distance. (like Local betweenness centrality within a range The calculation formula is as follows: (1) In the formula, This indicates the distance of the shortest path from node v in the network. The set of all nodes within the range; Represents a node To the node The total number of shortest paths between them; Represents a node To the node The number of paths that pass through node v in the shortest path between them.

[0028] A probability distribution function is constructed based on the local betweenness centrality scores of each node, ensuring a non-linear positive correlation between the sampling probability and the score. This approach retains high-potential intersections while also considering the diversity of spatial distribution. The specific calculation formula for the probability distribution function is shown below: (2) In the formula, Represents a node The probability of being selected as an initial candidate site in the sampling; Represents a node The local betweenness centrality score; This represents the sensitivity adjustment factor, used to control the concentration of the sampling distribution; This represents the total number of original road nodes to be sampled. A sensitivity adjustment factor is introduced. (For example, setting) This can non-linearly amplify the competitive advantage of high betweenness nodes. When At this time, the system will significantly favor transportation hub nodes. However, since the denominator includes the weighted contributions of all nodes, nodes with lower scores but unique spatial locations still have a certain probability of being selected. The function described above is used to perform probability sampling on the original node set, selecting several representative nodes as initial candidate sites, providing a discretized search space for subsequent feature extraction and refined optimization.

[0029] Furthermore, a set of Points of Interest (POIs) within the target area is obtained as the original demand carrier; spatial distribution data of population density is introduced as a weighting gain factor; based on the population density intensity of the area where each POI is located, the POI set is weighted and randomly sampled to determine the set of charging demand points representing the actual distribution of users; specifically including: Data on multiple Points of Interest (POIs) for the target area are obtained from an open-source map platform (such as Open Street Map, OSM) and stored in a structured CSV file. Simultaneously, high-resolution population density geographic raster data (TIF format file) corresponding to the target area is acquired. Then, using spatial overlay technology from a Geographic Information System (GIS), the POI location data is matched with the population density raster coordinates to retrieve and extract the population density value of the geographic grid where each POI is located.

[0030] Each Point of Interest (POI) is assigned a corresponding sampling probability weight based on its population density. A weighted random sampling algorithm is then used to perform non-uniform selection from the original POI set based on the weights of each POI. This step transforms the heterogeneity of population distribution into the spatial distribution characteristics of energy replenishment demand points, ultimately generating a set of energy replenishment demand points that represents the true energy replenishment demand distribution in the target area.

[0031] Furthermore, collaborative coverage efficiency features are extracted: within the framework of the spatial constraint system, the access cost and service collaboration efficiency of each candidate energy replenishment facility site are quantified; through spatial overlay and buffer analysis, the coverage contribution of each candidate site to surrounding demand points is calculated, thereby extracting collaborative coverage efficiency features reflecting the complementarity of spatial layout, specifically including: For each candidate site Identify the nearest one A set of demand points Use road network analysis tools (specifically, the Graph Hopper-Web 7.0 engine) to retrieve electronic map road network data for the target area; based on the latitude and longitude coordinates of the energy replenishment demand points and candidate sites, use routing path algorithms (such as Dijkstra's algorithm or A / B algorithm) to determine the optimal route. The algorithm performs real-time road network matching and calculation to obtain the demand points. to candidate sites The shortest path distance between them, and use it as the access cost. ; By analyzing the set The costs of all accesses within the site are summed to quantify the basic coverage capability of the site for surrounding needs.

[0032] If multiple candidate sites can jointly cover a specific number of The demand points are all below the preset maximum acceptable threshold, and the access cost to each site for these demand points is lower than the preset maximum acceptable threshold. If these sites are found to have a cooperative service relationship, then it can be determined that such sites have a cooperative service relationship. By analyzing the coverage overlap between sites, their spatial complementarity can be identified, avoiding excessive redundant competition for resources in the same high-demand area. For any candidate site j, the set of its cooperative partner sites is denoted as . .

[0033] The feature values ​​of each candidate site are quantified using a pre-defined collaborative coverage efficiency function, and the calculation formula is shown below: (3) In the formula, Indicates candidate sites The collaborative coverage efficiency feature components, Indicates the location A set of sites that have a cooperative relationship.

[0034] Furthermore, competitive cost attractiveness features are extracted: Within the framework of the spatial constraint system, for each candidate site, based on its access cost ranking within its respective collaborative service cluster and the corresponding collaborative coverage efficiency component, a function mapping relationship reflecting the competitive advantages and disadvantages among sites is constructed; the relative attractiveness intensity of candidate sites within the shared service area is quantified through this mapping relationship, thereby extracting competitive cost attractiveness features, specifically including: For each candidate site Extract the set with which it has collaborative service relationships. Total access cost of all sites Sort the total cost within the set in descending order and determine the location. Rank in the sequence .

[0035] The ranking reflects the cost advantage of a site within the same demand coverage area; the higher the ranking, the stronger its potential competitive appeal.

[0036] A nonlinear mapping function reflecting the competitive advantages and disadvantages between sites is established, and the specific calculation formula is shown below: (4) In the formula, Candidate sites Its competitive cost attractiveness characteristics; For set The number of candidate sites; This is a preset positive reference offset constant used to limit the range of eigenvalues; This is the extracted collaborative coverage efficiency component.

[0037] The static ranking of sites is achieved through the aforementioned competition model. Deeply integrate with the dynamic collaborative contribution ratio. Utilize the baseline offset term. The competition cost attraction feature component is extracted from the multidimensional spatial feature vector by performing a cancellation operation with the normalized competition loss term and quantifying the relative attractiveness of candidate sites within the shared service area.

[0038] Specifically, the process of generating an optimal site selection set by combining heuristic optimization algorithms, determining the selection frequency statistics of candidate facility sites within this set, and generating corresponding construction value scoring labels based on these statistics to obtain a labeled dataset includes: By repeatedly executing random sampling procedures for urban interest point sets and road node sets independently, several sets of simulated demand point sets and simulated candidate site sets with spatial distribution differences are generated. Each set of demand points and candidate sites constitutes an independent simulated positioning scenario. In various scenarios, the dung beetle optimization algorithm is invoked to perform multi-objective optimization, resulting in a set of optimal site selection schemes that reflect the overall global benefits. By statistically analyzing the frequency of occurrence and fitness ranking of each simulated candidate facility site in the set of optimal site selection schemes, a weighted evaluation score representing the construction potential of the site is calculated. The spatial feature vectors of the simulated candidate facility sites are paired with the evaluation scores to form a labeled dataset containing feature-label pairs.

[0039] In one specific embodiment of the present invention, multiple simulated positioning scenarios are generated within the spatially constrained system using random sampling technology. Each scenario contains a set of simulated demand points extracted from urban points of interest (POIs). and a set of simulated candidate sites extracted from the road network. Based on formulas (3) and (4), extract the spatial feature vector of the simulated demand point set; specifically as follows: By repeatedly executing a random sampling procedure on the original point set (POI set and road node set) independently, several sets of simulated demand point sets and simulated candidate site sets with spatial distribution differences are generated. Each combination of demand points and candidate sites constitutes an independent simulated positioning scenario. Since each sampling process is random, each scenario is different in terms of point coordinates, local demand density and spatial topology, thus providing diverse spatial distribution samples for the subsequent generalization training of the model.

[0040] Traverse all generated simulated localization scenarios, extract the spatial feature vectors of simulated candidate sites in each scenario according to formulas (3) and (4), and construct them as input samples for the offline labeled dataset.

[0041] Furthermore, for each simulation scenario, the dung beetle optimization algorithm is invoked to perform multiple rounds of independent optimization with the joint objective of minimizing access cost and maximizing demand coverage, obtaining multiple sets of site layout solutions that reflect the optimal overall global benefits; the calculation of the joint objective function is shown below: (5) In the formula, The set of demand points for service coverage of the sites selected by the algorithm The sum of access costs is represented in this invention using the path distance between the demand point and the location, and the selected location has a predetermined service coverage radius. ; The total number of demand points covered by the selected site service, i.e. Size.

[0042] This is for normalization. In the formula This represents the set of sites selected by the algorithm. This represents the maximum acceptable access cost for a user within the adjustable demand points, and it is consistent for all users. In the formula This represents the total number of demand points in this scenario.

[0043] The population size for dung beetle optimization algorithms can typically be set to... In this example The dimension of the population individuals is set to Maximum number of iterations of the algorithm , As a regulating factor, in this example The search behavior of dung beetle populations is simulated to perform nonlinear optimization within the candidate site space. This is implemented for each scenario. Multiple independent optimization processes are used to obtain multiple sets of layout solutions that reflect the global optimal benefits.

[0044] For each simulation scenario, a statistical set Each simulated candidate facility site is in The construction value score of each simulated candidate facility site is generated by weighting the frequency of occurrence of each optimal site selection scheme in the set of optimal site selection schemes and combining the fitness ranking of each scheme. This leads to the construction of a labeled dataset containing "feature-label" pairs. The calculation formula is as follows: (6) In the formula, For the plan In the Fitness ranking among the options; It is an adjustable scaling constant; As an indicative variable, if the site Appeared in the If the optimal addressing scheme is in the set of group optimal addressing schemes, the value is 1; otherwise, it is 0. Finally, the extracted simulated site feature vectors are compared with the... Pairing is performed to construct an offline labeled dataset.

[0045] Specifically, the process of training the neural network includes: A deep neural network architecture is constructed, with the multidimensional spatial feature vector of the initial candidate sites as the input layer and the construction value score as the output layer; The deep neural network is iteratively trained using the labeled dataset. By minimizing the difference between the predicted score and the labeled score, the model is able to identify high-value sites under complex spatial constraints. The trained network weights and bias parameters are stored to form an offline value evaluator.

[0046] In a specific embodiment of the present invention, a multi-layer fully connected deep neural network is constructed. The input layer dimension is consistent with the spatial feature vector, set to 2-dimensional; four hidden layers are set, with the number of neurons in each layer decreasing according to the gradient, and the Swish activation function is used in conjunction with batch normalization to enhance the model's ability to fit the complex nonlinear relationship between features and construction value; the output layer adopts a single-neuron regression architecture and is configured with the ReLU activation function, aiming to accurately predict and output the non-negative continuous construction value score of each candidate site.

[0047] The generated offline labeled dataset was used to iteratively train the neural network. The labeled dataset was divided into training and validation sets in an 8:2 ratio to ensure good generalization performance of the model. For training configuration, an initial learning rate of 0.0001 was set, and the Nadam optimizer was used to perform weight updates. To balance the stability of gradient updates with computational efficiency, the batch size for each iteration was set to 32. Mean squared error (MSE) is used as the loss function. The difference between the predicted score and the labeled score is minimized through the backpropagation algorithm. Mean absolute error (MAE) and root mean square error (RMSE) are introduced as performance monitoring indicators. The maximum number of iterations (Epochs) is set to 1000, and a dynamic callback mechanism is introduced: (1) Early Stopping: If the validation set loss does not decrease within 50 consecutive epochs, the training is terminated early and the model weights with the best performance on the validation set are restored; (2) Learning Rate Decay: If the validation set loss enters a plateau within 15 consecutive epochs, the learning rate is reduced by a factor of 0.5 (minimum to 0.5). This helps the model achieve more refined convergence in the later stages of training.

[0048] Specifically, the process of obtaining the high-value candidate site pool includes: The receiving planning side receives the initial candidate point set pre-selected according to actual needs, and uses the spatial feature vector extracted for each initial candidate point as input to the offline value evaluator that has been trained, outputting the corresponding continuous value, and using the continuous value as the construction value evaluation score that characterizes the construction potential. A threshold for assessing construction value was set, and all initial candidate sites were sorted in descending order based on their assessment scores. A quantile screening mechanism was used to identify and remove low-potential sites with scores below the threshold, and the remaining sites were used to obtain a high-value candidate pool.

[0049] In a specific embodiment of the present invention, the receiving planning side receives a set of initial candidate sites pre-selected according to actual needs, and inputs the spatial feature vectors extracted for each initial candidate site into an offline value evaluator that has been trained, outputs continuous numerical values ​​corresponding to each initial site, and uses them as construction value evaluation scores that characterize their construction potential.

[0050] A threshold for assessing construction value was set, and all initial candidate sites were sorted in descending order based on their assessment scores. A quantile screening mechanism was adopted to identify and eliminate low-potential sites with scores in the lowest 20% range, thereby effectively compressing the original solution space and retaining the remaining 80% of sites to construct a high-value candidate pool.

[0051] Specifically, a multi-objective optimization model is constructed that comprehensively considers planning benefits and user experience; the multi-objective optimization model takes maximizing demand coverage and minimizing the average access path distance as the joint objective function; a target search space is constructed based on the high-value candidate pool, and the dung beetle optimization algorithm is executed for iterative optimization; The optimal combination of energy replenishment facility sites is determined from the high-value candidate pool and mapped to a layout decision scheme with geospatial attributes, outputting the optimal network configuration that meets urban planning constraints.

[0052] In a specific embodiment of the present invention, a multi-objective optimization model that comprehensively considers planning benefits and user experience is constructed; the multi-objective optimization model takes maximizing demand coverage (planner side) and minimizing average access path distance (user side) as the joint objective function (formula (5)); a target search space is constructed based on the high-value candidate pool, and the dung beetle optimization algorithm is executed for iterative optimization; The optimal combination of energy replenishment facility sites is determined from the high-value candidate pool and mapped to a layout decision scheme with geospatial attributes. Finally, the optimal network configuration that meets urban planning constraints is output.

[0053] In this embodiment, the energy replenishment facilities include, but are not limited to, fast charging stations, battery swapping stations, or mobile charging bays. The spatial constraint system, acting as a bottom-level geographic filter, achieves rigid locking of urban ecological red lines, reserved spaces for power corridors, and existing building red lines through a digital map layer. In this way, the optimized layout scheme can be directly transformed into a site selection red line map with engineering feasibility.

[0054] Specifically, such as Figure 2 As shown, a power replenishment facility positioning system includes: Data preprocessing module: used to construct a spatial constraint system for the target region and extract multidimensional spatial feature vectors of initial candidate sites; Simulation training and annotation module: used to construct simulated localization scenarios and generate an annotated dataset containing value rating labels through heuristic optimization; Neural network evaluation module: used to build and train a deep neural network evaluator to achieve offline mapping from spatial features to construction value; Space reduction and filtering module: used to call the evaluator to quickly retrieve the value of real candidate sites and remove low-potential sites based on a preset threshold; The refined optimization module is used to perform secondary heuristic iterations in the high-value candidate pool and output the final energy replenishment facility layout scheme.

[0055] The detailed process of the method provided in the specific application embodiment is as follows: I. Experimental Scenario and Parameter Settings. This embodiment uses Wuhan City as the research scenario. Data was acquired through a multi-source geospatial data platform, and a spatial constraint system was established. Within the study area, 100 initial candidate sites for energy replenishment facilities and 50 energy replenishment demand points were selected. This embodiment sets the final construction scale at 25 sites, and based on the Wuhan city-wide scale and the distribution density of demand points, the effective service radius of a single energy replenishment facility is set at 3 km.

[0056] II. Evaluation Metrics. This example introduces the average user access path distance. Site coverage requirements Two key technical indicators are used to quantitatively evaluate the layout performance of the two-stage optimization method of this invention in ultra-large-scale scenarios.

[0057] (1) Average access path distance This metric represents the average commuting cost for all users effectively served by replenishment facilities, from their point of demand to the nearest facility. Its calculation formula is as follows: (7) In the formula, the relevant character definitions are consistent with the aforementioned steps; wherein, this indicator is obtained by statistically analyzing the covered point set. The sum of the actual road network weights from each demand point to its corresponding facility point, divided by the total number of people. This information is used to measure the ease of energy replenishment in a layout scheme.

[0058] (2) Demand coverage This indicator characterizes the responsiveness of the final layout plan to the total energy replenishment demand within the region, reflecting the utilization rate of public service resources. Its calculation formula is as follows: (8) In the formula, the relevant character definitions are consistent with the aforementioned steps; this indicator reflects the impact of the final site selection scheme on the entire set of demand points under the preset service radius constraint. The degree of effective response is used to measure the service inclusiveness of the layout plan.

[0059] III. Training Performance of the Offline Estimator. This embodiment trains the offline value estimator using a labeled dataset from Wuhan city. Experimental results show that the mean squared error (MSE) loss curves of the neural network converge smoothly on both the training and validation sets, without significant overfitting. Testing demonstrates that this estimator exhibits excellent accuracy in predicting land value under complex urban spatial constraints, with a goodness of fit (…). The score reached 0.96, proving that the pre-trained model can accurately capture the deep mapping relationship between geographic topological features and site selection potential, providing a reliable evaluation logic basis for subsequent rapid value retrieval and search space shrinkage.

[0060] Fourth, the trained offline value evaluator is invoked to perform reverse value retrieval on 100 initial candidate sites. Based on the multidimensional spatial topological attributes of the sites, their predicted value scores are output, and low-potential sites in the bottom 20% of the sequence are automatically eliminated according to a quantile screening mechanism, significantly narrowing the search space. Subsequently, a refined online optimization algorithm is used in the pool of retained high-value sites to finally output the optimal energy replenishment facility layout scheme with the best overall benefits. Experiments demonstrate that this "initial screening + fine-tuning" model effectively avoids search blind spots in geographical constraints, ensuring that the final scheme accurately targets high-demand, high-potential spatial clusters.

[0061] V. Performance Quantitative Comparison. To further verify the technical advantages of this invention, this embodiment compares the two-stage optimization method of "neural network pre-evaluation + heuristic fine ranking" with traditional single-optimization algorithms (PSO, DBO) under the same hardware environment and initial candidate space. To eliminate the interference of algorithm randomness, the performance indicators shown in Table 1 are the average values ​​of each method after 10 independent repeated experiments. The specific quantitative comparison results are shown in Table 1: Table 1. Performance comparison of the method of this invention and the benchmark algorithm under multidimensional evaluation indicators.

[0062] Experimental results show that, in the Wuhan city area scenario of this embodiment, compared with the benchmark algorithm that performs better, the location selection scheme generated by the method of this invention reduces the average access path distance for users. The distance decreased from 5.36km to 4.23km, a reduction of approximately 21.1%; meanwhile, demand coverage... The value improved significantly, from a peak of 0.55 to 0.72, representing a growth of approximately 30.9%. This demonstrates that the offline value evaluator, by pre-emptively sensing the potential value of candidate sites, effectively guides the optimization algorithm to focus on high-yield regions, avoiding the problem of traditional methods getting trapped in local optima through blind searching.

[0063] Regarding convergence speed, benchmark algorithms require over 220 iterations to converge due to the complex nonlinear constraints of a large initial candidate space. In contrast, the method of this invention significantly reduces the complexity of the solution space by implementing a space reduction strategy, achieving high-quality convergence in only 182 iterations, thus improving convergence efficiency by approximately 20% to 23%. Due to the reduced number of iterations and the clarification of the search target, the final computation time is drastically reduced to around 7 minutes, demonstrating excellent engineering practicality.

[0064] In summary, the two-stage optimization method provided by this invention not only ensures that the layout scheme has greater universality and convenience, but also significantly improves the planning efficiency in ultra-large-scale urban scenarios, and has significant engineering application value.

[0065] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0066] 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 locating energy replenishment facilities, characterized in that, include: Acquire vector data of urban administrative boundaries and vector data of water system distribution, and perform preprocessing; A spatial constraint system is defined based on the preprocessed data to delineate the distribution boundaries of energy replenishment facilities and energy replenishment demand. Initial candidate sites are selected based on the road network topology, and feature attributes reflecting the site service efficiency and competitiveness are extracted based on the spatial constraint system to generate corresponding multi-dimensional spatial feature vectors. A spatial feature vector of a set of simulated demand points is extracted by constructing a simulated positioning scenario. This vector is then combined with a heuristic optimization algorithm to generate corresponding construction value scoring labels, resulting in a labeled dataset. A neural network is trained using labeled datasets to establish a mapping relationship between spatial features and construction value, and the trained model parameters are stored as an offline value evaluator. An offline value evaluator was used to quickly retrieve and screen the initial candidate sites to obtain a pool of high-value candidate sites. Heuristic optimization is performed on the pool of high-value candidate sites to output the final energy replenishment facility layout scheme.

2. The method for locating a power replenishment facility according to claim 1, characterized in that, The city's administrative boundary vector data and water system distribution vector element data are obtained separately, and the geometric attributes and attribute statistics of each modal data are read using the corresponding geographic information engine. Using the city's administrative boundary vector data as a reference system, coordinate projection transformation is performed on the water system distribution vector element data. Perform topology checks on urban administrative boundary vector data and water system distribution vector element data, correct topology errors, obtain standardized spatial constraint metadata, and form a spatial constraint system that defines the permitted construction scope of facilities through spatial analysis operations.

3. The method for locating a power replenishment facility according to claim 1, characterized in that, The generation of the corresponding multidimensional spatial feature vector includes: Based on the urban road network topology of the target area, each road intersection node is extracted, the betweenness centrality feature of each road intersection node is calculated, and probability sampling is performed according to the betweenness centrality from high to low to select several nodes as initial candidate sites. The set of urban points of interest within the target area is obtained as the original demand carrier. Population density spatial distribution data is introduced as a weight gain factor. The set of urban points of interest is weighted and randomly sampled according to the regional population density intensity of each urban point of interest location to determine the set of energy replenishment demand points. Based on the aforementioned spatial constraint system, the access cost and service coordination efficiency of each candidate energy replenishment facility site are quantified; through spatial overlay and buffer analysis, the coverage contribution of each initial candidate site to surrounding demand points is calculated, and the coordination coverage efficiency features reflecting the complementarity of spatial layout are extracted. Based on the access cost ranking and corresponding collaborative coverage efficiency components within the collaborative service cluster to which each initial candidate site belongs, a function mapping relationship reflecting the competitive advantages and disadvantages among sites is constructed. The relative attractiveness of the initial candidate sites within the shared service area is quantified through the mapping relationship, and a multi-dimensional spatial feature vector is extracted.

4. The method for locating a power replenishment facility according to claim 1, characterized in that, The process of obtaining the labeled dataset includes: By repeatedly executing random sampling procedures for urban interest point sets and road node sets independently, several sets of simulated demand point sets and simulated candidate site sets with spatial distribution differences are generated. Each set of demand points and candidate sites constitutes an independent simulated positioning scenario. In various scenarios, the dung beetle optimization algorithm is invoked to perform multi-objective optimization, resulting in a set of optimal site selection schemes that reflect the overall global benefits. By statistically analyzing the frequency of occurrence and fitness ranking of each simulated candidate facility site in the optimal site selection scheme set, a weighted evaluation score representing the site construction potential is obtained. The spatial feature vectors of the simulated candidate facility sites are paired with the evaluation scores to form a labeled dataset containing feature-label pairs.

5. The method for locating a power replenishment facility according to claim 4, characterized in that, The process of training the neural network includes: A deep neural network architecture is constructed, with the multidimensional spatial feature vector of the initial candidate sites as the input layer and the construction value score as the output layer; The deep neural network was iteratively trained using the labeled dataset. The trained network weights and bias parameters are stored to form an offline value evaluator.

6. The method for locating a power replenishment facility according to claim 5, characterized in that, The process of obtaining the high-value candidate site pool includes: The receiving planning side receives the initial candidate point set pre-selected according to actual needs, and uses the spatial feature vector extracted for each initial candidate point as input to the offline value evaluator that has been trained, outputting the corresponding continuous value, and using the continuous value as the construction value evaluation score that characterizes the construction potential. A threshold for assessing construction value was set, and all initial candidate sites were sorted in descending order based on their assessment scores. A quantile screening mechanism was used to identify and remove low-potential sites with scores below the threshold, and the remaining sites were used to obtain a high-value candidate pool.

7. The method for locating a power replenishment facility according to claim 6, characterized in that, A multi-objective optimization model is constructed that comprehensively considers planning benefits and user experience; the multi-objective optimization model takes maximizing demand coverage and minimizing the average access path distance as the joint objective function; a target search space is constructed based on the high-value candidate pool, and the dung beetle optimization algorithm is executed for iterative optimization; The optimal combination of energy replenishment facility sites is determined from the high-value candidate pool and mapped to a layout decision scheme with geospatial attributes, outputting the optimal network configuration that meets urban planning constraints.

8. A positioning system for energy replenishment facilities, characterized in that, include: Data preprocessing module: used to construct a spatial constraint system for the target region and extract multidimensional spatial feature vectors of initial candidate sites; Simulation training and annotation module: used to construct simulated localization scenarios and generate an annotated dataset containing value rating labels through heuristic optimization; Neural network evaluation module: used to build and train a deep neural network evaluator to achieve offline mapping from spatial features to construction value; Space reduction and filtering module: used to call the evaluator to quickly retrieve the value of real candidate sites and remove low-potential sites based on a preset threshold; The refined optimization module is used to perform secondary heuristic iterations in the high-value candidate pool and output the final energy replenishment facility layout scheme.

Images