Charging station peak shaving planning method and system based on space-time coupling and POI collaborative decision-making

Abstract

The application discloses a charging station peak shaving planning method and system based on space-time coupling and POI collaborative decision. The method firstly accurately identifies high-coincidence stations that aggravate power grid peak and low-coincidence stations with peak shaving potential by calculating a space-time load coincidence degree index; then a demand prediction model is constructed by fusing POI features, distance attenuation and regional density correction to improve prediction accuracy; on this basis, a two-stage capacity planning scheme is collaboratively formulated, including a load transfer and expansion scheme for high-coincidence stations, an exit strategy for low-efficiency stations and a new station planning scheme for high-demand low-coincidence areas; finally, a multi-objective optimization model is established with peak shaving, cost and service as targets, an improved genetic algorithm is used for solution, and a recommended scheme is output through multi-attribute decision. The application realizes active collaborative optimization of charging load and power grid load, and significantly improves power grid safety and investment benefit.

Description

Technical Field

[0001] This invention belongs to the field of power system planning and operation optimization technology, specifically relating to a charging station peak shaving planning method and system based on spatiotemporal coupling and POI collaborative decision-making. Background Technology

[0002] As a key support for the electric vehicle industry, the scientific planning and layout of charging infrastructure directly affects power grid security, user charging experience, and operator economic benefits. However, the large-scale and random access of charging loads brings significant peak-valley regulation pressure to the distribution network. When peak charging times overlap with the inherent peak load periods of the power grid, it further exacerbates the peak load and threatens the safe and stable operation of the power grid.

[0003] Currently, charging station planning methods mainly revolve around "where to build and how large to build," relying heavily on static factors such as traffic flow and land costs, and employing integer programming or heuristic algorithms for spatial site selection and capacity allocation. While these methods have achieved some success, they generally suffer from three shortcomings: First, they neglect the temporal coupling characteristics, failing to conduct a refined time comparison between peak charging station loads and regional power grid loads during planning, potentially leading to new or expanded stations exacerbating the peak load on the power grid. Second, demand forecasting accuracy is limited; traditional methods rely on simple models and fail to fully utilize the rich spatial functions and population activity information contained in urban points of interest (POI) data, resulting in significant prediction bias. Third, the decision-making dimension is singular, mostly considering only the single action of new construction, lacking a systematic mechanism for coordinated optimization of the existing network through "expansion-new construction-retirement," leading to low investment efficiency and rigid resource allocation.

[0004] Therefore, there is an urgent need to propose a new method for charging station planning that can accurately quantify the spatiotemporal load relationship, deeply integrate multi-source data features, and support multi-dimensional dynamic decision-making, so as to achieve proactive coordination between charging load and grid load, effectively shaving peaks and filling valleys while ensuring service quality, and improving the safety and economy of grid operation. Summary of the Invention

[0005] The main objective of this invention is to provide a new method for charging station planning that can accurately quantify spatiotemporal load relationships, deeply integrate multi-source data features, and support multi-dimensional dynamic decision-making.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a charging station peak shaving planning method based on spatiotemporal coupling and POI collaborative decision-making, comprising the following steps: Step S1: Collect charging load data of existing charging stations, POI data of candidate road network nodes, and regional power grid load data. Perform outlier detection and interpolation repair on the charging load data and standardize the POI data. Calculate the spatiotemporal load overlap index of each charging station. This index is the annual weighted overlap ratio between the peak load time of the charging station and the peak load time of the regional power grid. Based on this index, the charging stations are divided into high overlap stations, low overlap stations, and medium overlap stations. Step S2: Construct a POI feature-driven charging demand prediction model. Utilize the normalized POI data and historical charging volume of existing charging stations, and train the model weights using a regularized ridge regression algorithm. For candidate nodes, calculate the linear combination of their normalized POI vector and model weights as the basic prediction value. Introduce an attenuation coefficient based on geographical distance and a correction coefficient based on the density of charging stations in the region to obtain the predicted charging demand value for the candidate nodes. Based on the peak hour distribution of existing stations near the candidate nodes, infer their predicted peak hour and predicted spatiotemporal overlap. Step S3: For high-overlapping stations, within a preset geographical search radius, low-overlapping stations located in the same power grid area are selected as expansion candidates; for each expansion candidate station, a quantitative score is performed from four dimensions: time staggered peak performance, remaining capacity ratio, spatial distance, and facility conditions, and the station with the highest score is selected as the expansion target; based on the M / M / c model in queuing theory, and with the preset average user waiting time as a constraint, the minimum number of charging piles required to transfer load from high-overlapping stations is calculated, and the expansion capacity of the expansion target station is determined accordingly; For existing charging stations, inefficiency scores are calculated based on their return on investment, average annual load rate, and density of surrounding stations. For stations with inefficiency scores exceeding the threshold, one of four strategies is selected for implementation based on decision tree rules: complete withdrawal, partial withdrawal, gradual withdrawal, and monitoring and observation. Before implementation, the number of affected users is assessed to avoid service gaps. Step S4: Select candidate nodes whose predicted charging demand is higher than a set threshold and whose predicted spatiotemporal overlap is lower than a set threshold as valid new candidate points; for each valid new candidate point, quantify and score it from five dimensions: demand potential, spatiotemporal staggered peak degree, spatial accessibility, network synergy and construction economy, and select the node that meets the minimum distance constraint between stations as the new station in descending order of score. For selected new sites, a two-stage capacity planning method is used to determine their construction capacity: In the first stage, the deterministic basic capacity is calculated based on the demand forecast, the preset future demand growth rate, and the target load rate; In the second stage, three demand scenarios—pessimistic, baseline, and optimistic—and their probabilities of occurrence are set, and the elastic capacity is calculated with the goal of minimizing the expected cost; The final construction capacity of the new site is the sum of the basic capacity and the elastic capacity, and is subject to preset upper and lower capacity limits. Step S5: Construct a multi-objective optimization model. This model includes three objective functions: minimizing regional weighted peak load, minimizing total investment cost, and minimizing service quality loss. It also sets grid capacity constraints, service coverage constraints, site capacity constraints, budget constraints, and mutual exclusion constraints. The improved non-dominated sorting genetic algorithm NSGA-II is used to solve the model. The improvements include using a hybrid strategy to initialize the population, adaptively adjusting the crossover and mutation probabilities, and introducing a constraint repair mechanism. Finally, the Pareto optimal solution set is output. Step S6: Use the TOPSIS (Topology for Approximating Ideal Solutions) method to evaluate and rank the Pareto optimal solution set, and select the solution with the highest relative approximation as the final recommended scheme; output the list of expanded sites, the list of newly built sites, and the list of decommissioned sites under this scheme, calculate the change in the total peak load of the region before and after the planning, and obtain the peak reduction rate evaluation result.

[0007] Optionally, in step S1, the calculation process of the spatiotemporal load overlap index is as follows: extract the daily average load curve for each month from the annual load data of the charging station, find the moment with the largest load in the curve as the peak moment of that month; compare the monthly peak moment with the monthly peak moment of the power grid in the region, and if the time difference between the two is within the preset time window, it is determined that the month overlaps; count the number of overlapping months in the 12 months of the year, and assign different weight coefficients to the months of different seasons, and obtain the annual overlap index after weighted summation.

[0008] Optionally, in step S2, the distance attenuation coefficient is calculated based on the geographical distance from the candidate node to the nearest residential area, commercial area, and transportation hub. The closer the distance, the larger the coefficient. The regional charging station density correction coefficient is calculated based on the ratio of the number of charging stations per unit area in the region to the average density of the entire region. The lower the density, the larger the correction coefficient.

[0009] Optionally, in step S3, the specific rules for classifying sites with inefficiency scores exceeding the threshold using a decision tree are as follows: when the site's return on investment is negative and the average annual load rate is lower than the first threshold, a complete exit strategy is implemented; when the site has high overlap and the average annual load rate is lower than the second threshold, a partial exit strategy is implemented, reducing its capacity by 50%; when the number of alternative sites within a preset distance of the site exceeds the redundancy threshold, a gradual exit strategy is implemented; the remaining cases are listed as monitoring and observation.

[0010] Optionally, in step S4, the process of calculating the elastic capacity in the second stage of the two-stage capacity planning method is as follows: calculate the minimum capacity required to meet the demand under three demand scenarios: pessimistic, baseline, and optimistic; calculate the expected capacity with the probability of occurrence of the three scenarios as weights; the elastic capacity is the part of the expected capacity that exceeds the basic capacity of the first stage.

[0011] Optionally, in step S5, the specific operations of the improved non-dominated sorting genetic algorithm include: during population initialization, half of the individuals are generated using a heuristic method based on the decision rules in steps S3 and S4 of claim 1, and the other half are generated randomly; the crossover probability decreases linearly with the number of iterations, and the mutation probability changes sinusoidally and periodically with the number of iterations; for individuals that violate the constraints, a special repair operator is executed to make them satisfy all the constraints.

[0012] Optionally, in step S6, the scheme comparison and verification includes: comparing the peak reduction rate obtained by this method with traditional site selection methods based only on spatial distance and POI density, methods that only consider the single action of new construction, and demand forecasting methods that do not use POI features, in order to demonstrate the advantages of this method in improving peak reduction effect, reducing cost, and improving demand forecasting accuracy.

[0013] Secondly, the present invention provides a charging station peak shaving planning system based on spatiotemporal coupling and POI collaborative decision-making, used to implement any of the charging station peak shaving planning methods based on spatiotemporal coupling and POI collaborative decision-making as described above. The system includes: The data acquisition and preprocessing module is configured to: acquire charging load data, POI data and grid load data, and perform anomaly detection, interpolation and standardization processing; The spatiotemporal analysis and site classification module is configured to: calculate the spatiotemporal load overlap index of charging stations and classify charging stations based on the index; The demand forecasting module is configured to: train a POI feature-driven regression forecasting model and predict the charging demand and peak times of candidate nodes. The existing site optimization module is configured to: screen and determine expansion targets and expansion capacity for highly overlapping sites, and evaluate inefficient sites and generate exit strategies; The new site planning module is configured to: filter candidate nodes for new sites, score site selection, and determine the construction capacity of new sites; The multi-objective optimization solution module is configured to: construct and solve the multi-objective optimization model, and output a set of Pareto optimal planning solutions; The solution decision and output module is configured to recommend the final solution from the Pareto solution set and output an executable plan list and evaluation report.

[0014] In another aspect, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements any of the charging station peak shaving planning methods based on spatiotemporal coupling and POI collaborative decision-making as described above.

[0015] On the other hand, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements any of the above-mentioned charging station peak shaving planning methods based on spatiotemporal coupling and POI collaborative decision-making.

[0016] The above technical solution has the following technical effects: This invention accurately identifies problem sites and potential sites by quantifying the spatiotemporal load coupling relationship, improves the accuracy of demand estimation by using a prediction model that integrates multi-source features, and establishes a multi-action collaborative mechanism for expansion, new construction, and decommissioning, as well as a multi-objective global optimization model, ultimately achieving the following technical effects: I. Achieved refined spatiotemporal decoupling and proactive coordination between charging load and grid load: By calculating the quantitative indicator of "spatiotemporal load overlap" and accurately classifying charging stations (high / medium / low overlap) based on statistical distribution, it is possible to clearly identify "problem sites" that aggravate the peak pressure on the grid and "advantageous sites" with natural peak shaving potential. This makes subsequent decisions on expansion, relocation, and new construction more targeted and promotes the staggering of charging load with grid peak in the time dimension from the source.

[0017] Second, it significantly improves the accuracy and spatial rationality of charging demand forecasting: It adopts a regularized ridge regression model and introduces POI interaction features, geographical distance attenuation coefficient and regional charging station density correction coefficient for multi-level correction. This makes demand forecasting not only dependent on the static number of POIs, but also considers the functional combination effect, spatial accessibility and regional supply and demand balance. The forecast results are closer to the actual distribution pattern, providing a reliable basis for scientific site selection and capacity determination.

[0018] Third, a collaborative decision-making mechanism covering the entire lifecycle of "existing optimization and incremental deployment" has been established: This invention not only plans new sites but also systematically integrates the optimization of "expansion" and "retirement" of the existing network. Load transfer for highly overlapping sites, precise expansion based on multi-dimensional scoring and queuing theory, and evaluation and retirement of inefficient sites based on multi-dimensional indicators together constitute a dynamic, closed-loop network optimization system. This system can revitalize existing resources, eliminate inefficient assets, and achieve overall optimization of investment costs and network performance.

[0019] IV. Achieving a Global Balance of Comprehensive Benefits through Multi-Objective Optimization and Intelligent Algorithms: The constructed multi-objective optimization model, centered on "peak shaving effect, investment cost, and service quality," comprehensively characterizes the complex demands of the planning problem. An improved NSGA-II algorithm (hybrid initialization, adaptive parameters, and constraint repair) is used for efficient solution, outputting a set of balanced Pareto optimal solutions. Finally, the TOPSIS method is used for decision ranking, providing decision-makers with data-driven, multi-objective-considered optimal recommendations, ensuring the overall optimality of the planning scheme.

[0020] V. High Operability and Evaluability: The final output solution is a detailed list including specific site identifiers, locations, capacity, investment details, and execution priorities, and provides clear quantitative evaluation indicators such as peak shaving rate. This method is not only a theoretical model, but also directly guides engineering practice and investment decisions. Comparison with traditional methods verifies its significant advantages in improving peak shaving rate, reducing costs, and improving service. Under the premise of ensuring charging service quality, it effectively reduces regional power grid peak load, optimizes investment costs, and enhances the scientific planning of charging infrastructure and the safety of power grid operation. Attached Figure Description

[0021] Figure 1 This is a flowchart of a charging station peak shaving planning method based on spatiotemporal coupling and POI collaborative decision-making according to an embodiment of the present invention; Figure 2 This is a structural block diagram of a charging station peak shaving planning system based on spatiotemporal coupling and POI collaborative decision-making, according to another embodiment of the present invention. Figure 3 This is a schematic diagram of an electronic device structure according to an embodiment of the present invention. Detailed Implementation

[0022] To further illustrate the various embodiments, the present invention provides accompanying drawings. These drawings are part of the disclosure of the present invention, primarily used to illustrate the embodiments and to explain the operating principles of the embodiments in conjunction with the relevant descriptions in the specification. With reference to these drawings, those skilled in the art should be able to understand other possible implementations and the advantages of the present invention. Components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0023] With the advancement of global energy transition and the "dual-carbon" goals, the electric vehicle industry is developing rapidly, and the number of electric vehicles on the road continues to rise. Charging infrastructure, as a key support for the development of the electric vehicle industry, directly impacts user charging experience, grid operation safety, and the economic benefits of operating companies through its scientific planning and layout. However, the large-scale integration of charging loads poses significant challenges to the power grid, especially when peak charging times coincide with peak grid load periods, further exacerbating peak grid pressure and threatening the safe and stable operation of the grid. Therefore, how to scientifically plan the layout of charging stations and achieve spatiotemporal coordination between charging load and grid load has become a crucial issue that urgently needs to be addressed.

[0024] Existing charging station planning methods primarily focus on spatial site selection and capacity configuration optimization. They typically establish site selection models based on factors such as traffic flow, user demand distribution, and land costs, employing methods like integer programming and genetic algorithms to solve for optimal site locations and construction scale. While these methods have achieved some success in addressing the questions of "where to build" and "how large to build," they generally suffer from the following shortcomings: First, they neglect the temporal characteristics of charging loads, failing to consider the temporal coupling between peak charging station loads and peak grid loads, potentially exacerbating rather than alleviating peak grid pressure. Second, they lack a systematic evaluation of the performance of existing charging station networks, failing to identify which sites exacerbate grid loads and which have peak-shaving potential, resulting in a lack of targeted planning decisions. Third, existing methods mainly focus on the site selection of new sites, insufficiently considering diversified methods such as expanding existing sites and eliminating inefficient sites to optimize network layout, leading to low investment efficiency and unreasonable resource allocation.

[0025] From a technical perspective, existing charging station planning technologies suffer from three key shortcomings. First, there is a lack of quantitative assessment methods for spatiotemporal load matching. While existing research focuses on charging load forecasting and grid load analysis, it rarely conducts temporal correlation analysis between the two. This makes it difficult to accurately identify which charging stations' peak hours highly overlap with the regional grid's peak hours, thus exacerbating peak pressure, or which stations have natural peak-shaving effects by offsetting peak hours with the grid. This neglect of spatiotemporal characteristics results in limited effectiveness of planning schemes in achieving peak shaving and valley filling goals, and may even be counterproductive. Second, the demand forecasting methods for candidate sites are simplistic and lack accuracy. Traditional methods primarily rely on traffic flow or simple distance decay models to predict demand, failing to fully utilize the rich information contained in urban spatial POI data. POI data reflects the functional attributes, population distribution, and activity intensity of a region, which are closely related to charging demand. However, existing methods do not delve deeply enough into POI characteristics and lack integration with historical charging data, leading to significant demand forecasting bias and limited site selection accuracy. Third, the decision-making space is limited, lacking a multi-action collaborative optimization mechanism. Existing methods typically only consider the location of new charging stations, lacking systematic consideration of strategies such as expanding low-load-rate stations to take over transferred loads, or closing or reducing the capacity of inefficient stations to release resources. This makes it impossible to achieve dynamic optimization of the charging station network and efficient allocation of resources, resulting in high overall investment costs and limited peak shaving effects.

[0026] In summary, there is an urgent need to propose a new charging station planning method that can quantitatively evaluate the matching relationship between charging station load and grid load from a spatiotemporal coupling perspective, accurately identify problem sites that aggravate grid peak loads and sites with peak-shaving potential; fully utilize POI spatial characteristics and historical charging data to construct a high-precision demand forecasting model, providing a reliable basis for candidate site evaluation; and establish a multi-dimensional decision space including expansion, new construction, and decommissioning, achieving peak-shaving and valley-filling goals for the charging station network through multi-action collaborative optimization. This method maximizes peak-shaving effects and minimizes investment costs while ensuring charging service quality, providing technical support for grid safety operation and the sustainable development of charging infrastructure.

[0027] The following is combined with Figure 1 The flowchart below provides a detailed explanation of the method of the present invention.

[0028] A first aspect of the present invention provides a charging station peak shaving planning method based on spatiotemporal coupling and POI collaborative decision-making, comprising the following steps: Step S1: Collect charging load data of existing charging stations, POI data of candidate road network nodes, and regional power grid load data. Perform outlier detection and interpolation repair on the charging load data and standardize the POI data. Calculate the spatiotemporal load overlap index of each charging station. This index is the annual weighted overlap ratio between the peak load time of the charging station and the peak load time of the regional power grid. Based on this index, the charging stations are divided into high overlap stations, low overlap stations, and medium overlap stations. Step S2: Construct a POI feature-driven charging demand prediction model. Utilize the normalized POI data and historical charging volume of existing charging stations, and train the model weights using a regularized ridge regression algorithm. For candidate nodes, calculate the linear combination of their normalized POI vector and model weights as the basic prediction value. Introduce an attenuation coefficient based on geographical distance and a correction coefficient based on the density of charging stations in the region to obtain the predicted charging demand value for the candidate nodes. Based on the peak hour distribution of existing stations near the candidate nodes, infer their predicted peak hour and predicted spatiotemporal overlap. Step S3: For high-overlapping stations, within a preset geographical search radius, low-overlapping stations located in the same power grid area are selected as expansion candidates; for each expansion candidate station, a quantitative score is performed from four dimensions: time staggered peak performance, remaining capacity ratio, spatial distance, and facility conditions, and the station with the highest score is selected as the expansion target; based on the M / M / c model in queuing theory, and with the preset average user waiting time as a constraint, the minimum number of charging piles required to transfer load from high-overlapping stations is calculated, and the expansion capacity of the expansion target station is determined accordingly; For existing charging stations, inefficiency scores are calculated based on their return on investment, average annual load rate, and density of surrounding stations. For stations with inefficiency scores exceeding the threshold, one of four strategies is selected for implementation based on decision tree rules: complete withdrawal, partial withdrawal, gradual withdrawal, and monitoring and observation. Before implementation, the number of affected users is assessed to avoid service gaps. Step S4: Select candidate nodes whose predicted charging demand is higher than a set threshold and whose predicted spatiotemporal overlap is lower than a set threshold as valid new candidate points; for each valid new candidate point, quantify and score it from five dimensions: demand potential, spatiotemporal staggered peak degree, spatial accessibility, network synergy and construction economy, and select the node that meets the minimum distance constraint between stations as the new station in descending order of score. For selected new sites, a two-stage capacity planning method is used to determine their construction capacity: In the first stage, the deterministic basic capacity is calculated based on the demand forecast, the preset future demand growth rate, and the target load rate; In the second stage, three demand scenarios—pessimistic, baseline, and optimistic—and their probabilities of occurrence are set, and the elastic capacity is calculated with the goal of minimizing the expected cost; The final construction capacity of the new site is the sum of the basic capacity and the elastic capacity, and is subject to preset upper and lower capacity limits. Step S5: Construct a multi-objective optimization model. This model includes three objective functions: minimizing regional weighted peak load, minimizing total investment cost, and minimizing service quality loss. It also sets grid capacity constraints, service coverage constraints, site capacity constraints, budget constraints, and mutual exclusion constraints. The improved non-dominated sorting genetic algorithm NSGA-II is used to solve the model. The improvements include using a hybrid strategy to initialize the population, adaptively adjusting the crossover and mutation probabilities, and introducing a constraint repair mechanism. Finally, the Pareto optimal solution set is output. Step S6: Use the TOPSIS (Topology for Approximating Ideal Solutions) method to evaluate and rank the Pareto optimal solution set, and select the solution with the highest relative approximation as the final recommended scheme; output the list of expanded sites, the list of newly built sites, and the list of decommissioned sites under this scheme, calculate the change in the total peak load of the region before and after the planning, and obtain the peak reduction rate evaluation result.

[0029] This embodiment provides a complete implementation process for charging station peak shaving planning. For example, the method can be configured with the following parameters during implementation: 546 existing charging stations, 371 candidate road network nodes, and 38 grid area divisions; the geographical search radius is set to 5 kilometers; the preset average user waiting time constraint is 10 minutes; and the upper and lower capacity limits for newly built stations are constrained to 180kW to 600kW. In actual operation, the system, based on the above steps, sequentially calls modules such as data preprocessing, overlap calculation, demand forecasting, existing resource optimization, new station planning, multi-objective optimization, and scheme decision-making, ultimately generating an executable file containing specific station IDs, coordinates, expansion capacity, new capacity, and exit instructions.

[0030] The multiple steps covered in this embodiment constitute a systematic engineering method for addressing the peak-valley problem in power grids. Specifically, it uses the temporal coupling relationship between charging station load and grid load (step S1) as the physical basis for all subsequent optimization decisions; it overcomes the disconnect between demand forecasting and spatial function in traditional methods by integrating spatial POI information and statistical learning models (step S2); and for the first time, it systematically integrates three types of decision-making actions within a single planning framework: "load transfer-expansion" and "inefficiency assessment-exit" for existing networks, and "site selection-flexible capacity setting" for incremental demand (steps S3 and S4). This combination of technical features transforms the planning process from a static, single-point site selection problem into a dynamic resource reallocation process that considers the spatiotemporal state and lifecycle cost of the entire network. Finally, a multi-objective optimization algorithm (step S5) is used for global optimization, and a multi-criteria decision-making method (step S6) is used to output a definitive solution, thereby ensuring the rigor and feasibility of the logical chain from problem identification to solution generation.

[0031] In some embodiments, the calculation process of the spatiotemporal load overlap index in step S1 is as follows: extract the daily average load curve for each month from the annual load data of the charging station, find the moment with the largest load in the curve as the peak moment of that month; compare the monthly peak moment with the monthly peak moment of the power grid in the region, and if the time difference between the two is within a preset time window, it is determined that the month overlaps; count the number of overlapping months in the 12 months of the year, and assign different weight coefficients to the months of different seasons, and obtain the annual overlap index after weighted summation.

[0032] In the above embodiments, the specific calculation of the spatiotemporal load overlap index can be further exemplified as follows: from the charging station 8760-hour load sequence Extract the monthly peak times for each month. Extract load data for all times of the month and calculate the 24-hour daily average load curve. (h=0,1,...,23); Determine monthly peak times; Define charging stations. With the region The function for determining overlap during peak hours in month m; setting a time tolerance window, and calculating charging stations. The annual spatiotemporal load overlap index is used to classify charging stations into three categories based on the overlap: high overlap station, low overlap station, and medium overlap station.

[0033] The essence of the above calculation process is to transform the qualitative concept of "time overlap" into a quantifiable and comparable continuous numerical indicator. Its technical uniqueness lies in two aspects: First, it uses "monthly comparison" rather than "annual single-moment comparison," which captures the seasonal variation of peak load and avoids overgeneralization. Second, it introduces a seasonal weighting coefficient, reflecting the differences in grid tension across seasons (such as peak summer cooling load), allowing the indicator to more accurately measure the true impact of charging stations on the grid during its most vulnerable periods. This refined measurement method provides a reliable numerical basis for subsequently classifying charging stations into "high, medium, and low" overlap categories, thus enabling the core objective of "peak shaving" to be decomposed and precisely located at specific network nodes, avoiding the blind allocation of planned resources.

[0034] In some embodiments, the distance attenuation coefficient in step S2 is calculated based on the geographical distance from the candidate node to the nearest residential area, commercial area, and transportation hub. The closer the distance, the larger the coefficient. The regional charging station density correction coefficient is calculated based on the ratio of the number of charging stations per unit area in the region to the average density of the entire region. The lower the density, the larger the correction coefficient.

[0035] For example, this embodiment provides a specific calculation example of the distance attenuation coefficient and the regional density correction coefficient. The introduction of these two correction coefficients makes a physically meaningful adjustment to the basic forecast value from two dimensions: geospatial and market supply and demand. The distance attenuation coefficient, based on the "distance deterrence" theory, quantifies the convenience of users reaching charging stations (reflected in the distance to three key destinations) into a reduction in potential demand through an exponential decay function, making the forecast results more consistent with user travel behavior. The regional density correction coefficient, based on the "supply and demand balance" principle, provides an upward adjustment to the demand forecast for areas with scarce charging stations (low density), guiding planning towards areas with weak service and helping to optimize the overall balance of network layout. This composite forecasting framework of "basic model forecast + spatial and market mechanism correction" overcomes the shortcomings of simple regression models in characterizing spatial heterogeneity and market externalities, improving the spatial rationality and policy orientation of the forecast results.

[0036] In some embodiments, in step S3, the specific rules for classifying sites with inefficiency scores exceeding a threshold using a decision tree are as follows: when the site's return on investment is negative and the average annual load rate is lower than a first threshold, a complete exit strategy is implemented; when the site has high overlap and the average annual load rate is lower than a second threshold, a partial exit strategy is implemented, reducing its capacity by 50%; when the number of alternative sites within a preset distance of the site exceeds a redundancy threshold, a gradual exit strategy is implemented; and the remaining cases are listed as monitoring and observation.

[0037] In this embodiment, specific numerical examples of decision tree classification rules are provided. The first threshold (low load rate threshold) can be set to 0.1, the second threshold (lower load rate threshold) can be set to 0.3, and the redundancy threshold can be set to 1.5 (i.e., an average of more than 1.5 alternative stations within a 1.5-kilometer radius). Accordingly, the decision rules are divided into: complete withdrawal, partial withdrawal (50% capacity reduction), gradual withdrawal, and monitoring and observation.

[0038] The design logic of the aforementioned decision tree rule lies in transforming the decision to exit inefficient sites from a vague, comprehensive judgment into a series of clear, automatically executable logical judgments. It distinguishes between three different types of "inefficiency": unsustainable economics and extremely low utilization (complete exit), significant harm to the grid and low utilization (partial exit to quickly reduce negative impacts), and pure resource redundancy (gradual exit for a smooth transition). This categorized approach avoids service interruptions or resource waste that might result from a "one-size-fits-all" approach. More importantly, it incorporates the grid security indicator of "overlap" into the core consideration of the exit decision, ensuring that resource exit not only considers economics but also directly serves the grid security goal of "peak shaving," achieving coordinated optimization across both economic operation and grid security dimensions.

[0039] In some embodiments, in step S4, the process of calculating the elastic capacity in the second stage of the two-stage capacity planning method is as follows: calculate the minimum capacity required to meet the demand under three demand scenarios: pessimistic, baseline, and optimistic; calculate the expected capacity with the probability of occurrence of the three scenarios as weights; the elastic capacity is the portion of the expected capacity that exceeds the basic capacity of the first stage.

[0040] In this embodiment, the essence of this two-stage planning method is to decompose capacity decision-making into a "deterministic part" and a "risk response part." The first stage determines the rigid capacity that must be guaranteed to meet basic demand based on the current best estimate (baseline forecast). The second stage formally describes the uncertainty of future demand by constructing a simplified probability distribution (pessimistic, baseline, optimistic) and solves for an additional flexible capacity with the objective of "minimizing expected cost." The special feature of this method is that it is neither a rigid plan that completely ignores risk nor an overly complex stochastic plan, but rather achieves a balance between operability and robustness. It gives the final construction capacity a certain degree of flexibility, maintaining good economic efficiency under different demand scenarios, reducing the risk of severe facility idleness or premature congestion due to demand forecast deviations, and improving the adaptability of long-term investment.

[0041] In some embodiments, the specific operation of the improved non-dominated sorting genetic algorithm in step S5 includes: during population initialization, half of the individuals are generated using a heuristic method based on the decision rules in steps S3 and S4 of claim 1, and the other half are generated randomly; the crossover probability decreases linearly with the number of iterations, and the mutation probability changes sinusoidally periodically with the number of iterations; for individuals that violate the constraints, a special repair operator is executed to make them satisfy all the constraints.

[0042] The improvements described in the above embodiments are designed to address the characteristics of charging station planning, a high-dimensional, multi-constraint, and mixed-variable optimization problem. Hybrid initialization injects high-quality solutions obtained through heuristic rules into the population, accelerating initial convergence and avoiding the inefficiency of completely random initialization under complex constraints. The decreasing crossover probability with iteration allows the algorithm to focus on exploration (broad search) in the early stages and utilization (fine-tuning) in the later stages; the periodic change in mutation probability helps to re-energize exploration when trapped in local optima. A specialized constraint repair operator ensures that all solutions generated during iteration are feasible, which is crucial for handling "hard constraints" such as grid capacity, avoiding meaningless searches and improving search efficiency. These improvements enable standard multi-objective evolutionary algorithms to more effectively handle the complex planning model in this application, which contains a large number of 0-1 variables and continuous variables, thereby outputting a high-quality, diverse set of Pareto optimal solutions within a reasonable timeframe.

[0043] In some embodiments, step S6, the scheme comparison and verification includes: comparing the peak reduction rate obtained by this method with traditional site selection methods based only on spatial distance and POI density, methods that only consider the single action of new construction, and demand forecasting methods that do not use POI features, in order to demonstrate the advantages of this method in improving peak reduction effect, reducing cost, and improving demand forecasting accuracy.

[0044] This embodiment describes the specific implementation method for scheme comparison and verification. After the planning of this method is completed, three comparison models can be run in parallel: Model A (traditional spatial method), which selects sites based solely on the total POI density of candidate nodes and their distance from demand hotspots, without calculating spatiotemporal overlap or performing multi-action coordination; Model B (single new construction method), which uses the POI demand forecasting of this method, but only allows new sites, without considering expansion or withdrawal; Model C (no POI feature method), which adopts the spatiotemporal overlap analysis and multi-action coordination framework of this method, but uses a simple regional electric vehicle ownership average allocation method for demand forecasting. The comparison indicators include: overall peak shaving rate, peak shaving effect per unit investment cost (kW / 10,000 yuan), and the average absolute percentage error of demand forecasting after the actual operation of new / expanded sites.

[0045] The purpose of the above embodiments is to systematically deconstruct and verify the contributions of each core component of this method. By comparing this method with Model A, the incremental value brought by "spatiotemporal load coupling analysis" and "multi-action coordination" can be isolated and demonstrated; compared with Model B, the advantages of "existing network optimization (expansion and decommissioning)" compared with simply adding resources can be demonstrated; and compared with Model C, the role of "POI feature-driven refined demand forecasting" in improving planning accuracy can be demonstrated. This multi-angle, controlled-variable comparison method can objectively reveal the performance improvement achieved by the combination of technical features proposed in this method compared with existing technical routes at various levels, thus providing empirical support for the effectiveness and innovation of the method, rather than relying solely on theoretical explanations.

[0046] For example, the following is an embodiment of a charging station peak shaving planning method based on spatiotemporal coupling and POI collaborative decision-making in practical applications, as detailed below: The method in this embodiment includes the following steps: I. Data Acquisition and Preprocessing Data was collected from 546 existing charging stations, 371 candidate road network nodes, and 38 regional power grid loads. The existing charging station data includes: charging stations... Geographic coordinates of (i=1,2,...,546) Hourly charging load sequence for the whole year of 2024 (Unit: kW) Vector of the number of POIs of 20 types within a 500m radius Region code and rated capacity (Unit: kW) Candidate road network node data includes: nodes Geographic coordinates of (j=1,2,...,371) Vector of the number of POIs in 20 categories within a 500m radius and its area code ; Regional power grid data includes: regional The monthly peak load times for the 12 months of 2024 for (r=1,2,...,38) (Value range 0-23 hours) and regional power grid capacity limit (Unit: kW) Outlier detection is performed on the charging load data to identify outliers with negative values ​​or exceeding 1.5 times the rated capacity. Cubic spline interpolation is used for repair, followed by smoothing with a moving average of size 3. Standardization is performed on the POI data for all sites and candidate nodes, and the global mean is calculated for the k-th class of POIs. and standard deviation Perform Z-score standardization and Min-Max normalization to the interval; II. Calculation of Spatiotemporal Load Overlap Degree and Classification of Charging Stations From charging station 8760-hour load sequence Extract the monthly peak times for each month. Extract load data for all times of the month and calculate the 24-hour daily average load curve. (h=0,1,...,23):

[0047] in Let m be the number of days in month m. This refers to the time step corresponding to the h-hour on the d-th day of the m-th month; charging station The peak time in month m is defined as:

[0048] Define charging station With the region The function for determining coincidence at the peak in month m:

[0049] in The hour is the time tolerance window. For the region At the moment of maximum load in month m; Calculation charging station Annual spatiotemporal load overlap index:

[0050] in For monthly weighting, set the weights for summer months (m=6,7,8). Winter months (m=12,1,2) settings Other month settings ; Calculate the overlap vector of 546 charging stations The statistic, the first quartile and the third and fourth quartiles Set adaptive classification threshold ; Charging stations are classified into three categories based on their degree of overlap: when... When classified as a high overlap station, When classified as a low overlap station, The station is classified as having moderate overlap. For each charging station, calculate the annual average load factor and the percentage of remaining capacity. The annual average load factor is defined as:

[0051] The remaining capacity percentage is defined as:

[0052] Site attributes are labeled by combining overlap and load rate: high overlap sites and Marked as "load transfer required", low overlap station with remaining capacity Marked as "expandable"; III. POI-Driven Demand Forecasting Model The annual average daily charging volume was calculated as the target variable from the 8760-hour load series of 546 existing charging stations:

[0053] in This represents 24 hours on day d. Constructing the POI feature matrix The i-th row is the charging station. Normalized POI vectors are used to construct the target vector. ; Generating POI interaction features includes (Residential building × subway station) (Office building × parking lot) and (Shopping mall × restaurants) can be expanded into an augmented feature matrix; Ridge regression is used to train the POI weight vector. and intercept The objective function is:

[0054] Where regularization coefficient After training, nonnegativity constraints are applied to the weight vector.

[0055] and L1 normalization ; For candidate nodes Calculate the basic demand forecast:

[0056] in The augmented POI feature vector of the candidate node; Calculate candidate nodes The shortest distance to various POIs is defined, and the distance attenuation correction factor is defined:

[0057] in , , They are nodes Distance to the nearest residential POI, commercial POI, and transport hub POI (in meters); Calculate the region to which the candidate node belongs Charging station density:

[0058] Where Area represents the region's area (in km²), and the average density across the entire region is calculated. Define the regional correction factor:

[0059] Calculate the final demand forecast for the candidate nodes:

[0060] The CLIP function constrains the results to the range of kWh / day. For candidate nodes Set the radius of influence m, filter the set of spatially adjacent sites ,in Geographic distance; Gaussian kernel spatial weights are calculated for each neighboring site:

[0061] in m is a scale parameter; Predict the probability that a candidate node will be the peak in the h-th hour of the m-th month:

[0062] in Using the indicator function, the hour with the highest probability is selected as the predicted peak time. ; Calculate the prediction overlap between candidate nodes and their respective regions:

[0063] IV. Capacity Expansion Decisions For each highly overlapping site marked "load transfer required" Set the search radius m, select low-overlapping sites within the same area as the candidate expansion site set. ; For each candidate site Calculate the four-dimensional comprehensive score, including the time-varying peaks dimension. Remaining capacity dimension (Calculated by formula (6)) Spatial accessibility dimension Facility conditions dimension The overall score is:

[0064] Choose the highest-rated site As a target for capacity expansion; from highly overlapping sites Estimated average arrival rate of the 8760-hour load sequence Set the transfer ratio to 0.4 and calculate the load transferred to the expansion site. The charging service rate is defined as the reciprocal of the average service duration. Therefore, an average of 1.5 hours corresponds to This notation follows the standard notation of queuing theory. Indicates arrival rate, This represents the service rate, facilitating integration with subsequent M / M / c models and capacity calculations. If the available grid-connected capacity of the target expansion site is... (Unit: kW), and the rated power of a single pile is The theoretically configurable number of piles is estimated as follows: The average waiting time is calculated based on the M / M / c queuing model. Minimum number of charging stations per hour (10 minutes) The capacity expansion is calculated as follows:

[0065] And round up to the nearest multiple of 60kW; V. Site Selection and Size Determination Decisions For target areas where new charging stations are needed Filter the set of valid candidate nodes: For each valid candidate node, a five-dimensional evaluation index is constructed to assess demand potential. Spatiotemporal peak deviation Spatial accessibility in Weighted distance to high-demand areas and network synergy in To determine the distance to the nearest existing station and the economics of construction Based on land and construction costs, a comprehensive score is calculated:

[0066] Candidate nodes are sorted in descending order of score, and non-redundant nodes that are more than 1500m away from the selected nodes are added to the site construction plan; for the selected new nodes... A two-stage volume determination method is adopted. The first stage calculates the deterministic basic capacity:

[0067] Where 1.2 represents the 3-year demand growth rate, 0.65 represents the target load factor, and 8 represents the average daily charging hours. This indicates rounding up to the nearest multiple of 120kW; the second stage defines the set of demand scenarios, including pessimistic scenarios. (need (Probability 0.2) Baseline Scenario (need (Probability 0.6) Optimistic Scenario (need Given a probability of 0.2, the optimal elastic capacity is calculated, and the final construction capacity is:

[0068] The limit is 180-600kW. VI. Resource Exit Decisions For each charging station Calculate economic efficiency indicators, annual returns ,in Electricity price, annual return on investment The operational efficiency indicator is the annual average load factor. (Calculated by formula (5)); the redundancy index is:

[0069] Construct a comprehensive inefficiency score:

[0070] in This represents the average rate of return on investment. for The site uses decision tree classification, when The strategy is to exit completely when... and The strategy is partial exit (reducing capacity by 50%). The strategy is to gradually exit if necessary, otherwise the strategy is to monitor and observe. VII. Global Multi-Objective Optimization Model Define decision variables including new decision (j=1,...,371), capacity expansion (i=1,...,546) and the decision to close (i=1,...,546); Construct a three-objective optimization model, with objective 1 being to minimize the weighted peak load of the 38 regions:

[0071] in For regional weights, The load curve is generated for the new site based on the demand forecast using formula (13); Objective 2 is to minimize the total investment cost:

[0072] Of these, 800 yuan / kW represents the unit cost of new construction, and 500 yuan / kW represents the unit cost of capacity expansion. Annual operating costs; Objective 3 is to minimize the service quality loss.

[0073] in The waiting time is calculated based on the M / M / c model, and Gap is the uncovered demand. The constraints include grid capacity constraints, service coverage constraints, site capacity upper and lower limits constraints, mutual exclusion constraints for new sites, budget constraints, load factor constraints, and mutual exclusion constraints for expansion and shutdown. VIII. Solving with an improved NSGA-II algorithm The initial population size is 200, the maximum number of iterations is 500, and a hybrid initialization strategy is used. In each iteration, fast non-dominated sorting and crowding distance calculation are performed, and tournament selection is used to generate parents, with crossover probability... Probability of mutation Single-point crossover is used for newly created variables, and arithmetic crossover is used for expanded variables; After outputting the Pareto optimal solution set, the TOPSIS method is used to select the optimal solution and calculate the distance between the positive and negative ideal solutions. and The relative similarity is:

[0074] choose The largest option is recommended. IX. Comparison of Solution Output and Effect Evaluation To comprehensively evaluate the effectiveness of this method, four comparative methods were set up for experimental comparison. Method 1 is the traditional spatial site selection method, which selects sites based solely on POI density and geographical distance, without considering spatiotemporal load overlap; Method 2 is the single new construction method, which considers demand forecasting but not spatiotemporal overlap; Method 3 is the method without POI features, which considers spatiotemporal overlap but does not use POI features for demand forecasting; Method 4 is the method of this invention, which comprehensively considers spatiotemporal overlap and POI features, and performs three actions: expansion, new construction, and withdrawal.

[0075] The peak reduction effects of the four methods are compared as follows: Method 1 has a peak reduction rate of 5.26%, Method 2 has a peak reduction rate of 7.70%, Method 3 has a peak reduction rate of 11.81%, and Method 4 (this embodiment) has a peak reduction rate of 13.82%. Compared with Method 1, this method improves the peak reduction effect by 162.7% and the absolute peak reduction rate by 8.56 percentage points; compared with Method 2, the peak reduction effect is improved by 79.5% and the absolute peak reduction rate is improved by 6.12 percentage points; compared with Method 3, the peak reduction effect is improved by 17.0% and the absolute peak reduction rate is improved by 2.01 percentage points, fully verifying the effectiveness of this method.

[0076] A second aspect of the present invention provides a charging station peak shaving planning system based on spatiotemporal coupling and POI collaborative decision-making, for implementing any of the charging station peak shaving planning methods based on spatiotemporal coupling and POI collaborative decision-making described above, the system comprising: The data acquisition and preprocessing module is configured to: acquire charging load data, POI data and grid load data, and perform anomaly detection, interpolation and standardization processing; The spatiotemporal analysis and site classification module is configured to: calculate the spatiotemporal load overlap index of charging stations and classify charging stations based on the index; The demand forecasting module is configured to: train a POI feature-driven regression forecasting model and predict the charging demand and peak times of candidate nodes. The existing site optimization module is configured to: screen and determine expansion targets and expansion capacity for highly overlapping sites, and evaluate inefficient sites and generate exit strategies; The new site planning module is configured to: filter candidate nodes for new sites, score site selection, and determine the construction capacity of new sites; The multi-objective optimization solution module is configured to: construct and solve the multi-objective optimization model, and output a set of Pareto optimal planning solutions; The solution decision and output module is configured to recommend the final solution from the Pareto solution set and output an executable plan list and evaluation report.

[0077] refer to Figure 2To facilitate understanding of the above examples, this embodiment describes the hardware and software architecture of the planning system implementing the method. The system can be deployed on one or more servers, with each module built as a microservice. The data acquisition and preprocessing module connects to the charging operation platform database, electronic map API, and power grid dispatching system. The spatiotemporal analysis and site classification module and the demand forecasting module rely on Python's Scikit-learn library to implement the core algorithms. The existing site optimization module and the new site planning module contain a large number of business logic judgments. The multi-objective optimization solution module is based on JMetal or a custom-implemented improved NSGA-II algorithm. The scheme decision and output module generates a planning list in JSON format and a visual evaluation report in HTML format, which are provided to planners through a web interface.

[0078] In the system described above, each software module is designed as a functional unit with clearly defined input, processing, and output interfaces, and its functions strictly correspond to the six core steps of the method. This modular design not only ensures consistency from algorithmic logic to software implementation but also brings operational flexibility: for example, the accuracy of the demand forecasting module can be tested independently, or the optimization module can be rerun after parameter adjustments without affecting other parts. By integrating different data sources from the charging network, geographic information network, and power grid, and processing them under a unified optimization framework, the system physically achieves the fusion and collaborative computation of information from multiple systems, including "vehicle-road-station-network." Ultimately, the system transforms the complex planning model into interactive and executable software operations, enabling the method to evolve from a theoretical solution into a practical tool supporting planning decisions.

[0079] In some embodiments, such as Figure 3 This application also provides an electronic device, including a processor 301, a memory 302 and a bus 303 and a computer program stored in the memory. When the processor executes the program, it implements any of the charging station peak shaving planning methods based on spatiotemporal coupling and POI collaborative decision-making as described above.

[0080] Furthermore, as an executable solution, the drone hangar authorized landing system based on dynamic encryption authentication can be a computer unit, which can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer unit may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above-described computer unit structure is merely an example and does not constitute a limitation on the computer unit. It may include more or fewer components, or combine certain components, or use different components. For example, the computer unit may also include input / output devices, network access devices, buses, etc., which are not limited in this embodiment of the invention.

[0081] Furthermore, as an executable solution, the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The processor is the control center of the computer unit, connecting various parts of the entire computer unit via various interfaces and lines.

[0082] In some embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the charging station peak shaving planning method based on spatiotemporal coupling and POI collaborative decision-making described above in the embodiments of the present invention.

[0083] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the computer unit by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory and non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device. It should be noted that the content contained in the computer-readable medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.

[0084] Although the invention has been specifically shown and described in conjunction with preferred embodiments, those skilled in the art should understand that various changes in form and detail may be made to the invention without departing from the spirit and scope of the invention as defined in the appended claims, all of which shall be within the scope of protection of the invention.

Claims

1. A charging station peak shaving planning method based on spatiotemporal coupling and POI collaborative decision-making, characterized in that, Includes the following steps: Step S1: Collect charging load data of existing charging stations, POI data of candidate road network nodes, and regional power grid load data; perform outlier detection and interpolation repair on the charging load data; perform standardization processing on the POI data; calculate the spatiotemporal load overlap index of each charging station, which is the annual weighted overlap ratio between the peak load time of the charging station and the peak load time of the regional power grid; and classify the charging stations into high overlap stations, low overlap stations, and medium overlap stations based on this index. Step S2: Construct a POI feature-driven charging demand prediction model. Utilize the normalized POI data and historical charging volume of existing charging stations, and train the model weights using a regularized ridge regression algorithm. For candidate nodes, a linear combination of their normalized POI vector and model weights is calculated as the basic predicted value. An attenuation coefficient based on geographical distance and a correction coefficient based on regional charging station density are then introduced to obtain the predicted charging demand value for the candidate nodes. The peak time distribution of candidate nodes near existing sites is used to infer their predicted peak time and predicted spatiotemporal overlap. Step S3: For the high overlap stations, within a preset geographical search radius, select the low overlap stations located in the same power grid area as expansion candidates; For each candidate station for capacity expansion, a quantitative score is given based on four dimensions: time staggered peak, remaining capacity percentage, spatial distance, and facility conditions. The station with the highest score is selected as the expansion target. Based on the M / M / c model in queuing theory, and with the preset average user waiting time as a constraint, the minimum number of charging piles required to transfer the load from the high overlap station is calculated, and the expansion capacity of the target station is determined accordingly. For existing charging stations, inefficiency scores are calculated based on their return on investment, average annual load rate, and density of surrounding stations. For stations with inefficiency scores exceeding the threshold, one of four strategies is selected for implementation based on decision tree rules: complete withdrawal, partial withdrawal, gradual withdrawal, and monitoring and observation. Before implementation, the number of affected users is assessed to avoid service gaps. Step S4: Select candidate nodes whose predicted charging demand is higher than a set threshold and whose predicted spatiotemporal overlap is lower than a set threshold as valid new candidate points; for each valid new candidate point, quantify and score it from five dimensions: demand potential, spatiotemporal staggered peak degree, spatial accessibility, network synergy and construction economy, and select the node that meets the minimum distance constraint between stations as the new station in descending order of score. For the selected new site, a two-stage capacity planning method is used to determine its construction capacity: In the first stage, a deterministic basic capacity is calculated based on the demand forecast, the preset future demand growth rate, and the target load rate; In the second stage, three demand scenarios—pessimistic, baseline, and optimistic—and their probabilities of occurrence are set, and the elastic capacity is calculated with the goal of minimizing the expected cost; The final construction capacity of the new site is the sum of the basic capacity and the elastic capacity, and is subject to preset upper and lower capacity limits. Step S5: Construct a multi-objective optimization model. This model includes three objective functions: minimizing regional weighted peak load, minimizing total investment cost, and minimizing service quality loss. It also sets grid capacity constraints, service coverage constraints, site capacity constraints, budget constraints, and mutual exclusion constraints. The improved non-dominated sorting genetic algorithm NSGA-II is used to solve the model. The improvements include using a hybrid strategy to initialize the population, adaptively adjusting crossover and mutation probabilities, and introducing a constraint repair mechanism. Finally, the Pareto optimal solution set is output. Step S6: The Pareto optimal solution set is evaluated and sorted using the TopSIS approximation method, and the solution with the highest relative approximation is selected as the final recommended scheme; the list of expanded sites, the list of newly built sites, and the list of decommissioned sites under this scheme are output, the change in the total peak load of the region before and after the planning is calculated, and the peak reduction rate evaluation result is obtained.

2. The method according to claim 1, characterized in that, In step S1, the calculation process of the spatiotemporal load overlap index is as follows: extract the daily average load curve of each month from the annual load data of the charging station, find the moment with the largest load in the curve as the peak moment of that month; compare the monthly peak moment with the monthly peak moment of the power grid in the region, and if the time difference between the two is within the preset time window, it is determined that the month overlaps; count the number of overlapping months in the 12 months of the year, and assign different weight coefficients to the months of different seasons, and obtain the annual overlap index after weighted summation.

3. The method according to claim 1, characterized in that, In step S2, the distance attenuation coefficient is calculated based on the geographical distance from the candidate node to the nearest residential area, commercial area, and transportation hub. The closer the distance, the larger the coefficient. The regional charging station density correction coefficient is calculated based on the ratio of the number of charging stations per unit area in the region to the average density of the entire region. The lower the density, the larger the correction coefficient.

4. The method according to claim 1, characterized in that, In step S3, the specific rules for classifying sites with inefficiency scores exceeding the threshold using a decision tree are as follows: when the site's return on investment is negative and the average annual load rate is lower than the first threshold, a complete exit strategy is implemented; when the site has high overlap and the average annual load rate is lower than the second threshold, a partial exit strategy is implemented, reducing its capacity by 50%; when the number of alternative sites within a preset distance of the site exceeds the redundancy threshold, a gradual exit strategy is implemented; the remaining cases are listed as monitoring and observation.

5. The method according to claim 1, characterized in that, In step S4, the process of calculating the elastic capacity in the second stage of the two-stage capacity planning method is as follows: calculate the minimum capacity required to meet the demand under three demand scenarios: pessimistic, baseline, and optimistic; calculate the expected capacity using the probability of occurrence of the three scenarios as weights; the elastic capacity is the portion of the expected capacity that exceeds the basic capacity of the first stage.

6. The method according to claim 1, characterized in that, In step S5, the specific operation of the improved non-dominated sorting genetic algorithm includes: during population initialization, half of the individuals are generated using a heuristic method based on the decision rules in steps S3 and S4 of claim 1, and the other half are generated randomly; the crossover probability decreases linearly with the number of iterations, and the mutation probability changes sinusoidally with the number of iterations; for individuals that violate the constraints, a special repair operator is executed to make them satisfy all the constraints.

7. The method as described in claim 1, characterized in that, In step S6, the scheme comparison and verification includes comparing the peak reduction rate obtained by this method with traditional site selection methods based only on spatial distance and POI density, methods that only consider the single action of new construction, and demand forecasting methods that do not use POI features, so as to demonstrate the advantages of this method in improving peak reduction effect, reducing cost, and improving demand forecasting accuracy.

8. A charging station peak shaving planning system based on spatiotemporal coupling and POI collaborative decision-making, characterized in that, The system for implementing the method of any one of claims 1 to 7 comprises: The system for implementing the method of any one of claims 1 to 7 comprises: The data acquisition and preprocessing module is configured to: acquire charging load data, POI data and grid load data, and perform anomaly detection, interpolation and standardization processing; The spatiotemporal analysis and site classification module is configured to: calculate the spatiotemporal load overlap index of charging stations and classify charging stations based on the index; The demand forecasting module is configured to: train a POI feature-driven regression forecasting model and predict the charging demand and peak times of candidate nodes. The existing site optimization module is configured to: screen and determine expansion targets and expansion capacity for highly overlapping sites, and evaluate inefficient sites and generate exit strategies; The new site planning module is configured to: filter candidate nodes for new sites, score site selection, and determine the construction capacity of new sites; The multi-objective optimization solution module is configured to: construct and solve the multi-objective optimization model, and output a set of Pareto optimal planning solutions; The solution decision and output module is configured to recommend the final solution from the Pareto solution set and output an executable plan list and evaluation report.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 7.

Images