Off-grid distributed hydrogen-electricity integrated system site selection method and related products

Abstract

This invention belongs to the field of hydrogen-electricity integration technology, specifically involving a site selection method and related products for off-grid distributed hydrogen-electricity integration systems. It acquires multi-source heterogeneous raw data of the target area, preprocesses it to form a clean data matrix, predicts the number of electric vehicles and hydrogen fuel cell vehicles based on this matrix, and generates energy demand forecasts and photovoltaic power generation potential assessment data. Based on this, a multi-objective optimization model is constructed with the objectives of minimizing total lifecycle cost, maximizing energy self-sufficiency, maximizing service coverage, and maximizing carbon emission reduction. Solving the model yields a Pareto optimal solution set, from which the site selection scheme with the best overall benefits is determined. This invention effectively solves the planning deviation problems caused by data fragmentation and single objectives in traditional methods through multi-factor coupling analysis, achieving synergistic optimization of economy, reliability, serviceability, and environmental protection.

Description

Technical Field

[0001] This invention relates to the field of hydrogen-electricity integration technology, specifically to a site selection method and related products for an off-grid distributed hydrogen-electricity integration system. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Against the backdrop of a global energy structure transition towards cleaner and lower-carbon energy, distributed energy systems that deeply integrate renewable energy sources such as photovoltaics and wind power with hydrogen energy and electric vehicles have become an important development direction. Off-grid integrated hydrogen-electricity systems, as a new type of comprehensive energy infrastructure integrating power generation, hydrogen production, energy storage, and charging / refueling, are of strategic significance for promoting deep decarbonization in the transportation sector, improving energy efficiency, and ensuring energy supply in remote areas through their scientific planning and layout. The site selection decisions for such systems heavily rely on the fusion and analysis of multi-source heterogeneous data within the region, including natural resource endowments, traffic characteristics, geospatial information, and socio-economic development status. Therefore, a technical methodology capable of coordinating multiple factors and achieving precise and collaborative planning is urgently needed.

[0004] However, existing site selection methods generally suffer from problems such as fragmented multi-factor analysis, inaccurate dynamic demand forecasting, and a single optimization objective, making it difficult to effectively support the complex planning requirements of off-grid hydrogen-electric integrated systems. Specifically, there is a lack of an integrated method that can effectively fuse multi-source heterogeneous raw data, simultaneously and accurately predict vehicle energy demand and photovoltaic power generation potential, and then construct and solve a multi-objective optimization model that comprehensively considers economics, self-sufficiency, service coverage, and environmental protection. This technological deficiency often leads to site selection schemes that compromise between life-cycle costs, energy independence, user service scope, and carbon emission reduction benefits, failing to achieve a truly comprehensive optimal solution. Summary of the Invention

[0005] Addressing the shortcomings of existing off-grid distributed hydrogen-electricity integrated system site selection methods, the core objective of this invention is to propose a multi-dimensional coupling and multi-objective collaborative optimization method for off-grid distributed hydrogen-electricity integrated systems, along with related products. This method overcomes the limitations of traditional single-dimensional planning, enabling safe, economical, efficient, and low-carbon deployment of off-grid hydrogen-electricity integrated systems in complex scenarios. It promotes deep synergy between hydrogen energy and renewable energy, and provides a replicable integrated "source-grid-load-storage" solution for new energy vehicle refueling networks.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a site selection method for an off-grid distributed hydrogen-electricity integrated system.

[0007] A site selection method for an off-grid distributed hydrogen-electricity integrated system includes the following steps: Acquire multi-source heterogeneous raw data of the target area and preprocess the multi-source heterogeneous raw data to obtain a clean data matrix; Based on the clean data matrix, predict the number of electric vehicles and hydrogen fuel cell vehicles in the target area; Based on the clean data matrix and existing stock, energy demand forecast data and photovoltaic power generation potential assessment data are generated. Based on energy demand forecast data and photovoltaic power generation potential assessment data, a multi-objective optimization model is constructed with the objectives of minimizing the total life cycle cost, maximizing energy self-sufficiency, maximizing service coverage, and maximizing carbon emission reduction. Solve the multi-objective optimization model to obtain the Pareto optimal solution set, and determine the location scheme with the best overall benefits from the Pareto optimal solution set.

[0008] In one implementation of the first aspect of the present invention, the multi-source heterogeneous raw data includes photovoltaic resource data, traffic data, geographic information data, and infrastructure data; Preprocessing of multi-source heterogeneous raw data includes: spatiotemporal alignment, missing value imputation, and anomaly detection to obtain a clean data matrix.

[0009] In one implementation of the first aspect of the present invention, predicting the number of electric vehicles and hydrogen fuel cell vehicles in a target area includes: Based on historical traffic data in the clean data matrix and combined with a policy support strength function, the number of electric vehicles and hydrogen fuel cell vehicles in operation is calculated.

[0010] In one implementation of the first aspect of the present invention, the energy demand forecast data includes electric vehicle charging load time-series forecast data, charging demand breakdown forecast data, hydrogen fuel cell vehicle hydrogen refueling demand time-series forecast data, and hydrogen refueling demand characteristic forecast data. The detailed forecast data for charging demand includes DC fast charging demand power, AC slow charging demand power, peak demand, flat demand, off-peak demand, number of vehicles charging at the same time, average charging time, and average state of charge when vehicles arrive. Hydrogen demand characteristics forecast data include hydrogen refueling rate demand, peak hydrogen refueling demand, and daily hydrogen refueling volume; The photovoltaic power generation potential assessment data includes short-term photovoltaic output forecast data and medium- and long-term photovoltaic power generation potential assessment data.

[0011] In one implementation of the first aspect of the present invention, a CNN-LSTM hybrid prediction model is used to generate energy demand prediction data and photovoltaic power generation potential assessment data, wherein the input of the CNN-LSTM hybrid prediction model is a clean data matrix; In the CNN-LSTM hybrid prediction model, local and global spatial features of the input data are extracted and compressed through a multi-scale convolutional neural network. The compressed features are then concatenated with time series data and input into a bidirectional long short-term memory network for temporal dependency modeling. Attention mechanisms are used to weight key time steps, and the final outputs are time series prediction data for electric vehicle charging load, detailed prediction data for charging demand, time series prediction data for hydrogen refueling demand of hydrogen fuel cell vehicles, hydrogen refueling demand feature prediction data, and short-term photovoltaic output prediction data.

[0012] In one implementation of the first aspect of the present invention, constructing a multi-objective optimization model includes: The total lifecycle cost is defined as the discounted sum of the initial investment cost, operating cost, and grid interaction cost of each candidate site over the planning period. Energy self-sufficiency rate is defined as the ratio of total photovoltaic power generation to total charging demand and total hydrogen production demand within the planning period, taking into account the efficiency of the electrolyzer. Service coverage is defined as the ratio of the sum of weighted traffic flows of covered traffic nodes to the total traffic flow. Here, a traffic node is an intersection or road segment in the road network, and the weighted traffic flow of a traffic node is its historical average traffic flow. If the distance from any traffic node to the nearest station among all candidate stations is less than or equal to the preset maximum service radius, then the traffic node is determined to be covered. Carbon emission reduction is defined as the total amount of carbon emissions reduced by replacing steam methane reforming with hydrogen production through water electrolysis within the planning period, taking into account the grid emission factor.

[0013] In one implementation of the first aspect of the present invention, solving the multi-objective optimization model includes: The multi-objective optimization model is decomposed into a main problem and sub-problems. The main problem is responsible for the site selection and capacity determination of candidate sites, while the sub-problems are based on the site selection and capacity determination of the main problem to simulate power grid operation. The first optimization algorithm is used to perform a global search on the main problem to obtain a coarse-grained solution, and the second optimization algorithm is used to perform a local fine search on the coarse-grained solution to obtain an initial solution. Based on the initial solution, the subproblems are solved to obtain the dual variable vector. The cutting plane is generated according to the dual variable vector and fed back to the main problem to update the solution space. The process is iteratively executed until convergence.

[0014] As a further limitation of the first aspect of the present invention, the first optimization algorithm is an improved pelican optimization algorithm. When the improved pelican optimization algorithm performs a global search on the main problem, it introduces an adaptive perturbation factor to dynamically adjust the movement of individuals towards the optimal individual. The adaptive perturbation factor decreases as the number of iterations increases, so as to balance the algorithm's global exploration and local development capabilities.

[0015] As a further limitation of the first aspect of the present invention, in each iteration of the improved pelican optimization algorithm, for each individual, if the random number is less than a preset threshold, an exploration strategy is executed, and the individual position is updated according to the adaptive perturbation factor, the optimal individual position of the population, and the average individual position of the population; otherwise, an exploration strategy is executed, and the individual position is updated according to the optimal individual position of the population and three random individual positions.

[0016] In one implementation of the first aspect of the present invention, determining the location scheme with the best overall benefits from the Pareto optimal solution set includes: standardizing multiple objective function values ​​of each scheme in the Pareto optimal solution set. The information entropy of each objective is calculated based on the standardized objective function value, and the objective weight of each objective is determined based on the information entropy. Based on objective weights, the weighted distances from each scheme to the positive and negative ideal solutions are calculated, and the closeness of each scheme is calculated based on the weighted distances. Grey relational analysis is introduced to correct the proximity, resulting in a comprehensive decision index.

[0017] As a further limitation of the first aspect of the present invention, when calculating the weighted distances of each scheme to the positive ideal solution and the negative ideal solution, the standardized objective function values ​​are weighted according to objective weights to obtain a weighted standardization matrix; The positive ideal solution is defined as the maximum value corresponding to each objective in the weighted normalization matrix, and the negative ideal solution is defined as the minimum value corresponding to each objective in the weighted normalization matrix. The distances from each scheme to the positive and negative ideal solutions are calculated using the Euclidean distance formula.

[0018] As a further limitation of the first aspect of the present invention, when introducing grey relational analysis to correct the closeness, the weighted normalized matrix of each scheme in the Pareto optimal solution set is used as the comparison sequence, and the weighted normalized vector corresponding to the positive ideal solution is used as the reference sequence. The grey relational coefficients of each comparison sequence and the reference sequence are calculated, and the grey relational coefficients of all targets are averaged to obtain the grey relational degree. Finally, the proximity degree and the grey relational degree are linearly weighted and fused to obtain the comprehensive decision index.

[0019] Secondly, the present invention provides a site selection system for an off-grid distributed hydrogen-electricity integrated system.

[0020] A site selection system for an off-grid distributed hydrogen-electricity integrated system includes: The data preprocessing unit is configured to: acquire multi-source heterogeneous raw data of the target area, and preprocess the multi-source heterogeneous raw data to obtain a clean data matrix; The vehicle ownership prediction unit is configured to predict the ownership of electric vehicles and hydrogen fuel cell vehicles in a target area based on a clean data matrix. The energy forecasting unit is configured to generate energy demand forecast data and photovoltaic power generation potential assessment data based on the clean data matrix and inventory. The model building unit is configured to: construct a multi-objective optimization model based on energy demand forecast data and photovoltaic power generation potential assessment data, with the objectives of minimizing the total life cycle cost, maximizing energy self-sufficiency, maximizing service coverage, and maximizing carbon emission reduction; The site selection decision unit is configured to: solve a multi-objective optimization model to obtain a Pareto optimal solution set, and determine the site selection scheme with the best overall benefits from the Pareto optimal solution set.

[0021] Thirdly, the present invention provides a computer device, comprising: a processor and a computer-readable storage medium; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the off-grid distributed hydrogen-electricity integrated system location method of the first aspect of the present invention.

[0022] Fourthly, the present invention provides a computer-readable storage medium storing a computer program adapted to be loaded by a processor and execute the off-grid distributed hydrogen-electricity integrated system location method of the first aspect of the present invention.

[0023] Compared with the prior art, the beneficial effects of the present invention are: This invention innovatively proposes a site selection method for off-grid distributed hydrogen-electricity integrated systems. By constructing a full-chain technical solution of "data-prediction-optimization-decision," it effectively solves the core problem in existing technologies where fragmented multi-factor analysis leads to the inability to obtain a comprehensive optimal solution. The method first fuses and preprocesses heterogeneous raw data from multiple sources, including photovoltaic, transportation, and geography, to form a high-quality clean data matrix. Then, based on this matrix, it simultaneously and accurately predicts the number of electric vehicles and hydrogen fuel cell vehicles in the region, generating matching refined energy demand and photovoltaic power generation potential data. On this basis, it innovatively constructs a multi-objective optimization model integrating life-cycle cost, energy self-sufficiency rate, service coverage, and carbon emission reduction, and solves for the Pareto optimal solution set using an efficient algorithm. Finally, a scientific multi-criteria decision-making method is used to select the site selection scheme with the best comprehensive benefits from the solution set. This scheme achieves an organic unity of economy, reliability, serviceability, and environmental protection, overcoming the planning bias caused by single objectives or fragmented data in traditional methods, and significantly improving the overall efficiency and sustainable development capability of off-grid hydrogen-electricity integrated systems.

[0024] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0025] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0026] Figure 1 A schematic diagram of an off-grid super hydrogen power system architecture is provided as an exemplary embodiment of the present invention; Figure 2 A system architecture diagram provided for an exemplary embodiment of the present invention; Figure 3 A flowchart illustrating an off-grid distributed hydrogen-electricity integrated system location method provided as an exemplary embodiment of the present invention; Figure 4 A schematic diagram of the CNN-LSTM hybrid prediction model provided as an exemplary embodiment of the present invention; Figure 5 A schematic diagram of multi-scale CNN feature extraction provided as an exemplary embodiment of the present invention; Figure 6 A schematic diagram of bidirectional LSTM timing modeling provided for an exemplary embodiment of the present invention. Figure 1 ; Figure 7 A schematic diagram of the attention mechanism layer provided in an exemplary embodiment of the present invention; Figure 8 A schematic diagram of bidirectional LSTM timing modeling provided for an exemplary embodiment of the present invention. Figure 2 ; Figure 9 A schematic diagram of a multi-task output layer provided in an exemplary embodiment of the present invention; Figure 10 A flowchart illustrating the improved Pelican optimization algorithm provided as an exemplary embodiment of the present invention; Figure 11 A flowchart of multi-criteria decision analysis is provided as an exemplary embodiment of the present invention; Figure 12 A schematic diagram of an off-grid distributed hydrogen-electricity integrated system location system provided as an exemplary embodiment of the present invention; Figure 13 A schematic diagram of a computer device provided for an exemplary embodiment of the present invention. Detailed Implementation

[0027] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0028] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0029] Traditional centralized grid-connected charging stations suffer from problems such as unstable energy supply, high operating costs, high expansion costs, and significant environmental impact. Meanwhile, hydrogen refueling stations face long construction cycles, high construction and operating costs, an incomplete hydrogen supply chain, and limited grid support. To address these issues, off-grid distributed hydrogen-electricity integrated systems have emerged. These systems store energy through hydrogen energy storage systems, effectively solving the problems of renewable energy integration and energy conversion, reducing carbon emissions and environmental noise impact, while also reducing dependence on the grid and mitigating the high cost of grid expansion. These devices can be deployed in various locations, including city centers, suburbs, and highway service areas, enabling flexible energy supply. The advantages of off-grid distributed hydrogen-electricity integrated systems lie in their ability to utilize the intermittency and volatility of renewable energy sources, providing a more stable power supply through energy storage regulation. Furthermore, their distributed nature reduces dependence on the traditional grid, lowering operating costs.

[0030] However, the practical application of off-grid distributed hydrogen-electricity integrated systems is still in the early stages of exploring multi-scenario applications of hydrogen energy. Their site selection methods still have some shortcomings: 1) Single planning factor: Traditional methods often focus on single dimensions such as geographical coverage or safety distance, failing to comprehensively consider the dynamic correlation between key factors such as pedestrian and vehicle traffic density and hydrogen storage and transportation routes; 2) Insufficient economic efficiency: Centralized hydrogen production models suffer from high transportation costs and significant safety hazards; 3) Stability challenges: The strong volatility and intermittency of photovoltaic power generation inherently contradict the requirement for continuous and stable operation of hydrogen production equipment. Traditional planning methods struggle to achieve global optimization among complex variables, resulting in low energy utilization; 4) Algorithm limitations: Traditional models often employ single-objective optimization or simple weighted methods, making it difficult to effectively handle multiple conflicting objective functions and prone to getting trapped in local optima. This is particularly true in the field of off-grid distributed hydrogen-electricity integrated systems (including photovoltaic hydrogen production-hydrogen storage-hydrogen fuel cells-energy storage systems-hydrogen / charging piles), where a site selection method that comprehensively considers multiple factors such as safety, pedestrian and vehicle traffic, and storage and transportation is lacking, making it difficult to improve system utilization, reduce construction costs, and optimize layout.

[0031] To address the shortcomings of existing site selection methods for off-grid distributed hydrogen-electricity integrated systems, the core objective of this invention is to propose a multi-dimensional, coupled, and multi-objective collaborative optimization planning method. This method overcomes the limitations of traditional single-dimensional planning by establishing a dynamic coupling model encompassing multiple factors such as pedestrian density, vehicle demand, hydrogen storage and transportation networks, renewable energy fluctuation characteristics, safety distance constraints, and grid interaction potential. By quantitatively analyzing the correlations and conflicts between these factors (such as the synergy between vehicle demand and storage / transportation route efficiency, and the contradiction between photovoltaic fluctuations and hydrogen production stability), the method achieves dynamic matching between system site selection and energy supply and demand. This enables safe, economical, efficient, and low-carbon deployment of off-grid hydrogen-electricity integrated systems in complex scenarios, promotes deep synergy between hydrogen energy and renewable energy, and provides a replicable integrated "source-grid-load-storage" solution for new energy vehicle refueling networks.

[0032] like Figure 1 As shown, the off-grid distributed hydrogen-electricity integrated system mainly includes a photovoltaic power generation system, a water electrolysis hydrogen production system, a hydrogen compression system, a hydrogen buffer device, a hydrogen storage device, an energy storage system, a fuel cell power generation system, a charging system, a refueling system, and necessary pipelines and electrical cables. It constructs a multi-energy system architecture comprising a photovoltaic array, an electrolyzer, a hydrogen storage tank, a hydrogen fuel cell, and an energy storage battery. This system operates off-grid, using photovoltaic power generation to drive water electrolysis for hydrogen production. The stored hydrogen is then used to generate electricity from the hydrogen fuel cell, meeting the charging / refueling needs of service areas. Excess electricity can be stored in the energy storage system.

[0033] Table 1 shows the core module composition and functions of the off-grid distributed hydrogen-electricity integrated system.

[0034] Table 1: Composition and Functions of Core System Modules

[0035] like Figure 2The diagram illustrates the architecture of an off-grid distributed hydrogen-electricity integrated system. Solid arrows represent the main data and control flows, while dashed arrows represent system interactions between modules within a layer. Data interaction is achieved between all layers, and each module includes specific functional components and algorithm cores. Specifically, from left to right, the system is divided into four layers: the energy supply layer, the conversion and storage layer, the service provision layer, and the intelligent decision-making layer. The energy supply layer relies on a geospatial analysis system to complete resource assessment and site selection optimization, connecting to both the photovoltaic power generation system and the auxiliary energy system. The photovoltaic power generation system and the auxiliary energy system transmit energy to the energy management system in the conversion and storage layer via energy conversion. In the conversion and storage layer, the hydrogen-electricity conversion system and the multi-energy storage system achieve coordinated control based on hydrogen-electricity balance and energy storage status. Both systems synchronously transmit operational data to the energy management system, which uses energy status feedback to achieve overall control of this layer. Subsequently, the energy management system distributes energy to the vehicle-road-network coordination layer of the service provision layer. The interface and service provision layer connect the charging subsystem and the hydrogen refueling subsystem, which rely on load allocation and power coordination to achieve supply and demand linkage. The two subsystems are uniformly connected to the vehicle-road network collaborative interface and transmit their operating status back to it. After receiving the energy allocation and operating status information, the vehicle-road network collaborative interface takes on the energy supply from the energy management system on the one hand, and feeds back the operating status to the optimization decision engine of the intelligent decision layer on the other hand. In the intelligent decision layer, the prediction and planning system outputs prediction data to the dynamic scheduling system. The dynamic scheduling system transmits scheduling instructions back to the prediction and planning system. The optimization decision engine summarizes all the information, generates control instructions, and sends them back to the vehicle-road network collaborative interface, thereby realizing data interaction, energy flow, and closed-loop control between the four layers of architecture.

[0036] like Figure 3As shown, the off-grid distributed hydrogen-electricity integrated system site selection method of this implementation is carried out in six stages. Stage 1 involves multi-source data collection and fusion, sequentially completing meteorological data collection, traffic data collection, geographic data collection, power grid data collection, multi-source data fusion, and data preprocessing. The processed data flows into Stage 2, the new energy vehicle growth forecasting stage, which sequentially undergoes historical data analysis, policy impact assessment, competition model construction, growth curve prediction, and uncertainty quantification. The output of Stage 2 then proceeds to Stage 3, energy demand forecasting. Relying on feature engineering and CNN-LSTM model training, photovoltaic power output forecasting, charging load forecasting, and hydrogen refueling load forecasting are achieved, and the forecasting model is validated. Stage 3 completes the forecasting process. If the input is in stage four, multi-objective optimization modeling is carried out sequentially, the objective function is defined, constraints are set, decision variables are determined, model parameters are configured, and the optimization problem is constructed. Then, in stage five, the hybrid algorithm is solved, and in sequence, population initialization, improved algorithm iteration, Benders decomposition, constraint verification, and convergence analysis are performed to generate the Pareto optimal solution set. The output solution set of stage five is sent to stage six, multi-criteria decision output, and in sequence, decision matrix construction, TOPSIS analysis, grey relational analysis, comprehensive result summary, candidate scheme ranking and sensitivity analysis are carried out, and finally the optimal location scheme is output. The optimization feedback generated by the sensitivity analysis and parameter verification of each stage can be back-feeded back to the corresponding front-end stage to realize the full-process iterative optimization.

[0037] More specifically, it includes the following processes: Phase 1: Multidimensional data acquisition and preprocessing.

[0038] Collect photovoltaic resource data (historical solar irradiance, meteorological data), traffic data (vehicle flow, vehicle type ratio), geographic information data (elevation, slope, land type), and infrastructure data (road network, existing site distribution) within the target area. Establish a distributed data acquisition network, integrate heterogeneous data from multiple sources such as meteorology, traffic, geography, power grid, and economy, and form a high-quality, consistent dataset through preprocessing technologies such as spatiotemporal alignment, intelligent missing value filling, and anomaly detection, thus creating a database for site selection analysis.

[0039] Original multi-source data matrix expression: (1); in, For multi-source data matrix, It is meteorological data, and ; Traffic flow data, and ; It is geospatial data, and ; For power grid operation data, and ; For new energy vehicle data, and , dimension N represents the number of spatial locations, M represents the number of monitored variable types, and T represents the time series length. Represents the real number space, Represents the number of meteorological data source monitoring points. This represents the number of traffic flow data source monitoring points. Represents the dimension of geospatial characteristics. This represents the number of monitoring points for geospatial data sources, and the number of monitoring points represents the number of monitoring points for power grid operation data sources. This represents the number of data source monitoring points for new energy vehicles.

[0040] Formula (1) is used to construct a unified expression framework for multi-source heterogeneous raw data, which provides a standardized data input basis for subsequent spatiotemporal alignment and feature fusion.

[0041] In this implementation, the preprocessing process includes: Perform spatiotemporal alignment: (2); in, This is a clean data matrix after spatiotemporal alignment. For spatiotemporal alignment operators, To establish a unified time series reference, the format is as follows: The time resolution is , To establish a unified spatial grid coordinate system and grid resolution, a spatial grid reference standard is adopted. 0.5m, This is the data matrix before spatiotemporal alignment.

[0042] Formula (2) provides a spatiotemporal alignment operator that unifies data from different sources and at different resolutions onto a consistent spatiotemporal reference, ensuring the comparability and consistency of data in both spatial and temporal dimensions.

[0043] Missing value imputation: (3); in, for The subsequent data matrix, A function to fill in missing values. This is a multi-interpolation chain equation method, with the number of iterations being... Convergence tolerance The interpolation form is a regression model based on random forest.

[0044] Formula (3) realizes multiple imputation based on random forest, intelligently fills in missing data, effectively maintains the integrity and statistical characteristics of the dataset, and avoids information loss.

[0045] Anomaly detection, specifically, includes: (4); in, This is an anomaly detection and removal function. This is an isolated forest anomaly detection method, with the number of trees being [number missing]. Sampling size Abnormal score threshold .

[0046] Formula (4) constructs an anomaly detection and removal mechanism based on isolated forests, which efficiently identifies and removes outliers in the data, improving the quality of the clean data matrix and the robustness of subsequent prediction models.

[0047] The data processing expression after preprocessing the original multi-source data matrix: (5); in, This is the preprocessed clean data matrix; This is a preprocessing function.

[0048] Formula (5) integrates the results of spatiotemporal alignment, missing value imputation and anomaly detection, forming a high-quality, highly consistent clean data matrix, providing a solid data foundation for the entire location selection method.

[0049] Phase Two: New Energy Growth Forecast.

[0050] An improved Logistic-Competition model is used to predict the growth trend of electric vehicles and hydrogen fuel cell vehicles. The new energy vehicle growth prediction function is as follows: (6); (7); in, and The number of electric vehicles and hydrogen fuel cell vehicles at time t are respectively (in vehicles). and The intrinsic growth rates (1 / year) for electric vehicles and hydrogen fuel cell vehicles, respectively. and These are the environmental carrying capacity for electric vehicles and hydrogen fuel cell vehicles, respectively, i.e., the market saturation size (vehicles). and These are the competition coefficients for electric vehicles and hydrogen fuel cell vehicles, respectively, reflecting the intensity of competition among technological routes. and The policy impact coefficients for electric vehicles and hydrogen fuel cell vehicles are used to quantify the effectiveness of policy support. This is a function for policy support, taking into account factors such as subsidies and infrastructure.

[0051] Formulas (6) and (7) construct an improved Logistic-Competition model that incorporates competition coefficients and policy impact coefficients, accurately depicting the dynamic growth and mutual constraints of the two types of new energy vehicle ownership.

[0052] Phase 3: Energy supply and demand forecasting modeling.

[0053] Based on CNN-LSTM hybrid neural network models, such as Figure 4 , Figure 5 , Figure 6 , Figure 7 , Figure 8 and Figure 9 As shown, an energy demand forecasting model is constructed by extracting deep features and temporal dependencies from multi-source spatiotemporal data, providing data support for subsequent optimization decisions. It outputs short-term photovoltaic output forecasts and medium- to long-term photovoltaic power generation potential assessments. Regarding energy demand, it outputs time-series forecasts of electric vehicle charging loads, including detailed charging demand forecasts such as power level classification (DC fast charging demand, AC slow charging demand power) and time-period distribution characteristics (peak demand, average demand, and off-peak demand), as well as charging behavior characteristic forecasts (number of vehicles charging simultaneously, average charging time, and average SOC upon vehicle arrival) to improve energy supply configuration optimization. It also outputs time-series forecasts of hydrogen refueling demand for hydrogen fuel cell vehicles, hydrogen refueling demand characteristic forecasts (hydrogen refueling rate demand, peak hydrogen refueling demand, and daily hydrogen refueling volume), and electrolyzer hydrogen production demand forecasts to rationally plan service facility capacity.

[0054] Specifically, Figure 4 In this model, after the data is fed into the input layer, it first undergoes multi-dimensional spatial feature mining through a multi-scale CNN feature extraction module. The extracted feature data then enters the bidirectional LSTM temporal modeling module to perform the first round of forward and backward temporal correlation analysis. Subsequently, the data flows into the attention mechanism layer to achieve adaptive allocation of key feature weights. The features after weight optimization are then fed into the next-level bidirectional LSTM temporal modeling module for secondary temporal depth fitting. The processed data is then transmitted to the multi-task output layer to generate multi-target prediction results. The output results propagate the error back to the input layer through a backpropagation process. Based on the error, the parameters of each layer of the entire network are iteratively optimized to complete the closed-loop training of the model.

[0055] Figure 5In this process, the input data is first fed into a first convolutional layer equipped with 64 3×1 convolutional kernels and ReLU activation for shallow feature extraction. The extracted results are then processed by a max pooling layer with a pooling size of 2×1 to reduce the data dimensionality. Subsequently, the data enters a second convolutional layer equipped with 128 5×1 convolutional kernels and ReLU activation to mine mid-level features. The feature scale is then compressed again by a max pooling layer with a pooling size of 2×1. After that, the data flows into a third convolutional layer with 256 7×1 convolutional kernels and ReLU activation to extract deep multi-scale features. The obtained features are compressed by global average pooling, generating a feature tensor with a dimension of batch_size×128. This feature tensor is then reshaped to obtain data with a size of batch_size×24+128. The reshaped data is fed back into the first convolutional layer to perform cyclic optimization of parameters, thus forming a closed-loop feature extraction process for multi-scale CNN.

[0056] Figure 6 In the process, the feature data first enters the first bidirectional LSTM layer with 128 hidden units and return_sequences set to True to complete the first round of bidirectional temporal feature extraction. The output data is then fed into the Dropout layer with dropout_rate set to 0.2 to achieve random deactivation and suppress model overfitting. After noise reduction, the data flows into the second bidirectional LSTM layer with 64 hidden units and return_sequences set to True to further mine forward and backward temporal correlation features. After passing through the Dropout layer with dropout_rate set to 0.2 for secondary regularization, the data finally enters the third bidirectional LSTM layer with 32 hidden units and return_sequences set to False to complete the final temporal feature condensation and output the single-step feature result.

[0057] Figure 7 In the process, the preceding temporal feature data is first fed into a multi-head attention mechanism module equipped with 8 attention heads and a key_dim value of 64. The multi-head parallel operation is used to differentiate the feature weights and filter key information. The feature data after weight optimization enters the layer normalization module to complete the data distribution regularization and eliminate the interference caused by the internal covariate offset. The normalized data is finally input into the context vector generation module, which integrates global correlation information to complete the construction and output of the final context feature vector.

[0058] Figure 8In the process, the feature data processed by the pre-module is first fed into the first fully connected layer with 256 neurons and ReLU activation function for feature mapping transformation. The output data then enters the Dropout layer with dropout_rate set to 0.3 to randomly discard neurons to achieve model regularization and suppress overfitting. The denoised data flows into the second fully connected layer with 128 neurons and ReLU activation function to continue to complete feature dimensionality compression and nonlinear fitting. Then, it goes through the Dropout layer with dropout_rate equal to 0.2 for secondary random deactivation optimization. Finally, the data is sent to the third fully connected layer with 64 neurons and ReLU activation function to complete the final feature condensation and dimensionality convergence.

[0059] Figure 9 In the middle, the output layer synchronously splits into three branch tasks. The first branch performs 24-hour hydrogen refueling demand prediction based on Huber loss and completes the corresponding loss calculation and weighted combination. The second branch uses MSE loss to realize 24-hour charging demand prediction and performs corresponding loss calculation and weighted combination. The third branch uses MSE loss to realize 24-hour photovoltaic power output prediction and performs loss calculation and weighted combination. The weighted losses of the three types of tasks are summed to obtain the total loss. The total loss is backpropagated through the backpropagation link with Adam optimizer to complete the iterative optimization training of the multi-task model.

[0060] Spatiotemporal distribution feature prediction includes spatial distribution of demand (demand intensity in transportation node areas), weekday / holiday pattern recognition, seasonal variation patterns, and impact assessment of special events to determine site selection and service radius; in terms of uncertainty quantification, it includes prediction interval estimation and prediction error estimation to improve robustness and risk control management; in terms of feature importance, it identifies key influencing factors including the contribution of meteorological factors (radiation, temperature, cloud cover, etc.), the contribution of traffic factors (flow rate, vehicle type ratio), the contribution of time factors (hour, week, season, etc.), and the contribution of economic policy factors.

[0061] More specifically, the input layer of the CNN-LSTM hybrid neural network model is: (8); in, Input the feature vector for the time step; The length of the time series; For feature dimensions.

[0062] Formula (8) defines the input layer structure of the CNN-LSTM hybrid model, which organizes the preprocessed multidimensional spatiotemporal data into a tensor format suitable for deep learning model processing.

[0063] Perform multi-scale CNN feature extraction, including: (1) Calculate local features: (9); in, Local spatial features; For activation functions; These are the weights of the first convolutional layer; The first convolutional layer is biased; This is a convolution operation.

[0064] Formula (9) uses convolution kernels to extract local spatial features of the input data, capturing short-range dependencies and local patterns between adjacent regions or variables.

[0065] (2) Calculate global features: (10); in, For global spatial features; These are the weights of the second convolutional layer; This is the bias for the second convolutional layer.

[0066] Formula (10) extracts the global spatial features of the input data through operations such as global average pooling, and grasps the overall distribution trend and macro pattern.

[0067] (3) Perform feature compression: (11); in, This is the compressed feature vector; This is the global average pooling function.

[0068] Formula (11) compresses local and global features, reduces feature dimensions, decreases computational complexity, and provides refined feature vectors for subsequent time series modeling.

[0069] Perform bidirectional LSTM timing modeling, including: (1) Perform forward propagation: (12); in, This represents the hidden state of the forward LSTM layer. For Long Short-Term Memory (LSTM) network operations; Represents the compression feature at time t; This represents the forward hidden state at time (t-1).

[0070] (2) Backpropagation: (13); in, This represents the hidden state of the backward LSTM layer. This represents the backward hidden state at time (t+1).

[0071] Formulas (12) and (13) construct a bidirectional LSTM structure, which captures the long-term dependencies of time series from both forward and backward directions, and fully understands the impact of historical and future information on the current state.

[0072] (3) Perform feature splicing: (14); in, This represents the bidirectional temporal characteristics after splicing.

[0073] Formula (14) concatenates the outputs of the forward and backward LSTMs, integrating bidirectional temporal information to form a more comprehensive and richer temporal feature representation.

[0074] The attention mechanism includes: (1) Calculate attention score: (15); in, Score for attention; This is the transpose of the attention weight vector; It is the hyperbolic tangent activation function; These are the weights for the attention layer. For the recursive weights of the attention layer; This is the output state from the previous moment.

[0075] (2) Calculate attention weights: (16); in, is the attention weight; exp is the exponential operation.

[0076] (3) Calculate the context vector: (17); in, This is the context vector.

[0077] Formulas (15) and (16) quantify the importance of features at different times by calculating attention scores and weights, enabling the model to focus on the key time steps that have the greatest impact on the prediction results; Formula (17) uses attention weights to perform weighted summation on the LSTM output, generating a context vector containing key temporal information, which improves the prediction accuracy.

[0078] The output layer (final prediction result) is: (18).

[0079] in, The model predicts the output; These are the output layer weights; This is the output layer bias.

[0080] Formula (18) defines a multi-task output layer that maps context vectors to the final prediction results of various energy demands and photovoltaic output, thus achieving multi-objective synchronous prediction.

[0081] Phase 4: Construction of multi-objective optimization model.

[0082] Construct a multi-objective collaborative optimization system to balance economy, reliability, accessibility and environmental impact.

[0083] Objective 1: Minimize total lifecycle costs.

[0084] (19); in, Cost per lifetime (RMB). Let $ be the initial investment cost (in yuan) for the j-th site. Let $t$ be the operating cost (in yuan / year) of the j-th site in year t. Let $t$ be the grid interaction cost (in yuan / year) for the j-th station in year t. For the number of candidate sites, The planning period is in years. is the discount rate.

[0085] Formula (19) quantifies the comprehensive economic cost of the system throughout its entire life cycle, provides an economic objective for the optimization model, and guides the scheme to evolve towards a lower cost.

[0086] Objective 2: Maximize energy self-sufficiency.

[0087] (20); in, For energy self-sufficiency rate, Let t be the total photovoltaic power generation in year t (kWh). Let t be the total amount of electricity purchased from the grid in year t (kWh). Let t be the total charging demand in year t (kWh). Let t be the total hydrogen production demand in year t (kWh, equivalent). This refers to the efficiency of the electrolytic cell.

[0088] Formula (20) defines the energy self-sufficiency rate index, which measures the system's ability to meet its own load using local renewable energy sources. It is a core indicator for evaluating off-grid reliability.

[0089] Objective 3: Maximize service coverage.

[0090] (twenty one); in, To improve service coverage, All are weighting coefficients. The number of traffic nodes. Let be the traffic flow (vehicles / day) at the i-th traffic node. This indicates whether the i-th traffic node is covered (1 indicates coverage, 0 indicates no coverage). Let k be the service area (km²) of the kth station. The total planned area (km²) Let be the distance (km) from the i-th traffic node to the j-th station. Where N is the maximum service radius (km) and N is the total number of stations.

[0091] Formula (21) constructs a service coverage model that combines traffic flow with spatial distance to scientifically evaluate the degree to which the site layout meets user needs and the accessibility of services.

[0092] Objective 4: Maximize carbon emission reductions.

[0093] (twenty two); in, Carbon emission reduction (kg CO2), The power grid emission factor (kg CO2 / kWh) Let t be the hydrogen production volume (kg) in year t. The emission factor (kg CO2 / kg H2) for hydrogen production from steam methane reforming. The emission factor for hydrogen production by water electrolysis (kg CO2 / kg H2).

[0094] Formula (22) quantifies the carbon emission reduction benefits achieved by replacing gray hydrogen with green hydrogen, integrates environmental protection goals into the optimization model, and promotes the system towards low-carbon development.

[0095] Phase 5: Solving using a combination of multiple algorithms.

[0096] The main problem is decomposed using the Benders algorithm, such as... Figure 10As shown, a hybrid optimization algorithm combining the improved Pelican Optimization Algorithm and the Particle Swarm Optimization Algorithm is used to solve the model. The Pelican Optimization Algorithm is responsible for the global search, while the Particle Swarm Optimization Algorithm performs the local fine search to obtain the Pareto optimal solution set. Specifically, the algorithm first initializes the parameters, then generates an initial population and evaluates its initial fitness. Next, it calculates the crowding distance and initiates the iteration process. Within each iteration, adaptive parameter adjustments are performed, followed by individual traversal. A path is split based on whether the Rand value is less than 0.6. If the condition is met, an exploration strategy is used to move the individual towards the optimal one and handle boundary constraints. If the condition is not met, an exploration strategy is used to perform random exploration and boundary constraint handling. After both path handling is completed, the fitness of the new position is evaluated. Based on whether the new position is better, the individual position is updated or kept in its original position. The algorithm then moves to the next individual and repeats the individual traversal process until all individuals have been traversed. Finally, a collaborative algorithm mechanism including global search, local exploration, constraint handling, and multi-objective balancing is used to handle power grid safety constraints. The algorithm then determines whether it has converged. If it has not converged, it moves to the next generation and returns to the initial iteration step to restart the loop. If the convergence condition is met, the Pareto optimal solution is output, and the entire algorithm process ends.

[0097] Main problem (site selection and capacity): (twenty three); in, For investment cost function, As an auxiliary variable (representing the lower bound estimate of the objective function value of the subproblem), Constraints of the main problem; These are the location and capacity decision variables.

[0098] Formula (23) formalizes the complex site selection and capacity determination problem into a main problem. Its objective function includes investment cost and auxiliary variables from subproblems, which facilitates decomposition and solution.

[0099] Sub-problem (Power grid operation): (twenty four); in, For the power grid operating cost function, For power grid operation constraints, As a decision variable for power grid operation, Represents the vector of unit operating cost coefficients. Represents the total number of variables in the runtime environment. Represents the constant coefficient matrix. Represents a fixed resource constant term. Represents the coupling coefficient matrix. This represents the fixed-location optimal solution obtained through iterative solving of the main problem. It represents the p-dimensional nonnegative real space.

[0100] Formula (24) formalizes the power grid operation simulation into a sub-problem, evaluates its operating cost and feasibility under a given location scheme, and provides a basis for Benders decomposition.

[0101] Benders cut generation: (25); in, The dual variable vector of the subproblem (reflecting the marginal cost of the constraint). This is either a Bender optimal cut or a feasible cut. This represents the decision vector.

[0102] Formula (25) generates the Benders cutting plane based on the dual variables of the subproblems, feeds back the running information of the subproblems to the main problem, and guides the main problem to search for a better solution.

[0103] Location update strategy: (26); in, This represents the new position of the i-th individual. This represents the current position of the i-th individual. , , , For random disturbance coefficients; The optimal position for an individual in the population; It is an adaptive disturbance factor; The average position of the population; It is a random number; , , The positions are three random individuals.

[0104] Formula (26) defines the position update strategy for individual pelicans, which is to find a better solution by moving towards the optimal individual or exploring randomly. It is the core mechanism of the algorithm for optimization.

[0105] Adaptive perturbation factor design: (27); in, This represents the current iteration number; This represents the maximum number of iterations. It is a sine function.

[0106] Formula (27) incorporates an adaptive perturbation factor that decreases with each iteration, dynamically balancing the algorithm's global exploration capability in the early stages and its local development capability in the later stages.

[0107] Phase Six: Multi-criteria Decision Analysis and Solution Recommendation.

[0108] The entropy weight-TOPSIS multi-criteria decision-making method, such as Figure 11 As shown, the process includes steps such as data standardization, weight determination, distance calculation, correlation analysis, and comprehensive decision-making, to select the location scheme with the best overall benefits from the Pareto optimal solution set. Specifically, the Pareto solution set decision matrix is ​​input, and the original data is first standardized. On one hand, the entropy method is used to calculate the feature weight, information entropy, and weights sequentially, and output the objective weight vector. On the other hand, the AHP method is used to construct the judgment matrix, perform consistency checks, and calculate the feature vectors to obtain the subjective weight vector. The two types of weights are then combined and enter the weight integration stage. The integrated weights are first used for TOPSIS analysis, which involves constructing a weighted standardization matrix, determining the ideal solution, calculating the distance, calculating the relative proximity, and outputting the proximity. Then, grey relational analysis is carried out simultaneously. By determining the reference sequence, calculating the correlation coefficient, and calculating the correlation degree, the correlation degree vector is obtained. Subsequently, the proximity and correlation degree vectors are integrated and the schemes are ranked. The ranking results enter the sensitivity analysis stage. The weight adjustment suggestions generated by the analysis can be fed back to the Pareto solution set decision matrix input step for iterative optimization. After the sensitivity analysis is completed, robustness checks are performed, and the parameter optimization suggestions generated by the checks are also fed back to the upstream process. After all the verification work is completed, the optimal location scheme and related decision analysis reports are output, and the entire decision-making process ends.

[0109] In this implementation, the entropy weight method is used to determine the objective weights. Specifically, the weight calculation includes: (28); in, For the first Scheme No. Target proportion; The standardized target value; The number of options.

[0110] Formula (28) calculates the proportion of each scheme in each objective after standardization, providing basic data for the subsequent calculation of information entropy.

[0111] The calculation of information entropy includes: (29); Formula (29) measures the dispersion of each objective by information entropy. The greater the dispersion, the more decision information the objective provides, and the higher its weight should be.

[0112] The calculation of entropy weight includes: (30); in, For the first Objective weighting of objectives; For the first Target information entropy; The target quantity.

[0113] Formula (30) transforms information entropy into objective weights, avoiding the arbitrariness of subjective weighting and making the multi-criteria decision-making process more scientific and objective.

[0114] The closeness was calculated using the TOPSIS method: (1) Calculate the ideal distance: (31); in, Let i be the distance from the i-th solution to the positive ideal solution; The j-th objective is the positive ideal value.

[0115] Formula (31) calculates the Euclidean distance from each scheme to the positive ideal solution (the virtual scheme composed of the optimal values ​​of each objective). The smaller the distance, the better the scheme.

[0116] (2) Calculate the negative ideal distance: (32); Among them, among them, Let i be the distance from the i-th solution to the negative ideal solution; The negative ideal value is for the j-th objective.

[0117] Formula (32) calculates the Euclidean distance from each scheme to the negative ideal solution (the virtual scheme composed of the worst values ​​of each objective). The larger the distance, the better the scheme.

[0118] (3) Calculate the relative closeness: (33); in, Let be the relative similarity of the i-th scheme.

[0119] Formula (33) integrates positive and negative ideal distances through relative proximity, and obtains an evaluation index between 0 and 1, which is used for sorting schemes.

[0120] Improvements were made through grey relational analysis: (34); in, Let be the grey relational coefficient of the j-th objective in the i-th scheme; The resolution coefficient.

[0121] Formula (34) calculates the grey relational coefficient between each scheme sequence and the positive ideal reference sequence, and measures the similarity of their changing trends.

[0122] Calculate the grey relational degree: (35); in, Let be the grey relational degree of the i-th scheme.

[0123] Formula (35) calculates the average of the correlation coefficients of all targets to obtain the grey correlation degree, which supplements the evaluation of the scheme from the perspective of sequence similarity.

[0124] Calculate the comprehensive decision-making index (i.e., the overall score): (36); in, The comprehensive decision-making index for the i-th scheme; This is the proximity weighting coefficient.

[0125] Formula (36) linearly weights and integrates TOPSIS proximity and grey relational degree to form a more robust and comprehensive decision-making index.

[0126] The technical constraints employed throughout the entire algorithm execution process in this implementation include: (1) Power balance constraint: (37); in, Photovoltaic power generation (kW) Power exchanged with the grid (kW), The energy storage discharge power (kW) Energy storage charging power (kW), The operating power of the electrolytic cell (kW) For other load power (kW).

[0127] Formula (37) establishes a real-time power balance constraint for the system, ensuring that the supply and demand of power generation, power consumption, energy storage and hydrogen production are matched at any time, thus guaranteeing the stable operation of the system.

[0128] (2) Constraints on energy storage operation: (38); (39); in, Battery load status (%) Minimum load electrical state (%) Maximum load electrical state (%) The battery life health status threshold (%). Battery health status (%).

[0129] Formulas (38) and (39) set the upper and lower limits of the state of charge and the health threshold of the energy storage battery, preventing overcharging and over-discharging, extending battery life, and ensuring energy storage safety.

[0130] (3) Equipment capacity constraints: (40); (41); in, This represents the upper limit of photovoltaic installation capacity (kW). This represents the upper limit of the electrolytic cell capacity (kW).

[0131] Formulas (40) and (41) impose upper limit constraints on the installation capacity of key equipment such as photovoltaic cells and electrolytic cells, ensuring that the scheme meets physical feasibility and site limitations.

[0132] The economic constraint system adopted throughout the entire algorithm execution process of this implementation includes: (1) Investment budget constraints: (42); in, Maximum investment budget (RMB); For the first Initial investment cost per site (RMB); Represents the total number of sites.

[0133] Formula (42) sets an upper limit on the total investment budget, ensuring that the optimization scheme is economically feasible and in line with the project investment plan.

[0134] (2) Investment payback period constraint: (43); in, The maximum permissible investment payback period (in years). The investment payback period (in years) is the investment recovery period.

[0135] Formula (43) introduces an investment recovery period constraint, which ensures that the project can recover costs within the expected time and controls investment risks.

[0136] (3) Internal rate of return constraint: (44); in, Internal rate of return (%) The minimum required internal rate of return (%).

[0137] Formula (44) sets a lower limit for the internal rate of return, ensuring that the project has basic profitability and investment attractiveness.

[0138] The spatial constraint system used throughout the entire algorithm execution process in this implementation includes: (1) Land area constraints: (45); in, For the first The area occupied by each station (m²) 2 ), Total usable land area (m²) 2 ).

[0139] Formula (45) constrains the site area, ensuring that the site selection scheme complies with the actual limitations of available land resources.

[0140] (2) Service radius constraint: (46); in, Represents site service variables. Represents the distance between stations (km). Represents the maximum station service distance (km). This represents the total number of sites.

[0141] Formula (46) ensures that the site layout can effectively cover the target area by constraining the service radius, thus meeting the user's service accessibility requirements.

[0142] (3) Safety distance constraints: (47); in, Represents the geographical coordinates of site j. Represents the minimum safe distance between stations (km). The geographic coordinates of site k.

[0143] Formula (47) sets the minimum safe distance between stations, which meets the safety standards for hydrogen storage and use and avoids safety risks.

[0144] The environmental constraint system used throughout the entire algorithm execution process in this implementation includes: (1) Carbon emission constraints: (48); in, The maximum permissible carbon emissions (kg CO2); For the first Annual electricity consumption of the power grid (kWh).

[0145] Formula (48) imposes an upper limit on the total carbon emissions of the system, ensuring that the scheme meets the environmental protection policy requirements of the region or project.

[0146] (2) Noise constraint: (49); in, Equivalent noise level (dB), This represents the maximum permissible noise level (dB).

[0147] Formula (49) sets an upper limit for noise emissions, controls the noise pollution of the surrounding environment caused by the operation of the equipment, and meets environmental protection standards.

[0148] (3) Ecological constraints: (50); in, A collection of ecological protection zones; Representative number is The coordinates of the candidate construction site locations.

[0149] Formula (50) protects the ecological environment by prohibiting site selection within the set of ecological protection zones through ecological constraints, which is in line with the principle of sustainable development.

[0150] In summary, this invention aims to overcome existing technological bottlenecks and provide a scientific, accurate, efficient, and reliable site selection optimization method. It proposes a three-stage methodology of "prediction-optimization-decision," establishes a collaborative planning theoretical framework for the coupling of multiple systems including vehicles, roads, networks, and hydrogen, constructs an improved Logistic-Competition growth model and a CNN-LSTM hybrid prediction model, achieving accurate prediction across multiple time scales, designs a hybrid optimization strategy combining an improved Pelican optimization algorithm and a Benders decomposition algorithm, solving the problem of solving high-dimensional nonlinear optimization problems, establishes a four-layer bidirectional collaborative system architecture, achieves reliable operation under completely off-grid conditions, develops an entropy weight-TOPSIS-GRA integrated decision-making method, and provides a scientific and objective scheme evaluation system.

[0151] Figure 12 An off-grid distributed hydrogen-electricity integrated system location system is shown, comprising: The data preprocessing unit 1201 is configured to: acquire multi-source heterogeneous raw data of the target area, and preprocess the multi-source heterogeneous raw data to obtain a clean data matrix; The vehicle stock prediction unit 1202 is configured to predict the stock of electric vehicles and hydrogen fuel cell vehicles in a target area based on a clean data matrix. Energy forecasting unit 1203 is configured to generate energy demand forecast data and photovoltaic power generation potential assessment data based on the clean data matrix and inventory. Model building unit 1204 is configured to: construct a multi-objective optimization model based on energy demand forecast data and photovoltaic power generation potential assessment data, with the objectives of minimizing the total life cycle cost, maximizing energy self-sufficiency, maximizing service coverage, and maximizing carbon emission reduction; The site selection decision unit 1205 is configured to: solve a multi-objective optimization model to obtain a Pareto optimal solution set, and determine the site selection scheme with the best overall benefits from the Pareto optimal solution set.

[0152] It is understood that the aforementioned units can be individually or entirely merged into one or more other units, or some of the units can be further divided into multiple functionally smaller units. This achieves the same operation without affecting the technical effects of the embodiments of the present invention. The aforementioned units are based on logical functional division. In practical applications, the function of one unit can be implemented by multiple units, or the function of multiple units can be implemented by one unit. In other embodiments of the present invention, the system may also include other units. In practical applications, these functions can also be implemented with the assistance of other units, and can be implemented collaboratively by multiple units.

[0153] According to another embodiment of the present invention, the system of this embodiment can be constructed by running a computer program (including program code) capable of performing the steps involved in the corresponding method of the present invention on a general-purpose computing device, such as a computer, which includes processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). The computer program can be recorded on, for example, a computer-readable recording medium, loaded into the aforementioned computing device through the computer-readable recording medium, and run therein.

[0154] Figure 13 A computer device is shown, which includes a processor 1301, a communication interface 1302, and a computer-readable storage medium 1303. The processor 1301, communication interface 1302, and computer-readable storage medium 1303 can be connected via a bus or other means.

[0155] The communication interface 1302 is used to receive and send data. The computer-readable storage medium 1303 can be stored in the memory of the electronic device. The computer-readable storage medium 1303 is used to store computer programs, which include program instructions. The processor 1301 is used to execute the program instructions stored in the computer-readable storage medium 1303.

[0156] The processor 1301 is the computing and control core of the electronic device. It is suitable for implementing one or more instructions, specifically for loading and executing one or more instructions to achieve the corresponding method flow or corresponding function.

[0157] Processor 1301 is configured to perform the following procedure: Acquire multi-source heterogeneous raw data of the target area and preprocess the multi-source heterogeneous raw data to obtain a clean data matrix; Based on the clean data matrix, predict the number of electric vehicles and hydrogen fuel cell vehicles in the target area; Based on the clean data matrix and existing stock, energy demand forecast data and photovoltaic power generation potential assessment data are generated. Based on energy demand forecast data and photovoltaic power generation potential assessment data, a multi-objective optimization model is constructed with the objectives of minimizing the total life cycle cost, maximizing energy self-sufficiency, maximizing service coverage, and maximizing carbon emission reduction. Solve the multi-objective optimization model to obtain the Pareto optimal solution set, and determine the location scheme with the best overall benefits from the Pareto optimal solution set.

[0158] This invention also provides a computer-readable storage medium, which is a memory device in an electronic device for storing programs and data. It is understood that the computer-readable storage medium here may include both built-in storage media in the electronic device and extended storage media supported by the electronic device. The computer-readable storage medium provides storage space for storing the processing system of the electronic device.

[0159] Furthermore, this storage space also contains one or more instructions suitable for loading and execution by the processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory; alternatively, it can also be at least one computer-readable storage medium located remotely from the aforementioned processor.

[0160] In one embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes the one or more instructions stored in the computer-readable storage medium to perform the following process: Acquire multi-source heterogeneous raw data of the target area and preprocess the multi-source heterogeneous raw data to obtain a clean data matrix; Based on the clean data matrix, predict the number of electric vehicles and hydrogen fuel cell vehicles in the target area; Based on the clean data matrix and existing stock, energy demand forecast data and photovoltaic power generation potential assessment data are generated. Based on energy demand forecast data and photovoltaic power generation potential assessment data, a multi-objective optimization model is constructed with the objectives of minimizing the total life cycle cost, maximizing energy self-sufficiency, maximizing service coverage, and maximizing carbon emission reduction. Solve the multi-objective optimization model to obtain the Pareto optimal solution set, and determine the location scheme with the best overall benefits from the Pareto optimal solution set.

[0161] The present invention also provides a computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the following process: Acquire multi-source heterogeneous raw data of the target area and preprocess the multi-source heterogeneous raw data to obtain a clean data matrix; Based on the clean data matrix, predict the number of electric vehicles and hydrogen fuel cell vehicles in the target area; Based on the clean data matrix and existing stock, energy demand forecast data and photovoltaic power generation potential assessment data are generated. Based on energy demand forecast data and photovoltaic power generation potential assessment data, a multi-objective optimization model is constructed with the objectives of minimizing the total life cycle cost, maximizing energy self-sufficiency, maximizing service coverage, and maximizing carbon emission reduction. Solve the multi-objective optimization model to obtain the Pareto optimal solution set, and determine the location scheme with the best overall benefits from the Pareto optimal solution set.

[0162] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can implement the described functions using different methods for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0163] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic cable, digital cable) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data processing device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0164] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A site selection method for an off-grid distributed hydrogen-electricity integrated system, characterized in that, The process includes the following: Acquire multi-source heterogeneous raw data of the target area and preprocess the multi-source heterogeneous raw data to obtain a clean data matrix; Based on the clean data matrix, predict the number of electric vehicles and hydrogen fuel cell vehicles in the target area; Based on the clean data matrix and existing stock, energy demand forecast data and photovoltaic power generation potential assessment data are generated. Based on energy demand forecast data and photovoltaic power generation potential assessment data, a multi-objective optimization model is constructed with the objectives of minimizing the total life cycle cost, maximizing energy self-sufficiency, maximizing service coverage, and maximizing carbon emission reduction. Solve the multi-objective optimization model to obtain the Pareto optimal solution set, and determine the location scheme with the best overall benefits from the Pareto optimal solution set.

2. The site selection method for off-grid distributed hydrogen-electricity integrated systems as described in claim 1, characterized in that, The multi-source heterogeneous raw data includes photovoltaic resource data, transportation data, geographic information data, and infrastructure data; Preprocessing of multi-source heterogeneous raw data includes: spatiotemporal alignment, missing value imputation, and anomaly detection to obtain a clean data matrix.

3. The site selection method for off-grid distributed hydrogen-electricity integrated systems as described in claim 1, characterized in that, Predict the number of electric vehicles and hydrogen fuel cell vehicles in the target area, including: Based on historical traffic data in the clean data matrix and combined with a policy support strength function, the number of electric vehicles and hydrogen fuel cell vehicles in operation is calculated.

4. The site selection method for off-grid distributed hydrogen-electricity integrated systems as described in claim 1, characterized in that, Energy demand forecast data includes time-series forecast data for electric vehicle charging load, detailed forecast data for charging demand, time-series forecast data for hydrogen refueling demand for hydrogen fuel cell vehicles, and characteristic forecast data for hydrogen refueling demand. The detailed forecast data for charging demand includes DC fast charging demand power, AC slow charging demand power, peak demand, flat demand, off-peak demand, number of vehicles charging at the same time, average charging time, and average state of charge when vehicles arrive. Hydrogen demand characteristics forecast data include hydrogen refueling rate demand, peak hydrogen refueling demand, and daily hydrogen refueling volume; The photovoltaic power generation potential assessment data includes short-term photovoltaic output forecast data and medium- and long-term photovoltaic power generation potential assessment data.

5. The site selection method for off-grid distributed hydrogen-electricity integrated systems as described in claim 1, characterized in that, A CNN-LSTM hybrid prediction model was used to generate energy demand forecast data and photovoltaic power generation potential assessment data. The input of the CNN-LSTM hybrid prediction model was a clean data matrix. In the CNN-LSTM hybrid prediction model, local and global spatial features of the input data are extracted and compressed through a multi-scale convolutional neural network. The compressed features are then concatenated with time series data and input into a bidirectional long short-term memory network for temporal dependency modeling. Attention mechanisms are used to weight key time steps, and the final outputs are time series prediction data for electric vehicle charging load, detailed prediction data for charging demand, time series prediction data for hydrogen refueling demand of hydrogen fuel cell vehicles, hydrogen refueling demand feature prediction data, and short-term photovoltaic output prediction data.

6. The site selection method for off-grid distributed hydrogen-electricity integrated systems as described in claim 1, characterized in that, Construct a multi-objective optimization model, including: The total lifecycle cost is defined as the discounted sum of the initial investment cost, operating cost, and grid interaction cost of each candidate site over the planning period. Energy self-sufficiency rate is defined as the ratio of total photovoltaic power generation to total charging demand and total hydrogen production demand within the planning period, taking into account the efficiency of the electrolyzer. Service coverage is defined as the ratio of the sum of weighted traffic flows of covered traffic nodes to the total traffic flow. Here, a traffic node is an intersection or road segment in the road network, and the weighted traffic flow of a traffic node is its historical average traffic flow. If the distance from any traffic node to the nearest station among all candidate stations is less than or equal to the preset maximum service radius, then the traffic node is determined to be covered. Carbon emission reduction is defined as the total amount of carbon emissions reduced by replacing steam methane reforming with hydrogen production through water electrolysis within the planning period, taking into account the grid emission factor.

7. The site selection method for off-grid distributed hydrogen-electricity integrated systems as described in claim 1, characterized in that, Solving multi-objective optimization models, including: The multi-objective optimization model is decomposed into a main problem and sub-problems. The main problem is responsible for the site selection and capacity determination of candidate sites, while the sub-problems are based on the site selection and capacity determination of the main problem to simulate power grid operation. The first optimization algorithm is used to perform a global search on the main problem to obtain a coarse-grained solution, and the second optimization algorithm is used to perform a local fine search on the coarse-grained solution to obtain an initial solution. Based on the initial solution, the subproblems are solved to obtain the dual variable vector. The cutting plane is generated according to the dual variable vector and fed back to the main problem to update the solution space. The process is iteratively executed until convergence.

8. The site selection method for off-grid distributed hydrogen-electricity integrated systems as described in claim 7, characterized in that, The first optimization algorithm is the improved Pelican Optimization Algorithm. When performing a global search on the main problem, the improved Pelican Optimization Algorithm introduces an adaptive perturbation factor to dynamically adjust the movement of individuals towards the optimal individual. The adaptive perturbation factor decreases as the number of iterations increases, in order to balance the algorithm's global exploration and local development capabilities.

9. The site selection method for an off-grid distributed hydrogen-electricity integrated system as described in claim 8, characterized in that, In each iteration of the improved Pelican optimization algorithm, if the random number is less than a preset threshold for each individual, the development strategy is executed, and the individual position is updated according to the adaptive perturbation factor, the optimal individual position of the population, and the average position of the population. Otherwise, execute the exploration strategy, updating the individual position based on the optimal individual position in the population and three random individual positions.

10. The site selection method for an off-grid distributed hydrogen-electricity integrated system as described in claim 1, characterized in that, Determining the optimal location scheme with the best overall benefits from the Pareto optimal solution set includes: standardizing the multiple objective function values ​​of each scheme in the Pareto optimal solution set; The information entropy of each objective is calculated based on the standardized objective function value, and the objective weight of each objective is determined based on the information entropy. Based on objective weights, the weighted distances from each scheme to the positive and negative ideal solutions are calculated, and the closeness of each scheme is calculated based on the weighted distances. Grey relational analysis is introduced to correct the proximity, resulting in a comprehensive decision index.

11. The site selection method for an off-grid distributed hydrogen-electricity integrated system as described in claim 10, characterized in that, When calculating the weighted distances of each scheme to the positive and negative ideal solutions, the standardized objective function values ​​are weighted according to objective weights to obtain the weighted standardization matrix; The positive ideal solution is defined as the maximum value corresponding to each objective in the weighted normalization matrix, and the negative ideal solution is defined as the minimum value corresponding to each objective in the weighted normalization matrix. The distances from each scheme to the positive and negative ideal solutions are calculated using the Euclidean distance formula.

12. The site selection method for off-grid distributed hydrogen-electricity integrated systems as described in claim 11, characterized in that, When introducing grey relational analysis to correct the closeness, the weighted normalized matrix of each scheme in the Pareto optimal solution set is used as the comparison sequence, and the weighted normalized vector corresponding to the positive ideal solution is used as the reference sequence. The grey relational coefficients of each comparison sequence and the reference sequence are calculated, and the grey relational coefficients of all targets are averaged to obtain the grey relational degree. Finally, the proximity degree and the grey relational degree are linearly weighted and fused to obtain the comprehensive decision index.

13. A site selection system for an off-grid distributed hydrogen-electricity integrated system, characterized in that, include: The data preprocessing unit is configured to: acquire multi-source heterogeneous raw data of the target area, and preprocess the multi-source heterogeneous raw data to obtain a clean data matrix; The vehicle ownership prediction unit is configured to predict the ownership of electric vehicles and hydrogen fuel cell vehicles in a target area based on a clean data matrix. The energy forecasting unit is configured to generate energy demand forecast data and photovoltaic power generation potential assessment data based on the clean data matrix and inventory. The model building unit is configured to: construct a multi-objective optimization model based on energy demand forecast data and photovoltaic power generation potential assessment data, with the objectives of minimizing the total life cycle cost, maximizing energy self-sufficiency, maximizing service coverage, and maximizing carbon emission reduction; The site selection decision unit is configured to: solve a multi-objective optimization model to obtain a Pareto optimal solution set, and determine the site selection scheme with the best overall benefits from the Pareto optimal solution set.

14. A computer device, characterized in that, include: Processor and computer-readable storage media; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the off-grid distributed hydrogen-electricity integrated system location method as described in any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and executed as described in any one of claims 1 to 12 for the location method of an off-grid distributed hydrogen-electricity integrated system.

Images