Urban expansion area transformer substation layout method and system based on artificial intelligence algorithm

Abstract

The invention discloses a city expansion area transformer substation layout method and system based on an artificial intelligence algorithm. The method comprises the following steps: predicting future electrical load demand data of an area by mining an association rule between geographic space data and historical load data of the city expansion area; analyzing the prediction data by using a large language model, partitioning the city expansion area according to an analysis result, obtaining each sub-area and load difference data thereof, combining the data with historical power distribution network grid structure data of the city expansion area, and processing the data through the large language model to obtain an initial layout scheme of the transformer substation; by simulating the power supply capacity of each sub-region at different load growth rates, the initial layout scheme of the transformer substation is optimized, an optimization scheme is obtained and executed, and the optimal layout of the transformer substation is obtained by comprehensively applying artificial intelligence algorithms such as a large language model and a machine learning algorithm dynamic planning algorithm. Accurate prediction of the electrical load demand of the urban expansion area and effective optimization of the transformer substation layout scheme are realized.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method and system for arranging substations in an urban expansion area based on an artificial intelligence algorithm. Background Art

[0002] In urban expansion areas, distribution network planning faces complex challenges in substation site selection and capacity configuration. Currently, substation layout plans are usually determined based on existing load forecast data, but the load growth trend in the edge areas of urban expansion is difficult to accurately predict, resulting in limited accuracy of existing planning plans.

[0003] It can be seen that when the uncertainty of load growth in the planning area leads to dynamic changes in the supply and demand balance relationship, how to determine the substation layout and capacity configuration plan has become a technical problem that technical personnel in this field need to solve urgently. Summary of the invention

[0004] The present invention provides a method and system for substation layout in an urban expansion area based on an artificial intelligence algorithm, which solves the problem of how to determine the substation layout and capacity configuration plan when the uncertainty of load growth in the planning area leads to dynamic changes in the supply and demand balance relationship.

[0005] In order to solve the above technical problems, the first aspect of the present invention provides a method for substation layout in an urban expansion area based on an artificial intelligence algorithm, comprising: Obtaining geographic spatial data and historical load data of the urban expansion area, and using a large language model to mine the association rules between the geographic spatial data and the historical load data, and obtaining key features to predict the power load demand data of the urban expansion area within a target period; Analyzing the power load demand data by using the large language model to obtain load change characteristics to partition and cluster the urban expansion area and obtain load difference data of each sub-area; Collecting historical distribution network structure data of the urban expansion area and combining it with the load difference data to process it through the large language model to obtain an initial substation layout plan; Analyzing the historical load data by a machine learning algorithm to obtain a load growth probability distribution of each sub-region within the target period to simulate the power supply capacity of each sub-region under different load growth rates; A dynamic programming algorithm is used to analyze the power supply capacity of each of the sub-areas to optimize the initial substation layout plan, obtain the substation optimized layout plan and execute it.

[0006] A second aspect of the present invention provides a system for substation layout in urban expansion area based on artificial intelligence algorithm, comprising: A demand forecasting module is used to obtain geographic spatial data and historical load data of the urban expansion area, and use a large language model to mine the association rules between the geographic spatial data and the historical load data to obtain key features to predict the power load demand data of the urban expansion area within a target period; A partition clustering module is used to analyze the power load demand data through the large language model to obtain load change characteristics to partition and cluster the urban expansion area to obtain each sub-area and its load difference data; A scheme generation module, used for collecting historical distribution network structure data of the urban expansion area, and combining it with the load difference data, so as to obtain an initial substation layout scheme through processing by the large language model; A capacity simulation module is used to analyze the historical load data through a machine learning algorithm to obtain the load growth probability distribution of each sub-area within the target period, so as to simulate the power supply capacity of each sub-area under different load growth rates; The scheme optimization module is used to analyze the power supply capacity of each sub-area by using a dynamic programming algorithm to optimize the initial layout scheme of the substation, obtain the optimized layout scheme of the substation and execute it.

[0007] Compared with the prior art, the embodiments of the present invention have the following advantages: (1) By mining the correlation between geographic spatial data and historical load data through a large language model, the power load demand in urban expansion areas can be more accurately predicted. The urban expansion areas are divided and clustered according to the load change characteristics to obtain the load difference data of each sub-area, thereby optimizing the substation layout and capacity configuration and improving resource utilization efficiency. (2) Analyze historical load data through machine learning algorithms and simulate the power supply capacity of each sub-region under different load growth rates to ensure the stability and reliability of the power supply system. Use dynamic programming algorithms to analyze power supply capacity to optimize the initial layout of substations, provide scientific basis for urban planners and decision makers, and improve decision-making efficiency. (3) By comprehensively using artificial intelligence algorithms such as large language models and machine learning algorithms, it is possible to accurately predict the electricity load demand in urban expansion areas and optimize the substation layout and capacity configuration plans, which will help improve resource utilization efficiency and ensure the stability and reliability of the power supply system, providing a scientific basis for urban planners and decision makers. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the implementation mode will be briefly introduced below. Obviously, the drawings described below are only some implementation modes of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0009] Figure 1 It is a flow chart of a method for substation layout in an urban expansion area based on an artificial intelligence algorithm provided by a certain embodiment of the present invention; Figure 2 It is a structural diagram of a substation layout system for an urban expansion area based on an artificial intelligence algorithm provided by a certain embodiment of the present invention. DETAILED DESCRIPTION

[0010] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings and embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0011] In the description of this application, the terms "first", "second", "third", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the feature. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0012] In the description of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two components. The terms "vertical", "horizontal", "left", "right", "upper", "lower" and similar expressions used herein are for illustrative purposes only, and do not indicate or imply that the system or component referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0013] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood by specific circumstances.

[0014] At present, with the expansion of the city, the load in the area will also show a rapid growth trend, but the uncertainty of load growth will make it difficult to stabilize the supply and demand balance in the area. Therefore, in the early stage of planning, the substation layout scheme determined based on the existing load forecast data is often difficult to adapt to the long-term load growth demand in the area: on the one hand, when the actual load growth exceeds expectations, the power supply capacity of the existing substation will face bottlenecks, and the new substations are restricted by land resources and long construction periods; on the other hand, if the substation capacity is over-reserved, it will lead to excessive initial investment and low equipment utilization, which will affect the economic efficiency of the power grid. This contradiction is particularly prominent in the marginal areas of urban expansion. The load growth trend in this area is difficult to accurately predict, but it is an important bearing area for future urban development. Therefore, planners need to consider moderate advanced layout when selecting substation sites, but how to determine the reserved capacity and construction sequence so that it can meet the future load growth needs without causing resource waste has become a technical difficulty. Under the influence of load growth uncertainty, how to dynamically adjust the substation layout plan so that it can adapt to different load growth scenarios and maintain a reasonable investment scale is a key issue that needs to be solved in the planning of distribution network in urban expansion areas.

[0015] Based on this, in one embodiment, if Figure 1 As shown, the first aspect of the present invention provides a method for substation layout in an urban expansion area based on an artificial intelligence algorithm, comprising: S1. Obtaining geographic spatial data and historical load data of an urban expansion area, and using a large language model to mine the association rules between the geographic spatial data and the historical load data, and obtaining key features to predict the power load demand data of the urban expansion area within a target period; S2. Analyze the power load demand data by using the large language model to obtain load change characteristics to partition and cluster the urban expansion area, and obtain load difference data of each sub-area and its load difference; S3, collecting historical distribution network structure data of the urban expansion area, and combining it with the load difference data, so as to process it through the large language model and obtain an initial layout plan of the substation; S4. Analyze the historical load data by a machine learning algorithm to obtain a load growth probability distribution of each sub-region within the target period to simulate the power supply capacity of each sub-region under different load growth rates; S5. Analyze the power supply capacity of each sub-area using a dynamic programming algorithm to optimize the initial substation layout plan, obtain the substation optimized layout plan and execute it.

[0016] Specifically, the present invention obtains geographical features such as land use type, building density, transportation network, population thermal distribution and urban planning texts (such as urban development plans, etc.) of urban expansion areas through satellite remote sensing images, GIS systems, etc. as geographic space features, and collects substation / feeder level time series load data (regional power consumption, load curve, peak load, etc.), user type distribution (residential / commercial / industrial), etc. from SCADA systems, etc. as historical load data, and then performs data alignment processing on these data in a time-space coordinate system and matches them to geographic space units (such as grids); then the structured data is converted into natural language descriptions, and artificial intelligence algorithms, such as deepseek and other advanced large language models, are used to mine association rules, so that deepseek and other advanced large language models use their integrated multiple algorithms to extract features from the input data, and after obtaining key features, the Prompt design or Transformer model of the deepseek large language model is combined to generate the load curve of the urban expansion area in the future preset target period, that is, the power load demand data.

[0017] The electricity load demand data is analyzed through the deepseek large language model, and its peak-to-valley difference, load density, growth slope, seasonal volatility, etc. are extracted to form load change characteristics represented by high-dimensional vectors. The urban expansion area is partitioned and clustered through clustering algorithms and the semantic rules of the deepseek large language model to obtain the data of each sub-area (such as industrial high-load area, residential low-growth area, etc.) and its load difference.

[0018] The historical grid data (substation coordinates, transformer capacity, line impedance, power supply radius, line impedance, etc., i.e., historical planning schemes for urban expansion areas) are associated with the load difference data of sub-regions, and the deepseek large language model is used to call the power grid design rule library (such as IEEE standards) to generate candidate site locations and capacities, i.e., the initial layout plan of substations.

[0019] The historical load data of urban expansion areas are analyzed through machine learning algorithms to predict the probability distribution of load growth in each sub-area during a preset target period in the future. Monte Carlo is used to generate samples to simulate the equipment utilization of each sub-area under various load growth rates, and then the power supply capacity of each sub-area under various different load growth rates is obtained.

[0020] Finally, the dynamic programming algorithm is used to analyze the power supply capacity of each sub-area to optimize the initial layout plan of the substation. By constructing the objective function, state variables and decision variables and performing recursive optimization, the optimized layout plan of the substation is obtained and executed.

[0021] The present invention utilizes large language models such as DeepSeek to learn and understand massive amounts of data, and is able to grasp the inherent laws of load changes and make intelligent predictions about future trends. This cutting-edge artificial intelligence technology is combined with a variety of artificial intelligence algorithms to generate a more dynamic and flexible substation layout optimization strategy, which can meet future load growth needs while also taking into account investment benefits, achieving a balance between the forward-looking and economical nature of planning. This solution improves the scientificity and adaptability of power grid planning by deeply combining the semantic understanding of large language models with the numerical calculation capabilities of traditional algorithms, and realizes full-chain intelligence from data to decision-making, making it suitable for high-uncertainty urban expansion scenarios.

[0022] In one embodiment, step S1 includes: The urban expansion area is divided into a number of grid units, and the geographic spatial data and historical load data of each grid unit are obtained, and the historical load data of each grid unit are grouped by similarity using a clustering algorithm based on dynamic time warping to obtain a grid unit load grouping result; Extracting industrial land distribution data from the geographic spatial data, and using the association rule mining algorithm in the large language model to quantify the support and confidence between the industrial land distribution data and the load grouping results of each grid unit, to obtain an association rule set characterizing the industrial land type and the power load fluctuation pattern in the urban expansion area; Extracting economic indicator time series data corresponding to the industrial land type from the geographic spatial data according to the association rule set, combining it with the industrial land distribution data and the historical load data, and processing it through a third-order tensor decomposition algorithm to obtain key features characterizing the power load and industrial economic development; Based on the key features, a recursive neural network is used to predict the power load in the urban expansion area and obtain the power load demand data of the urban expansion area in the target period.

[0023] Specifically, the present invention divides the urban expansion area into several grid units based on the population density data of the urban expansion area. For example, the population density of the grid unit of the core business district of a certain urban area reaches 25,000 people / square kilometer during the daytime on weekdays, and the population density drops to 8,000 people / square kilometer at night, while the grid unit of the residential area presents the opposite population density change characteristics; or the urban expansion area is divided into regular grid units of 1km×1km using GIS tools to ensure that each grid covers a continuous geographical space; through remote sensing image analysis or network public data, the industrial land type, building density, population thermal distribution and other geographic spatial data of each grid are extracted, and the hourly load data provided by the power company is spatially interpolated and allocated according to the grid geographic coordinates to form a time series load data set for each grid; then the DTW distance matrix is ​​calculated for the load curve of each grid to solve the phase offset problem of load fluctuation on the time axis (such as the peak and valley difference between weekdays and holidays), and a hierarchical clustering algorithm is used in combination with the DTW distance to generate groups with similar load patterns to obtain grid unit load grouping results. The similarity of these electricity load time series curves is measured by dynamic time warping algorithm, and the grid units can be clustered into different electricity consumption characteristic types such as commercial-dominated, residential-dominated, and mixed.

[0024] The types of industrial land are screened from the geographic spatial data (for example, the distribution data of industrial land shows that commercial land accounts for more than 60% of the commercial-dominated grid units, office land accounts for about 25%, and residential land accounts for 75% of the residential-dominated grid units). The area proportions of different land types in each grid are counted, and the grid unit load grouping results and industrial land data are converted into natural language descriptions and input into the DeepSeek large language model for prompt design and support and confidence calculation. Through association rule mining, it is found that when the proportion of commercial land exceeds 50%, the confidence of the bimodal characteristics of weekday electricity load is 0.85 and the support is 0.72, so as to generate a set of association rules that characterize the relationship between industrial land types and electricity load fluctuation patterns in urban expansion areas. The grid labels can also be encoded as transaction items (such as "grid ID_industry type_load group"), the minimum support threshold is set, and then the FP-Growth algorithm is used to mine frequent item sets, and strong association rules are screened according to lift (such as lift>1.5) and confidence (such as conf>0.7), thereby obtaining a set of association rules that characterize the relationship between industrial land types and electricity load fluctuation patterns in urban expansion areas.

[0025] Based on the grid positions and industrial land type labels in the association rule set, the economic indicator time series data of the region are obtained (such as GDP growth rate, industrial added value, etc., and the economic indicator time series data show that the quarterly month-on-month growth rate of commercial retail sales in the commercial-dominated grid units has a significant positive correlation with the growth rate of electricity load). Then, the third-order tensor decomposition algorithm is used to construct a third-order tensor with the dimension of "grid×time×feature". Its characteristics include: economic indicators, industrial distribution (land type ratio), load data (weekly average load, etc.), and the alternating least squares method is used to decompose the tensor to extract multiple potential factors. After verifying the decomposition stability through core consistency diagnosis, it is used as the key feature to characterize electricity load and industrial economic development. At the same time, the characteristic vector obtained after the third-order tensor decomposition shows that there is a stable co-evolution relationship between commercial land expansion, retail sales growth and electricity load growth. It is also possible to construct a third-order tensor containing time (such as monthly granularity), space (grid units, such as 100 grids), and characteristic indicators (proportion of industrial land, GDP growth rate, electricity load value), and use KNN interpolation to complete the missing economic indicators (such as some grids have no GDP data). Then, CP decomposition is used to decompose the original tensor into three factor matrices and a core tensor, and the factor combination with the highest weight in the core tensor is extracted (such as the correlation coefficient between the night load in commercial land-intensive areas and the regional service industry GDP is 0.73) as the key feature input prediction model.

[0026] A multi-dimensional input vector is constructed by integrating tensor decomposition factors, historical load time series data (previous 24 months) and economic indicators (GDP, population). The sliding window method is used to generate training samples. A recursive neural network with two hidden layers (128 and 64 neurons) and a Dropout layer (ratio 0.2) is used to prevent overfitting to predict the electricity load in urban expansion areas. The loss function is Huber Loss, and the hyperparameters are selected through Bayesian optimization. Finally, the key features are input into the trained recursive neural network for processing, and the electricity load demand data of the urban expansion area in the future preset target period, including the growth trend of electricity demand and the peak and valley conditions of electricity demand, are output.

[0027] The present invention captures the heterogeneity of local power consumption patterns (such as the load difference between commercial and residential areas) through grid unit division to improve prediction accuracy; uses DTW clustering to identify nonlinear time series similarities (such as sudden load changes on holidays) to avoid the sensitivity of Euclidean distance to phase shift; uses explicit modeling of industry-load causal association to provide physical explainability for association rule mining, supports the transformation of power grid planning from "statistical correlation" to "mechanism driven", and thus improves the confidence of rules; uses third-order tensor decomposition to capture the nonlinear coupling relationship of "economic indicators-industry distribution-load fluctuations", and crosses multi-dimensional features to enhance prediction robustness.

[0028] In one embodiment, step S2 includes: Parsing the semantic text in the power load demand data by the semantic word segmenter in the large language model to obtain load change features including load numerical features and text features; Extracting regional spatial layout data of the urban expansion area based on the load change characteristics, quantifying the Euclidean distance similarity of the urban expansion area through a hierarchical clustering algorithm, and performing regional division based on the similarity calculation results to obtain a plurality of sub-areas; The historical load data of each of the sub-regions are processed by a conditional random field to construct a time series state transition probability matrix including load density and load peak-to-valley ratio, and the load growth pattern of each of the sub-regions is probabilistically inferred based on each of the time series state transition probability matrices to obtain a load growth probability distribution of each of the sub-regions; The periodicity and trend characteristics of the load peak-to-valley difference of each sub-region are extracted from each load growth probability distribution, and processed by a time series decomposition algorithm to obtain the load difference data of each sub-region.

[0029] Specifically, the present invention constructs a feature semantic analysis corpus by extracting semantic features of total load, load density, load peak-to-valley ratio, and load growth rate from the prediction results, and then uses the semantic segmenter in the deepseek large language model to perform structured analysis on the load prediction description text through named entity recognition to obtain load change features consisting of load numerical features (load value, growth rate, time, etc., such as peak load increment: 80MW, time: 2024, etc.) and text features (industry keywords, etc., such as industry type: food street, area: commercial area, grid unit).

[0030] Based on the regional labels in the load change characteristics, the regional spatial layout data of the urban expansion area is extracted and used as input data together with the load change characteristics. Then, the hierarchical clustering algorithm is used to quantify the Euclidean distance similarity of the urban expansion area through the input characteristics (such as load density, industry type, etc.), and the region is divided based on the similarity calculation results to obtain several sub-regions. In the urban core area expansion planning, the new area often presents different spatial clustering characteristics. For example, the planned land area of ​​the eastern new district of a certain city reaches 35 square kilometers, of which commercial land accounts for 35%, industrial land accounts for 25%, and residential land accounts for 40%. The hierarchical clustering based on the load density and growth rate characteristics shows that the area can be divided into 3 load characteristic sub-areas, among which the load density prediction value of the commercial area is 6000 kW / km2, the industrial area is 4500 kW / km2, and the residential area is 2500 kW / km2.

[0031] The conditional random field is used to divide the historical load data of each sub-region into discrete states, and the characteristic function is designed. The state transition characteristics are used to count the transition frequency between states in the history. Based on the observed characteristics, the load density, peak-to-valley ratio and other observations are associated, and the weight parameters of the characteristic function are solved by the maximum likelihood estimation method. The L-BFGS algorithm is used for optimization, and then a time-series state transition probability matrix including load density and load peak-to-valley ratio is constructed. Based on the time-series state transition probability matrix of each sub-region, the Viterbi algorithm is used to predict the state sequence probability distribution of the next k steps (such as the next 12 months), and the load growth probability distribution of each sub-region is obtained. The historical load data analysis of these sub-regions shows that the average annual growth rate of load density in the commercial area is 12%, and the daily load peak-to-valley ratio is 2.1, the average annual growth rate of load density in the industrial area is 8%, and the daily load peak-to-valley ratio is 1.5, and the average annual growth rate of load density in the residential area is 6%, and the daily load peak-to-valley ratio is 1.8. The state transition probability constructed by the conditional random field shows that the probability of load density growth in the commercial area will remain above 0.85 in the next three years.

[0032] The load growth probability distribution of each sub-region is decomposed by STL to extract the periodicity and trend characteristics of the load peak-to-valley difference of each sub-region. The slope of the trend term, the difference between the maximum and minimum values ​​of the periodic term, and the standard deviation of the residual term are quantified through the time series decomposition algorithm to calculate the characteristic difference index of each sub-region, and obtain the load difference data of each sub-region including load density, load peak-to-valley difference and load growth rate.

[0033] The present invention parses text and numerical load data through a semantic word segmenter to fully explore the implicit impact of unstructured information on load changes; adopts hierarchical clustering combined with Euclidean distance and semantic features to achieve regional segmentation that is more in line with the actual load distribution; captures the load state transfer law through conditional random fields to quantify the growth risk under uncertainty scenarios; and uses time series decomposition to separate the periodicity and trend of load peak-to-valley differences to provide a basis for differentiated power grid planning.

[0034] In one embodiment, step S2 further includes: Establish a feature text dictionary including total load, load density, load growth rate and load peak-to-valley ratio according to the power load demand data, and extract numerical features from the feature text dictionary through the word segmentation rules in the large language model to obtain a regional load forecast numerical matrix; Based on the regional load prediction numerical matrix, land use type data of the urban expansion area is obtained, and based on the land use type data, the Euclidean distance similarity between the load density and the growth rate is calculated by a spectral clustering algorithm to perform unsupervised classification on the urban expansion area to obtain a plurality of sub-areas; The historical load data of each sub-region is processed by principal component analysis to obtain the principal component eigenvector of the load density of each sub-region to obtain the load growth data of each sub-region, and regression fitting is performed on each load growth data to obtain the load growth potential coefficient of each sub-region; The load peak and valley data of each sub-region are processed by Fourier transform to obtain the load peak and valley characteristic sequence of each sub-region and combine it with each load growth potential coefficient. The contour drawing method based on kernel density estimation is used for processing to quantify the load density gradient, load growth trend line and load peak and valley change band to obtain the load difference data of each sub-region.

[0035] Specifically, the present invention establishes a feature text dictionary including the total load, load density, load growth rate and load peak-to-valley ratio associated with the grid unit based on the electricity load demand data (such as the peak load density reaches 5500 kilowatts / square kilometer, an increase of 12.5% ​​over the same period last year, and the daily peak-to-valley difference is 2800 kilowatts), and stores the feature names, dimensions, calculation rules and examples in JSON format, and then performs numerical extraction through a large language model: identifying keywords (such as total load, peak-to-valley ratio, etc.) and associated values ​​in the feature text dictionary, and extracting numerical features based on formula parsing in the dictionary to generate a regional-feature matrix, and obtaining a regional load prediction numerical matrix.

[0036] Based on the regional characteristics or grid units in the regional load prediction numerical matrix, the land use type data of the urban expansion area (such as land use nature, plot area, volume ratio, building density, etc.) are extracted, and the land use type is encoded as a categorical variable (such as industry = 1, commercial = 2, etc.) and associated with the load matrix according to the regional ID; and the Euclidean distance similarity of the characteristic vectors such as load density and load growth rate of these encoded data is calculated through the spectral clustering algorithm, and the Gaussian kernel function is converted to construct a normalized Laplace matrix, and the characteristic decomposition is performed on it, and the first k characteristic vectors (k is the preset clustering number) are taken for K-means clustering to obtain several sub-areas.

[0037] The historical load data matrix of each sub-region is used as input for covariance matrix calculation, and the principal component eigenvectors of the load density of each sub-region are obtained as input. After being processed by the ridge regression model, the load growth potential data of each sub-region are output; among them, the load density distribution data are analyzed by principal component analysis, showing that the first principal component reflects the overall load level of the region, accounting for 65% of the total variance contribution rate, and the second principal component reflects the uneven distribution of load, accounting for 25% of the total variance contribution rate. These two principal components can explain 90% of the spatial distribution characteristics of load density; and the historical load growth data show that the average annual growth rate of load in the commercial area has remained above 13% in the past five years, the industrial area is around 7.5%, and the residential area is around 5.5%. The growth potential coefficients obtained by regression fitting are 1.8, 1.3 and 1.1 respectively, reflecting the differences in development potential in different regions.

[0038] The load peak and valley data of each sub-region are processed by Fourier transform to obtain the load peak and valley characteristic sequence of each sub-region, which is combined with the load growth potential coefficient of each sub-region to design a multidimensional kernel function: load density gradient (spatial kernel (Gaussian kernel) + load density value), load growth trend line (time kernel + growth rate) and load peak and valley change band (frequency domain kernel + Fourier amplitude). The contour drawing method based on kernel density estimation is used for processing, and the kernel density estimation value is calculated for the gridded space points. The equivalent points are connected to form gradient lines, trend lines and change bands to obtain the load difference data of each sub-region. Among them, the load peak and valley characteristic analysis is carried out. The analysis found that the peak-to-valley ratio of the commercial area on weekdays is 2.2, which drops to 1.6 on holidays, the peak-to-valley ratio of the industrial area is stable at around 1.4 throughout the year, the peak-to-valley ratio of the residential area reaches 1.9 in summer, and around 1.5 in other seasons. The obvious daily and seasonal cycle characteristics are extracted through Fourier transform; and the contour map generated based on kernel density estimation shows that the load density decreases from the city center to the periphery, and the contour line spacing is densest in the commercial area, reflecting that the load density gradient is the largest. The load growth trend line points to the main development axis of the city, and the load peak-to-valley change zone shows an obvious circle structure, with the peak-to-valley difference in the core area being the largest and gradually decreasing outward.

[0039] The present invention unifies the multi-source load data format by constructing a feature text dictionary to improve data fusion and model generalization capabilities; uses spectral clustering combined with land use type and load characteristics to perform space-load collaborative classification to achieve more accurate regional division; extracts core growth drivers through principal component analysis, and uses regression models to quantify development potential; integrates Fourier transform and kernel density estimation to visualize spatiotemporal dynamics, generate multidimensional load difference maps, and support refined power grid planning.

[0040] In one embodiment, step S3 includes: Extracting the spatial distribution characteristics of the power load of each sub-region from each load difference data by using a convolutional neural network, and processing the spatial distribution characteristics of each power load by using a spatial coordinate mapper to generate a load density thermal distribution matrix of each sub-region; Acquire historical distribution network grid structure data of the urban expansion area, and use support vector regression algorithm to optimize the power supply radius and load coverage rate in the historical distribution network grid structure data to obtain a candidate substation location sequence; Based on the land use restriction conditions of each of the sub-areas, the candidate substation site sequence is preprocessed by the large language model, and a multi-objective optimization operation is performed on the preprocessed candidate substation site sequence by using a genetic algorithm to obtain a substation site selection coordinate set; The load density thermal distribution matrix of each sub-area is quantified by a hierarchical clustering algorithm to obtain the maximum load capacity of the power supply area to obtain a standard transformer capacity sequence from the equipment specification library, and a transformer combination scheme is calculated based on the standard transformer capacity sequence using a dynamic programming algorithm to obtain a main transformer capacity configuration table; Under the constraints of power supply reliability level requirements and load transfer requirements, the number of outgoing line loops is determined through a graph theory algorithm based on the main transformer capacity configuration table to obtain a substation outgoing line plan, which is combined with the substation site selection coordinate set and the main transformer capacity configuration table to obtain an initial substation layout plan.

[0041] Specifically, the load difference data of each sub-region (load density, growth rate, peak-to-valley ratio, etc.) are grid-encoded into a multi-channel matrix as input data and input into the convolutional neural network for processing, so that the spatial features of the regional electricity load are extracted through a sliding window of 500 meters × 500 meters, and the load spatial feature vector of each grid is output (such as industrial area characteristics = high density, low volatility, residential area characteristics = medium density, high volatility). The coordinate mapper is used to associate the grid longitude and latitude coordinates (the boundary coordinates of each sub-region can also be directly used) with the feature vector, and a spatial index is constructed to perform Gaussian kernel smoothing on the load density of each grid to generate a load density thermal distribution matrix with a resolution of 100m × 100m.

[0042] The historical distribution network structure data of the urban expansion area (such as the power supply radius, load coverage, line impedance, transformer load rate and other information of the historical planning scheme) are obtained, and the multi-objective optimization of minimizing the power supply radius and maximizing the load coverage is carried out. The radial basis function (RBF) kernel is used, and the parameters are optimized through grid search to output the candidate substation location sequence; among them, the substation site selection optimization also needs to consider the geographical constraints, such as the nature of the construction land, traffic accessibility, topography, etc. A candidate site is located near the intersection of the city's main roads, covers an area of ​​0.8 hectares, has a flat terrain, convenient transportation, a power supply radius coverage rate of 95%, and the lowest operating cost. It is determined as the optimal site by the genetic algorithm.

[0043] Taking the candidate site coordinates and land use restrictions (such as ecological red lines and residential avoidance requirements) as input, the DeepSeek large language model is used to perform rule parsing, and the policy text (such as "the substation must be ≥200m away from the residential area") is converted into a geographic fence constraint. The candidate points that violate the constraints (such as points located in the ecological protection area) are removed to realize the processing of the candidate substation site sequence, and binary coding is used to indicate whether the substation is selected (such as 1=selected, 0=unselected). The land acquisition and line laying costs of the processed candidate substation site sequence are used as the cost to construct the fitness function for evolutionary operations, and the Pareto optimal solution set is obtained. The substation site coordinate set is selected by the entropy weight method.

[0044] The load density values ​​of the load density thermal distribution matrix base of each sub-region are aggregated into several capacity levels according to the Euclidean distance through the hierarchical clustering algorithm, and the 95% quantile load value of each cluster is taken as the upper limit of the power supply area capacity to obtain the maximum load capacity of the power supply area. Then, based on the maximum load capacity of the power supply area, the standard transformer capacity sequence is obtained from the equipment specification library, and the sub-region is taken as the stage, the cumulative configuration capacity and the remaining budget are taken as the state, and the transformer model is selected as the decision to define the state. Then, based on these states, transformer capacity and regional load demand, a state transfer equation is constructed, and the minimization of total cost (equipment purchase + operation and maintenance) is taken as the optimization goal to be solved using the dynamic programming algorithm to obtain the main transformer capacity configuration table; among them, based on the load characteristics, the dynamic programming algorithm gives the main transformer capacity configuration plan of 2 63 MVA transformers. Considering the expected load growth and spare capacity requirements, the rated capacity of the transformer meets 125% of the maximum load, and it has the load transfer capability when a single main transformer fails.

[0045] The outgoing line loop must meet the N-1 criterion (when any line fails, the remaining lines can carry all the loads) and the load transfer path must be connected in the topology as reliability constraints. The substation and the load point are modeled as graph nodes, the lines are edges, and the weight = line capacity / length. The initial radial grid is constructed through the minimum spanning tree to ensure the lowest cost, and redundant lines are added until the N-1 check is met to calculate the outgoing line loop. The number and path of the outgoing lines, that is, the substation outgoing line plan, are output, and the initial layout plan of the substation is obtained by combining it with the substation site selection coordinate set and the main transformer capacity configuration table. Among them, in terms of outgoing line loop configuration, considering the power supply reliability, the proportion of primary load is as high as 40%. The minimum spanning tree calculated by the graph theory algorithm contains 8 trunk nodes. Combined with the load transfer requirements, 10 outgoing lines are configured, of which 6 are used as normal operating loops and 4 are used as backup contact loops. Load transfer can be quickly realized when the main transformer fails or the line is under maintenance.

[0046] The present invention extracts the thermal distribution characteristics of load density through convolutional neural networks to accurately characterize the spatial heterogeneity of regional electricity demand; integrates support vector regression and genetic algorithm to perform collaborative optimization of multi-dimensional objectives such as power supply radius, coverage rate, and land use constraints; generates a transformer capacity configuration table based on hierarchical clustering and dynamic programming to achieve dynamic matching of equipment selection and load growth; uses graph theory algorithms to generate line plans to balance reliability and economy to reduce the risk of power outages; the plan fully considers multiple dimensions such as load distribution characteristics, geographical location constraints, and power supply reliability requirements, ensures the scientificity and economy of power grid planning through multi-level optimization, and forms a complete site planning solution.

[0047] In one embodiment, step S4 includes: Extracting trend items, period items, and random items from the historical load data of each sub-region by a time series decomposition algorithm, and processing the extracted results by a kernel density estimation method to obtain a load growth probability distribution matrix of each sub-region; Based on the load growth probability distribution matrices, a long short-term memory network is used to predict the load growth data at monthly, quarterly and annual time scales to obtain a multi-time scale load growth trend vector; According to the multi-time scale load growth trend vector, the load level of each sub-area under multiple growth rate scenarios is sampled by Monte Carlo method to obtain the main transformer load probability distribution to quantify the average utilization hours, maximum load rate and load duration of the equipment, and obtain the equipment utilization characteristic matrix; The equipment utilization characteristic matrix is ​​fitted by polynomial regression to quantify the power supply capacity variation trend of each sub-region at multiple time scales, and the power supply capacity of each sub-region at different load growth rates is obtained.

[0048] Specifically, the present invention uses the historical load time series of each sub-region as input, extracts the trend item (reflecting long-term growth), cycle item (seasonal fluctuation), and random item (noise or sudden event impact) of each sub-region through the time series decomposition algorithm, and uses the growth rate sequence of the extracted trend item data as input, processes it using the Gaussian kernel function, and outputs the load growth probability density function to construct the load growth probability distribution matrix of each sub-region; wherein, the time series decomposition of the load data shows obvious multi-level characteristics, such as the historical load data of a commercial district, the trend item shows an average annual growth trend of 8.5%, the cycle item contains obvious intraday double peak characteristics, the peak-to-valley difference reaches 45%, and the random item fluctuates greatly during holidays, with a standard deviation of 15%. The probability distribution matrix obtained by kernel density estimation shows that the cumulative probability of load growth in this area between 6% and 10% reaches 0.75.

[0049] Taking the load growth probability distribution matrix of each sub-region, historical cycle items and external variables (such as economic indicators, population growth rate, etc.) as input, the long short-term memory network is used to predict the load growth data at multiple time scales such as monthly, quarterly and annual, so as to splice the prediction results of different time scales and output the multi-time scale load growth trend vector; among them, the multi-time scale prediction results show the growth characteristics of different cycles. The monthly data are obviously affected by seasonal factors. The average monthly growth rate in summer reaches 12%, and it drops to about 5% in winter. The quarterly data shows a step-type growth related to the industry production cycle. The annual data shows a steady upward trend as a whole, and the average annual growth rate remains at 8%.

[0050] Based on the multi-time scale load growth trend vector and the load growth probability distribution matrix of each sub-region, 1000 groups of random growth rate sequences (such as pessimistic / neutral / optimistic scenarios) were generated by the Monte Carlo method, and the main transformer load rate was calculated for each group of scenarios to obtain the main transformer load probability distribution to quantify the average utilization hours, maximum load rate and load duration of each main transformer equipment, and obtain the equipment utilization rate characteristic matrix (each row corresponds to a sub-region, and each column is an indicator value), that is, the main transformer utilization rate of each sub-region; among them, the main transformer load data showed significant differences under different growth scenarios. Under the high growth scenario, the maximum load rate within 5 years reached 85%, and under the medium growth scenario, the maximum load rate within 5 years reached 95%. The probability of the load factor exceeding 80% is 75% in the high-growth scenario and 65% in the low-growth scenario; the Monte Carlo sampling results show that the probability of the load factor exceeding 80% is 0.35 in the high-growth scenario, and this probability drops to 0.15 in the medium-growth scenario; as for the equipment utilization evaluation, a multi-dimensional indicator system is adopted. The average annual utilization hours of the main transformer in the high-load area reaches 5,500 hours, the maximum load rate is 85%, and the full-load operation time accounts for 25%. The matrix composed of these characteristic indicators reflects the trend that the equipment operating pressure gradually increases with the increase of load, and the power supply capacity change trend shows an obvious nonlinear relationship with the equipment utilization. When the equipment utilization rate exceeds 75%, the speed of power supply capacity improvement slows down significantly.

[0051] Taking the equipment utilization characteristic matrix of each sub-region as input and the power supply capacity (defined as the maximum load growth rate that can be carried) as the target variable, the optimal order (usually 2-3 orders) is determined through cross-validation to construct a regression model to fit the power supply capacity curve under different load growth rates, and output the power supply capacity matrix of each sub-region under different load growth rates; among them, the polynomial regression fitting results show that under the high load growth scenario, the power supply capacity will increase by only 85% of the load growth rate after 3 years; when the equipment utilization rate reaches 80%, the power supply reliability rate drops to 99.95%, and the voltage compliance rate drops to 98%. If the high-speed growth trend is maintained, the power supply capacity constraint will become a key factor restricting regional development within 5 years.

[0052] The present invention separates the trend, cycle and random components of load changes through time series decomposition, thereby improving the interpretability and accuracy of the prediction; uses long short-term memory networks (LSTM) to capture the multi-level laws of monthly, quarterly and annual load growth to adapt to the phased characteristics of urban expansion; uses Monte Carlo simulation to quantify the probability of extreme scenarios of equipment load (such as overload risk) to support grid resilience planning; reveals the dynamic relationship between power supply capacity and load growth based on polynomial regression, and can guide the dynamic adjustment of transformer capacity and operation and maintenance strategies; this solution realizes a complete chain from data analysis to decision support through a technical closed loop of "decomposition-prediction-simulation-quantification", which is particularly suitable for grid adaptability planning scenarios under the background of rapid increase in load growth rate.

[0053] In one embodiment, step S5 includes: Based on the power supply capacity of each sub-area, a multi-stage optimization function with the minimum investment cost as the goal and the equipment utilization rate not lower than the preset threshold as the constraint is solved by a dynamic programming algorithm to obtain a substation construction time planning table; According to the transformer capacity specification constraints, a genetic algorithm is used to optimize the main transformer capacity configuration table in the initial substation layout plan to obtain a main transformer capacity configuration plan; Based on the main transformer capacity configuration plan, equipment load rate data is obtained to calculate the annual average utilization hours and maximum load rate of the equipment through a multi-objective optimization algorithm to obtain a substation capacity expansion plan; Extracting grid structure data from the historical distribution network grid structure data according to the substation capacity expansion plan, so as to calculate the substation power supply radius and load coverage rate through a graph theory algorithm, and obtain a substation layout optimization plan; The substation site optimization plan, the substation outgoing line plan, the substation construction time planning table, the main transformer capacity configuration plan and the substation capacity expansion plan are combined to obtain a substation optimization layout plan and execute it.

[0054] Specifically, the present invention obtains load forecast data and corresponding power supply capacity data under three growth scenarios of high, medium and low based on the historical load data of each sub-region, and divides the planning period into several stages through a dynamic programming algorithm, takes the capacity of the substations built in each period, the accumulated investment cost, and the equipment utilization rate as state variables, takes the capacity and location of the newly built / expanded substations in each period as decision variables, solves a multi-stage optimization function with the minimum investment cost as the goal and the equipment utilization rate not less than a preset threshold as a constraint, and outputs a substation construction time planning table by recursively solving the optimal decision for each stage; wherein, in the substation construction plan, different load growth scenarios show significant differences, with an average annual growth rate of 12% in the high growth scenario, 8% in the medium growth, and 5% in the low growth. The dynamic programming algorithm uses the equipment utilization rate of not less than 65% as a constraint, and calculates that two substations need to be built within three years in the high growth scenario, and one substation needs to be built within five years in the medium growth scenario.

[0055] The transformer capacity selection is encoded with real numbers, and the capacity must match the standard specifications (such as 50MVA multiples) and the total capacity must not be less than 1.2 times the maximum load in the area (N-1 criterion redundancy) are injected as constraints. A fitness function is constructed based on the load coverage rate (covered load / total load) and the capacity utilization rate (actual load / total capacity) to perform evolutionary operations, so as to optimize the main transformer capacity configuration table and obtain the main transformer capacity configuration plan; among them, the transformer capacity configuration plan directly affects the investment benefits. The load forecast of a substation power supply area shows that the maximum load will reach 80 MVA within 5 years. The optimal configuration plan given by the genetic algorithm is to install two 50 MVA main transformers in advance, increase the capacity to three 50 MVA in the fourth year, and maintain the average equipment utilization rate above 75%.

[0056] According to the main transformer capacity configuration plan, the historical load rate data of each transformer equipment in the plan is obtained, and the objective function is to maximize the annual average utilization hours of the equipment (reflecting economy) and minimize the maximum load rate (reducing overload risk). The NSGA-II algorithm is used for non-dominated sorting and stratification according to the target value, and the Pareto frontier solution is retained. The congestion calculation is performed to ensure that the solution set is evenly distributed in the target space, and finally a set of capacity expansion plans is output; among them, the equipment operation data shows that the annual average utilization hours of the main transformer of the existing substation is 5200 hours, and the maximum load rate reaches 82%. The capacity expansion timing determined by multi-objective optimization shows that the expansion will be started when the annual average utilization hours exceed 5500 hours or the maximum load rate exceeds 85%.

[0057] According to the substation capacity expansion plan, the grid structure data (i.e., substation configuration location, quantity, transmission lines, etc.) are extracted from the historical distribution network grid structure data. The substations and load centers in the grid structure data are taken as nodes, and the transmission lines and their weights = line length × unit cost are taken as edges. The shortest path algorithm is used to calculate the shortest distance from the substation to each load point. If the distance is not greater than the power supply radius threshold (such as 5km), it is determined to be covered, and the distance calculation and coverage determination process are iteratively optimized to adjust the substation location until the coverage rate is ≥95%. The coordinates of the newly added substation and the optimized power supply radius are output to obtain the substation layout optimization plan; among them, in the site layout optimization, considering that the typical power supply radius of a 110 kV substation is 3 kilometers, the load coverage rate is required to be no less than 95%. The calculation results of the graph theory algorithm show that two sites are required in the commercial area with a station distance of 2.5 kilometers, one site in the industrial area with a power supply radius of 3.5 kilometers, and one site in the residential area with a power supply radius of 4 kilometers.

[0058] Finally, check the timing consistency between the time plan and the capacity expansion plan (such as whether the expansion is completed after the construction) for conflict monitoring, and accumulate the investment of each sub-plan to ensure that the total budget is not exceeded. After the verification is passed, combine the substation layout optimization plan, substation outlet plan, substation construction time plan, main transformer capacity configuration plan and substation capacity expansion plan to obtain the substation optimization layout plan (such as building two substations with a capacity of 100 MVA each in the first phase, and the power supply reliability reaches 99.95%. In the second phase, expand one substation and increase the capacity of the original station, the power supply reliability remains above 99.93%, and the voltage qualification rate is always higher than 98.5%) and implement it.

[0059] The present invention covers the full-cycle decision-making of substation construction timing, capacity configuration, and expansion planning through the collaboration of dynamic programming and genetic algorithms; adopts a multi-objective optimization algorithm to quantify equipment utilization and load rate, taking into account investment costs and grid resilience; integrates geographic space constraints and time stage goals through graph theory algorithms to achieve dynamic adaptive layout; combines multi-stage optimization results to ensure the feasibility of the scheme under complex constraints; this scheme realizes the intelligent upgrade of the grid from micro-equipment configuration to macro-space-time planning through algorithm collaboration and multi-objective optimization, providing a scientific, flexible, and feasible planning tool for highly dynamic urban power grids.

[0060] In one embodiment, step S5 further includes: Based on the historical load data of each of the sub-regions, a load forecasting function is constructed through a recurrent neural network to calculate the load growth probability under multiple scenarios, and a multi-scenario load forecasting matrix is ​​obtained; According to the multi-scenario load forecasting matrix, the load data of different periods under each scenario is used as a state variable, and the construction time of the new substation and the transformer capacity are used as decision variables. The weighted combination of equipment utilization rate and the power supply capacity of each sub-area is used as the objective function, and the dynamic programming algorithm is used to solve it to obtain the substation construction decision space; Taking the substation construction decision space as input, establishing an optimization function with the combination of construction time and transformer capacity as the decision vector through a heuristic algorithm, performing weighted calculation on the average equipment utilization rate and power supply reliability rate, and outputting a substation construction timetable; A sorting algorithm is used to prioritize the construction scale of each stage in the substation construction time sequence, and a construction scale allocation sequence is generated under the constraints of voltage qualification rate and power supply reliability rate, and a phased construction planning table is obtained to optimize the initial layout plan of the substation, and the optimized layout plan of the substation is obtained and executed.

[0061] Specifically, the present invention takes the historical load data (day / month / year), economic indicators (GDP, population), etc. of each sub-region as input, and inputs them into the recurrent neural network for training, with the goal of minimizing the cross entropy loss between the predicted load and the actual load, so that the trained recurrent neural network model takes the historical load data of each sub-region as input, and outputs the load growth probability under multiple scenarios, with a medium growth scenario: the load grows linearly according to the historical trend; a high growth scenario: the economic acceleration drives the load to increase by 12% annually; a low growth scenario: policy restrictions lead to an annual increase of ≤5%, and a three-dimensional matrix (scenario×time×sub-region) is obtained, that is, a multi-scenario load prediction matrix; wherein, the construction of the load prediction function needs to consider a variety of influencing factors, such as a new district plan shows that a commercial complex of 350,000 square meters, a high-tech industrial park of 500,000 square meters, and a residential community of 250,000 square meters will be built in the next five years. Based on this, the high growth scenario predicts an average annual growth rate of 15%, a medium growth scenario of 10%, and a low growth scenario of 6%, and the probability of occurrence of each scenario is 0.3, 0.5, and 0.2, respectively.

[0062] Based on the multi-scenario load forecasting matrix, the load forecast data of different periods under each scenario, the capacity and location distribution of the built substations, and the equipment utilization rate are used as state variables. The construction time and transformer capacity of the new substation are used as decision variables. The objective function is to maximize the comprehensive benefits of the weighted sum of equipment utilization and power supply capacity. The state transfer equation is constructed and solved recursively by dynamic programming, that is, the optimal value function of each state is reversely calculated from the end of the planning period, and the optimal decision path is recorded to output the substation construction decision space (the set of all feasible decisions); among them, the substation construction decision space includes two dimensions: the capacity of new substations and the capacity of expansion.

[0063] Taking the combination of construction time and transformer capacity as decision variables, weighted equipment utilization and power supply reliability as objective functions, the construction time and capacity in the substation construction decision space are encoded, roulette wheel selection is used to retain individuals with high fitness, arithmetic crossover is used to generate offspring time and capacity, and Gaussian mutation is used to perturb the time points to output the Pareto optimal solution set, so as to manually select the final substation construction schedule; among them, the establishment of the objective function comprehensively considers the average equipment utilization and power supply reliability, among which the weight of the average equipment utilization is 0.6, and the weight of the power supply reliability is 0.4. The substation construction schedule shows that two 110 kV substations will be built in the first phase, the capacity of the original station will be increased and a new substation will be built in the second phase, and the construction of all four substations will be completed in the third phase.

[0064] Taking load gap (demand-existing capacity, voltage qualification rate (>95% is qualified), and power supply reliability rate (>99% is high priority) as sorting indicators, the sorting algorithm is used to prioritize the construction scale of each stage in the substation construction time sequence table, and the TOPSIS method is used to calculate the comprehensive score. The mandatory priority construction of areas with unqualified voltage qualification rate and the single-stage investment not exceeding 30% of the total budget are used as hard constraints to generate a construction scale allocation sequence, and the phased construction planning table is output to be added to the initial layout plan of the substation for optimization, and the substation optimization layout plan is obtained and implemented; among them, the load transfer capacity analysis in the grid topology structure shows that the adjacent substations 25% of the capacity needs to be reserved for mutual supply. When the load rate of the main transformer exceeds 75%, the capacity expansion is started. The maximum reverse load capacity of the existing 220 kV substation to the 110 kV substation is 40 MVA; the capacity configuration plan after comprehensive optimization, such as the first phase, both substations use 2 63 MVA main transformers, the annual utilization hours of the equipment are controlled within 5000 hours, and the peak-to-valley difference rate does not exceed 0.4. In the second phase, the new substation adopts a combination of 3 50 MVA main transformers, and the original station is increased to 3 63 MVA. In the third phase, the new station adopts 2 63 MVA main transformers. This phased construction plan not only meets the load growth demand, but also ensures the economic operation of the equipment.

[0065] The present invention generates a multi-scenario load growth probability matrix through RNN, covering multiple possibilities under uncertainty and enhancing planning resilience; combines dynamic programming with heuristic algorithms to achieve coordinated optimization of long-term construction decisions and short-term capacity configuration; integrates technical constraints through sorting algorithms to ensure that key projects are implemented first; the phased construction planning table directly guides project implementation and reduces the risk of deviation between planning and execution; the scheme achieves full-process optimization from load forecasting to construction implementation through algorithm chain collaboration and multi-objective decision-making, providing reliable technical support for high-uncertainty urban power grid planning.

[0066] In one embodiment, after step S5, the method further includes: Obtaining load change data and equipment operation data when executing the substation optimization layout plan; The load change data is sequence modeled by a recurrent neural network to obtain a load growth dynamic prediction sequence, and the equipment operation data is processed by a clustering algorithm to obtain an equipment utilization state vector; Based on the equipment utilization state vector, a fuzzy inference algorithm is used to quantify the remaining life of the equipment, and an equipment utilization classification result is obtained to adjust the substation optimization layout plan, and a construction plan adjustment plan is obtained and executed.

[0067] Specifically, the present invention collects load change data in real time when simulating the execution of the substation optimization layout plan in the urban expansion area, such as hourly load, peak-to-valley difference, growth rate and external related data, such as meteorological data, holiday marks, economic indicators, etc., as well as equipment operation data, such as transformer load rate, oil temperature, winding temperature, insulation aging index, number of switch cabinet operations, fault alarm records, line current, voltage deviation and active power loss, etc., and cleans and aligns these data.

[0068] The historical load data and external variables (temperature, holidays) are used as covariates to generate training sample sequences for training the recurrent neural network. The bidirectional LSTM layer of the network is used to capture the bidirectional temporal dependency of load changes (such as morning and evening peaks), and its attention mechanism is used to determine the impact of weighted key time points (such as extreme weather days). The output layer is used to multi-step predict the load curve for the next 72 hours. During the training process, MSE and peak-to-valley difference penalty terms are used as loss functions, and the early stopping method is used to prevent overfitting to optimize the training of the recurrent neural network. The real-time load change data is input into the trained recurrent neural network model as input, and the load growth dynamic prediction sequence is output. The load rate mean / variance, temperature rise rate, cumulative operating time, etc. in the equipment operation data are used as numerical features, and the fault history label (such as "overload times>3" encoding) is used as the input. 1) is used as a category feature to construct feature engineering, and the K-means algorithm is used to cluster the equipment based on feature engineering to obtain the equipment utilization state vector; among them, load dynamic prediction can also be continuously modeled using a sliding time window in weeks. Real-time monitoring data show that the load growth rate of a commercial substation in the past four weeks was 2.8%, 3.2%, 3.5%, and 3.8%, respectively, showing a gradually accelerating trend. The recurrent neural network predicts that the growth rate in the next four weeks will remain above 3.5%, and the load peak is expected to exceed 85 MVA; equipment operation status monitoring covers multiple key parameters. The temperature of the A-phase winding of the main transformer is maintained at 55 degrees Celsius, an increase of 5 degrees over the same period last year. The vibration value is 0.8 mm per second, which is within the normal range, but the load rate has reached 78%. The clustering analysis results show that the transformer is in a high-load and high-temperature operation state.

[0069] The equipment utilization state vector (load rate, temperature, aging index, etc. of transformer equipment) is used as input variable, and the remaining life level (short / medium / long) and recommended operation (maintenance / expansion / replacement) of the equipment are used as output variables. The triangular / trapezoidal function is used to quantify the fuzzy concept (such as "high load rate" is defined as >85%), and the equipment state vector is mapped into a fuzzy set (such as load rate 78% → "high" membership 0.6, "medium" membership 0.4), and the activation strength of each rule is calculated (taking the minimum input membership value). The weighted average method is used to generate the remaining life quantization value (such as remaining life = 5 years) to obtain the equipment utilization classification result to dynamically adjust the substation optimization layout plan and execute it. For example, for the substation optimization layout plan, short-life equipment is given priority to be included in the expansion plan or replaced; the load distribution of medium-life equipment is optimized (such as transferring part of the load to the adjacent substation, etc.); the long-life equipment maintains the current capacity and monitors the operating status; after such adjustment, the construction plan adjustment plan can be obtained (such as bringing the original expansion plan in the 5th year forward to the 3rd year, etc.) for execution.

[0070] The present invention monitors load and equipment data in real time to perform data-driven dynamic optimization, thereby improving the adaptability and real-time performance of the planning scheme; quantifies the remaining life of equipment through fuzzy reasoning to support preventive maintenance and dynamic capacity adjustment; adopts dynamic correction of construction planning based on prediction and equipment status to reduce investment waste and operational risks; the scheme realizes a full-link closed loop from power grid operation monitoring to dynamic planning optimization through real-time data perception, intelligent algorithm reasoning and closed-loop feedback control, providing intelligent protection for the reliability and economy of urban power grids.

[0071] In the embodiments of the present application, based on the problem of how to determine the substation layout and capacity configuration plan when the uncertainty of load growth in the planning area leads to dynamic changes in the supply and demand balance relationship, a substation layout method for urban expansion areas based on artificial intelligence algorithms is designed. The method combines the geographic spatial data and historical load data of the urban expansion area, and uses a large language model to mine the correlation between the two to predict the power load demand of the urban expansion area in a specific period of time in the future; based on the predicted power load demand and the load change characteristics of the urban expansion area, combined with its historical distribution network grid structure data, a large language model is used to formulate a substation layout and capacity configuration plan; the power supply capacity of the urban expansion area under different load growth rates is simulated by a machine learning algorithm, and the initial plan formulated by the finger is optimized by a dynamic programming algorithm, and an optimized plan is obtained and executed; by comprehensively using a large language model and a variety of algorithms, a substation layout and capacity configuration plan for the planning area with load growth uncertainty is constructed, which provides a scientific and efficient method for power network planning in urban expansion areas, and helps to improve the city's power supply capacity and power efficiency.

[0072] It should be noted that although the steps in the above flowchart are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders.

[0073] In another embodiment, if Figure 2 As shown, the second aspect of the present invention provides a system for substation layout in urban expansion area based on artificial intelligence algorithm, comprising: The demand forecasting module 10 is used to obtain the geographic space data and historical load data of the urban expansion area, and use the large language model to mine the association rules between the geographic space data and the historical load data to obtain key features to predict the power load demand data of the urban expansion area within the target period; A partition clustering module 20 is used to analyze the power load demand data through the large language model to obtain load change characteristics to partition and cluster the urban expansion area to obtain each sub-area and its load difference data; A scheme generating module 30 is used to collect the historical distribution network structure data of the urban expansion area, and combine it with the load difference data to process it through the large language model to obtain the initial layout scheme of the substation; The capacity simulation module 40 is used to analyze the historical load data through a machine learning algorithm to obtain the load growth probability distribution of each sub-area within the target period, so as to simulate the power supply capacity of each sub-area under different load growth rates; The scheme optimization module 50 is used to analyze the power supply capacity of each sub-area by using a dynamic programming algorithm to optimize the initial substation layout scheme, obtain the substation optimized layout scheme and execute it.

[0074] It should be noted that each module in the above-mentioned urban expansion area substation layout system based on artificial intelligence algorithm can be fully or partially implemented by software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules. For the specific definition of a system for the layout of urban expansion areas substations based on artificial intelligence algorithms, please refer to the definition of a method for the layout of urban expansion areas substations based on artificial intelligence algorithms above. The two have the same functions and effects, which will not be repeated here.

[0075] In summary, the present invention relates to the field of information technology, and discloses a method and system for substation layout in an urban expansion area based on an artificial intelligence algorithm, which predicts the future electricity load demand data of the area by mining the correlation law between the geographic spatial data and historical load data of the urban expansion area; uses a large language model to analyze the predicted data, and divides the urban expansion area into zones according to the analysis results, obtains the data of each sub-area and its load difference, combines it with the historical distribution network grid structure data of the urban expansion area, and processes it with the large language model to obtain an initial layout plan of the substation; optimizes the initial layout plan of the substation by simulating the power supply capacity of each sub-area under different load growth rates, obtains an optimized plan and executes it, and realizes accurate prediction of the electricity load demand of the urban expansion area and effective optimization of the substation layout plan by comprehensively using large language models and artificial intelligence algorithms such as machine learning algorithms and dynamic programming algorithms, while realizing the intelligentization and dynamic optimization of urban power planning, and improving the scientificity and adaptability of power grid planning.

[0076] Each embodiment in this specification is described in a progressive manner, and the same or similar parts of each embodiment can be directly referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. It should be noted that the technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0077] The above-mentioned embodiments only express several preferred implementation modes of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in the technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be regarded as the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be based on the protection scope of the claims.

Claims

1. A method for substation layout in urban expansion area based on artificial intelligence algorithm, characterized in that: include: Obtaining geographic spatial data and historical load data of the urban expansion area, and using a large language model to mine the association rules between the geographic spatial data and the historical load data, and obtaining key features to predict the power load demand data of the urban expansion area within a target period; Analyzing the power load demand data by using the large language model to obtain load change characteristics to partition and cluster the urban expansion area and obtain load difference data of each sub-area; Collecting historical distribution network structure data of the urban expansion area and combining it with the load difference data to process it through the large language model to obtain an initial substation layout plan; Analyzing the historical load data by a machine learning algorithm to obtain a load growth probability distribution of each sub-region within the target period to simulate the power supply capacity of each sub-region under different load growth rates; A dynamic programming algorithm is used to analyze the power supply capacity of each of the sub-areas to optimize the initial substation layout plan, obtain the substation optimized layout plan and execute it.

2. The method for substation layout in urban expansion area based on artificial intelligence algorithm according to claim 1 is characterized in that: The obtaining of geographic spatial data and historical load data of the urban expansion area, and using a large language model to mine the association rules between the geographic spatial data and the historical load data, and obtaining key features to predict the power load demand data of the urban expansion area within a target period, includes: The urban expansion area is divided into a number of grid units, and the geographic spatial data and historical load data of each grid unit are obtained, and the historical load data of each grid unit are grouped by similarity using a clustering algorithm based on dynamic time warping to obtain a grid unit load grouping result; Extracting industrial land distribution data from the geographic spatial data, and using the association rule mining algorithm in the large language model to quantify the support and confidence between the industrial land distribution data and the load grouping results of each grid unit, to obtain an association rule set characterizing the industrial land type and the power load fluctuation pattern in the urban expansion area; Extracting economic indicator time series data corresponding to the industrial land type from the geographic spatial data according to the association rule set, combining it with the industrial land distribution data and the historical load data, and processing it through a third-order tensor decomposition algorithm to obtain key features characterizing the power load and industrial economic development; Based on the key features, a recursive neural network is used to predict the power load in the urban expansion area and obtain the power load demand data of the urban expansion area in the target time period.

3. The method for substation layout in urban expansion area based on artificial intelligence algorithm according to claim 1 is characterized in that: The large language model is used to analyze the power load demand data to obtain load change characteristics to partition and cluster the urban expansion area, and obtain each sub-area and its load difference data, including: Parsing the semantic text in the power load demand data by the semantic word segmenter in the large language model to obtain load change features including load numerical features and text features; Extracting regional spatial layout data of the urban expansion area based on the load change characteristics, quantifying the Euclidean distance similarity of the urban expansion area through a hierarchical clustering algorithm, and performing regional division based on the similarity calculation results to obtain a plurality of sub-areas; The historical load data of each of the sub-regions are processed by a conditional random field to construct a time series state transition probability matrix including load density and load peak-to-valley ratio, and the load growth pattern of each of the sub-regions is probabilistically inferred based on each of the time series state transition probability matrices to obtain a load growth probability distribution of each of the sub-regions; The periodicity and trend characteristics of the load peak-to-valley difference of each sub-region are extracted from each load growth probability distribution, and processed by a time series decomposition algorithm to obtain the load difference data of each sub-region.

4. The method for substation layout in urban expansion area based on artificial intelligence algorithm according to claim 1 is characterized in that: The analyzing the power load demand data by the large language model to obtain load change characteristics to partition and cluster the urban expansion area to obtain each sub-area and its load difference data also includes: Establish a feature text dictionary including total load, load density, load growth rate and load peak-to-valley ratio according to the power load demand data, and extract numerical features from the feature text dictionary through the word segmentation rules in the large language model to obtain a regional load forecast numerical matrix; Based on the regional load prediction numerical matrix, land use type data of the urban expansion area is obtained, and based on the land use type data, the Euclidean distance similarity between the load density and the growth rate is calculated by a spectral clustering algorithm to perform unsupervised classification on the urban expansion area to obtain a plurality of sub-areas; The historical load data of each sub-region is processed by principal component analysis to obtain the principal component eigenvector of the load density of each sub-region to obtain the load growth data of each sub-region, and regression fitting is performed on each load growth data to obtain the load growth potential coefficient of each sub-region; The load peak and valley data of each sub-region are processed by Fourier transform to obtain the load peak and valley characteristic sequence of each sub-region and combine it with each load growth potential coefficient. The contour drawing method based on kernel density estimation is used for processing to quantify the load density gradient, load growth trend line and load peak and valley change band to obtain the load difference data of each sub-region.

5. The method for substation layout in urban expansion area based on artificial intelligence algorithm according to claim 1 is characterized in that: The collecting of historical distribution network structure data of the urban expansion area and combining it with the load difference data to process it through the large language model to obtain an initial substation layout plan includes: Extracting the spatial distribution characteristics of the power load of each sub-region from each load difference data by using a convolutional neural network, and processing the spatial distribution characteristics of each power load by using a spatial coordinate mapper to generate a load density thermal distribution matrix of each sub-region; Acquire historical distribution network grid structure data of the urban expansion area, and use support vector regression algorithm to optimize the power supply radius and load coverage rate in the historical distribution network grid structure data to obtain a candidate substation location sequence; Based on the land use restriction conditions of each of the sub-areas, the candidate substation site sequence is preprocessed by the large language model, and a multi-objective optimization operation is performed on the preprocessed candidate substation site sequence by using a genetic algorithm to obtain a substation site selection coordinate set; The load density thermal distribution matrix of each sub-area is quantified by a hierarchical clustering algorithm to obtain the maximum load capacity of the power supply area to obtain a standard transformer capacity sequence from the equipment specification library, and a transformer combination scheme is calculated based on the standard transformer capacity sequence using a dynamic programming algorithm to obtain a main transformer capacity configuration table; Under the constraints of power supply reliability level requirements and load transfer requirements, the number of outgoing line loops is determined through a graph theory algorithm based on the main transformer capacity configuration table to obtain a substation outgoing line plan, which is combined with the substation site selection coordinate set and the main transformer capacity configuration table to obtain an initial substation layout plan.

6. The method for substation layout in urban expansion area based on artificial intelligence algorithm according to claim 1 is characterized in that: The analyzing the historical load data by a machine learning algorithm to obtain the load growth probability distribution of each sub-region within the target period to simulate the power supply capacity of each sub-region under different load growth rates includes: Extracting trend items, period items, and random items from the historical load data of each sub-region by a time series decomposition algorithm, and processing the extracted results by a kernel density estimation method to obtain a load growth probability distribution matrix of each sub-region; Based on the load growth probability distribution matrices, a long short-term memory network is used to predict the load growth data at monthly, quarterly and annual time scales to obtain a multi-time scale load growth trend vector; According to the multi-time scale load growth trend vector, the load level of each sub-area under multiple growth rate scenarios is sampled by Monte Carlo method to obtain the main transformer load probability distribution to quantify the average utilization hours, maximum load rate and load duration of the equipment, and obtain the equipment utilization characteristic matrix; The equipment utilization characteristic matrix is ​​fitted by polynomial regression to quantify the power supply capacity variation trend of each sub-region at multiple time scales, and the power supply capacity of each sub-region at different load growth rates is obtained.

7. The method for substation layout in urban expansion area based on artificial intelligence algorithm according to claim 5 is characterized in that: The adopting of a dynamic programming algorithm to analyze the power supply capacity of each of the sub-areas to optimize the initial substation layout plan, obtain the substation optimized layout plan and execute it, including: Based on the power supply capacity of each sub-area, a multi-stage optimization function with the minimum investment cost as the goal and the equipment utilization rate not lower than the preset threshold as the constraint is solved by a dynamic programming algorithm to obtain a substation construction time planning table; According to the transformer capacity specification constraints, a genetic algorithm is used to optimize the main transformer capacity configuration table in the initial substation layout plan to obtain a main transformer capacity configuration plan; Based on the main transformer capacity configuration plan, equipment load rate data is obtained to calculate the annual average utilization hours and maximum load rate of the equipment through a multi-objective optimization algorithm to obtain a substation capacity expansion plan; Extracting grid structure data from the historical distribution network grid structure data according to the substation capacity expansion plan, so as to calculate the substation power supply radius and load coverage rate through a graph theory algorithm, and obtain a substation layout optimization plan; The substation site optimization plan, the substation outgoing line plan, the substation construction time planning table, the main transformer capacity configuration plan and the substation capacity expansion plan are combined to obtain a substation optimization layout plan and execute it.

8. The method for substation layout in urban expansion area based on artificial intelligence algorithm according to claim 5 is characterized in that: The method of using a dynamic programming algorithm to analyze the power supply capacity of each sub-area to optimize the initial substation layout plan, obtain and execute the substation optimized layout plan, and further includes: Based on the historical load data of each of the sub-regions, a load forecasting function is constructed through a recurrent neural network to calculate the load growth probability under multiple scenarios, and a multi-scenario load forecasting matrix is ​​obtained; According to the multi-scenario load forecasting matrix, the load data of different periods under each scenario is used as a state variable, and the construction time of the new substation and the transformer capacity are used as decision variables. The weighted combination of equipment utilization rate and the power supply capacity of each sub-area is used as the objective function, and the dynamic programming algorithm is used to solve it to obtain the substation construction decision space; Taking the substation construction decision space as input, establishing an optimization function with the combination of construction time and transformer capacity as the decision vector through a heuristic algorithm, performing weighted calculation on the average equipment utilization rate and power supply reliability rate, and outputting a substation construction timetable; A sorting algorithm is used to prioritize the construction scale of each stage in the substation construction time sequence, and a construction scale allocation sequence is generated under the constraints of voltage qualification rate and power supply reliability rate, and a phased construction planning table is obtained to optimize the initial layout plan of the substation, and the optimized layout plan of the substation is obtained and executed.

9. The method for substation layout in urban expansion area based on artificial intelligence algorithm according to claim 1 is characterized in that: The method of using a dynamic programming algorithm to analyze the power supply capacity of each sub-area to optimize the initial substation layout plan, obtain the substation optimized layout plan and execute it, further comprising: Obtaining load change data and equipment operation data when executing the substation optimization layout plan; The load change data is sequence modeled by a recurrent neural network to obtain a load growth dynamic prediction sequence, and the equipment operation data is processed by a clustering algorithm to obtain an equipment utilization state vector; Based on the equipment utilization state vector, a fuzzy inference algorithm is used to quantify the remaining life of the equipment, and an equipment utilization classification result is obtained to adjust the substation optimization layout plan, and a construction plan adjustment plan is obtained and executed.

10. A system for substation layout in urban expansion area based on artificial intelligence algorithm, characterized in that: include: A demand forecasting module is used to obtain geographic spatial data and historical load data of the urban expansion area, and use a large language model to mine the association rules between the geographic spatial data and the historical load data to obtain key features to predict the power load demand data of the urban expansion area within a target period; A partition clustering module is used to analyze the power load demand data through the large language model to obtain load change characteristics to partition and cluster the urban expansion area to obtain each sub-area and its load difference data; A scheme generation module, used for collecting historical distribution network structure data of the urban expansion area, and combining it with the load difference data, so as to obtain an initial substation layout scheme through processing by the large language model; A capacity simulation module is used to analyze the historical load data through a machine learning algorithm to obtain the load growth probability distribution of each sub-area within the target period, so as to simulate the power supply capacity of each sub-area under different load growth rates; The scheme optimization module is used to analyze the power supply capacity of each sub-area by using a dynamic programming algorithm to optimize the initial layout scheme of the substation, obtain the optimized layout scheme of the substation and execute it.

Images