A Parametric 3D Power Grid Framework Intelligent Construction Method

Abstract

The present application provides a method for intelligent construction of a parameterized three-dimensional power grid network framework, which relates to the field of power technology. The method includes steps such as first collection, second collection, third collection, model construction, demand prediction, power grid division, first setting, and architecture construction. The present application can dynamically monitor the charging demand fluctuations of electric vehicles. At the same time, according to the charging demands and regional characteristics of different regions, a clustering algorithm is used to divide the power grid. Based on the power grid division results, load centers are set and a power grid architecture is constructed. The constructed power grid architecture can better cope with the power load fluctuations and power load regional differences brought about by the popularization of electric vehicles, enhance the dynamic response ability of the power grid, and improve the stability of the power grid.

Description

Technical Field

[0001] This application relates to the field of power technology, and in particular to a method for intelligently constructing a parametric three-dimensional power grid framework. Background Art

[0002] With the acceleration of the urbanization process and the rapid development of the electric vehicle (EV) industry, the traditional power grid architecture is facing huge challenges. The widespread popularity of electric vehicles has led to a surge in charging demand, especially in high-density urban areas. The load of the power grid shows high time-variability and regionality, and the original power grid architecture often has difficulty coping with the fluctuations of the charging load.

[0003] The Chinese invention patent with the publication date of March 30, 2022 and the publication number of CN115238943A provides an active distribution network planning method. This patent obtains the basic data of the active distribution network; according to the obtained data, constructs an active distribution network planning model with the minimum annual comprehensive cost as the objective function, considering the uncertainty of renewable energy reserve and the delay of electric vehicle reserve; linearizes the non-linear active distribution network planning model, and uses the mixed integer linear programming algorithm to solve the linearized active distribution network planning model to obtain the planning scheme of the active distribution network.

[0004] For the above technical solution, the goal is to minimize the annual comprehensive cost, considering the reserve uncertainty of renewable energy and electric vehicles, and solving the distribution network planning problem through the mixed integer linear programming algorithm. However, the above solution focuses on the statistics and planning of historical data, ignoring the dynamic impact of dynamic factors such as holidays and traffic flow on the charging load of electric vehicles, and cannot timely respond to the time-variability and regionality of the load presented by the widespread application of electric vehicles to the power grid, resulting in the appearance of local overload or unreasonable allocation of power grid resources. Summary of the Invention

[0005] In order to better cope with the load fluctuations and regional differences brought about by the popularization of electric vehicles, enhance the dynamic response of the power grid, and improve the stability and efficiency of the power grid, this application provides a method for intelligently constructing a parametric three-dimensional power grid framework.

[0006] In the first aspect, this application provides a method for intelligently constructing a parametric three-dimensional power grid framework, adopting the following technical solution:

[0007] A method for intelligently constructing a parametric three-dimensional power grid framework includes the following steps:

[0008] First collection: Collect geographical data, power grid equipment data, and regional characteristic data of each region according to the geographical information system; the power grid equipment data includes charging pile location data;

[0009] Second data collection: Collect the first real-time data of each area, where the first real-time data includes holiday situations, regional traffic flow, and car owners' travel habits;

[0010] Third data collection: Collect the first historical data of each area, where the first historical data includes the historical charging data of each charging pile, historical regional traffic flow, historical holiday situations, and historical car owners' travel habits;

[0011] Model construction: Construct a charging demand prediction model;

[0012] Demand prediction: Input the first real-time data into the charging demand prediction model to obtain the charging demand prediction data of each charging pile;

[0013] Power grid division: Cluster the charging piles based on the location data of each charging pile, the charging demand prediction data of each charging pile, and the regional characteristic data. Obtain several sub-areas according to the clustering results, and take the sum of the charging demand prediction data of the charging piles within each sub-area as the predicted load of each sub-area;

[0014] First setting: Divide the sub-areas according to the preset grid size, divide the predicted load of each sub-area based on the area of each grid to obtain the predicted load of each grid, obtain the geometric center coordinates of each grid, calculate the coordinates of the load center of each sub-area using the weighted centroid method, and take the predicted load of each sub-area as the load capacity of the load center;

[0015] Grid architecture construction: Connect the load centers of each sub-area based on the load capacity of the load centers of each sub-area to construct a power grid architecture.

[0016] By adopting the above technical solutions, a charging demand prediction model is constructed to predict the charging demand of each charging pile at different time periods, which can accurately predict the future charging demand, identify the charging peak and trough periods in advance, help achieve reasonable scheduling of power resources, and reduce the situation of the power grid being overloaded due to sudden load fluctuations. In addition, based on the location data of each charging pile, the charging demand prediction data of each time period, and the regional characteristic data, the clustering algorithm is used to divide the power grid into regions. According to the charging pile location and the corresponding charging demand prediction data, it is divided into multiple regions, and the regions with high and low charging demands and the regions with similar time fluctuations in charging demand can be identified, so as to allocate power resources targeted, reduce the situations of load overload and resource idleness in each region, help plan the power grid capacity more reasonably, thereby optimizing the allocation of power resources, and improving the adaptability of the power grid to load changes. Moreover, based on the divided power grid, load centers are set, and appropriate electrical equipment and transmission lines are selected based on the set load centers to construct a power grid architecture, which can accurately master the predicted load changes of each sub-area, adjust the power distribution in a timely manner, optimize the load distribution of the power grid, and improve the stability and security of the power grid.

[0017] Optionally, after the step of performing the third collection and before the step of performing model construction, it further includes:

[0018] First regional division: According to the regional characteristic data of each region, use the clustering algorithm to perform the first regional division on each region to obtain a new regional grouping result, denoted as the first traffic region;

[0019] First data matching: Map the historical travel habits of vehicle owners and the historical regional traffic flow into each first traffic region to obtain the new historical travel habits of vehicle owners and the new historical regional traffic flow in each first traffic region, denoted as the first historical vehicle travel data and the first historical traffic flow data respectively;

[0020] Second data matching: Use the historical holiday situation to divide the first historical vehicle travel data and the first historical traffic flow data respectively to obtain the non-holiday historical vehicle travel data, holiday historical vehicle travel data, non-holiday historical traffic flow data, and holiday historical traffic flow data. Denote the non-holiday historical vehicle travel data and the non-holiday historical traffic flow data as the first traffic data, and denote the holiday historical vehicle travel data and the holiday historical traffic flow data as the second traffic data;

[0021] Calculate the transfer probability: Based on the first traffic data and the second traffic data respectively, use the Markov chain method to calculate the first vehicle transfer probability and the second vehicle transfer probability of each first traffic region. Denote the first vehicle transfer probability and the second vehicle transfer probability of the first traffic region as the comprehensive transfer probability of the first traffic region;

[0022] Vehicle flow dynamic modeling: Collect the comprehensive transfer probabilities of each first traffic region to construct a vehicle flow dynamic model;

[0023] Obtain the vehicle flow distribution: Input the regional traffic flow into the vehicle flow dynamic model to obtain the predicted vehicle flow distribution data of each region, and use the predicted vehicle flow distribution data as the new regional traffic flow.

[0024] By adopting the above technical solution, a vehicle flow dynamic model is constructed. Combining the comprehensive transfer probability, it can dynamically simulate the vehicle flow distribution between regions. By predicting the future traffic flow, the prediction range of the charging demand prediction model can be extended to several days or even weeks, which is applicable to the charging demand planning and grid load scheduling on a longer time scale. It helps to identify the peak charging periods of charging piles in advance, perform advance scheduling on the grid load, and reduce the situation of local grid overload caused by concentrated load. Moreover, the vehicle flow dynamic model can reflect the changes in real-time and long-term traffic flow trends, improve the adaptability of the charging demand prediction model to complex scenarios, enhance the accuracy and adaptability of charging demand prediction and grid load management, and improve the stability and efficiency of the grid.

[0025] Optionally, after performing the step of the first regional division and before performing the step of the first data matching, it further includes:

[0026] Construct a traffic network diagram: Map the travel habits of vehicle owners and regional traffic flows in each first traffic region to obtain the travel habits of vehicle owners and regional traffic flows in each first traffic region. Using each first traffic region as a node and the travel habits of vehicle owners in each first traffic region and the corresponding regional traffic flow as edges, construct a traffic network diagram;

[0027] Second regional division: Based on the traffic network diagram, use a graph partitioning algorithm to perform a second regional division on each first traffic region, and use the result after the second regional division as the new first traffic region.

[0028] By adopting the above technical solution, a traffic network diagram and a graph partitioning algorithm are introduced in the regional division process, making the regional division more reasonable, and also helping the traffic flow dynamic model to adapt to the traffic network characteristics of different cities or regions, improving the scalability of the traffic flow dynamic model, helping the traffic flow dynamic model to more accurately reflect the traffic flows inside and outside each region, and improving the accuracy of the traffic flow dynamic model prediction.

[0029] Optionally, after performing the step of calculating the transition probability and before performing the step of traffic flow dynamic modeling, it further includes:

[0030] Fourth data collection: Collect the historical condition data corresponding to the first historical vehicle owner travel data and the first historical traffic flow data, where the historical condition data includes historical weather conditions and historical electricity prices;

[0031] Define a Bayesian network: According to the historical travel habits of vehicle owners in each first traffic region and the corresponding historical condition data, define the Bayesian network variables and the causal relationships between the variables; the Bayesian network variables include sub-variables and parent variables;

[0032] Construct a conditional probability table: Perform a full permutation of the values of the parent variables of the sub-variables to obtain several conditional combinations, calculate the probability values of each conditional combination based on the historical condition data and the historical travel habits of vehicle owners, and collect the probability values of each conditional combination to obtain a conditional probability table;

[0033] Optimize the transition probability: Based on the historical condition data of each first traffic region, obtain the conditional data corresponding to the first traffic data and the second traffic data, query the conditional probability table according to the obtained conditional data to obtain the conditional probability, use the conditional probability to optimize the comprehensive transition probability of the first traffic region, and use the optimized comprehensive transition probability as the new comprehensive transition probability;

[0034] After performing the steps of dynamic modeling of traffic flow and before performing the steps of obtaining traffic flow distribution, it further includes:

[0035] Fifth data collection: Collect real-time condition data, where the real-time condition data includes weather conditions and electricity prices;

[0036] Data annotation: Perform time-series matching on weather conditions, electricity prices, and regional traffic flow using timestamps, label the weather conditions and electricity price tags for the regional traffic flow, and use the labeled regional traffic flow as the new regional traffic flow.

[0037] By adopting the above technical solution, introducing the Bayesian network and conditional probability table, considering the influence of external factors on vehicle owner behavior and traffic flow, improving the prediction accuracy and adaptability of the traffic flow dynamic model, enhancing the accuracy of traffic flow prediction, not only improving the accuracy and real-time performance of the charging demand prediction model, but also helping the power grid and charging pile operators to more efficiently schedule resources, optimize services, reducing the situation of supply-demand imbalance caused by sudden condition changes, enabling the power grid to more effectively cope with complex and changing traffic environments.

[0038] Optionally, after performing the steps of model construction and before performing the steps of demand prediction, it further includes:

[0039] Construct a sample training set: Based on the first historical data, construct a sample training set;

[0040] Encoder training: Input the sample training set into the charging demand prediction model, and the Transformer encoder encodes the sample data in the sample training set;

[0041] Decoder prediction: The Transformer decoder decodes and predicts the encoded sample data to obtain the charging demand prediction data;

[0042] Iterative optimization: Define the mean squared error loss function, and iteratively optimize the charging demand prediction model by minimizing the mean squared error loss function, and use the optimized charging demand prediction model as the new charging demand prediction model.

[0043] By adopting the above technical solution, the self-attention mechanism of the Transformer model can capture long-range dependencies in the sample data, identify long-term trends and periodic fluctuations in the sample data, handle the charging demand differences of different charging piles at different times, and further improve the adaptability of the charging demand prediction model.

[0044] Optionally, after performing the steps of power grid division and before performing the steps of the first setting, it further includes:

[0045] First judgment: Judge whether there are charging piles in the sub-region:

[0046] If so, perform the steps of the second judgment;

[0047] If not, perform the steps of the first setting;

[0048] Second judgment: Determine whether the number of charging piles existing in the sub-region is greater than 1:

[0049] If so, calculate the load weight of each charging pile according to the charging demand prediction data of each charging pile, and according to the positions of all the charging piles in the current sub-region and the corresponding load weights, calculate the coordinates of the load center of the current sub-region by using the weighted centroid method, and use the predicted load of the current sub-region as the load capacity of the load center, and perform the steps of architecture construction;

[0050] If not, use the coordinates of the charging pile as the load center of the current sub-region, use the predicted load of the current sub-region as the load capacity of the load center, and perform the steps of architecture construction.

[0051] By adopting the above technical solution, in the case where there is no charging pile in the sub-region, the load center is directly set, which simplifies the processing flow; when there are multiple charging piles, the load weight is calculated through the charging data, considering the load weight and position of each charging pile, and the weighted centroid method is used to set the load center, making the load center closer to the actual charging demand distribution, ensuring that the position of the load center can best represent the charging demand distribution of the region, improving the rationality, flexibility and accuracy of the first setting, helping to optimize the power resource allocation, reduce the operation cost, make the power transmission more efficient, reduce unnecessary power waste, and improve the power supply efficiency and stability of the power grid.

[0052] Optionally, after performing the steps of grid division and before performing the steps of the first setting, it further includes:

[0053] Calculate the load density: According to the geographical data of each sub-region, obtain the area of each sub-region, calculate the load density of each time period of each sub-region according to the area of each sub-region and the predicted load of each charging pile in each sub-region at each time period, and aggregate the load density of each time period of the sub-region to obtain the load density feature vector of each sub-region;

[0054] Secondary grid division: Use the clustering algorithm to cluster each sub-region into different preliminary sub-regions by using the load density feature vector of each sub-region, and aggregate each preliminary sub-region to obtain a preliminary region set;

[0055] Construct a regional connection graph: According to the grid equipment data of each region, use each preliminary sub-region as a graph node and the transmission lines between each preliminary sub-region as edges to construct a regional connection graph;

[0056] Region optimization: Based on the region connection graph, the minimum cut algorithm is used for optimization, and each optimized preliminary sub-region is used as a new sub-region.

[0057] By adopting the above technical solution, during the power grid division process, by calculating the load density eigenvector of the sub-region and comprehensively considering the influence of the region area and the charging load distribution, the division result of the sub-region is more in line with the actual requirements. In addition, based on the obtained preliminary region set, a region connection graph is constructed, and the division of the sub-region is further optimized through the minimum cut algorithm, enabling the power grid to more flexibly adapt to the demand changes when facing the charging demand fluctuations in different regions, and improving the adaptability of the power grid. Moreover, the optimization based on considering the load density and the transmission network structure helps to formulate a more scientific long-term development plan for the power grid.

[0058] Optionally, after performing the steps of architecture construction, it further includes:

[0059] Third judgment: Obtain the actual charging load of each sub-region, calculate the difference between the actual charging load and the predicted load of each sub-region, denoted as the first difference, and judge whether the first difference of the sub-region is higher than the preset first load threshold:

[0060] If so, mark the sub-region with the first difference higher than the first load threshold as an overloaded region, and perform the steps of the fourth judgment;

[0061] If not, do nothing;

[0062] Fourth judgment: Denote the absolute value of the first difference of the remaining sub-regions except the overloaded regions as the second difference, and judge whether the second difference of the remaining sub-regions except the overloaded regions is higher than the preset second load threshold:

[0063] If so, mark the sub-region with the second difference higher than the second load threshold as an idle region, calculate the difference between the second difference of the idle region and the preset third load threshold, denoted as the allocation load, allocate the allocation load of the idle region to the overloaded region, and at the same time reduce the charging power of the charging piles in the overloaded region;

[0064] If not, reduce the charging power of the charging piles in the overloaded region.

[0065] By adopting the above technical solution, setting multiple load thresholds and performing difference judgments can handle different load scenarios, accurately control the risks of load overloading and excessive load, and reduce the situation of long-term overloading operation by dynamically adjusting the charging power and load distribution, which helps to extend the service life of the charging piles and power grid equipment and reduce the equipment maintenance cost.

[0066] In summary, the present application includes at least one of the following beneficial technical effects:

[0067] 1. Build a charging demand prediction model to predict the charging demands of each charging pile at different time periods, which can accurately predict future charging demands, identify peak and off-peak charging periods in advance, help achieve reasonable power resource scheduling, and reduce the overload of the power grid due to sudden load fluctuations. In addition, based on the location data of each charging pile, the predicted charging demand data for each time period, and the regional characteristic data, use the clustering algorithm to divide the power grid into regions. According to the charging pile locations and the corresponding predicted charging demand data, it can be divided into multiple regions, and regions with high and low charging demands and regions with similar time fluctuations in charging demands can be identified, so as to allocate power resources targeted, reduce the situations of load overload and resource idleness in each region, help plan the power grid capacity more reasonably, thereby optimizing the power resource allocation, and improving the adaptability of the power grid to load changes. Moreover, based on the divided power grid, set load centers, and select appropriate electrical equipment and transmission lines based on the set load centers to build a power grid architecture, which can accurately grasp the predicted load changes in each sub-region, adjust the power distribution in a timely manner, optimize the load distribution of the power grid, and enhance the stability and security of the power grid.

[0068] 2. Build a traffic flow dynamic model, combined with the comprehensive transfer probability, which can dynamically simulate the traffic flow distribution between regions. By predicting future traffic flows, the prediction scope of the charging demand prediction model can be extended to several days or even weeks, which is suitable for charging demand planning and power grid load scheduling on a longer time scale, helps identify peak charging periods of charging piles in advance, perform advance scheduling of the power grid load, and reduce the overload of local power grids due to concentrated loads. Moreover, the traffic flow dynamic model can reflect the changes in real-time and long-term traffic flow trends, improve the adaptability of the charging demand prediction model to complex scenarios, enhance the accuracy and adaptability of charging demand prediction and power grid load management, and improve the stability and efficiency of the power grid.

[0069] 3. During the power grid division process, by calculating the load density eigenvector of the sub-region and comprehensively considering the influence of the regional area and the charging load distribution, the division results of the sub-region are more in line with the actual requirements. In addition, based on the constructed regional connection graph and the obtained preliminary regional set, further optimize the sub-region division through the minimum cut algorithm, so that the power grid can more flexibly adapt to demand changes when facing charging demand fluctuations in different regions, and improve the adaptability of the power grid. Brief Description of the Drawings

[0070] Figure 1 is the flowchart of Embodiment 1 of the present application;

[0071] Figure 2 is the flowchart of S41 model training in Embodiment 1 of the present application;

[0072] Figure 3It is the flowchart of S31 traffic flow prediction in Embodiment 2 of the present application;

[0073] Figure 4 It is the flowchart of S317 conditional probability optimization in Embodiment 2 of the present application. Detailed implementation manners

[0074] The following is a further detailed description of the present application in conjunction with Figures 1 to 4 to further elaborate on the present application.

[0075] Embodiment 1: This embodiment discloses a method for intelligently building a parametric three-dimensional power grid framework. As Figure 1 shown, the building method includes: collecting geographical data, power grid equipment data, regional feature data, first real-time data, and first historical data of each region, and constructing a charging demand prediction model based on the Transformer model; inputting the first real-time data into the charging demand prediction model to obtain the charging demand prediction data of each charging pile, clustering the charging piles based on the location data of each charging pile, the charging demand prediction data of each charging pile, and the regional feature data, obtaining several sub-regions according to the clustering result, taking the sum of the charging demand prediction data of the charging piles in each sub-region as the predicted load of each sub-region, dividing each sub-region into grids, calculating the coordinates of the load center of each sub-region based on the geometric center coordinates of the grids, taking the predicted load of each sub-region as the load capacity of the load center, and connecting the load centers of each sub-region based on the load capacity of the load centers of each sub-region to construct a power grid architecture. This embodiment includes the following steps:

[0076] S1 First collection: Collecting geographical data, power grid equipment data, and regional feature data of each region according to the geographic information system; the power grid equipment data includes the location data of the charging piles.

[0077] Collecting the geographical data of each region by using the geographic information system, where the geographical data includes the regional boundary, regional terrain, regional coordinates, regional area, and regional shape. The power grid equipment data includes the path, capacity, and electrical characteristics of the transmission line, and the power grid equipment data also includes electrical equipment, and the electrical equipment includes the location data of the charging piles. The regional feature data includes the population density within the region and the building functions within each region, and the building functions include commercial areas, industrial areas, office areas, and residential areas.

[0078] In this embodiment, each region is each administrative region of the area.

[0079] S2 Second collection: Collecting the first real-time data of each region, where the first real-time data includes holiday situations, regional traffic flow, and car owner travel habits.

[0080] The regional traffic flow includes the traffic flow data of electric vehicles within the region and the traffic flow change data of electric vehicles in each time period. The regional traffic flow can be obtained through the traffic management platform. The holiday situation includes weekdays and non-weekdays. Non-weekdays include statutory holidays (including Spring Festival, Tomb-Sweeping Festival, Labor Day, Dragon Boat Festival, Mid-Autumn Festival, National Day, etc.), floating holidays (including Children's Day, etc.), Saturdays and Sundays. Weekdays generally refer to Monday to Friday. In some cases, there are overlaps between holidays, between holidays and Saturdays and Sundays, and between holidays and weekdays. At this time, the dates that overlap with holidays are all marked as the corresponding holidays. For example, the Dragon Boat Festival includes Friday, Saturday and Sunday, which are collectively called the Dragon Boat Festival. The holiday situation is imported through the electronic calendar system. The travel habits of vehicle owners include the travel frequency, travel time period and travel route of electric vehicles, which can be obtained through in-vehicle GPS devices or the traffic management platform.

[0081] S3 Third collection: Collect the first historical data of each region. The first historical data includes the historical charging data of each charging pile, historical regional traffic flow, historical holiday situation and historical travel habits of vehicle owners.

[0082] The historical charging data of each charging pile includes the historical charging pile status, historical charging power and the corresponding historical charging duration. The historical regional traffic flow includes the historical traffic flow data of electric vehicles within the region and the historical traffic flow change data of electric vehicles in each time period. The historical travel habits of vehicle owners include the travel frequency, travel time period and travel route of electric vehicles, which can be obtained through in-vehicle GPS devices or the traffic management platform. The historical holiday situation is imported through the electronic calendar system.

[0083] In this embodiment, after collecting the above data, it is necessary to preprocess the data. The steps of the data preprocessing include data cleaning and data normalization.

[0084] Data cleaning: Perform operations including but not limited to denoising, handling missing values and removing outliers on the geographical data, grid equipment data, regional characteristic data, first real-time data and first historical data of each region collected.

[0085] Reduce the noise interference in the data through methods such as smoothing processing, frequency domain filtering and outlier detection denoising. Handle missing values through methods such as interpolation method, mean filling, nearest neighbor interpolation, regression filling and deletion method. Abnormal points can be identified through visualization means such as box plots, scatter plots and histograms, and outliers can be removed.

[0086] Data normalization: Normalize the geographical data, grid equipment data, regional characteristic data, first real-time data and first historical data of each region collected.

[0087] S4 Model Construction: Based on the Transformer model, a charging demand prediction model is constructed. The charging demand prediction model includes an encoder and a decoder. In this embodiment, the charging demand prediction model is provided with a self-attention mechanism.

[0088] S41 Model Training: It includes S411 First Data Processing, S412 Second Data Processing, S413 Constructing a Sample Training Set, S414 Encoder Training, S415 Decoder Prediction, and S416 Iterative Optimization, as Figure 2 shown.

[0089] S411 First Data Processing: Based on the historical charging durations in the historical charging data of each charging pile, different charging time periods of the charging piles are divided. Using the historical charging durations, the divided charging time periods, and the historical charging power, calculate the historical charging demands of each charging pile in different time periods, denoted as the first historical charging data.

[0090] S412 Second Data Processing: Temporally match the historical holiday situations and the historical regional traffic flow through timestamps. Then, label the historical regional traffic flow with the corresponding historical holiday situation labels, and denote the labeled regional traffic flow as the first historical labeled data.

[0091] S413 Constructing a Sample Training Set: Extract the time features of the first labeled historical data and the first historical charging data. Take the data at the same time as a set of input data, denoted as a sample data. Collect all sample data to obtain a sample training set.

[0092] S414 Encoder Training: Input the sample data in the sample training set into the charging demand prediction model. The Transformer encoder transforms the input sample data into a context-related feature representation through multiple self-attention layers and a feed-forward neural network, and takes the output of the Transformer encoder (i.e., the context-related feature representation) as input and passes it to the Transformer decoder.

[0093] S415 Decoder Prediction: The Transformer decoder decodes according to the context-related feature representation output by the Transformer encoder to predict the future charging demand and obtain the charging demand prediction data.

[0094] S416 Iterative Optimization: Define the mean squared error loss function. Using the defined mean squared error loss function, calculate the loss value of the charging demand prediction model. Use the backpropagation algorithm to calculate the gradient of the model parameters of the charging demand prediction model. Use the gradient descent method to update the model parameters of the charging demand prediction model until the preset number of iterative training times is completed or the calculated loss value of the charging demand prediction model no longer decreases. Complete the model training to obtain an optimized charging demand prediction model, and use the optimized charging demand prediction model as the new charging demand prediction model.

[0095] The formula for the mean squared error loss function is as follows:

[0096] 。

[0097] Where, represents the mean squared error loss function of the charging demand prediction model, represents the number of samples in the sample training set, represents the th actual charging demand of the sample, represents the predicted charging demand data of the th sample predicted by the model.

[0098] S42 First Marking: Perform temporal matching on the holiday situation and regional traffic flow through timestamps, label the corresponding holiday situation tags for the regional traffic flow, and record the marked regional traffic flow as the first marked data.

[0099] S5 Demand Prediction: Input the first marked data into the charging demand prediction model for model inference to obtain the charging demand prediction data for each charging pile at each time period.

[0100] S6 Grid Division: Based on the location data of each charging pile, the charging demand prediction data for each time period, and the regional feature data, use weighted distance clustering (K-means clustering or hierarchical clustering) or DBSCAN clustering algorithm to cluster the charging piles. Take each clustering result as a sub-region, and each unclustered region is also taken as a sub-region to obtain several sub-regions. Take the sum of the charging demand prediction data of the charging piles in each sub-region as the predicted load of each sub-region.

[0101] In this embodiment, the K-means clustering method is used to divide the power grid:

[0102] Record each region as the original region, use the geographic information system to obtain the geographical location and regional feature data of each original region, and perform standardization processing on the geographical location and regional feature data of each original region.

[0103] Extract the time features of the charging demand prediction data for each charging pile at each time period to obtain the charging demand time feature data for each charging pile per day.

[0104] Perform one-hot encoding on the building functions in the regional feature data for each original region to obtain the encoded building functions. Construct a linear regression model, select the charging demand prediction data of the charging piles as the target variable, analyze the relationship between the charging demand prediction data of each charging pile per day and the population density and building functions in the regional feature data, and obtain the regression weights of the population density and the regression weights of each building function.

[0105] The formula for the linear regression model is:

[0106] .

[0107] .

[0108] Among them, represents the charging demand prediction data of the charging pile, represents the model intercept, represents the regression weight of the population density, represents that the building function is the regression weight of, represents that the building function is the regression weight of, represents that the building function is the regression weight of, represents that the building function is the regression weight of, represents the total regression weight of the building function, represents the error term.

[0109] Among them, = 1.

[0110] Obtain the location data of each charging pile, and calculate the geographical distance between each pair of charging piles through the Haversine formula or Euclidean distance.

[0111] In this embodiment, the Euclidean distance is used to calculate the geographical distance between each pair of charging piles. The Euclidean distance formula is:

[0112] .

[0113] Among them, represents the geographical distance between charging pile and charging pile , represents the longitude of charging pile , represents the longitude of charging pile , Represents a charging pile 's latitude, Represents a charging pile 's latitude.

[0114] Calculate the weighted distance of the regional characteristics of the areas where each pair of charging piles are located. The formula for the weighted distance of regional characteristics is:

[0115] .

[0116] Among them, Represents the area where the charging pile is located and the area where the charging pile is located The weighted distance of the regional characteristics between them, Represents the regression weight of population density, Represents the area where the charging pile is located 's population density, Represents the area where the charging pile is located 's population density, Represents the total regression weight of building functions, Represents the area where the charging pile is located 's building function, Represents the area where the charging pile is located 's building function, where = 1.

[0117] Based on the geographical distance and the weighted distance of regional characteristics between each pair of charging piles, calculate the weighted Euclidean distance between each charging pile.

[0118] The formula for the weighted Euclidean distance is:

[0119] .

[0120] Among them, Represents the charging pile and the charging pile The weighted Euclidean distance between them.

[0121] Adopt the K-means clustering algorithm, use the calculated weighted Euclidean distance to cluster the areas where the charging piles are located. Take each clustering result as a sub-region, and take each unclustered area as a sub-region to obtain several sub-regions. Use the sum of the charging demand prediction data of all charging piles in each sub-region as the predicted load of each sub-region.

[0122] S7 First setting: Divide each sub-region according to a preset grid size, divide each sub-region into multiple grids to form a grid structure, and each grid represents a small geographical unit. Divide the predicted load of each sub-region according to the area of each grid to obtain the predicted load of each grid in the sub-region. Use a geographic information system to obtain the geometric center coordinates of each grid in the sub-region, and calculate the coordinates of the load center of each sub-region using the first weighted centroid formula. Take the predicted load of each sub-region as the load capacity of the load center.

[0123] In this embodiment, the grid size can be set according to actual needs.

[0124] The first weighted centroid formula is:

[0125] ;

[0126] .

[0127] Among them, represents the longitude of the load center in the current sub-region, represents the latitude of the load center in the current sub-region, represents the th predicted load of the grid in the current sub-region, represents the th longitude of the grid in the current sub-region, represents the th latitude of the grid in the current sub-region, represents the number of grids in the current sub-region.

[0128] Example 1: The total area of a sub-region is 95 km², the predicted load of the sub-region is 400 kW, and the preset grid size is 20 km². Divide the sub-region.

[0129] 1. Divide the grid

[0130] Divide the sub-region according to the preset grid size and process the remaining area. The sub-region is divided into 4 standard grids and one remaining area. Grid A is 20 km², Grid B is 20 km², Grid C is 20 km², Grid D is 20 km², and the remaining area is denoted as Grid E, and Grid E is 15 km².

[0131] 2. Allocate the predicted load

[0132] The predicted load of each grid is allocated according to the grid area. Therefore, the predicted loads obtained by Grid A, Grid B, Grid C, and Grid D are:

[0133] .

[0134] The predicted load of grid E is: .

[0135] 3. Calculate the load center

[0136] Let the geometric center coordinates (unit: km) of each grid be as follows:

[0137] Grid A: ; Grid A: ; Grid C: ; Grid D: ; Grid E: .

[0138] Then, calculate the load center coordinates according to the first weighted centroid formula:

[0139] ;

[0140] ;

[0141] Therefore, the coordinates of the load center of this sub-region are , and the load capacity of the load center is 400 kW.

[0142] S8 architecture construction: Based on the load capacity of the load center of each sub-region, select the corresponding electrical equipment, and connect the load centers of each sub-region to each other through transmission lines to construct the power grid architecture.

[0143] S81 Select electrical equipment: Based on the load capacity of the load center of each sub-region, select the type and capacity of the substation and transformer.

[0144] Based on the load capacity of the load center of each sub-region, refer to the pre-set load capacity level correspondence table to obtain the load level of the load center of each sub-region, and select the substation type according to the load level of the load center of each sub-region. For example, for sub-regions with a high load level of the load center, select to build large centralized substations, and for sub-regions with a low load level of the load center, select to build distributed substations or small transformer networks. The substation includes main transformers, standby transformers, distribution devices, and voltage control and regulation equipment.

[0145] Take 20% of the load capacity of the load center in each sub-region as the redundant load for the corresponding sub-region, and take the sum of the load capacity of the load center in each sub-region and the corresponding redundant load as the capacity of the main transformer in each sub-region. The capacity of the standby transformer is the same as that of the main transformer, and the standby transformer is used to provide redundant support when the main transformer fails. The distribution device includes distribution busbars, circuit breakers, switchgear, etc., which are used to distribute the power of the substation to each region and ensure the safety and stability of the power supply. The voltage control and regulation equipment includes voltage regulators, capacitors, voltage regulation devices, etc., which are used to control the voltage within a predetermined range, ensure the stable operation of the power grid, and avoid the impact of too high or too low voltage on electrical equipment and user electricity consumption.

[0146] S82 Determine the power grid topology: Design the power grid topology based on the coordinates and load capacity of the load centers in each sub-region.

[0147] Preset a load screening threshold, compare the load capacity of each load center with the load screening threshold, record the load center with a load capacity greater than the preset load screening threshold as the first load center, and record the load center with a load capacity less than or equal to the preset load screening threshold as the second load center. Connect each first load center with a main line, connect each second load center with each first load center with a branch line, and form a ring network structure with the main line and the branch line to obtain the power grid topology. At the same time, use software tools such as PSS / E, DIgSILENT, etc. to perform load flow simulation on the obtained power grid topology, analyze the current and voltage distribution of the entire power grid, ensure that the voltage fluctuates stably within the specified range, and avoid the impact of too high or too low voltage on the safe operation of the power grid.

[0148] S9 Load regulation: It includes S91 the third judgment and S92 the fourth judgment.

[0149] S91 The third judgment: Obtain the actual charging load of each sub-region, calculate the difference between the actual charging load and the predicted load of each sub-region, record it as the first difference, and judge whether the first difference of the sub-region is higher than the preset first load threshold.

[0150] If so, record the sub-region with the first difference higher than the first load threshold as the overloaded region and execute S92 the fourth judgment.

[0151] If not, do nothing.

[0152] S92 The fourth judgment: Record the absolute value of the first difference of the remaining sub-regions except the overloaded region as the second difference, and judge whether the second difference of the remaining sub-regions except the overloaded region is higher than the preset second load threshold.

[0153] If so, mark the sub-region where the second difference is higher than the second load threshold as the idle region, calculate the difference between the second difference of the idle region and the preset third load threshold, and mark it as the allocation load. Allocate the allocation load of the idle region to the overloaded region, and at the same time reduce the charging power of the charging piles in the overloaded region to the first preset charging power.

[0154] If not, reduce the charging power of the charging piles in the overloaded region to the second preset charging power.

[0155] In the fourth judgment of S92, allocating the allocation load of the idle region to the overloaded region includes S921 the first sorting, S922 the second sorting, S923 the first allocation, and S924 the second allocation.

[0156] S921 the first sorting: Sort the overloaded regions in descending order based on the first difference of the overloaded regions to obtain the overloaded region load difference sorting table.

[0157] S922 the second sorting: Calculate the distances between the load centers of the overloaded region ranked first and the load centers of each idle region, and sort them in ascending order to obtain the idle region distance sorting table of the overloaded region ranked first.

[0158] S923 the first allocation: Prioritize selecting the idle region ranked first in the idle region distance sorting table to allocate load for this overloaded region. If the allocation load is still less than the first difference of this overloaded region at this time, then select the idle region ranked second in the order of the idle region distance sorting table of this overloaded region to allocate load for this overloaded region until the total allocation load is equal to the first difference of this overloaded region.

[0159] S924 the second allocation: According to the overloaded region load difference sorting table, perform S922 the second sorting to S923 the first allocation for each overloaded region in turn until the load allocation task for each overloaded region is completed. If the allocation load allocated by the idle region is not sufficient to complete the load allocation tasks of all overloaded regions, then reduce the charging power of the charging piles in the overloaded regions where the load allocation tasks are not completed to the second preset charging power.

[0160] Example: A power grid includes 4 sub-regions, namely region a, region b, region c, and region d, and each sub-region has multiple charging piles.

[0161] The predicted load of region a is 100 kWh, the actual charging load of region a is 150 kWh, the predicted load of region b is 180 kWh, the actual charging load of region b is 110 kWh, the predicted load of region c is 200 kWh, the actual charging load of region c is 220 kWh, the predicted load of region d is 90 kWh, and the actual charging load of region d is 80 kWh.

[0162] The first load threshold is 0 kWh, the second load threshold is 10 kWh, and the third load threshold is 15 kWh.

[0163] 1. According to the calculation, the first difference in area a is 50 kWh, the first difference in area b is -70 kWh, the first difference in area c is 20 kWh, and the first difference in area d is -10 kWh.

[0164] 2. Compare the first differences of each area with the first load threshold. The first differences of area a and area c are greater than the first load threshold, so area a and area c are overloaded areas.

[0165] 3. Calculate the second differences of area b and area d, which are 70 kWh and 10 kWh respectively. Compare the second differences of area b and area d with the second load threshold. The second difference of area b is greater than the second load threshold, so area b is an idle area. Calculate the difference between the second difference of area b and the third load threshold, and the allocated load of area b is 55 kWh.

[0166] 4. Sort area a and area c in descending order according to the first differences of area a and area c. The first is area a, and the second is area c. Then reduce the charging power of the charging piles in area a and area c to the first preset charging power.

[0167] 5. Area b first allocates 50 kWh of load to area a, and then allocates the remaining 5 kWh of load to area c, and further reduces the charging power of the charging piles in area c to the second preset charging power.

[0168] This embodiment is a power grid architecture constructed based on the charging requirements of charging piles.

[0169] In this embodiment, a charging demand prediction model is constructed to predict the charging demands of each charging pile at different time periods, which can accurately predict future charging demands, identify peak and off-peak charging periods in advance, contribute to the rational scheduling of power resources, and reduce the overload of the power grid due to sudden load fluctuations. In addition, based on the location data of each charging pile, the charging demand prediction data for each time period, and the regional characteristic data, the power grid is divided into regions using a clustering algorithm. According to the charging pile locations and the corresponding charging demand prediction data, it is divided into multiple regions, and regions with high and low charging demands and regions with similar time fluctuations in charging demands can be identified, so as to allocate power resources targeted, reduce the situations of load overload and resource idleness in each region, contribute to more reasonable planning of the power grid capacity, thereby optimizing the allocation of power resources and improving the adaptability of the power grid to load changes. Moreover, a load center is set based on the divided power grid, and appropriate electrical equipment and transmission lines are selected based on the set load center to construct a power grid architecture, which can accurately grasp the predicted load changes in each sub-region, adjust the power distribution in a timely manner, optimize the load distribution of the power grid, and enhance the stability and security of the power grid.

[0170] Embodiment 2: The differences from Embodiment 1 are as follows:

[0171] As Figure 3 shown, after the third collection in S3 is executed and before the model construction in S4, it further includes S31 traffic flow prediction. S31 traffic flow prediction includes S311 first regional division, S312 construction of a traffic network diagram, S313 second regional division, S314 first data matching, S315 second data matching, S316 calculation of transfer probabilities, S317 conditional probability optimization, S318 dynamic modeling of vehicle flow, and S319 obtaining the vehicle flow distribution.

[0172] S311 first regional division: According to the population density and building functions of each region, the K-means clustering algorithm is used to perform the first regional division on each region to obtain a new regional grouping result, denoted as the first traffic region.

[0173] S312 construction of a traffic network diagram: Map the travel habits of vehicle owners and the regional traffic flow in each first traffic region to obtain the travel habits of vehicle owners and the regional traffic flow in each first traffic region. According to the travel habits of vehicle owners and the regional traffic flow in each first traffic region, taking each first traffic region as a node and defining the connection relationship of the edges with the travel habits of vehicle owners and the corresponding regional traffic flow in each first traffic region, a traffic network diagram is constructed.

[0174] S313 second regional division: Based on the traffic network diagram, the Louvain algorithm is used to perform the second regional division on each first traffic region, and the result after the second regional division is used as the new first traffic region.

[0175] S314 First data matching: Map the travel habits of historical vehicle owners and the historical regional traffic flow to each first traffic area to obtain the new travel habits of historical vehicle owners and the new historical regional traffic flow in each first traffic area, which are respectively recorded as the first historical vehicle owner travel data and the first historical traffic flow data.

[0176] S315 Second data matching: Use the historical holiday situation to divide the first historical vehicle owner travel data and the first historical traffic flow data respectively to obtain the non-holiday historical vehicle owner travel data, holiday historical vehicle owner travel data, non-holiday historical traffic flow data, and holiday historical traffic flow data. Record the non-holiday historical vehicle owner travel data and the non-holiday historical traffic flow data as the first traffic data, and record the holiday historical vehicle owner travel data and the holiday historical traffic flow data as the second traffic data.

[0177] S316 Calculate the transfer probability: Based on the first traffic data and the second traffic data respectively, use the Markov chain method to calculate the first vehicle transfer probability and the second transfer probability of each first traffic area, and perform a weighted sum of the first vehicle transfer probability and the second transfer probability of the first traffic area. Record the weighted sum result as the comprehensive transfer probability of the first traffic area.

[0178] The formula for calculating the vehicle transfer probability is:

[0179] 。

[0180] Among them, represents the vehicle transfer probability from the first traffic area to the first traffic area , the traffic flow from the first traffic area to the first traffic area , represents the sum of the traffic flows from the first traffic area to all first traffic areas, is the index, indicating traversing all first traffic areas.

[0181] S317 Conditional probability optimization: Includes S3171 Fourth data collection, S3172 Define the Bayesian network, S3173 Construct the conditional probability table, and S3174 Optimize the transfer probability, as Figure 4 shown.

[0182] S3171 Fourth data collection: Collect the historical conditional data corresponding to the first historical vehicle owner travel data and the first historical traffic flow data. The historical conditional data includes historical weather conditions and historical electricity prices.

[0183] Historical weather data can be obtained through the official website of the China Meteorological Administration, and historical electricity prices can be obtained through power companies.

[0184] S3172 Define the Bayesian network: Define the Bayesian network variables and the causal relationships between variables according to the historical conditional data corresponding to the historical travel habits of vehicle owners in each first traffic area; the Bayesian network variables include sub-variables and parent variables. The parent variables are weather conditions and electricity prices, and the sub-variables are vehicle owner travel habits.

[0185] S3173 Construct the conditional probability table: Perform a full permutation of the values of the parent variables of the sub-variables to obtain a number of conditional combinations (such as weather condition: sunny, electricity price: low electricity price, travel habit: 0.6 frequent travel; 0.4 no travel), and calculate the probability values of each conditional combination based on the historical conditional data and the historical travel habits of vehicle owners. Collect the probability values of each conditional combination to obtain the conditional probability table.

[0186] S3174 Optimize the transition probability: Based on the historical conditional data of each first traffic area, obtain the conditional data corresponding to the first traffic data and the second traffic data, query the conditional probability table according to the obtained conditional data to obtain the conditional probability, and use the conditional probability to optimize the comprehensive transition probability of the first traffic area using the Bayesian method. Take the optimized comprehensive transition probability as the new comprehensive transition probability.

[0187] The Bayesian formula is:

[0188] 。

[0189] Among them, represents the comprehensive transition probability under the conditional data , represents the conditional probability of the conditional data , represents the comprehensive transition probability before optimization, represents the total probability of the conditional data , = 1, which is the normalization coefficient.

[0190] S318 Dynamic modeling of traffic flow: Collect the comprehensive transition probabilities of each first traffic area to construct a dynamic traffic flow model.

[0191] Define the traffic flow distribution of each first traffic area at the initial moment as . Among them, represents the traffic flow of all first traffic areas at the initial moment, represents the traffic flow of the th first traffic area at the initial moment.

[0192] The comprehensive transfer probabilities of each first traffic area are aggregated to obtain a comprehensive transfer probability matrix, and a traffic flow dynamic model is constructed based on the comprehensive probability matrix.

[0193] 。

[0194] Among them, represents the traffic flow at time , represents the comprehensive transfer probability matrix at time , represents the traffic flow at time .

[0195] S3181 Fifth data collection: Collect real-time condition data, where the real-time condition data includes weather conditions and electricity prices.

[0196] S3182 Data annotation: The weather conditions, electricity prices, and regional traffic flow are temporally matched using timestamps, and the weather conditions and electricity price labels are annotated for the regional traffic flow. The annotated regional traffic flow is used as the new regional traffic flow.

[0197] S319 Obtain traffic flow distribution: Input the regional traffic flow into the traffic flow dynamic model to obtain the predicted traffic flow distribution data for each region. The predicted traffic flow distribution data is used as the new regional traffic flow.

[0198] In this embodiment, the predicted traffic flow distribution data is used as the new regional traffic flow, and the charging demand for each time period of each charging pile is predicted in combination with the corresponding holiday situation.

[0199] In this embodiment, the regions are the administrative regions of the area.

[0200] In this embodiment, by constructing a traffic flow dynamic model and combining the comprehensive transfer probability, the traffic flow distribution between regions can be dynamically simulated. By predicting future traffic flows, the prediction scope of the charging demand prediction model can be extended to several days or even weeks, which is applicable to the charging demand planning and grid load scheduling on a longer time scale. This helps to identify the peak charging periods of charging piles in advance, schedule the grid load in advance, and reduce the overload of the local grid due to concentrated loads. In addition, by introducing a traffic network diagram and a graph partitioning algorithm in the region partitioning process, the region partitioning becomes more reasonable, which also helps the traffic flow dynamic model to adapt to the traffic network characteristics of different cities or regions, improving the scalability of the traffic flow dynamic model. Moreover, by introducing a Bayesian network and a conditional probability table and considering the influence of external factors on the behavior of vehicle owners and traffic flows, the prediction accuracy and adaptability of the traffic flow dynamic model are improved, enhancing the accuracy of traffic flow prediction. This not only improves the accuracy and real-time performance of the charging demand prediction model, but also helps the grid and charging pile operators to schedule resources more efficiently, optimize services, and reduce the supply-demand imbalance caused by sudden changes in conditions, enabling the grid to more effectively respond to complex and changing traffic environments.

[0201] Embodiment 3: The differences from Embodiment 1 are as follows:

[0202] Before executing S7 First Setting after executing S6 Grid Partitioning, it further includes:

[0203] S611 First Judgment: Judge whether there is a charging pile in the sub-region. If so, execute S612 Second Judgment; if not, execute S7 First Setting.

[0204] S612 Second Judgment: Judge whether the number of charging piles existing in the sub-region is greater than 1.

[0205] If so, calculate the charging pile load weight according to the charging demand prediction data of each charging pile. According to the positions and corresponding load weights of all the charging piles in the current sub-region, calculate the coordinates of the load center of the current sub-region through the second weighted centroid formula, and take the predicted load of the current sub-region as the load capacity of the load center, and execute the steps of S8 Architecture Construction.

[0206] If not, take the coordinates of the charging pile as the load center of the current sub-region, take the predicted load of the current sub-region as the load capacity of the load center, and execute S8 Architecture Construction.

[0207] The formula for calculating the charging pile load weight is:

[0208] .

[0209] Wherein, represents the charging pile The load weight, represents the charging demand prediction data of the charging pile and represents the predicted load of the current sub-region.

[0210] The second weighted centroid formula is:

[0211] ;

[0212] .

[0213] Among them, represents the longitude of the load center in the current sub-region, represents the latitude of the load center in the current sub-region, represents the load weight of the charging pile , represents the longitude of the charging pile in the current sub-region, represents the latitude of the charging pile in the current sub-region, and

[0214] represents the number of charging piles in the current sub-region.

[0215] In this embodiment, according to the distribution of charging piles in the sub-region, considering the load weight and location of each charging pile, the location of the load center is set to ensure that the location of the load center can best represent the charging demand distribution of the region, which can effectively allocate grid resources, make power transmission more efficient, reduce unnecessary power waste, and improve the power supply efficiency and stability of the grid.

[0216] After performing the step of S6 grid division and before performing S7 first setting, it further includes:

[0217] S621 Calculate the load density: According to the geographical data of each sub-region, obtain the area of each sub-region, and calculate the load density of each time period of each sub-region based on the area of each sub-region and the predicted load of each charging pile in each sub-region at each time period. Aggregate the load density of each time period of the sub-region to obtain the load density feature vector of each sub-region.

[0218] S622 Secondary grid division: Use the K-means clustering algorithm to cluster each sub-region into different preliminary sub-regions by using the load density feature vector of each sub-region, and aggregate each preliminary sub-region to obtain a preliminary region set.

[0219] S623 Construct the regional connection graph: Based on the power grid equipment data of each region, use each preliminary sub-region as a graph node, use the transmission lines between each preliminary sub-region as edges, and set the edge weights according to the capacity and distance of the transmission lines between each preliminary sub-region to construct the regional connection graph. The edge weight is the ratio of the capacity of the transmission line to the distance of the transmission line.

[0220] S624 Region optimization: Based on the regional connection graph, use the minimum cut algorithm for optimization, and use the optimized regions as new sub-regions.

[0221] Arbitrarily specify two nodes in the regional connection graph as the source point and the sink point respectively, use the maximum flow algorithm (such as the Ford-Fulkerson algorithm) to calculate the maximum flow from the source point to the sink point, divide each node in the regional connection graph into a source point set and a sink point set according to the maximum flow result, so that the sum of the weights of the edges crossing the source point set and the sink point set is the smallest, obtain the minimum cut result in the regional connection graph, re-divide the preliminary sub-regions according to the minimum cut result, and use the re-divided regions as new sub-regions.

[0222] In this embodiment, the regions are administrative regions.

[0223] In this embodiment, during the power grid division process, by calculating the load density eigenvector of the sub-region and comprehensively considering the influence of the regional area and the charging load distribution, the division result of the sub-region is more in line with the actual demand. In addition, based on the obtained preliminary region set and the constructed regional connection graph, the division of the sub-region is further optimized by the minimum cut algorithm, so that the power grid can more flexibly adapt to the demand changes when facing the charging demand fluctuations in different regions, and the adaptability of the power grid is improved. Moreover, optimizing on the basis of considering the load density and the transmission network structure helps to formulate a more scientific long-term development plan for the power grid.

[0224] The above are all the preferred embodiments of this application. The protection scope of this application is not limited by this. Therefore, all equivalent changes made according to the structure, shape, and principle of this application should be covered within the protection scope of this application.

Claims

1. A parameterized three-dimensional power grid intelligent construction method, characterized in that: include: First collection: collect geographical data, power grid equipment data and regional characteristic data of each region based on the geographic information system; Grid equipment data includes charging station location data; Second collection: collect the first real-time data of each area, including holiday conditions, regional traffic flow and car owners' travel habits; Third collection: collecting the first historical data of each area, the first historical data includes the historical charging data of each charging pile, the historical regional traffic flow, the historical holiday conditions and the historical travel habits of car owners; First area division: Based on the regional characteristic data of each area, a clustering algorithm is used to perform the first area division on each area to obtain a new area grouping result, which is recorded as the first traffic area; First data matching: mapping the historical vehicle owner travel habits and historical regional traffic flow to each first traffic area, obtaining new historical vehicle owner travel habits and new historical regional traffic flow in each first traffic area, which are recorded as first historical vehicle owner travel data and first historical traffic flow data respectively; Second data matching: using historical holiday conditions to divide the first historical vehicle owner travel data and the first historical traffic flow data, respectively, to obtain non-holiday historical vehicle owner travel data, holiday historical vehicle owner travel data, non-holiday historical traffic flow data, and holiday historical traffic flow data, and record the non-holiday historical vehicle owner travel data and non-holiday historical traffic flow data as the first traffic data, and record the holiday historical vehicle owner travel data and holiday historical traffic flow data as the second traffic data; Calculate the transfer probability: based on the first traffic data and the second traffic data, use the Markov chain method to calculate the first vehicle transfer probability and the second vehicle transfer probability of each first traffic area, and record the first vehicle transfer probability and the second vehicle transfer probability of the first traffic area as the comprehensive transfer probability of the first traffic area; Traffic flow dynamic modeling: collect the comprehensive transfer probabilities of each first traffic area and build a traffic flow dynamic model; Obtaining traffic flow distribution: inputting regional traffic flow into the traffic flow dynamic model to obtain the predicted traffic flow distribution data of each region, and using the predicted traffic flow distribution data as the new regional traffic flow; Model construction: Build a charging demand prediction model; Constructing a sample training set: constructing a sample training set based on the first historical data; Encoder training: The sample training set is input into the charging demand prediction model, and the Transformer encoder encodes the sample data in the sample training set; Decoder prediction: The Transformer decoder decodes and predicts the encoded sample data to obtain charging demand prediction data; Iterative optimization: define a mean square error loss function, iteratively optimize the charging demand prediction model by minimizing the mean square error loss function, and use the optimized charging demand prediction model as a new charging demand prediction model; Demand forecasting: inputting the first real-time data into the charging demand forecasting model to obtain charging demand forecasting data of each charging pile; Grid division: Cluster the charging piles based on their location data, charging demand forecast data and regional feature data. According to the clustering results, several sub-areas are obtained. The sum of the charging demand forecast data of the charging piles in each sub-area is used as the forecast load of each sub-area. The first setting is: divide the sub-areas according to the preset grid size, divide the predicted load of each sub-area based on the area of ​​each grid, obtain the predicted load of each grid, obtain the geometric center coordinates of each grid, use the weighted centroid method to calculate the coordinates of the load center of each sub-area, and use the predicted load of each sub-area as the load capacity of the load center; Architecture construction: Based on the load capacity of the load centers in each sub-region, the load centers in each sub-region are interconnected to build the power grid architecture.

2. The method for intelligently constructing a parameterized three-dimensional power grid according to claim 1, characterized in that: After performing the first area division step and before performing the first data matching step, the method further includes: Constructing a traffic network diagram: mapping the travel habits of car owners and regional traffic flow in each first traffic area, obtaining the travel habits of car owners and regional traffic flow in each first traffic area, and constructing a traffic network diagram with each first traffic area as a node and the travel habits of car owners in each first traffic area and the corresponding regional traffic flow as an edge; Second area division: Based on the traffic network graph, a graph division algorithm is used to perform a second area division on each first traffic area, and the result of the second area division is used as the new first traffic area.

3. The method for intelligently constructing a parameterized three-dimensional power grid according to claim 1, characterized in that: After executing the step of calculating the transfer probability and before executing the step of dynamic modeling of traffic flow, the method further includes: Fourth data collection: collecting historical condition data corresponding to the first historical vehicle owner travel data and the first historical traffic flow data, the historical condition data including historical weather conditions and historical electricity prices; Defining a Bayesian network: defining Bayesian network variables and causal relationships between variables based on the historical travel habits of vehicle owners in each first traffic area and the corresponding historical condition data; the Bayesian network variables include child variables and parent variables; Construct a conditional probability table: fully arrange the parent variable values ​​of the child variables to obtain several condition combinations, calculate the probability value of each condition combination based on historical condition data and historical vehicle owner travel habits, and collect the probability values ​​of each condition combination to obtain a conditional probability table; Optimize the transfer probability: based on the historical condition data of each first traffic area, obtain the condition data corresponding to the first traffic data and the second traffic data, query the condition probability table according to the obtained condition data, obtain the condition probability, use the condition probability to optimize the comprehensive transfer probability of the first traffic area, and use the optimized comprehensive transfer probability as the new comprehensive transfer probability; After executing the step of dynamic modeling of vehicle flow, and before executing the step of obtaining vehicle flow distribution, the method further includes: Fifth, data collection: collect real-time condition data, including weather conditions and electricity prices; Data labeling: Use timestamps to match weather conditions, electricity prices, and regional traffic flows, label the regional traffic flows with weather conditions and electricity prices, and use the labeled regional traffic flows as new regional traffic flows.

4. The method for intelligently constructing a parameterized three-dimensional power grid according to claim 1, characterized in that: After executing the step of dividing the power grid and before executing the step of first setting, the method further includes: First judgment: determine whether there is a charging pile in the sub-area: If yes, then execute the second judgment step; If not, then execute the first setting step; Second judgment: judge whether the number of charging piles in the sub-area is greater than 1: If yes, the load weight of each charging pile is calculated according to the charging demand forecast data of each charging pile, the coordinates of the load center of the current sub-area are calculated by the weighted centroid method according to the positions of all charging piles in the current sub-area and the corresponding load weights, the predicted load of the current sub-area is used as the load capacity of the load center, and the step of architecture construction is executed; If not, the coordinates of the charging pile are used as the load center of the current sub-area, the predicted load of the current sub-area is used as the load capacity of the load center, and the architecture construction step is executed.

5. The method for intelligently constructing a parameterized three-dimensional power grid according to claim 1, characterized in that: After executing the step of dividing the power grid and before executing the step of first setting, the method further includes: Calculate load density: According to the geographical data of each sub-region, obtain the area of ​​each sub-region, calculate the load density of each sub-region in each period according to the area of ​​each sub-region and the predicted load of each charging pile in each sub-region in each period, aggregate the load density of each sub-region in each period, and obtain the load density characteristic vector of each sub-region; Secondary division of power grid: clustering algorithm is used to cluster each sub-region into different preliminary sub-regions using the load density feature vector of each sub-region, and the preliminary sub-regions are aggregated to obtain a preliminary region set; Constructing a regional connection graph: Based on the power grid equipment data of each region, each preliminary sub-region is used as a graph node, and the transmission lines between the preliminary sub-regions are used as edges to construct a regional connection graph; Region optimization: Based on the region connection graph, the minimum cut algorithm is used for optimization, and the optimized preliminary sub-regions are used as new sub-regions.

6. The method for intelligently constructing a parameterized three-dimensional power grid according to claim 5, characterized in that: After executing the architecture building steps, it also includes: Third judgment: obtain the actual charging load of each sub-area, calculate the difference between the actual charging load and the predicted load of each sub-area, record it as the first difference, and judge whether the first difference of the sub-area is higher than the preset first load threshold: If yes, the sub-region where the first difference is higher than the first load threshold is recorded as an overload region, and the fourth determination step is executed; If not, no action will be taken; Fourth judgment: record the absolute value of the first difference of the remaining sub-areas except the overload area as the second difference, and judge whether the second difference of the remaining sub-areas except the overload area is higher than the preset second load threshold: If yes, the sub-area where the second difference is higher than the second load threshold is recorded as an idle area, the difference between the second difference of the idle area and the preset third load threshold is calculated and recorded as the deployed load, the deployed load of the idle area is deployed to the overloaded area, and the charging power of the charging piles in the overloaded area is reduced; If not, the charging power of the charging piles in the overload area is reduced.

Images