CN-122020113-A - Prediction method considering space power load space-time two-dimensional characteristics

CN122020113ACN 122020113 ACN122020113 ACN 122020113ACN-122020113-A

Abstract

A prediction method considering space power load space-time two-dimensional characteristics belongs to the technical field of power distribution network planning. According to the method, the electric power geographic information system GIS is established, the space electric power load meshing technology is used for converting the I-type cell load data into the II-type cell load data, and then the GAT, the deep TCN and the multi-head attention mechanism are used for establishing a space load prediction model, wherein the GAT and the DeepTCN respectively acquire the space dimension characteristic and the time dimension characteristic of the II-type cell load, the multi-head attention mechanism is used for realizing deep fusion of the space-time characteristics of the II-type cell load, the space load prediction result based on the II-type cell is output through the full connection layer, the accuracy and the effectiveness of the method are verified through the calculation analysis result, the problem that the prediction model excessively depends on a certain dimension characteristic can be effectively avoided, the prediction precision is improved, and more accurate data support can be provided for the lean planning of the urban power grid.

Inventors

JIANG ZHUO
XIAO BAI
ZHANG FUSHENG

Assignees

北华大学
东北电力大学

Dates

Publication Date: 20260512
Application Date: 20251229

Claims (8)

1. The prediction method considering the space power load time-space two-dimensional characteristics is characterized by comprising the following steps of: step one, establishing a power geographic information system GIS; Step two, generating a space load space-time information matrix based on II type cells; step three, extracting space dimension characteristics of II type cell loads based on a graph attention network GAT; Extracting time dimension characteristics of class II cell load based on a deep time convolution network DeepTCN; Step five, fusing load space-time characteristics of II type cells; and step six, constructing a prediction model considering space power load time-space double-dimensional characteristics, and predicting the power load time-space double-dimensional characteristics.
2. The method for predicting space-time two-dimensional characteristics considering space power load according to claim 1, wherein the specific method for establishing the power geographic information system GIS is as follows: ① The method comprises the steps of collecting and arranging basic data information in a region to be predicted, establishing a power geographic information system GIS by utilizing the arranged basic data information, wherein the basic data information comprises satellite pictures, a street map, a grid structure, geographic positions of a power plant and a transformer substation, a specific path of a power transmission line, a 10kV feeder line power supply range, land type information and historical load data, wherein the power geographic information system GIS is used for establishing a registration base map, a land information layer and a class II cell layer of the region to be predicted, generating a space load space-time information matrix based on class II cells, extracting space dimension characteristics of class II cell loads, extracting time dimension characteristics of class II cell loads, carrying out fusion of class II cell load space-time characteristics and prediction of power load space-time double dimension characteristics, and the grid structure is a network formed by interconnection among a power supply, the power transmission line, a bus and a transformer; ② In a power geographic information system GIS, firstly, measuring the longitude and latitude values of at least 4 points above a set distance threshold in a base map, namely a satellite picture or a street map to be registered, and then establishing a registration base map of a region to be predicted by using geographic information system type information and historical load data through the satellite picture, the street map, a grid structure, geographic positions of a power plant and a transformer substation, specific paths of a power transmission line and land type information; Dividing the region to be predicted according to different set land types in the registered base map, and establishing a land information layer; dividing a region to be predicted according to a power supply range of a10 kV feeder line, establishing a class I cell layer, dividing the region to be predicted according to a square regular grid with the same size and 300m side length, and establishing a class II cell layer.
3. The method of claim 2, wherein the set different land types include residential land, commercial land, cultural entertainment land, industrial land, administrative land, greening land, municipal facility land and special land.
4. The method for predicting space-time two-dimensional characteristics considering space power load according to claim 1, wherein the specific method for generating the space-time information matrix based on class II cells in the second step is as follows: ① Converting the space power load data based on the class I cell into the space power load data based on the class II cell at a corresponding moment by adopting a space power load meshing technology; ② And taking each type II cell as a node, taking load data of each type II cell as characteristic information of the node, taking an adjacent relation among the type II cells as an edge, and generating a space load space-time information matrix based on the type II cells, wherein the matrix comprises load values and corresponding space position distribution information of all the type II cells.
5. The method for predicting space-time two-dimensional features considering space power load according to claim 1, wherein the specific method for extracting the space-dimensional features of class II cell load is as follows: ① Generating an adjacent matrix A by using a space load space-time information matrix based on II type cells; Two class II cells are regarded as two cell nodes, and the two cells have a common boundary in space distribution, so that the two class II cells are regarded as adjacent relation, an adjacent matrix A epsilon R N×N is utilized to characterize the connection relation between the class II cells, wherein R is a real set, N is the total number of the class II cells, the two cell nodes are respectively represented by a cell node v p and a cell node v q , the value a pq = 1 of the p-th row and the q-th column in the adjacent matrix A, Conversely, if the cell node v p is not adjacent to the cell node v q , a pq =0; ② Calculating an attention coefficient; the attention coefficients of the target class II cell node v i and the neighbor class II cell node v j are calculated through an attention mechanism, and the calculation formula is shown in formula (1): e ij ＝σ 1 (α T [Wh i ||Wh j ]) (1); Wherein e ij is the attention coefficient of a class II cell node v i and a neighbor class II cell node v j , sigma 1 (·) is a LeakyReLU activation function, alpha T is a leachable parameter vector, W is a weight parameter matrix, ||represents vector concatenation, h i 、h j is the characteristics of class II cell nodes v i and v j respectively, and i and j represent the ith cell and the jth cell in all class II cells respectively; ③ Normalizing the attention weight; In order to enable the attention coefficients of different nodes to have comparability, normalizing the attention coefficients of the different nodes by using a softmax function to obtain normalized attention coefficients alpha ij ; ④ Weighting and aggregating neighbor node characteristics; In order to better learn the characteristic information among the class II cells, weighting and summing the characteristics of the neighboring class II cell nodes by using a multi-head attention mechanism, respectively calculating the characteristics of the class II cell nodes by using K independently working attention mechanisms, and performing splicing operation on the results to obtain the final class II cell node characteristics, wherein the calculation formula is shown in a formula (2): Wherein h i ' is the feature of class II cell node v i after weighted aggregation, k=1, 2..K, K is the attention number, sigma 2 (·) is a sigmoid activation function, j ε L i, L i is the set of all neighbor class II cell nodes of class II cell node v i ; The method comprises the steps of determining a normalized attention coefficient of a kth attention head, wherein W k is a weight parameter matrix of the kth attention head, and h j is the characteristic of a class II cell node v j .
6. The method for predicting the space-time two-dimensional characteristics under consideration of the space electric load according to claim 1, wherein the specific method for extracting the time-dimensional characteristics of the class II cell load in the fourth step is as follows: ① Performing expansion causal convolution calculation; The deep time convolution network DeepTCN expands the receptive field through the expansion coefficient d, captures the long-term time sequence development rule of the II-type cell load, and performs time sequence expansion on the II-type cell load data to perform expansion causal convolution calculation, wherein the calculation formula is shown in a formula (3): Wherein y t (l) is a class II cell feature vector output by a layer 1 network at a t moment, m=0, 1,.. M-1, M is the convolution kernel size, the M weight parameter matrix of a layer 1 network convolution kernel of W m (l) , y t-dm (l-1) is a class II cell feature vector output by a layer 1 network at a t-dm moment, and d is an expansion coefficient; ② Calculating the time dimension characteristics of the II type cell load; The deep time convolution network DeepTCN extracts the time sequence characteristic of the II-type cell load by combining the expansion causal convolution and the jump connection, wherein a residual block is formed by the two expansion causal convolutions, the long-term dependence of the II-type cell load is captured by stacking a plurality of residual blocks, then the input data of the II-type cell load and the output characteristic of the residual block are linearly overlapped by the jump connection, further the time sequence characteristic of the II-type cell load is obtained, the jump connection calculation formula is shown as a formula (4), y′ t ＝y t +x t (4); In the formula, y' t is a time sequence feature vector of the II-type cell load output by the t-time model, y t is an output feature vector of a t-time residual block, and x t is an original feature vector of the II-type cell load at the t-time.
7. The method for predicting the space-time two-dimensional characteristics considering the space power load according to claim 1, wherein the specific method for fusing the II-type cell load space-time characteristics in the fifth step is as follows: Constructing a space-time fusion layer by using a multi-head attention mechanism, and deeply extracting the spatial features and the temporal features of class II cellular loads; Firstly, respectively carrying out linear transformation on a spatial feature matrix and a time feature matrix of class II cell load to generate Q, K, V matrixes required by all attention heads; Secondly, for each attention head, obtaining attention weight by calculating dot product of a matrix Q and a matrix K and performing scaling and softmax operation; Then, weighting and summing the matrix V by using the attention weight to obtain the output characteristic of each head; And finally, splicing the outputs of all the attention heads, and performing linear transformation to obtain a fused II-type cell load space-time characteristic matrix.
8. The prediction method considering space power load space-time two-dimensional characteristics according to any one of claims 1 to 7, wherein the specific method for constructing a prediction model considering space power load space-time two-dimensional characteristics and predicting power load space-time two-dimensional characteristics is as follows: Establishing a prediction model considering space power load space time two-dimensional characteristics by using a graph attention network GAT, a depth time convolution network DeepTCN and a multi-head attention mechanism; ① The method comprises the steps of inputting a space load space-time matrix based on II type cells as a prediction model, carrying out weighted aggregation on node characteristics of each II type cell and node characteristics of adjacent cells by using GAT to extract the space dimension characteristics of the II type cells, and extracting time sequence information of the II type cell load by stacking residual blocks and jump connection by using DeepTCN expansion causal convolution to capture long-term dependence layer by layer to extract the time dimension characteristics of the II type cell load; ② Weighting and polymerizing the space characteristics and the time sequence characteristics of the II-type cell load by utilizing a multi-head attention mechanism of the space-time fusion layer to obtain a space-time characteristic matrix of the II-type cell load; ③ And outputting a space load prediction result based on the II-type cells through a full connection layer set by the prediction model.

Description

Prediction method considering space power load space-time two-dimensional characteristics Technical Field The invention belongs to the technical field of power distribution network planning, and particularly relates to a prediction method considering space power load space-time two-dimensional characteristics. Background The space load prediction (Spatial Load Forecasting, SLF) is the prediction of the future load size and position in the prediction area, is the basis of urban power grid planning, and the SLF result can provide important references for the allocation of the capacity of the electrical equipment and the determination of the installation place, so that the planning construction of the urban power grid is more economical and reliable; Most of the existing spatial load prediction methods focus on improving data quality or optimizing a prediction model to better mine time sequence development rules of cell loads, and lack deep and full mining and analysis on spatial correlation among cell loads, so that a novel technical scheme is needed in the prior art to solve the problem. Disclosure of Invention The invention aims to solve the technical problem of providing a prediction method considering space-time two-dimensional characteristics of space power load, which is used for solving the technical problem that the space influence between II-type cell loads is insufficiently considered in the current research, and carrying out deep fusion on key space-time characteristics of the II-type cell loads on the basis, so as to avoid the excessive dependence of a prediction model on certain dimensional characteristics. The prediction method considering the space power load time-space two-dimensional characteristics comprises the following steps in sequence: step one, establishing a power geographic information system GIS; Step two, generating a space load space-time information matrix based on II type cells; step three, extracting space dimension characteristics of II type cell loads based on a graph attention network GAT; Extracting time dimension characteristics of class II cell load based on a deep time convolution network DeepTCN; Step five, fusing load space-time characteristics of II type cells; and step six, constructing a prediction model considering space power load time-space double-dimensional characteristics, and predicting the power load time-space double-dimensional characteristics. The specific method for establishing the power geographic information system GIS comprises the following steps: ① The method comprises the steps of collecting and arranging basic data information in a region to be predicted, establishing a power geographic information system GIS by utilizing the arranged basic data information, wherein the basic data information comprises satellite pictures, a street map, a grid structure, geographic positions of a power plant and a transformer substation, a specific path of a power transmission line, a 10kV feeder line power supply range, land type information and historical load data, wherein the power geographic information system GIS is used for establishing a registration base map, a land information layer and a class II cell layer of the region to be predicted, generating a space load space-time information matrix based on class II cells, extracting space dimension characteristics of class II cell loads, extracting time dimension characteristics of class II cell loads, carrying out fusion of class II cell load space-time characteristics and prediction of power load space-time double dimension characteristics, and the grid structure is a network formed by interconnection among a power supply, the power transmission line, a bus and a transformer; ② In a power geographic information system GIS, firstly, measuring the longitude and latitude values of at least 4 points above a set distance threshold in a base map, namely a satellite picture or a street map to be registered, and then establishing a registration base map of a region to be predicted by using geographic information system type information and historical load data through the satellite picture, the street map, a grid structure, geographic positions of a power plant and a transformer substation, specific paths of a power transmission line and land type information; Dividing the region to be predicted according to different set land types in the registered base map, and establishing a land information layer; dividing a region to be predicted according to a power supply range of a10 kV feeder line, establishing a class I cell layer, dividing the region to be predicted according to a square regular grid with the same size and 300m side length, and establishing a class II cell layer. The different types of sites set include residential sites, commercial sites, cultural entertainment sites, industrial sites, administrative sites, greening sites, municipal facilities and special sites. The specific method for generating the sp