A transformer area level power short-term load prediction method and system
By combining multi-source data features and spatiotemporal map data with the CASTSGCN model, the problem of neglecting spatial correlation in substation-level load forecasting is solved, thus improving forecast accuracy. It is suitable for load forecasting of low-voltage distribution substations.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA ELECTRIC POWER RESEARCH INSTITUTE CO LTD
- Filing Date
- 2022-11-30
- Publication Date
- 2026-05-19
AI Technical Summary
Existing transformer area-level load forecasting methods neglect the potential spatial correlation between users' electricity consumption behaviors within the transformer area, making it difficult to improve forecast accuracy.
The Conditional Adaptive Spatiotemporal Synchronous Graph Convolutional Neural Network (CASTSGCN) model is adopted. By combining multi-source conditional data features and spatiotemporal graph data, the temporal and spatial correlations of user electricity consumption behavior within the distribution area are explored. The load patterns of low-voltage distribution areas are then trained and predicted.
It significantly improves the accuracy of short-term load forecasting in low-voltage distribution substations, and is applicable to low-voltage distribution substations with spatiotemporal correlation, thus having significant application value and promising prospects for promotion.
Smart Images
Figure CN115730740B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of power system load forecasting technology, and particularly relates to a method and system for short-term power load forecasting at the distribution area level. Background Technology
[0002] Currently, short-term load forecasting research mainly focuses on provincial or municipal regional power grids. However, load forecasting focusing on low-voltage distribution transformer areas faces challenges such as small power supply areas, large load fluctuations, and insufficient stability of forecast results. Therefore, traditional forecasting methods based on coarse-grained data are difficult to apply to distribution transformer-level load forecasting. With the widespread adoption of online monitoring equipment for distribution transformers, the problem of collecting and storing massive amounts of distribution transformer load data has been solved, thus providing a solid data foundation for distribution transformer-level load forecasting based on fine-grained data.
[0003] Existing transformer substation-level load forecasting methods mainly rely on algorithms such as random forests, support vector machines, artificial neural networks, and long short-term memory neural networks to construct forecasting models. However, these methods focus on uncovering the temporal correlation of electricity load sequences within a transformer substation, while neglecting the potential spatial correlation between user electricity consumption behaviors within the same substation (for example, users within the same substation share the same geographical space, weather conditions, holiday information, electricity pricing policies, and other comprehensive factors). Consequently, to some extent, this makes it difficult to further improve the accuracy of transformer substation-level load forecasting. Summary of the Invention
[0004] The main objective of this invention is to overcome the shortcomings and deficiencies of the prior art and provide a method and system for short-term power load forecasting at the distribution substation level. This effectively solves the problem that current low-voltage distribution substation load forecasting ignores the potential spatial correlation between the electricity consumption behaviors of users within the substation, and significantly improves the accuracy of low-voltage distribution substation load forecasting.
[0005] To achieve the above objectives, the present invention adopts the following technical solution:
[0006] According to one aspect of the present invention, a method for short-term power load forecasting at the distribution area level is provided, the method comprising the following steps:
[0007] Based on the three-phase load sequence of low-voltage distribution transformers, characteristic sequences and adjacency matrices of corresponding nodes are established, and spatiotemporal map data and graph data samples for load prediction of low-voltage distribution transformer areas are constructed.
[0008] Based on multi-source condition data, corresponding multi-source condition data features are constructed. The multi-source condition data includes week type, time information, holidays, temperature, humidity, wind speed, and air pressure.
[0009] Based on the constructed spatiotemporal graph data and the features of the multi-source conditional data, a conditionally adaptive spatiotemporal synchronization graph convolutional neural network model (CASTSGCN) is trained to mine the load patterns of low-voltage distribution substations.
[0010] The trained CASTSGCN model is used to predict the three-phase load of the low-voltage distribution transformer, and the predicted load values of the three phases of the distribution transformer at the next time moment are obtained respectively.
[0011] By aggregating the load forecast values of the three phases of the distribution transformer, the total load forecast value of the low-voltage distribution transformer area at the next moment is obtained.
[0012] Preferably, the step of establishing the characteristic sequence and adjacency matrix of the corresponding nodes based on the three-phase load sequence of the low-voltage distribution transformer, and constructing spatiotemporal map data and map data samples for load prediction of low-voltage distribution transformer areas includes:
[0013] The three-phase load sequence is used as the feature sequence of each node in the graph structure data, and the correlation coefficient between the phase load sequences is calculated.
[0014] Construct an adjacency matrix A based on the correlation coefficient, where...
[0015] Or A ij =ρ ij
[0016]
[0017] The graph data sample is:
[0018] in, For sample input, For sample output, Let A be the graph signal matrix composed of node features at time t. ij Let ξ be the element in the i-th row and j-th column of the adjacency matrix A, and let ρ be the threshold. ij Represents the load sequence X i and X j Pearson correlation coefficient, COV(X) i ,X j ) represents X i and X j covariance, and They represent X respectively i and X j The standard deviation.
[0019] Preferably, the construction of corresponding multi-source conditional data features based on multi-source conditional data includes:
[0020] Construct the following sequences: weekday type sequence W, time index sequence D, holiday marker sequence H, temperature feature sequence E, humidity feature sequence M, wind speed feature sequence P, and air pressure feature sequence Q. For time t, the conditional data feature sequence L consists of these seven sequences, represented as follows:
[0021] L = [W,D,H,E,M,P,Q].
[0022] Preferably, the training condition adaptive spatiotemporal synchronization graph convolutional neural network CASTSGCN model includes:
[0023] The parameters of the CASTSGCN model are set, including: historical load sequence input length, conditional data feature dimension, input transformation layer dimension, spatiotemporal embedding layer dimension, number of graph convolutional layers, sliding window length, graph convolutional layer activation function, output mapping layer dimension, output mapping layer activation function, learning rate, loss function, learning decay rate, batch size, and training period.
[0024] Preferably, the process of aggregating the load forecast values of the three phases of the distribution transformer to obtain the total load forecast value of the low-voltage distribution transformer area at the next moment includes:
[0025] The formula for calculating the total load forecast is as follows:
[0026]
[0027] In the formula, P next This represents the total load forecast for the low-voltage distribution substation area. These are the predicted load values for phases A, B, and C of the low-voltage distribution transformer at the next future moment, as predicted by the CASTSGCN model.
[0028] According to another aspect of the present invention, the present invention also provides a distribution area-level short-term power load forecasting system, the system comprising:
[0029] The module is used to build the characteristic sequence and adjacency matrix of the corresponding nodes based on the three-phase load sequence of the low-voltage distribution transformer, and to construct the spatiotemporal map data and map data samples for load prediction of low-voltage distribution transformer areas.
[0030] Based on multi-source condition data, corresponding multi-source condition data features are constructed. The multi-source condition data includes week type, time information, holidays, temperature, humidity, wind speed, and air pressure.
[0031] The training module is used to train the conditionally adaptive spatiotemporal synchronization graph convolutional neural network (CASTSGCN) model based on the constructed spatiotemporal graph data and the features of the multi-source conditional data, in order to mine the load patterns of low-voltage distribution substations.
[0032] The prediction module is used to apply the trained CASTSGCN model to predict the three-phase load of the low-voltage distribution transformer and obtain the predicted load values of the three phases of the distribution transformer at the next moment.
[0033] The aggregation module is used to aggregate the load forecast values of the three phases of the distribution transformer to obtain the total load forecast value of the low-voltage distribution transformer area at the next moment.
[0034] Preferably, the construction module establishes the characteristic sequence and adjacency matrix of the corresponding nodes based on the three-phase load sequence of the low-voltage distribution transformer, and constructs spatiotemporal map data and map data samples for load prediction of low-voltage distribution transformer areas, including:
[0035] The three-phase load sequence is used as the feature sequence of each node in the graph structure data, and the correlation coefficient between the phase load sequences is calculated.
[0036] Construct an adjacency matrix A based on the correlation coefficient, where...
[0037] Or A ij =ρ ij
[0038]
[0039] The graph data sample is:
[0040] in, For sample input, For sample output, Let A be the graph signal matrix composed of node features at time t. ij Let ξ be the element in the i-th row and j-th column of the adjacency matrix A, and let ρ be the threshold. ij Represents the load sequence X i and X j Pearson correlation coefficient, COV(X) i ,X j ) represents X i and X j covariance, and They represent X respectively i and X j The standard deviation.
[0041] Preferably, the construction module constructs corresponding multi-source conditional data features based on multi-source conditional data, including:
[0042] Construct the following sequences: weekday type sequence W, time index sequence D, holiday marker sequence H, temperature feature sequence E, humidity feature sequence M, wind speed feature sequence P, and air pressure feature sequence Q. For time t, the conditional data feature sequence L consists of these seven sequences, represented as follows:
[0043] L = [W,D,H,E,M,P,Q].
[0044] Preferably, the training module trains the conditionally adaptive spatiotemporal synchronization graph convolutional neural network (CASTSGCN) model, which includes:
[0045] The parameters of the CASTSGCN model are set, including: historical load sequence input length, conditional data feature dimension, input transformation layer dimension, spatiotemporal embedding layer dimension, number of graph convolutional layers, sliding window length, graph convolutional layer activation function, output mapping layer dimension, output mapping layer activation function, learning rate, loss function, learning decay rate, batch size, and training period.
[0046] Preferably, the aggregation module aggregates the load forecast values of the three phases of the distribution transformer to obtain the total load forecast value of the low-voltage distribution transformer area at the next moment, including:
[0047] The formula for calculating the total load forecast is as follows:
[0048]
[0049] In the formula, P next This represents the total load forecast for the low-voltage distribution substation area. These are the predicted load values for phases A, B, and C of the low-voltage distribution transformer at the next future moment, as predicted by the CASTSGCN model.
[0050] Beneficial Effects: This invention utilizes graph neural networks to not only uncover the temporal correlation between historical load sequences of distribution transformer areas and the spatial correlation between user electricity consumption behavior within those areas, but also to explore the impact of multiple external factors such as weekday type, time information, and weather on the load of distribution transformer areas. This significantly improves the accuracy of short-term load forecasting for low-voltage distribution transformer areas. Therefore, this invention is suitable for short-term load forecasting of low-voltage distribution transformer areas with strong spatiotemporal correlations. In the context of implementing unit-based systems, refined planning, and operation and maintenance of distribution networks, using low-voltage distribution transformer areas as the research object for load forecasting has significant application value and promising prospects for business scenarios such as gridded load development trend and planning demand analysis, identification of weak links in distribution networks, and adjustment of power system operation modes.
[0051] The features and advantages of the present invention will become clear from the following accompanying drawings and a detailed description of specific embodiments thereof. Attached Figure Description
[0052] The accompanying drawings, which form part of this specification, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:
[0053] Figure 1 This is a flowchart of the method for short-term power load forecasting at the substation level;
[0054] Figure 2 This is a schematic diagram of the adjacency matrix and spatiotemporal graph structure;
[0055] Figure 3 This is a schematic diagram of the spatiotemporal map data structure for load forecasting of low-voltage distribution substations;
[0056] Figure 4 This is a schematic diagram of a conditionally adaptive spatiotemporal synchronization graph convolutional neural network structure;
[0057] Figure 5 This is a schematic diagram of a district-level short-term power load forecasting system. Detailed Implementation
[0058] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.
[0059] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this invention is for describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.
[0060] Example 1
[0061] Figure 1 This is a flowchart of the short-term load forecasting method for power distribution zones. (For example...) Figure 1 As shown, this embodiment provides a method for short-term power load forecasting at the distribution area level. The method includes the following steps:
[0062] S1: Based on the three-phase load sequence of the low-voltage distribution transformer, establish the characteristic sequence and adjacency matrix of the corresponding nodes, and construct the spatiotemporal map data and map data samples for load prediction of low-voltage distribution transformer areas.
[0063] Preferably, the step of establishing the characteristic sequence and adjacency matrix of the corresponding nodes based on the three-phase load sequence of the low-voltage distribution transformer, and constructing spatiotemporal map data and map data samples for load prediction of low-voltage distribution transformer areas includes:
[0064] The three-phase load sequence is used as the feature sequence of each node in the graph structure data, and the correlation coefficient between the phase load sequences is calculated.
[0065] Construct an adjacency matrix A based on the correlation coefficient, where...
[0066] Or A ij =ρ ij
[0067]
[0068] The graph data sample is:
[0069] in, For sample input, For sample output, Let A be the graph signal matrix composed of node features at time t. ij Let ξ be the element in the i-th row and j-th column of the adjacency matrix A, and let ρ be the threshold. ij Represents the load sequence X i and X j Pearson correlation coefficient, COV(X) i ,X j ) represents X i and X j covariance, and They represent X respectively i and X j The standard deviation.
[0070] Specifically, see Figure 2 First, the three-phase load sequence is used as the feature sequence of each node in the graph structure data. Then, the correlation coefficient between the load sequences of each phase is calculated as follows:
[0071]
[0072] In the formula: ρ ij Represents the load sequence X i and X j Pearson correlation coefficient, COV(X) i ,X j ) represents X i and X j covariance, and They represent X respectively i and X j The standard deviation.
[0073] The first method for constructing the adjacency matrix is as follows: If the correlation coefficient between the corresponding historical load sequences of two nodes is not less than a certain threshold, then a connection is considered to exist between the two nodes, and the corresponding element in adjacency matrix A is set to 1; otherwise, no connection is considered to exist between the two nodes, and the corresponding element in adjacency matrix A is set to 0. The specific calculation formula is as follows:
[0074]
[0075] In the formula, A ij Let ξ be the element in the i-th row and j-th column of the adjacency matrix A, and let ξ be the threshold.
[0076] The second method for constructing the adjacency matrix is as follows: The correlation coefficient between the corresponding historical load sequences of two nodes is used as the corresponding element of the adjacency matrix. The specific calculation formula is as follows:
[0077] A ij =ρ ij (3)
[0078] (2) Based on the three-phase load sequence and the corresponding adjacency matrix, a spatiotemporal map data for low-voltage distribution area load prediction is jointly constructed.
[0079] (3) Constructing graphical data samples for load forecasting of low-voltage distribution substations. in, For sample input, For sample output, Let A be the graph signal matrix composed of node features at time t, where A is the adjacency matrix and T is the length of the history sequence. The spatiotemporal graph data structure for load forecasting of low-voltage distribution substations is as follows: Figure 3 As shown, the characteristic of each node is the load value at the corresponding time.
[0080] S2: Based on multi-source condition data, construct corresponding multi-source condition data features, including weekday type, time information, holidays, temperature, humidity, wind speed, and air pressure.
[0081] Preferably, the construction of corresponding multi-source conditional data features based on multi-source conditional data includes:
[0082] Construct the following sequences: weekday type sequence W, time index sequence D, holiday marker sequence H, temperature feature sequence E, humidity feature sequence M, wind speed feature sequence P, and air pressure feature sequence Q. For time t, the conditional data feature sequence L consists of these seven sequences, represented as follows:
[0083] L = [W,D,H,E,M,P,Q].
[0084] Specifically, for the graph data sample at time t, corresponding conditional data features are constructed, including seven multi-source data features such as week type, time information, holiday information, temperature, humidity, wind speed, and air pressure, as shown below:
[0085] 1) For the past T time points, the weekday sequence W is represented as:
[0086] W = [w t-T+1 ,...,w t-1 ,wt (4)
[0087] Among them, w t ∈[1,7] represents the weekday type of time t;
[0088] 2) For the past T time points, the time index sequence D is represented as:
[0089] D = [d t-T+1 ,...,d t-1 ,d t (5)
[0090] Where, d t ∈[1,F] is the time index of time t, and F is the sampling frequency of the load data, which is generally 24, 48 or 96;
[0091] 3) For the past T time points, the holiday marker sequence H is represented as:
[0092] H = [h] t-T+1 ,...,h t-1 ,h t (6)
[0093] Among them, h t Let t be the holiday marker for time t, with a value of 1 or 2, where 1 indicates a non-holiday and 2 indicates a holiday.
[0094] 4) For the past T time points, the temperature feature sequence E is represented as:
[0095] E = [e t-T+1 ,...,e t-1 ,e t (7)
[0096] Among them, e t The temperature at time t;
[0097] 5) For the past T time points, the humidity characteristic sequence M is represented as:
[0098] M = [m t-T+1 ,...,m t-1 ,m t (8)
[0099] Where, m t Let be the humidity at time t;
[0100] 6) For the past T time points, the wind speed characteristic sequence P is represented as:
[0101] P = [p] t-T+1 ,...,p t-1 ,p t (9)
[0102] Where, p t Let be the wind speed at time t;
[0103] 7) For the past T time points, the air pressure characteristic sequence Q is represented as:
[0104] Q = [q] t-T+1 ,...,q t-1 ,q t (10)
[0105] Where, q t Let be the air pressure at time t;
[0106] Therefore, for time t, the conditional data feature sequence L consists of the above 7 sequences, represented as:
[0107] L=[W,D,H,E,M,P,Q] (11)
[0108] S3: Based on the constructed spatiotemporal graph data and the features of the multi-source conditional data, train the conditionally adaptive spatiotemporal synchronization graph convolutional neural network model CASTSGCN to mine the load patterns of low-voltage distribution substations.
[0109] Preferably, the training condition adaptive spatiotemporal synchronization graph convolutional neural network CASTSGCN model includes:
[0110] The parameters of the CASTSGCN model are set, including: historical load sequence input length, conditional data feature dimension, input transformation layer dimension, spatiotemporal embedding layer dimension, number of graph convolutional layers, sliding window length, graph convolutional layer activation function, output mapping layer dimension, output mapping layer activation function, learning rate, loss function, learning decay rate, batch size, and training period.
[0111] Specifically, the CASTSGCN model is trained based on the graph data samples constructed in steps 1 and 2 and the corresponding conditional data features. The CASTSGCN model structure is as follows: Figure 4 As shown in Table 1, the corresponding parameter settings are as follows.
[0112] Table 1. Parameters of the CASTSGCN Model
[0113]
[0114] S4: Using the trained CASTSGCN model, predict the three-phase load of the low-voltage distribution transformer and obtain the predicted load values of the three phases of the distribution transformer at the next time step.
[0115] Specifically, the trained CASTSGCN model is applied to predict the three-phase load of the low-voltage distribution transformer, yielding... in, These are the predicted load values for phases A, B, and C of the low-voltage distribution transformer at the next moment.
[0116] S5: Aggregate the load forecast values of the three phases of the distribution transformer to obtain the total load forecast value of the low-voltage distribution transformer area at the next moment.
[0117] Preferably, the process of aggregating the load forecast values of the three phases of the distribution transformer to obtain the total load forecast value of the low-voltage distribution transformer area at the next moment includes:
[0118] The formula for calculating the total load forecast is as follows:
[0119]
[0120] In the formula, P next This represents the total load forecast for the low-voltage distribution substation area. These are the predicted load values for phases A, B, and C of the low-voltage distribution transformer at the next future moment, as predicted by the CASTSGCN model.
[0121] Specifically, the load forecast values of the three phases of the low-voltage distribution transformer are aggregated to obtain the total load forecast value of the low-voltage distribution transformer area at the next future moment. The specific calculation formula is as follows:
[0122]
[0123] In the formula, P next This represents the total load forecast for the low-voltage distribution area.
[0124] This embodiment utilizes graph neural networks to not only uncover the temporal correlation between historical load sequences of distribution transformer areas and the spatial correlation between user electricity consumption behavior within those areas, but also to explore the impact of multiple external factors such as weekday type, time information, and weather on the load of distribution transformer areas. This significantly improves the accuracy of short-term load forecasting for low-voltage distribution transformer areas. Therefore, this embodiment is suitable for short-term load forecasting of low-voltage distribution transformer areas with strong spatiotemporal correlations. In the context of implementing unit-based systems, refined planning, and operation and maintenance of distribution networks, using low-voltage distribution transformer areas as the research object for load forecasting has significant application value and promising prospects for business scenarios such as gridded load development trend and planning demand analysis, identification of weak links in distribution networks, and adjustment of power system operation modes.
[0125] Example 2
[0126] Figure 5 This is a schematic diagram of a district-level short-term power load forecasting system. (For example...) Figure 5 As shown, this embodiment provides a distribution area-level short-term power load forecasting system, the system comprising:
[0127] Module 501 is used to build the characteristic sequence and adjacency matrix of the corresponding nodes based on the three-phase load sequence of the low-voltage distribution transformer, and to build spatiotemporal map data and map data samples for load prediction of low-voltage distribution transformer areas.
[0128] Based on multi-source condition data, corresponding multi-source condition data features are constructed. The multi-source condition data includes week type, time information, holidays, temperature, humidity, wind speed, and air pressure.
[0129] Training module 502 is used to train the conditionally adaptive spatiotemporal synchronization graph convolutional neural network CASTSGCN model based on the constructed spatiotemporal graph data and the features of the multi-source conditional data, and to mine the load patterns of low-voltage distribution substations.
[0130] The prediction module 503 is used to apply the trained CASTSGCN model to predict the three-phase load of the low-voltage distribution transformer and obtain the predicted load values of the three phases of the distribution transformer at the next moment.
[0131] The aggregation module 504 is used to aggregate the load forecast values of the three phases of the distribution transformer to obtain the total load forecast value of the low-voltage distribution transformer area at the next moment.
[0132] Preferably, the construction module 501 establishes the characteristic sequence and adjacency matrix of the corresponding nodes based on the three-phase load sequence of the low-voltage distribution transformer, and constructs spatiotemporal map data and map data samples for low-voltage distribution transformer area load prediction, including:
[0133] The three-phase load sequence is used as the feature sequence of each node in the graph structure data, and the correlation coefficient between the phase load sequences is calculated.
[0134] Construct an adjacency matrix A based on the correlation coefficient, where...
[0135] Or A ij =ρ ij
[0136]
[0137] The graph data sample is:
[0138] in, For sample input, For sample output, Let A be the graph signal matrix composed of node features at time t. ij Let ξ be the element in the i-th row and j-th column of the adjacency matrix A, and let ρ be the threshold. ij Represents the load sequence X i and X j Pearson correlation coefficient, COV(X) i ,Xj ) represents X i and X j covariance, and They represent X respectively i and X j The standard deviation.
[0139] Preferably, the construction module 501 constructs corresponding multi-source conditional data features based on multi-source conditional data, including:
[0140] Construct the following sequences: weekday type sequence W, time index sequence D, holiday marker sequence H, temperature feature sequence E, humidity feature sequence M, wind speed feature sequence P, and air pressure feature sequence Q. For time t, the conditional data feature sequence L consists of these seven sequences, represented as follows:
[0141] L = [W,D,H,E,M,P,Q].
[0142] Preferably, the training module 502 trains the conditionally adaptive spatiotemporal synchronization graph convolutional neural network CASTSGCN model, which includes:
[0143] The parameters of the CASTSGCN model are set, including: historical load sequence input length, conditional data feature dimension, input transformation layer dimension, spatiotemporal embedding layer dimension, number of graph convolutional layers, sliding window length, graph convolutional layer activation function, output mapping layer dimension, output mapping layer activation function, learning rate, loss function, learning decay rate, batch size, and training period.
[0144] Preferably, the aggregation module 504 aggregates the load forecast values of the three phases of the distribution transformer to obtain the total load forecast value of the low-voltage distribution transformer area at the next moment, including:
[0145] The formula for calculating the total load forecast is as follows:
[0146]
[0147] In the formula, P next This represents the total load forecast for the low-voltage distribution substation area. These are the predicted load values for phases A, B, and C of the low-voltage distribution transformer at the next future moment, as predicted by the CASTSGCN model.
[0148] The specific implementation process of the functions implemented by each module in this embodiment 2 is the same as the implementation process of each step in embodiment 1, and will not be repeated here.
[0149] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0150] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0151] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0152] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A method for short-term power load forecasting at the distribution area level, characterized in that, The method includes the following steps: Based on the three-phase load sequence of low-voltage distribution transformers, characteristic sequences and adjacency matrices of corresponding nodes are established, and spatiotemporal map data and graph data samples for load prediction of low-voltage distribution transformer areas are constructed. Based on multi-source condition data, corresponding multi-source condition data features are constructed. The multi-source condition data includes week type, time information, holidays, temperature, humidity, wind speed, and air pressure. Based on the constructed spatiotemporal graph data and the features of the multi-source conditional data, a conditionally adaptive spatiotemporal synchronization graph convolutional neural network model (CASTSGCN) is trained to mine the load patterns of low-voltage distribution substations. The trained CASTSGCN model is used to predict the three-phase load of the low-voltage distribution transformer, and the predicted load values of the three phases of the distribution transformer at the next time moment are obtained respectively. By aggregating the load forecast values of the three phases of the distribution transformer, the total load forecast value of the low-voltage distribution transformer area at the next moment is obtained; The process of establishing characteristic sequences and adjacency matrices for corresponding nodes based on the three-phase load sequences of low-voltage distribution transformers, and constructing spatiotemporal map data and map data samples for low-voltage distribution transformer area load prediction includes: The three-phase load sequence is used as the feature sequence of each node in the graph structure data, and the correlation coefficient between the phase load sequences is calculated. Construct an adjacency matrix A based on the correlation coefficient, where... ,or The graph data sample is: , in, For sample input, For sample output, for t A graph signal matrix composed of node features at any given time. A ij The adjacency matrix A is the first... i Line 1 j Column elements, For the threshold, Represents load sequence and Pearson correlation coefficient, express and covariance, and They represent and Standard deviation; The training condition adaptive spatiotemporal synchronization graph convolutional neural network CATSGCN model includes: The parameters of the CASTSGCN model are set, including: historical load sequence input length, conditional data feature dimension, input transformation layer dimension, spatiotemporal embedding layer dimension, number of graph convolutional layers, sliding window length, graph convolutional layer activation function, output mapping layer dimension, output mapping layer activation function, learning rate, loss function, learning decay rate, batch size, and training period. The aggregation of the load forecast values of the three phases of the distribution transformer to obtain the total load forecast value of the low-voltage distribution transformer area at the next moment includes: The formula for calculating the total load forecast is as follows: In the formula, This represents the total load forecast for the low-voltage distribution substation area. , , These are the predicted load values for phases A, B, and C of the low-voltage distribution transformer at the next future moment, as predicted by the CASTSGCN model.
2. The method according to claim 1, characterized in that, The construction of corresponding multi-source conditional data features based on multi-source conditional data includes: Construct weekday type sequences respectively Time index sequence Holiday marker sequence Temperature characteristic sequence E, humidity characteristic sequence M, wind speed characteristic sequence P, and air pressure characteristic sequence Q, for t At time t, the conditional data feature sequence L consists of the above 7 sequences, represented as: 。 3. A distribution area-level short-term power load forecasting system, characterized in that, The system includes: The module is used to build the characteristic sequence and adjacency matrix of the corresponding nodes based on the three-phase load sequence of the low-voltage distribution transformer, and to construct the spatiotemporal map data and map data samples for load prediction of low-voltage distribution transformer areas. Based on multi-source condition data, corresponding multi-source condition data features are constructed. The multi-source condition data includes week type, time information, holidays, temperature, humidity, wind speed, and air pressure. The training module is used to train the conditionally adaptive spatiotemporal synchronization graph convolutional neural network (CASTSGCN) model based on the constructed spatiotemporal graph data and the features of the multi-source conditional data, in order to mine the load patterns of low-voltage distribution substations. The prediction module is used to apply the trained CASTSGCN model to predict the three-phase load of the low-voltage distribution transformer and obtain the predicted load values of the three phases of the distribution transformer at the next moment. The aggregation module is used to aggregate the load forecast values of the three phases of the distribution transformer to obtain the total load forecast value of the low-voltage distribution transformer area at the next moment. The construction module establishes the characteristic sequence and adjacency matrix of corresponding nodes based on the three-phase load sequence of the low-voltage distribution transformer, and constructs spatiotemporal map data and map data samples for low-voltage distribution transformer area load prediction, including: The three-phase load sequence is used as the feature sequence of each node in the graph structure data, and the correlation coefficient between the phase load sequences is calculated. Construct an adjacency matrix A based on the correlation coefficient, where... ,or The graph data sample is: , in, For sample input, For sample output, for t A graph signal matrix composed of node features at any given time. A ij The adjacency matrix A is the first... i Line 1 j Column elements, For the threshold, Represents load sequence and Pearson correlation coefficient, express and covariance, and They represent and Standard deviation; The training module trains the CASTSGCN model of an adaptive spatiotemporal synchronization graph convolutional neural network, including: The parameters of the CASTSGCN model are set, including: historical load sequence input length, conditional data feature dimension, input transformation layer dimension, spatiotemporal embedding layer dimension, number of graph convolutional layers, sliding window length, graph convolutional layer activation function, output mapping layer dimension, output mapping layer activation function, learning rate, loss function, learning decay rate, batch size, and training period. The aggregation module aggregates the load forecast values of the three phases of the distribution transformer to obtain the total load forecast value of the low-voltage distribution transformer area at the next moment, including: The formula for calculating the total load forecast is as follows: In the formula, This represents the total load forecast for the low-voltage distribution substation area. , , These are the predicted load values for phases A, B, and C of the low-voltage distribution transformer at the next future moment, as predicted by the CASTSGCN model.
4. The system according to claim 3, characterized in that, The construction module, based on multi-source conditional data, constructs corresponding multi-source conditional data features, including: Construct weekday type sequences respectively Time index sequence Holiday marker sequence Temperature characteristic sequence E, humidity characteristic sequence M, wind speed characteristic sequence P, and air pressure characteristic sequence Q, for t At time t, the conditional data feature sequence L consists of the above 7 sequences, represented as: 。