CN-122026338-A - Space-time fusion photovoltaic power generation prediction method, system, equipment and medium

CN122026338ACN 122026338 ACN122026338 ACN 122026338ACN-122026338-A

Abstract

The invention discloses a space-time fusion photovoltaic power generation prediction method, a system, equipment and a medium, and relates to the technical field of power systems, wherein the method comprises the steps of performing space-time and time conversion on acquired photovoltaic power generation data, generating fusion embedded representation and performing block processing to obtain block embedded representation; the method comprises the steps of carrying out conversion on a blocking embedded representation input space-time attention layer to obtain target characterization vectors of all nodes, outputting initial power generation prediction results, calculating Euclidean distances between all target nodes and all cluster centers of the fused space-time embedded representation and carrying out dynamic tuning to determine the cluster center corresponding to each target node, aggregating all nodes in the cluster centers to obtain cluster time sequence representation, coordinating the cluster time sequence representation and the time sequence characterization vectors, distributing weights for all prediction experts to output the target power generation prediction results, determining an optimal power generation plan of a target power system according to the target power generation prediction results, and improving the reliability of scheduling.

Inventors

FENG JUN
WANG YIDAN
Deng Hanzhi
CHEN WEI
WANG TENGJIAO
ZHU XIAOJIE
GU WEI
WANG LUMIN
Lou Zenan

Assignees

国网浙江省电力有限公司信息通信分公司
国网浙江省电力有限公司
北京大学(青岛)计算社会科学研究院

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (10)

1. A space-time fusion photovoltaic power generation prediction method is characterized by comprising the following steps: performing space-time and time conversion on the acquired target photovoltaic power generation data, generating a fusion embedded representation, and performing block processing to obtain a block embedded representation; the block embedding representation is input into a preset space-time attention layer for transformation, so that a target characterization vector corresponding to each node is obtained, and an initial power generation prediction result is output; Extracting space-time embedded representation from the fusion embedded representation, calculating Euclidean distance between each target node corresponding to the space-time embedded representation and each cluster center in the pre-constructed cluster representation, and performing dynamic tuning to determine the cluster center corresponding to each target node; Aggregating all nodes in the cluster centers to obtain cluster time sequence representations corresponding to the cluster centers; assigning weights to each prediction expert constructed by the initial power generation prediction result in cooperation with the cluster time sequence representation and the corresponding target characterization vector so as to output a target power generation prediction result; And determining an optimal power generation plan of the target power system according to the target power generation prediction result.
2. The method for predicting the photovoltaic power generation by space-time fusion according to claim 1, wherein the performing space-time and time conversion on the obtained target photovoltaic power generation data to generate a fusion embedded representation and performing block processing to obtain a block embedded representation comprises: Converting the target photovoltaic power generation data from the space-time characteristic angle to generate the space-time embedded representation corresponding to each node; Converting the target photovoltaic power generation data from the aspect of time characteristics to generate corresponding time embedded representation; Integrating the space-time embedded representation and the time embedded representation to obtain the fusion embedded representation; And carrying out blocking processing on the fusion embedded representation along the time dimension to obtain the blocking embedded representation.
3. The method for predicting the photovoltaic power generation by using space-time fusion according to claim 1, wherein the step of transforming the block embedded representation input preset space-time attention layer to obtain the target characterization vector corresponding to each node so as to output the initial power generation prediction result comprises the following steps: the block embedding representation is input into a preset time sequence attention layer to perform first transformation to obtain time sequence representation vectors corresponding to all nodes, the time sequence representation vectors are subjected to first linear transformation, and corresponding time sequence power generation prediction results are output; inputting the time sequence representation vector into a preset space attention layer for second transformation to obtain a space-time representation vector corresponding to each node, performing second linear transformation on the space-time representation vector, and outputting to obtain a corresponding space-time power generation prediction result; And integrating the time sequence power generation prediction result and the space-time power generation prediction result to obtain the initial power generation prediction result.
4. The method for predicting the photovoltaic power generation by space-time fusion according to claim 1, wherein the dynamic tuning process comprises: performing cluster-like allocation on each target node according to the Euclidean distance, and designing a first loss function according to the Euclidean distance; and dynamically optimizing a class cluster distribution process by using the first loss function to obtain the class cluster center corresponding to each target node.
5. The method of claim 3, wherein the assigning weights to the prediction experts constructed from the initial power generation prediction result to output a target power generation prediction result in cooperation with the cluster-like time sequence representation and the corresponding target characterization vector comprises: designing a time sequence prediction expert according to the time sequence power generation prediction result, and designing a space-time prediction expert according to the space-time power generation prediction result; And inputting the time sequence representation of the class cluster and the time sequence representation vector as comprehensive decision indexes into a special control network for training, distributing corresponding expert weights for the time sequence prediction expert and the space-time prediction expert, carrying out weighted fusion, and outputting the target power generation prediction result.
6. The method of claim 5, wherein the assigning weights to the prediction experts constructed from the initial power generation prediction result to output a target power generation prediction result in cooperation with the cluster-like time sequence representation and the corresponding target characterization vector, further comprises: And in the training process, introducing a second pre-designed loss function to adjust and update the expert gating network parameters.
7. The method for predicting power generation by using space-time fusion according to claim 1, wherein determining an optimal power generation plan of a target power system according to the target power generation prediction result comprises: Taking the minimum total power generation loss as a target, taking the output of target photovoltaic power generation equipment as a decision variable, and constructing a tide calculation model; and solving the tide calculation model by taking the target power generation prediction result as a boundary condition, and outputting to obtain the optimal power generation plan corresponding to the target photovoltaic power generation equipment.
8. A space-time fusion photovoltaic power generation prediction system, comprising: the embedded generation module is used for carrying out space-time and time conversion on the obtained target photovoltaic power generation data, generating a fusion embedded representation and carrying out block processing to obtain a block embedded representation; The initial prediction module is used for embedding the blocks into a space-time attention layer which is input into a preset representation for transformation to obtain target characterization vectors corresponding to all nodes so as to output an initial power generation prediction result; The class cluster distribution module is used for extracting space-time embedded representations from the fusion embedded representations, calculating Euclidean distances between each target node corresponding to the space-time embedded representations and each class cluster center in the pre-constructed class cluster representations, and dynamically optimizing to determine the class cluster center corresponding to each target node; The aggregation module is used for aggregating all the nodes in the cluster centers to obtain cluster time sequence representations corresponding to the cluster centers; the prediction output module is used for allocating weights to each prediction expert constructed by the initial power generation prediction result in cooperation with the cluster time sequence representation and the corresponding target characterization vector so as to output a target power generation prediction result; And the power generation plan generation module is used for determining an optimal power generation plan of the target power system according to the target power generation prediction result.
9. A computer device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, the processor implementing the spatiotemporal fusion photovoltaic power generation prediction method of any of claims 1 to 7 when the computer program is executed.
10. A computer readable storage medium, wherein the computer readable storage medium stores a computer program, and wherein the apparatus in which the computer readable storage medium is located implements the spatiotemporal fusion photovoltaic power generation prediction method according to any one of claims 1 to 7 when the computer program is executed.

Description

Space-time fusion photovoltaic power generation prediction method, system, equipment and medium Technical Field The invention relates to the technical field of power systems, in particular to a space-time fusion photovoltaic power generation prediction method, a system, equipment and a medium. Background At present, the distributed photovoltaic power generation has increasingly large duty ratio, and how to accurately predict the photovoltaic power generation amount and the load in a power grid, so that the accurate scheduling of the power system is a key for guaranteeing the safe and stable operation of the power system. However, the existing means for dispatching the power system based on the photovoltaic power generation and the load prediction are often focused on single time sequence dependency modeling, so that the accuracy of the photovoltaic power generation and the load prediction is limited, the dispatching quality is reduced, and the stable operation of the power system is affected. Therefore, how to effectively predict photovoltaic power generation and ensure the scheduling quality of a power system is a technical problem to be solved by those skilled in the art. Disclosure of Invention The invention provides a space-time fusion photovoltaic power generation prediction method, a system, equipment and a medium, which solve the problem of how to guide a prediction expert through cluster characterization and improve the accuracy of a prediction result. In order to solve the technical problems, an embodiment of the present invention provides a space-time fusion photovoltaic power generation prediction method, including: performing space-time and time conversion on the acquired target photovoltaic power generation data, generating a fusion embedded representation, and performing block processing to obtain a block embedded representation; the block embedding representation is input into a preset space-time attention layer for transformation, so that a target characterization vector corresponding to each node is obtained, and an initial power generation prediction result is output; Calculating Euclidean distances between each target node corresponding to the space-time embedded representation in the fusion embedded representation and each class cluster center in the pre-constructed class cluster representation, and dynamically adjusting and optimizing to determine the class cluster center corresponding to each target node; Aggregating all nodes in the cluster centers to obtain cluster time sequence representations corresponding to the cluster centers; the class cluster time sequence representation and the target characterization vector are cooperated, weight is distributed to each prediction expert constructed by the initial power generation prediction result, and a target power generation prediction result is output; And determining an optimal power generation plan of the target power system according to the target power generation prediction result. Further, the performing space-time and time conversion on the obtained target photovoltaic power generation data to generate a fusion embedded representation and performing block processing to obtain a block embedded representation includes: Converting the target photovoltaic power generation data from the space-time characteristic angle to generate the space-time embedded representation corresponding to each node; Converting the target photovoltaic power generation data from the aspect of time characteristics to generate corresponding time embedded representation; Integrating the space-time embedded representation and the time embedded representation to obtain the fusion embedded representation; And carrying out blocking processing on the fusion embedded representation along the time dimension to obtain the blocking embedded representation. Further, the step of embedding the blocks into the space-time attention layer to transform the input representation to obtain a target characterization vector corresponding to each node, so as to output an initial power generation prediction result, includes: the block embedding representation is input into a preset time sequence attention layer to perform first transformation to obtain time sequence representation vectors corresponding to all nodes, the time sequence representation vectors are subjected to first linear transformation, and corresponding time sequence power generation prediction results are output; inputting the time sequence representation vector into a preset space attention layer for second transformation to obtain a space-time representation vector corresponding to each node, performing second linear transformation on the space-time representation vector, and outputting to obtain a corresponding space-time power generation prediction result; And integrating the time sequence power generation prediction result and the space-time power generation prediction result to obtain the initial power generatio