CN-122000868-A - Wind-solar-storage micro-grid load prediction method, device, equipment and storage medium

CN122000868ACN 122000868 ACN122000868 ACN 122000868ACN-122000868-A

Abstract

The invention relates to a wind-solar-storage micro-grid load prediction method, a device, equipment and a storage medium, wherein the method comprises the steps of obtaining wind-solar-storage micro-grid data; the load prediction model comprises a global attention mechanism and an expansion convolution network, wherein a load sequence in a wind and light storage micro-grid data set is input into the load prediction model, the load sequence is subjected to position coding through the expansion convolution network to obtain a time sequence load sequence with enhanced position information, global context vectors and local self-attentions of each position in the time sequence load sequence are spliced through the global attention mechanism to obtain a grid load prediction result, the load prediction model comprising the global attention mechanism and the expansion convolution network is constructed, the load dynamic time sequence change is adapted through the expansion convolution network, global feature capture is enhanced through the global attention mechanism, the accuracy and the stability of wind and light storage micro-grid load prediction are improved, and reliable load prediction support is provided for safe and efficient operation of the wind and light storage micro-grid.

Inventors

SHI YING
Kuang Yingxian
WU XIANG

Assignees

武汉理工大学

Dates

Publication Date: 20260508
Application Date: 20251231

Claims (10)

1. The wind-solar micro-grid load prediction method is characterized by comprising the following steps of: acquiring wind-solar storage micro-grid data of a period to be predicted; Constructing a load prediction model, wherein the load prediction model comprises a global attention mechanism and an expansion convolution network; And inputting the load sequence in the wind-solar-storage micro-grid data set into the load prediction model, performing position coding on the load sequence through the expansion convolution network to obtain a time sequence load sequence with enhanced position information, and splicing the global context vector and the local self-attention of each position in the time sequence load sequence through the global attention mechanism to obtain a grid load prediction result.
2. The wind-solar micro-grid load prediction method according to claim 1, wherein the splicing the global context vector and the local self-attention of each position in the time-series load sequence by the global attention mechanism to obtain the power grid load prediction result comprises the following steps: calculating each key vector in the time sequence load sequence through the global attention mechanism to obtain global attention weight; Processing the time sequence load sequence through a local self-attention mechanism to obtain the self-attention of each position; performing weighted average on the value vectors of all positions in the time sequence load sequence and the global attention weight to obtain a global context vector; splicing the global context vector and the self-attention to obtain a global attention sequence; and outputting the global attention sequence to obtain a power grid load prediction result.
3. The wind-solar micro-grid load prediction method according to claim 1, wherein the expansion convolution network comprises a residual module, the load sequence is subjected to position coding through the expansion convolution network to obtain a time sequence load sequence with enhanced position information, and the method comprises the following steps: And extracting dynamic time sequence characteristics of the load sequence through the expansion convolution network, and transmitting cross-layer information through the residual error module in the extraction process to obtain the time sequence load sequence with enhanced position information.
4. The wind-solar micro-grid load prediction method according to claim 2, wherein the load prediction model further comprises a convolution gating recursion unit, the outputting the global attention sequence to obtain a grid load prediction result comprises: And carrying out time sequence updating and outputting on adjacent load data in the global attention sequence through the convolution gating recursion unit to obtain a power grid load prediction result.
5. The wind-solar micro power grid load prediction method according to claim 1, wherein the training process of the load prediction model comprises the following steps: Acquiring historical wind-solar-storage micro-grid data, wherein the historical wind-solar-storage micro-grid data comprises static data and dynamic data; Filling and normalizing the missing value of the dynamic data in the historical wind-solar micro-grid data to obtain preprocessed dynamic data; Based on the preprocessing dynamic data and the static data, constructing a historical wind-solar micro-grid data set; and training the load prediction model based on the historical wind-solar-storage micro-grid data set to obtain a trained load prediction model.
6. The wind-solar-storage micro-grid load prediction method according to claim 5, wherein the training the load prediction model based on the historical wind-solar-storage micro-grid dataset to obtain a trained load prediction model comprises: dividing the historical wind-solar-storage micro-grid data set into a training set and a testing set; after training the load prediction model through the training set, testing the load prediction model through the testing set to obtain a predicted value; Verifying deviation between the actual load value and the predicted load value of the wind and solar energy storage micro-grid in the test set through accuracy evaluation indexes to obtain a verification result, wherein the accuracy evaluation indexes comprise average absolute errors, root mean square errors and fractions; And when the verification result meets the verification condition, obtaining a trained load prediction model.
7. The wind-solar micro grid load prediction method according to claim 2, wherein the calculation of the global attention weight is as follows: in the formula, For global attention weights, n is the length of the time-series load sequence, In the form of a global query vector, For the key vector corresponding to the i-th position in the time-series load sequence, Is the key vector corresponding to the j-th position in the time sequence load sequence.
8. The utility model provides a wind-solar energy storage micro-grid load prediction device which is characterized in that the device comprises: the data acquisition module is used for acquiring wind-solar storage micro-grid data of a period to be predicted; the model construction module is used for constructing a load prediction model, wherein the load prediction model comprises a global attention mechanism and an expansion convolution network; the load prediction module is used for inputting the load sequence in the wind-solar storage micro-grid data set into the load prediction model, carrying out position coding on the load sequence through the expansion convolution network to obtain a time sequence load sequence with enhanced position information, and splicing the global context vector and the local self-attention of each position in the time sequence load sequence through the global attention mechanism to obtain a grid load prediction result.
9. An electronic device comprising a processor, a memory and a computer program stored on the memory and capable of running on the processor, which when executed by the processor, implements the steps of the wind and solar energy storage micro grid load prediction method according to any one of claims 1-7.
10. A computer readable storage medium, characterized in that it has stored thereon a computer program which, when executed by a processor, implements the steps of the wind and solar energy storage micro grid load prediction method according to any of claims 1-7.

Description

Wind-solar-storage micro-grid load prediction method, device, equipment and storage medium Technical Field The invention relates to the technical field of load prediction of power systems, in particular to a wind-solar storage micro-grid load prediction method, device, equipment and storage medium. Background The wind-solar energy storage micro-grid is used as a multi-energy complementary system of wind energy, photovoltaic energy and energy storage, can effectively balance the fluctuation and instability of renewable energy sources, and has important significance for improving the economy and reliability of a power system. Under the background of renewable energy development, accurate load prediction is used as a core support technology, and an important basis is provided for optimizing power generation resource allocation and reducing operation cost of a modern power system. However, with the rapid development of the power system and the influence of various factors, the load presents characteristics of complex nonlinearity, time sequence dependency and the like, and the energy supply and demand balance of the micro-grid is a significant challenge. A more accurate load prediction method is needed to address this challenge. In recent years, a transducer model and an attention mechanism are paid to the field of time sequence prediction. By introducing an attention mechanism, key time sequence characteristics and weather associated information in the model focusing load data can be enabled, the capturing capacity of the global dependency relationship is enhanced, and further prediction accuracy is improved. However, the current load prediction method based on the transducer still faces the bottleneck that the traditional attention mechanism is easily limited to local information, global trend of the load is difficult to fully mine, and the fixed position coding cannot adapt to dynamic time sequence change of the load. Therefore, there is an urgent need to provide a method, a device, equipment and a storage medium for predicting a wind-solar micro-grid load, which solve the technical problems that the traditional attention mechanism in the prior art is easily limited to local information, global trend of the load is difficult to fully excavate, and the fixed position code cannot adapt to dynamic time sequence change of the load. Disclosure of Invention In view of the foregoing, it is necessary to provide a method, an apparatus, a device and a storage medium for predicting a wind-solar micro-grid load, so as to solve the technical problems that the conventional attention mechanism in the prior art is easily limited to local information, global trend of the load is difficult to fully excavate, and the fixed position code cannot adapt to dynamic time sequence change of the load. In order to solve the above problems, in a first aspect, the present invention provides a wind-solar micro-grid load prediction method, including: acquiring wind-solar storage micro-grid data of a period to be predicted; Constructing a load prediction model, wherein the load prediction model comprises a global attention mechanism and an expansion convolution network; And inputting the load sequence in the wind-solar-storage micro-grid data set into the load prediction model, performing position coding on the load sequence through the expansion convolution network to obtain a time sequence load sequence with enhanced position information, and splicing the global context vector and the local self-attention of each position in the time sequence load sequence through the global attention mechanism to obtain a grid load prediction result. In a possible implementation manner, the splicing, by the global attention mechanism, the global context vector and the local self-attention of each position in the time-series load sequence to obtain a power grid load prediction result includes: calculating each key vector in the time sequence load sequence through the global attention mechanism to obtain global attention weight; Processing the time sequence load sequence through a local self-attention mechanism to obtain the self-attention of each position; performing weighted average on the value vectors of all positions in the time sequence load sequence and the global attention weight to obtain a global context vector; splicing the global context vector and the self-attention to obtain a global attention sequence; and outputting the global attention sequence to obtain a power grid load prediction result. In one possible implementation manner, the expansion convolution network comprises a residual module, and the step of performing position coding on the load sequence through the expansion convolution network to obtain a time sequence load sequence with enhanced position information comprises the following steps: And extracting dynamic time sequence characteristics of the load sequence through the expansion convolution network, and transmitting cross-lay