CN-122026337-A - Wide-area photovoltaic power prediction method and system based on space-time residual feature fusion

CN122026337ACN 122026337 ACN122026337 ACN 122026337ACN-122026337-A

Abstract

The invention belongs to the field of photovoltaic prediction, and provides a wide-area photovoltaic power prediction method and a wide-area photovoltaic power prediction system based on space-time residual feature fusion, wherein the wide-area photovoltaic power prediction method and the wide-area photovoltaic power prediction system are based on regional macroscopic meteorological features, and a plurality of trained primary prediction models are utilized for parallel prediction and then are aggregated to obtain a primary aggregation predicted value and a predicted residual sequence; the method comprises the steps of constructing corresponding space-time residual characteristics according to a predicted residual sequence, fusing the space-time residual characteristics with time-aligned regional macroscopic meteorological characteristics to obtain an enhanced characteristic set, predicting by using a trained secondary correction model based on the enhanced characteristic set to obtain a final predicted power correction amount, and taking the sum of the final predicted power correction amount and a primary aggregation predicted value as a regional photovoltaic total power predicted value. The invention explicitly learns and corrects systematic prediction bias caused by regional geographic dispersion, thereby reducing dependence on underlying mass data.

Inventors

DU YUFENG
ZHANG YUCHEN
NIE XIUSHAN
SONG YANG
CHEN GUANZHONG
WANG JIN
SHANG ZHENHAO
LIN YUQING
YU GUANGZHOU

Assignees

山东建筑大学

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (10)

1. The wide-area photovoltaic power prediction method based on space-time residual feature fusion is characterized by comprising the following steps of: The method comprises the steps of obtaining multi-source heterogeneous data of a plurality of subareas in a target area, preprocessing the multi-source heterogeneous data to obtain an area-level historical database, and extracting area macroscopic weather characteristics based on the area-level historical database; Based on regional macroscopic weather features, predicting by utilizing a plurality of trained primary prediction models in parallel and then aggregating to obtain a primary aggregation predicted value and a predicted residual sequence; Constructing corresponding space-time residual error characteristics according to the predicted residual error sequence, and fusing the space-time residual error characteristics with the time-aligned regional macroscopic meteorological characteristics to obtain an enhanced characteristic set; based on the enhanced feature set, predicting by using the trained secondary correction model to obtain a final predicted power correction; And taking the sum of the final predicted power correction and the first-order aggregation predicted value as a regional photovoltaic total power predicted value.
2. The wide-area photovoltaic power prediction method based on space-time residual feature fusion according to claim 1, wherein the steps of obtaining multi-source heterogeneous data of a plurality of sub-areas in a target area and preprocessing the multi-source heterogeneous data to obtain an area-level historical database, and extracting area macroscopic meteorological features based on the area-level historical database comprise the steps of: acquiring actual measurement meteorological data, remote sensing inversion data, weather forecast data and total power historical data of regional photovoltaic power generation of a plurality of subareas in a target region, and taking the actual measurement meteorological data, remote sensing inversion data, weather forecast data and total power historical data of regional photovoltaic power generation of the target region as multi-source heterogeneous data of the plurality of subareas; preprocessing multi-source heterogeneous data of a plurality of subareas to obtain a regional level historical database; and extracting regional average surface irradiance, irradiance space variation coefficients, regional average temperature and humidity, regional cloud shielding evolution indexes and regional historical power statistical characteristics based on a regional-level historical database to form a regional-level characteristic vector sequence serving as regional macroscopic meteorological characteristics.
3. The wide-area photovoltaic power prediction method based on the temporal-spatial residual feature fusion according to claim 2, wherein the area average surface irradiance is an arithmetic average of surface irradiance of all valid grid points or sites in the target area at a certain moment; The irradiance space variation coefficient is the ratio of the standard deviation of irradiance in the area at the moment to the average surface irradiance of the area; The area average temperature and humidity are the area average ambient temperature and average humidity.
4. The wide-area photovoltaic power prediction method based on space-time residual feature fusion according to claim 1, wherein the prediction residual sequences corresponding to the first-level aggregation predicted values and different-level predicted models are obtained by performing parallel prediction and then aggregating by using a plurality of trained first-level predicted models based on regional macroscopic meteorological features, and the method comprises the following steps: based on regional macroscopic meteorological features, a plurality of trained primary prediction models are utilized to conduct prediction in parallel, and initial prediction values of regional photovoltaic total power corresponding to different primary prediction models are obtained; the method comprises the steps that initial predicted values of the photovoltaic total power of areas corresponding to different levels of predicted models are aggregated to obtain a level-one aggregation predicted value; taking the difference value between the initial predicted value of the regional photovoltaic total power corresponding to the different-level prediction model and the corresponding actual regional photovoltaic total power measured value as a prediction residual error of the different-level prediction model; And constructing a prediction residual sequence based on the prediction residues of all the primary prediction models.
5. The wide-area photovoltaic power prediction method based on temporal-spatial residual feature fusion of claim 1, wherein the temporal-spatial residual features comprise model-differential residual features, regional output dispersion proxy features, and residual timing statistics.
6. The wide-area photovoltaic power prediction method based on the fusion of space-time residual characteristics according to claim 5, wherein the model differential residual characteristics are standard deviations of prediction residuals of all primary prediction models at each time point; The regional output dispersion agent features are products of irradiance space variation coefficients of each time point and prediction residual absolute values of a primary prediction model with optimal current time point; the residual time sequence statistical characteristic is to calculate the statistics of the prediction residual of each primary prediction model in a sliding time window.
7. Wide area photovoltaic power prediction system based on space-time residual feature fusion, characterized by comprising: The data processing module is configured to acquire multi-source heterogeneous data of a plurality of subareas in the target area and preprocess the multi-source heterogeneous data to obtain an area-level historical database, and extract area macroscopic weather characteristics based on the area-level historical database; the first-level prediction model is configured to be based on regional macroscopic meteorological features, and a plurality of trained first-level prediction models are utilized to conduct prediction in parallel and then are aggregated to obtain a first-level aggregation predicted value and a predicted residual sequence; The space-time residual feature engineering module is configured to construct corresponding space-time residual features according to the predicted residual sequence, and fuse the space-time residual features with the time-aligned regional macroscopic meteorological features to obtain an enhanced feature set; the secondary correction module is configured to predict by using the trained secondary correction model based on the enhanced feature set to obtain a final predicted power correction; And the prediction fusion module is configured to take the sum of the final predicted power correction quantity and the primary aggregation predicted value as a regional photovoltaic total power predicted value.
8. A computer readable storage medium having stored thereon a computer program, which when executed by a processor performs the steps in the wide area photovoltaic power prediction method based on spatio-temporal residual feature fusion of any of claims 1-6.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor performs the steps in the wide area photovoltaic power prediction method based on spatio-temporal residual feature fusion of any of claims 1-6 when the program is executed.
10. A computer program product, characterized in that it comprises a computer program which, when executed by a processor, implements the steps in the wide-area photovoltaic power prediction method based on spatio-temporal residual feature fusion according to any of claims 1-6.

Description

Wide-area photovoltaic power prediction method and system based on space-time residual feature fusion Technical Field The invention belongs to the technical field of photovoltaic prediction, and particularly relates to a wide-area photovoltaic power prediction method and system based on space-time residual feature fusion. Background The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art. Photovoltaic power generation is an important component of clean energy, and the installed capacity of the photovoltaic power generation continuously and rapidly increases. Because the photovoltaic output is obviously influenced by meteorological factors, has intermittence, volatility and randomness, and brings challenges to the operation and market transaction of the power system, the high-precision power prediction has important significance for power grid consumption, scheduling optimization and operation safety. Currently, photovoltaic power prediction mainly aims at a single power station, local historical data and weather forecast are utilized, and a prediction model is built through a physical model, a statistical method or a machine learning algorithm. With the large-scale deployment of photovoltaics over a wide area, regional level prediction needs are increasingly prominent. The existing region prediction technology is mainly developed along two directions, namely, an optimized single-point prediction model is improved in precision by adopting strategies such as integrated learning, modal decomposition, residual error correction and the like, wherein residual error is mostly used for real-time correction or training data screening, and a dynamic weight and self-adaptive training mechanism is introduced, and a model structure or super parameters are adjusted through error evaluation so as to enhance model adaptability. However, the scheme still has the defects in wide-area prediction application, the existing method is not used for fully excavating the space-time law of the region contained in the residual error, the residual error processing mode is simple, modeling aiming at the geographic dispersion smoothing effect is lacked, the output space asynchronism in the region and the influence of the output space asynchronism on the total power prediction are ignored by simply summing the prediction results of all stations, in addition, the existing general prediction framework is not tightly combined with the physical process of the total output of the macroscopic weather driving region, and hierarchical design matched with the existing general prediction framework is lacked. Disclosure of Invention In order to solve the problems, the invention provides a wide-area photovoltaic power prediction method and a wide-area photovoltaic power prediction system based on space-time residual feature fusion, the invention constructs space-time residual features capable of reflecting regional output space heterogeneity, and designing a two-stage cascade prediction framework, wherein a second-stage model specifically learns how to refine the first-stage prediction by using the residual characteristics, thereby realizing the deep modeling of the integral output mode of the region. According to some embodiments, the first scheme of the invention provides a wide-area photovoltaic power prediction method based on space-time residual feature fusion, which adopts the following technical scheme: a wide-area photovoltaic power prediction method based on space-time residual feature fusion comprises the following steps: The method comprises the steps of obtaining multi-source heterogeneous data of a plurality of subareas in a target area, preprocessing the multi-source heterogeneous data to obtain an area-level historical database, and extracting area macroscopic weather characteristics based on the area-level historical database; Based on regional macroscopic weather features, predicting by utilizing a plurality of trained primary prediction models in parallel and then aggregating to obtain a primary aggregation predicted value and a predicted residual sequence; Constructing corresponding space-time residual error characteristics according to the predicted residual error sequence, and fusing the space-time residual error characteristics with the time-aligned regional macroscopic meteorological characteristics to obtain an enhanced characteristic set; based on the enhanced feature set, predicting by using the trained secondary correction model to obtain a final predicted power correction; And taking the sum of the final predicted power correction and the first-order aggregation predicted value as a regional photovoltaic total power predicted value. Further, obtaining multi-source heterogeneous data of a plurality of subareas in a target area and preprocessing the multi-source heterogeneous data to obtain an area-level historical database, extractin