CN-122000880-A - Photovoltaic power generation power prediction method, system, equipment and medium

CN122000880ACN 122000880 ACN122000880 ACN 122000880ACN-122000880-A

Abstract

The invention provides a photovoltaic power generation power prediction method, a system, equipment and a medium, which belong to the technical field of power generation power prediction, and the method comprises the steps of obtaining a historical photovoltaic power time sequence and corresponding historical meteorological data; decomposing a historical photovoltaic power time sequence by adopting a complete set empirical mode decomposition additive noise algorithm to obtain a plurality of eigenmode functions and a residual subsequence, globally optimizing key super-parameters of a prediction model by adopting an optimization algorithm with the overall prediction error of a training feature set containing a plurality of weather types as a target, processing the overall optimized prediction model to obtain a preliminary prediction sequence of the photovoltaic power, and modeling and correcting the residual error corresponding to the preliminary prediction sequence based on the error compensation module to obtain a final photovoltaic power prediction sequence. The problem that the prediction accuracy of the photovoltaic power prediction model is not high when extreme weather appears alternately is solved.

Inventors

WEN XIN
LI ZHUOQUN
DOU XIANG
ZHANG CHEN
LIU JIAQI
ZHANG WEIMIAO
LIU KERUI
LI HAOYANG

Assignees

兰州交通大学

Dates

Publication Date: 20260508
Application Date: 20260126

Claims (10)

1. A method for predicting photovoltaic power generation power, the method comprising: acquiring historical photovoltaic power time sequence data of a photovoltaic power station and corresponding historical meteorological data thereof; Performing adaptive modal decomposition on the historical photovoltaic power time series data to obtain a plurality of eigenmode function subsequences with different time scale characteristics and a residual subsequence; Constructing an enhancement feature set based on the eigenmode function subsequence, the residual subsequence, and corresponding historical meteorological data; Constructing a training data set comprising all weather types, taking the prediction error of the training data set as a target, and adopting a Tornado optimization algorithm with Coriolis force to perform global optimization on the super-parameters of the prediction model based on the neural network architecture so as to obtain the global optimal model super-parameters; The enhanced feature set is input into a prediction model configured with global optimal super parameters to obtain a preliminary prediction sequence of photovoltaic power, a residual sequence between the preliminary prediction value sequence and an actual value is calculated, and the residual sequence is subjected to secondary decomposition to obtain a plurality of error modal components; and adding the preliminary predicted sequence and the error compensation value to obtain a final photovoltaic power predicted value sequence.
2. The method of claim 1, wherein the historical photovoltaic power time series data is subjected to an adaptive modal decomposition using a full set empirical modal decomposition additive noise algorithm.
3. The method of claim 1, wherein the global optimization of the hyper-parameters of the neural network architecture-based predictive model is performed using a coriolis force tornado optimization algorithm.
4. The method according to claim 1, wherein the prediction model of the neural network architecture is a neural network model based on a transform architecture, the prediction model obtains a mapping of power output at a future time through learning historical data by an encoder and a decoder, the prediction model configured according to the global optimum super parameter processes the input enhanced feature set, outputs a preliminary prediction sequence of photovoltaic power, specifically inputs the enhanced feature set into the encoder of the prediction model to extract space-time dependent features, inputs the space-time dependent features into the decoder of the prediction model, obtains power values of a plurality of time steps in the future through recursive or forward computation, and maps to form the preliminary prediction sequence.
5. The method according to claim 1, characterized in that said residual sequence is subjected to a quadratic decomposition, in particular a modal decomposition of said error sequence using a collective empirical mode decomposition algorithm.
6. The method according to claim 1, wherein the time sequence prediction is performed on the plurality of error modal components by using a gating loop unit neural network to obtain an error compensation value, specifically, each error modal component is predicted by a preset neural network, and then the prediction results of each preset neural network are summed to obtain a total error compensation value.
7. The method according to claim 1, wherein the time sequence prediction is performed on the plurality of error modal components by using a gating cyclic unit neural network, specifically, the plurality of error modal components are taken as a multivariate time sequence, and are input into the gating cyclic unit neural network together for prediction, so as to obtain a total error compensation value sequence.
8. A photovoltaic power generation power prediction system, comprising: the acquisition module is used for acquiring historical photovoltaic power time series data of the photovoltaic power station and corresponding historical meteorological data; The decomposition module is used for carrying out self-adaptive modal decomposition on the historical photovoltaic power time series data to obtain a plurality of eigen-modal function subsequences with different time scale characteristics and a residual subsequence; the construction module is used for constructing an enhanced feature set based on the intrinsic mode function subsequence, the residual subsequence and corresponding historical meteorological data; the optimization module is used for constructing a training data set comprising all weather types, taking the prediction error of the training data set as a target, and adopting a tornado optimization algorithm with coriolis force to perform global optimization on the super-parameters of the prediction model based on the neural network architecture so as to obtain the global optimal model super-parameters; The prediction module is used for inputting the enhanced feature set into a prediction model configured with global optimal super parameters to obtain a preliminary prediction sequence of photovoltaic power, calculating a residual sequence between the preliminary prediction value sequence and an actual value, performing secondary decomposition on the residual sequence to obtain a plurality of error modal components, and adopting a gating circulation unit neural network to learn and predict a time sequence mode of the error modal components to obtain an error compensation value; And the result output module is used for adding the preliminary predicted sequence and the error compensation value to obtain a final photovoltaic power predicted value sequence.
9. A computer readable storage medium, characterized in that the storage medium stores a computer program which, when executed by a processor, implements the method of any of the preceding claims 1-7.
10. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of any of the preceding claims 1 to 7 when the program is executed.

Description

Photovoltaic power generation power prediction method, system, equipment and medium Technical Field The invention belongs to the technical field of power generation power prediction, and particularly relates to a photovoltaic power generation power prediction method, a system, equipment and a medium. Background The output of the photovoltaic power station is commonly influenced by factors such as solar radiation, temperature, cloud cover, sand dust, wind speed and the like, and the photovoltaic power station presents typical strong non-stable, multi-scale, multi-mode and coupled time series characteristics. In the background of the aggravation of global climate change, periodic climate phase change caused by large-scale climate modes such as el Nino-southern surge (ENSO) and the like can lead to severe switching of regional weather among multiple extreme modes such as sunny, heavy rain, sand storm and the like in a short period. In such a complex climate scene, the realization of high-precision prediction of photovoltaic output is a key for improving the capability of a power system for absorbing renewable energy and the operation toughness. Currently, researchers often use a hybrid model framework that combines signal decomposition techniques with deep learning models in order to improve prediction accuracy. Many studies on hybrid models still focus mainly on optimizing the model structure itself to improve the data fitting accuracy, and there is a lack of in-depth research on how the meteorological physical mechanisms systematically affect the model performance under different weather conditions. This disadvantage is particularly pronounced in variable weather patterns driven by large scale climate modalities such as ENSO. The accurate modeling of photovoltaic output under distinct extreme weather conditions is not stable in performance of the existing model in a climate transition period due to the lack of a modeling method for inherent robustness to climate rapid change in the prior art, and reliability and robustness in engineering application are difficult to guarantee. In addition, most researches focus on point prediction, neglect the key link of prediction error correction, and the practicability is affected by uncorrectable systematic deviation of a predicted value. It can be seen that the current photovoltaic power prediction technology has obvious defects in the following aspects that the existing hybrid prediction model lacks the capability of self-adaptive matching and global generalization of multiple weather physical mechanisms in a climate transition period, so that the prediction accuracy of the model is not high when extreme weather alternates. Disclosure of Invention In order to solve the problem that the prediction accuracy of a photovoltaic power prediction model is not high when extreme weather appears alternately, the invention provides a photovoltaic power generation power prediction method, a system, equipment and a medium. In a first aspect of an embodiment of the present invention, there is provided a photovoltaic power generation power prediction method, including the steps of: acquiring historical photovoltaic power time sequence data of a photovoltaic power station and corresponding historical meteorological data thereof; Performing adaptive modal decomposition on the historical photovoltaic power time series data to obtain a plurality of eigenmode function subsequences with different time scale characteristics and a residual subsequence; Constructing an enhancement feature set based on the eigenmode function subsequence, the residual subsequence, and corresponding historical meteorological data; Constructing a training data set comprising all weather types, taking the prediction error of the training data set as a target, and adopting a Tornado optimization algorithm with Coriolis force to perform global optimization on the super-parameters of the prediction model based on the neural network architecture so as to obtain the global optimal model super-parameters; The enhanced feature set is input into a prediction model configured with global optimal super parameters to obtain a preliminary prediction sequence of photovoltaic power, a residual sequence between the preliminary prediction value sequence and an actual value is calculated, and the residual sequence is subjected to secondary decomposition to obtain a plurality of error modal components; and adding the preliminary predicted sequence and the error compensation value to obtain a final photovoltaic power predicted value sequence. And further, carrying out self-adaptive modal decomposition on the historical photovoltaic power time series data, wherein the self-adaptive modal decomposition adopts a complete set empirical mode decomposition additive noise algorithm. Further, the super-parameters of the prediction model based on the neural network architecture are subjected to global optimization, wherein the global optimi