CN-121981330-A - Short-term photovoltaic power prediction method and system driven by data mechanism in combined mode

CN121981330ACN 121981330 ACN121981330 ACN 121981330ACN-121981330-A

Abstract

A method and a system for predicting short-term photovoltaic power driven by a data mechanism combination comprise the steps of obtaining geographic latitude, solar declination angle and solar time angle of a photovoltaic power station to be predicted, calculating the solar altitude angle according to the geographic latitude, the solar declination angle and the solar time angle, determining the projection direction of the sun in a horizontal plane to obtain the solar azimuth angle, calculating ideal irradiance of a photovoltaic panel plane based on the solar altitude angle and the solar azimuth angle in combination with photovoltaic panel arrangement parameters of the photovoltaic power station to be predicted, introducing the ideal irradiance of the photovoltaic panel plane to establish a photovoltaic power prediction probability neural network PNN model, establishing a photovoltaic power prediction neural network PVFNN model on the PNN model in combination with a long-term memory network LSTM, and realizing short-term photovoltaic power prediction by utilizing the PVFNN model. According to the method, the irradiance model under ideal weather is organically embedded into the neural network, learning in the data driving process is restrained, and a prediction result with interpretability can be obtained.

Inventors

HUANG JINGJING
LI YUXUAN
YAO JINGYU
Qiao Jichen
WANG XIAOYU
REN ZHIGANG
ZHANG YIFAN
WANG SHUO
GAO XUMING
XIE YURONG

Assignees

西安交通大学
华电电力科学研究院有限公司

Dates

Publication Date: 20260505
Application Date: 20260130

Claims (10)

1. A data mechanism co-driven short-term photovoltaic power prediction method, comprising: obtaining the geographical latitude, the solar declination angle and the solar time angle of a photovoltaic power station to be predicted; Calculating a solar altitude according to the geographic latitude, the solar declination and the solar time angle, and determining the projection direction of the sun in the horizontal plane to obtain a solar azimuth angle, wherein the solar altitude represents the inclination angle of the sun relative to the equatorial plane of the earth; based on the solar altitude and the solar azimuth, calculating ideal irradiance of a photovoltaic panel plane by combining photovoltaic panel arrangement parameters of the photovoltaic power station to be predicted; an ideal irradiance of a photovoltaic panel plane is introduced to establish a photovoltaic power prediction probability neural network PNN model; and building a photovoltaic power prediction neural network PVFNN model by combining a long-term memory network LSTM on the PNN model, and realizing short-term photovoltaic power prediction by utilizing the PVFNN model.
2. The method for predicting short-term photovoltaic power driven by a combination of data mechanisms according to claim 1, wherein the step of calculating a solar altitude according to a geographical latitude, a solar declination angle and a solar hour angle is characterized in that the calculation expression of the solar altitude is as follows: In the formula, Is the solar altitude; The geographical latitude of the photovoltaic power station to be predicted is; is the declination angle of the sun; Is the solar time angle.
3. The method for predicting short-term photovoltaic power driven by a combination of data mechanisms according to claim 2, wherein the step of determining the projection direction of the sun in the horizontal plane to obtain the solar azimuth angle is characterized in that the calculation expression of the solar azimuth angle is as follows: In the formula, Is the azimuth angle of the sun; Indicating that the true solar time is in the morning and the sun is in the southeast direction; indicating that true solar time is in afternoon and solar time is in the southwest direction.
4. A method for short term photovoltaic power generation by combined driving of data mechanisms according to claim 3, wherein in said step of calculating ideal irradiance of a photovoltaic panel plane based on solar altitude and solar azimuth in combination with photovoltaic panel placement parameters of a photovoltaic power plant to be predicted, said photovoltaic panel placement parameters of the photovoltaic power plant to be predicted include an angle between the photovoltaic panel and horizontal ground And the included angle between the projection of the normal line of the plane of the photovoltaic panel on the horizontal plane and the direction of the right south ; The incident angle is calculated according to the following expression Incident angle of The included angle between the solar ray and the plane normal of the photovoltaic module is as follows: The method comprises the steps of obtaining the external radiation intensity of the atmosphere, which is perpendicular to solar rays, and combining an atmospheric air quality model to calculate and obtain the direct irradiance S R reaching the ground, and calculating the ideal irradiance received by the surface of the photovoltaic module according to the following formula : 。
5. The data mechanism joint driving short term photovoltaic power prediction method according to claim 1, wherein the step of creating a photovoltaic power prediction probabilistic neural network PNN model by introducing ideal irradiance of a photovoltaic panel plane comprises: will be ideal irradiance Carrying out per unit output according to the installed capacity or the historical maximum irradiance of the power station, and normalizing to obtain an ideal irradiance factor k i ; constructing a triangular transformation operator, and performing phase expansion on an ideal irradiance factor k i by utilizing sine and cosine operators, wherein the expression of the triangular transformation operator is as follows: the ideal irradiance calculation expression obtained after the transformation of the triangular transformation operator is as follows: In the formula, Is the angular frequency of the daily cycle; , Each representing an expansion coefficient determined by the ideal irradiance S R obtained on the photovoltaic panel; Representing the photovoltaic panel layout parameters and the solar altitude angle of a photovoltaic power station to be predicted And solar azimuth angle The determined expansion coefficient, the photovoltaic panel arrangement parameters of the photovoltaic power station to be predicted comprise the included angle between the photovoltaic panel and the horizontal ground And the included angle between the projection of the normal line of the plane of the photovoltaic panel on the horizontal plane and the direction of the right south ; The expansion coefficient is obtained after parameterization treatment: the parameters in the above formula are rewritten into a matrix form to obtain: After integration, the following yields: wherein A, B, M, N are each a matrix of learnable parameters.
6. The data mechanism joint driving short term photovoltaic power prediction method according to claim 5, wherein at an ideal irradiance Introducing a nonlinear activation function into a computational expression of (a) Enhancing the adaptability to aperiodic fluctuations, resulting in: In the formula, , All represent learning parameters of the nonlinear activation function neural network.
7. The method for predicting the short-term photovoltaic power driven by the combination of data mechanisms according to claim 1, wherein a long-term memory network LSTM is combined on a PNN model, a photovoltaic power prediction neural network PVFNN model is built, the short-term photovoltaic power prediction is realized by utilizing a PVFNN model, the capturing capacity of the PNN model for time sequence data is enhanced by utilizing the LSTM, and simultaneously, non-rational disturbance is captured in a deeper level, and the model output is as follows: In the formula, Representing power time sequence data of an input daily headroom curve after VMD modal decomposition and outputting obtained result through LSTM, wherein the daily headroom curve is based on geographic latitude of a photovoltaic power station to be predicted Included angle between photovoltaic panel and horizontal ground And the included angle between the projection of the normal line of the plane of the photovoltaic panel on the horizontal plane and the direction of the right south Establishing; And (5) representing a short-term photovoltaic power result obtained through photovoltaic power prediction probability neural network PNN model prediction.
8. A data mechanism co-driven short term photovoltaic power prediction system comprising: The data acquisition module is used for acquiring the geographical latitude, the solar declination angle and the solar time angle of the photovoltaic power station to be predicted; the solar angle calculation module is used for calculating a solar altitude according to the geographic latitude, the solar declination angle and the solar time angle, and determining the projection direction of the sun in the horizontal plane to obtain a solar azimuth angle; The ideal irradiance calculating module is used for calculating the ideal irradiance of the plane of the photovoltaic panel based on the solar altitude angle and the solar azimuth angle by combining the photovoltaic panel arrangement parameters of the photovoltaic power station to be predicted; the probability neural network model building module is used for building a photovoltaic power prediction probability neural network PNN model by introducing ideal irradiance of a photovoltaic panel plane; And the power prediction neural network establishing and outputting module is used for establishing a photovoltaic power prediction neural network PVFNN model by combining a long-term memory network LSTM on the PNN model and realizing short-term photovoltaic power prediction by utilizing the PVFNN model.
9. An electronic device, comprising: A memory storing at least one instruction, and A processor executing instructions stored in the memory to implement the data mechanism jointly driven short term photovoltaic power prediction method of any of claims 1 to 7.
10. A computer readable storage medium having stored therein at least one instruction for execution by a processor in an electronic device to implement the data mechanism joint driven short term photovoltaic power prediction method of any of claims 1 to 7.

Description

Short-term photovoltaic power prediction method and system driven by data mechanism in combined mode Technical Field The invention belongs to the technical field of power prediction, and particularly relates to a data mechanism combined driving short-term photovoltaic power prediction method and system. Background Along with the continuous improvement of the permeability of photovoltaic power generation in a power system, short-term photovoltaic power prediction has become a key technology for guaranteeing safe and stable operation of a power grid and promoting new energy consumption. The current mainstream prediction method mainly relies on a deep learning model driven by pure data, and an end-to-end mapping model is established by analyzing the statistical relationship between historical power data and meteorological data. Typical pure data driving methods include a long short term memory network (LSTM) based time series prediction model, a time series convolutional network (TCN) based feature extraction method, and a series of improvements thereof. By training a large amount of data, the method can capture complex modes and nonlinear relations in the data to a certain extent, and can show good prediction accuracy in the distribution range of the training data. However, the existing prediction model completely depends on data statistics characteristics, and physical constraint of photovoltaic power generation cannot be introduced, so that generalization capability is obviously reduced when a scene uncovered by training data is encountered. Secondly, the 'black box' characteristic of deep learning causes the decision process to lack physical interpretation, the prediction result is difficult to be associated with the actual photovoltaic system running state, and the reliability in scheduling decision is reduced. In addition, the method needs large-scale high-quality training data, is sensitive to missing values, noise and measurement errors, and a newly built power station cannot be directly applied due to lack of historical data, so that engineering application of the method is restricted to a certain extent. Disclosure of Invention The invention aims to provide a short-term photovoltaic power prediction method and a short-term photovoltaic power prediction system driven by a data mechanism aiming at the problems in the prior art, and aims to solve the problems of insufficient generalization capability, poor interpretability and easiness in overfitting existing in the conventional short-term photovoltaic power prediction model based on pure data driving by organically embedding an irradiance model under ideal weather into a neural network and restricting learning in the data driving process to obtain a prediction result with interpretability. In order to achieve the above purpose, the present invention has the following technical scheme: In a first aspect, a method for predicting short-term photovoltaic power driven by a data mechanism in combination is provided, including: obtaining the geographical latitude, the solar declination angle and the solar time angle of a photovoltaic power station to be predicted; Calculating a solar altitude according to the geographic latitude, the solar declination and the solar time angle, and determining the projection direction of the sun in the horizontal plane to obtain a solar azimuth angle, wherein the solar altitude represents the inclination angle of the sun relative to the equatorial plane of the earth; based on the solar altitude and the solar azimuth, calculating ideal irradiance of a photovoltaic panel plane by combining photovoltaic panel arrangement parameters of the photovoltaic power station to be predicted; an ideal irradiance of a photovoltaic panel plane is introduced to establish a photovoltaic power prediction probability neural network PNN model; and building a photovoltaic power prediction neural network PVFNN model by combining a long-term memory network LSTM on the PNN model, and realizing short-term photovoltaic power prediction by utilizing the PVFNN model. As a preferable scheme, the step of calculating the solar altitude according to the geographic latitude, the solar declination angle and the solar hour angle comprises the following calculation expression: In the formula, Is the solar altitude; The geographical latitude of the photovoltaic power station to be predicted is; is the declination angle of the sun; Is the solar time angle. As a preferred solution, the step of determining the projection direction of the sun in the horizontal plane, and obtaining the sun azimuth angle, wherein the calculation expression of the sun azimuth angle is as follows: In the formula, Is the azimuth angle of the sun; Indicating that the true solar time is in the morning and the sun is in the southeast direction; indicating that true solar time is in afternoon and solar time is in the southwest direction. As a preferable scheme, in the step of calculating th