Research on a Short term Power Prediction Method for Photovoltaic Power Generation

This article studies a short-term power prediction model for large-scale photovoltaic power generation, focusing on short-term prediction of photovoltaic power generation. In this article, the concept of virtual prediction is innovatively introduced, and external influencing factors are comprehensively considered to form a prediction model algorithm library. A new model for predicting electricity consumption based on time series dynamic adjustment algorithms and weight ratios is studied to achieve optimal prediction results. With the goal of minimizing errors, we select the most recent historical data for hypothesis prediction, screen the best models and parameters, and achieve accurate short-term power prediction for large-scale photovoltaic power generation; At the same time, photovoltaic power stations that can obtain complete meteorological data will be used as reference power stations to obtain historical meteorological data with the highest similarity to the current target meteorological parameters, and a similar day dataset will be constructed; Then, the kernel function extreme learning mechanism optimized by particle swarm optimization is used to build a benchmark photovoltaic power plant power prediction model, while comprehensively considering the factors affecting the purchased electricity quantity. Based on the diversity and complexity of the variation patterns of the predicted quantity, and taking into account the impact of historical electricity purchase data and economic, meteorological, holiday and other factors on the predicted quantity, the previous single factor prediction model has been changed. Finally, we will review the commonly used methods for current electricity forecasting models, including univariate linear regression, similarity extrapolation, holiday point-to-point, ratio smoothing, overlapping curves, and other forecasting methods. We will construct a single forecasting model for different types of customers and form an algorithm library for forecasting models to meet the forecasting needs of different types of users.