An Ultra-Short-Term PV Power Prediction Method Based on Meteorological Factors with Weather Fluctuation Level and Historical Power Datasets

Photovoltaic (PV) power generation has attracted widespread attention due to its environmental friendliness and cost-effectiveness. However, the intermittency and unpredictability of PV power production pose challenges to the reliable operation of the electric power system (EPS), especially in complex weather conditions where the output power of PV becomes even more difficult to predict. Therefore, the development of an accurate ultra-short-term PV power forecasting system is crucial in assisting power system operators in maintaining grid stability and enabling effective energy trading among market participants. This study proposes a novel ultra-short-term power forecasting model that combines the GRU neural network and K-means clustering method. The model integrates the time-series data of various meteorological parameters over a certain period, considering the periodicity and continuity of weather changes. Additionally, it incorporates power time series data to extract information about the operational characteristics of the PV plant. Furthermore, the model defines several variables that measure the level of weather fluctuations, enabling effective classification of weather types on a short-time scale and providing references for the forecasting process. The experimental results demonstrate that the proposed method can improve the accuracy of ultra-short-term photovoltaic power prediction, and its advantages are more prominent in complex weather conditions.