Ultra-short term photovoltaic power prediction

Abstract

The prediction algorithm of photovoltaic power generation is based on the data of actual measurement or weather forecast, and the corresponding prediction model is established according to the characteristics of the geographical location of photovoltaic power plants. According to the prediction length, we can study the power prediction in different time periods, in which the ultra-short-term power prediction time ranges from several minutes to four hours. According to the actual data, this design selects the two characteristic values of irradiance and temperature as input, establishes the photovoltaic power prediction models based on BP neural network, support vector regression, KNN and LSTM respectively, and compares the results to get their advantages and disadvantages. The results show that the LSTM neural network's power prediction curve fits the actual photovoltaic power curve better, that is, LSTM neural network is more suitable for ultra-short-term photovoltaic power prediction.