Solar power generation forecasting based on AI

Abstract

As the world aims to reduce carbon emissions, there is a strong future trend of adopting clean energy for power generation. Solar photovoltaic (PV) power energy has grown to be one of the most promising choice of clean energy because of the abundance of solar energy. However, PV power generation greatly depends on weather conditions and lead to PV power generation to be unpredictable. This can affect the stability of power systems. Therefore, this paper aims to study solutions to predict PV power generation with only historical PV power generation data. In addition, this paper investigates the effect of the size of lookback window on the forecasting accuracy of various models. This paper looks into the different types of models used to forecast PV power generation and focuses on existing research regarding AI based models. AI based models show great potential in handling nonlinear and complex relationship between weather conditions and PV power generation. Some examples of AI based models studied in this paper include ANN, LSTM and transformer model. The models are trained and validated using two datasets containing hourly energy data from two solar sites in North America from 01/01/2015 to 01/01/2016. Forecasting results of the two datasets are presented and analysed. Lastly, conclusions will be made about which model and choice of lookback window size are most suitable for each dataset.