Using Artificial Intelligence to Accurately Forecast Solar Energy Including Air Quality in the Models

Accurate power forecasting is an important factor in photovoltaic (PV) solar array systems, due to the volatile nature of the technology. This paper proposes an advanced statistical method for accurate solar power forecasting 12-24 hours in advance using artificial intelligence techniques. The method incorporates previous power output from a solar panel array, local weather data, sun position, and irradiance levels as important inputs. In addition, a unique feature of this paper is the incorporation and study of the impact that local air pollution levels have on the forecast model. The results show that the Random Forest method produces a high-quality model with a very low error rate for medium term forecasts of 12-24 hours into advance, compared to other, more complex models such as a recurrent neural network. The study also highlights the influence that air quality pollution levels have as an input predictor to the model. The influence of air quality pollution is greatest from particle pollutants during winter months. The model proposed in this paper can be used as an efficient and easily interpretable forecasting technique for operational planning for a solar array.