Enhancing Machine Learning Based Solar Generation Forecasting with Time Data Utilization

Solar energy is expected to have a key role in the transition to a cleaner and environment-friendlier energy system, but given its highly weather dependent and intermittent nature, it creates operational challenges for the electric power grid because of the uncertainty involved in the output. Thus renewable energy generation forecasting becomes an increasingly important task in regard to not just the reliable operation of the energy system, but also from a financial standpoint. Here it is analyzed how the information carried by the timestamp can be utilized in order to improve solar generation forecasting accuracy. We devised and implemented our own time feature input, which we tested on several techniques based on machine learning for solar energy generation forecasting. In this paper, we demonstrate that, though often overlooked, the information carried by the timestamp is significant and can greatly improve the forecasting accuracy.