Probabilistic forecasts of wind power generation in regions with complex topography using deep learning methods: An Arctic case

Abstract

The energy market relies on forecasting capabilities of both demand and power generation that need to be kept in dynamic balance. Nowadays, contracts and auctions of renewable energy in a liberalized electricity market heavily rely on forecasting future power generation. The highly intermittent nature of renewable energy sources increases the uncertainty about future power generation. Since point forecast does not account for such uncertainties, it is necessary to rely on probabilistic forecasts. This work first introduces probabilistic forecasts with deep learning. Then, we show how deep learning models can be used to make probabilistic forecasts of day-ahead power generation from a wind power plant located in Northern Norway. The performance, in terms of the quality of the prediction intervals, is compared to different deep learning models and sets of covariates. The findings show that the accuracy of the predictions improves by 37% when historical data on measured weather and numerical weather predictions (NWPs) were included as exogenous variables. In particular, historical data allows the model to auto-correct systematic biases in the NWPs. Finally, we observe that when using only NWPs or only measured weather as exogenous variables, worse performances are obtained. The work shows the importance of understanding which variables must be included to improve the prediction performance, which is of fundamental value for the energy market that relies on accurate forecasting capabilities.