Primož Potočnik
adaptive forecasting, short-term, energy consumption, natural gas
Short term forecasting of natural gas consumption in daily resolution for the next gas consumption day is considered in this paper. Various forecasting models are constructed and compared, including linear models and neural network models that are trained with either static or adaptive algorithms. The models are trained on two winter seasons and then tested on the next three winter seasons where the last season shows considerable nonstationary behaviour. Best forecasting one-day-ahead results are obtained by applying the adaptive linear model. This model performs accurately also for nonstationary data and exceeds in the robustness and accuracy the neural network based models.
Important Links:
Go Back
