International Institute of Forecasters
International Symposium on Forecasting
Marquette University’s GasDay Project specializes in short-term load forecasting of natural gas demand. Traditionally, this forecasting is done using artificial neural networks and linear regression. This paper examines the viability of using DNNs as component models in the GasDay ensemble. The ensemble of interest is evaluated using weighted MAPE on 88 natural gas data sets and compared to the current GasDay ensemble. The DNN-enhanced ensemble performs better on this metric than the current ensemble.