Sustainable Natural Gas Price Forecasting with DEEPAR

Accurately forecasting natural gas prices is essential for efficient energy system management in the competitive market. However, the inconsistent data frequency and nonlinear fluctuation features cause challenges to reliable predictions. A novel natural gas price prediction model, the Optimized DeepAR model, is proposed to address this challenge. This model combines a deep auto-regressive neural network (DeepAR) with grid search optimization (GSO). DeepAR utilizes Long Short-Term Memory (LSTM) and a probabilistic time series approach. Our model enhances accuracy by integrating exogenous attributes from National Oceanic and Atmospheric Administration (NOAA) time series data. It provides a 95% confidence level probabilistic price range with a Root Mean Squared Error (RMSE) of 0.2021. This model provides valuable insights for stakeholders and serves as a tool to estimate natural gas market prices, assisting in decision-making within the competitive market. The approach used in this study enhances forecasting performance, enabling efficient management of the energy system.