Long-term natural gas peak demand forecasting in Tunisia Using machine learning

Abstract

Natural gas peak demand forecasting is crucial for efficient network infrastructure spending and stock planning. Herein, long-term forecasting is studied, using data of the Tunisian Company of Electricity and Gas (STEG) as case study. Gas peak flow data preprocessing is elaborated as preliminary step. Ridge regressor, support vector regressor (SVR) and K-nearest neighbors (K-NN) are implemented to make long-term forecasting with different resolutions (daily, monthly). Based on the best performing base models, two types of ensemble models are implemented: simple average and weighted average. The study provides important results to decision-makers in order to optimize the energy policies.