Forecasting Indonesia’s Coal Consumption Using Artificial Neural Network

Abstract

Indonesia is one of the world’s largest coal producers and exporters, with high domestic coal consumption, making reliable coal consumption forecasting essential for energy planning and policy development. This study applies an Artificial Neural Network (ANN) model using a Multilayer Perceptron (MLP) architecture to forecast coal consumption in Indonesia using population, Gross Domestic Product (GDP), exports, and imports as input variables. The model was trained using Levenberg–Marquardt (trainlm) and Bayesian Regularization (trainbr) algorithms with data normalized to the ranges of [0,1] and [-1,1]. Model performance was evaluated using correlation coefficient (R) and Mean Absolute Percentage Error (MAPE). Data normalized to the range [-1,1] when combined with the trainbr algorithm produced higher R values and lower MAPE. The optimal configuration, consisting of 9 hidden neurons, achieved an R value of 0.9991 and a MAPE of 2.90%. These results highlight the importance of data normalization and training algorithm selection in improving MLP model reliability. Forecasting results suggest a gradual increase in coal consumption, reaching approximately 343.96 million tonnes by 2032, and underscore the need for adaptive strategies in coal mining and coal-consuming sectors to maintain energy supply stability, address environmental sustainability challenges, and support national energy planning and policy decisions in the context of Indonesia's long-term decarbonization goals.

Access to Document

Other files and links

• Link to publication in Scopus
• Open Access Version Available

Fingerprint