Long-term energy demand forecasting in Thailand using ensemble prediction model

Predicting the electricity demand of a country will help with design, planning, and future energy management. Instead of creating a forecasting model depending only on historical energy demand and single time-series or machine learning models, sophisticated statistical and intelligent energy prediction models are proposed with related factors to enhance prediction performance and could be useful to energy managers. We have designed and implemented a data-driven method to predict the electricity monthly energy profile for Thailand. Our preliminary work exploits a data set of monthly energy demand from Thailand from 2011 to 2019. Our method combines the most relevant factors forecasted on future value from 2020 to 2022 to obtain a prediction of the long-term future energy demand in the next 3-year period accordingly. This research has proposed to utilize the combination of Machine Learning models (ML models) to optimally forecast the energy demand in Thailand. The various ML models are explored in which the individual and the combination of ML models are each optimized and evaluated for their best achievable performances. Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) are utilized to compare models’ performances. A total of 4 ML models are executed, which include Artificial Neural Network (ANN), Support Vector Regression (SVR), Decision Tree (DT), Random Forest (RF) Ensemble, and proposed two heuristic models of Vote Ensemble learning. The results show that, by means of choosing a suitable combination to produce an ensemble model, the Vote Ensemble model (VEM) with a neural net and decision tree algorithms (NND) could perform well over individual learnings with the lowest RMSE for training and testing of 509.68 and 523.49 and the lowest training and testing MAPE of 2.43% and 4.21% accordingly while also using the lowest execution time of 4 minutes and 01 seconds for training and 41 seconds for testing of NND ensemble model. We present empirical evidence that our procedures can be superior in its forecasting performance when compared to other econometric forecasting methods.