Proposed hybrid ARIMA-ETS model for mid-term heat load prediction

Mid-term load forecasting is essential for scheduling operational tasks, optimizing energy consumption, and boosting system output. It requires capturing complex patterns as it spans from several weeks to months; therefore, the model must be able to capture short-term as well as long-term patterns in the dataset. Handling the dataset's temporal variation while maintaining the model's adaptability to evolving trends over a longer time horizon is a challenge. To overcome this challenge, a hybrid solution has been proposed to accurately determine 1-month and 3-months of hourly heat load prediction. The proposed hybrid model is intended to take advantage of the two forecasting models, namely ARIMA and ETS. The optimal values of the ARIMA model's autoregressive, moving average, and integrated components are evaluated, which allows the model to capture longer linear time horizons in the dataset accurately. Concurrently, the ETS model focuses on simulating errors, trend, and seasonal components in the dataset, and these components are also tuned to train an accurate hybrid model. A forward feature selection approach is employed to select the optimum number of features. The performance of another ten machine learning algorithms is compared to the accuracy of the proposed hybrid model. The proposed hybrid model showed accurate results for 1-month prediction with MAE = 3.54, MSE = 41.74, and RMSE = 6.45 and accurate 3-months of prediction results with MAE = 12.38, MSE = 216.52, and RMSE = 14.66; therefore, it could serve a productive guiding role in DHN energy distribution.