Multi-step Power Consumption Forecasting in Thailand Using Dual-Stage Attentional LSTM

Abstract

Our task is to forecast the next day’s power consumption in the half-hour interval for a total of 48 intervals. There are many studies that proposed models for power consumption forecasting problems but most of the previously proposed techniques focus on impact from different time step to power consumption, the importance of different features was not considered in these works. The Dual-Stage Attention-Based Recurrent Neural Network (DA-RNN) is the state-of-the-art in time series forecasting that considers both varying impacts from different time features for one-step-ahead forecasting. On the contrary, our work focuses on multi-step-ahead forecasting. In this paper, we aim to apply Dual-Stage Attentional LSTM for a multi-step-ahead power consumption forecasting problem. Experiments were conducted on Thailand’s power consumption data consisting of 5 control areas, weather data, and day type (weekday/weekend). We use RMSE and MAPE as evaluation metrics, the results showed that the Dual-Stage Attentional LSTM outperformed traditional models in both metrics.