Apply reinforcement learning as model selection to predict power consumption

Electricity forecasting is critical to a country's economy, particularly in developing countries. Accuracy in power forecasting is essential to support growth in electricity demand, provide adequate power planning, and prevent power system outages. However, the accuracy of available forecasting methods greatly depends on the characteristics of the data and the nature of the problem. As a result, determining which forecasting approach is optimal for a given scenario is challenging. This study presents a novel approach using Double Deep Q-network (Double DQN) as a model selection to specify the most accurate forecasting model for each time point. The monthly peak electricity consumption in Thailand from 2003 to 2017 provided by the Electricity Generating Authority of Thailand (EGAT) is used to illustrate and verify the model. The proposed model selection using Double DQN surpasses all the individual models in terms of prediction accuracy calculated by mean squared error (MSE) and provides a method to leverage the strength of each forecasting model.