Leveraging Deep Q-Learning to maximize consumer quality of experience in smart grid

Abstract

The smart grid system has addressed the problems of traditional power grid by not only meeting energy demand in real-time but also limiting its wastage. The two key objectives of a smart grid system are to ensure a higher Quality of Experience (QoE) for consumers and to reduce consumer costs using dynamically varying pricing concepts. However, these two objective parameters oppose each other as maximizing the consumer QoE requires the availability of sufficient electric power at any given time, which in turn increases power purchase cost. In this paper, we introduce an efficient power management system architecture of a smart grid and develop an Optimal Energy Allocation and Prediction system based on Deep Q-Leaning, namely OEAP-DQL, that brings a trade-off between the two. The developed OEAP-DQL system is a multi-objective linear programming (MOLP) problem that predicts consumer electricity demand by exploiting four different weighted and regressive moving average forecasting methods in the action space to accurately capture dynamically varying customer demand behaviors. Furthermore, iterative exploitation of multiple learning methods decreases forecasting error and intelligent stored power management helps the OEAP-DQL smart grid operator (SGO) to enhance its profit. The results of our simulation experiments show that the OEAP-DQL system outperforms the state-of-the-art works in terms of QoE and cost.