PDQN: User Preference-Based Charging Station Recommendation

With the advent of consumer electronics, Electric Vehicles (EVs) are gradually becoming representative carriers for the next generation of consumer electronics. EVs are gaining market share compared to traditional fuel vehicles. This paper works on charging station (CS) recommendation to improve the charging experience. However, there is uncertainty that EV users follow exactly the CS recommendation from operation system, and this thereby degrades the performance of CS recommendation system. To address this issue, we extend the Markov Decision Process (MDP) model by integrating user preference (e∙, prefer the nearest CS). Then, we propose a neural network-based framework to solve the established model, in which we use a Long Short-Term Memory (LSTM) module to capture user preference features and an attention module to establish the relationship between the preference features and system state features. Based on these modules, a user Preference-based Deep Q Network (PDQN) is designed to solve the established CS recommendation model. Our simulation results demonstrate that our proposition can bring better performance in terms of average charging time and rewards, by comparing it with the nearest mechanism, the minimum queuing mechanism and the DQN mechanism without preference.