Boosting Latent Inference of Resident Preference from Electricity Usage - A Demonstration on Online Advertisement Strategies

Lo Pang-Yun Ting, Po-Hui Wu, Jhe-Yun Jhang, Kai-Jun Yang, Yen-Ju Chen, Kun-Ta Chuang

Abstract: The electricity demand has increased due to the rapid development of the digital economy. The mechanisms of demand-side management are thus proposed to reduce the consumption while electricity companies forecast the appearance of excessive peak load which may incur the instability of electrical grids. However, DSM solutions are generally devised as the way of compulsively controlling home appliances but the interruption is not a pleasurable mechanism. To address this issue, we figure out an advertising strategy based on the residential electricity data acquired from smart meters. By recommending preference-related coupons to residents, we can induce residents to go outside to use the coupon while helping the peak load reduction with pleasure, leading to the win-win result between users and electricity companies. In this paper, we propose a novel framework, called DMAR, which combines the directed inference and the mediated inference to infer residents’ preferences based on their electricity usage. Through experimental studies on the real data of smart meters from the power company, we demonstrate that our method can outperform other baselines in the preference inference task. Meanwhile, we also build a line bot system to implement our advertisement service for the real-world residents. Both offline and online experiments show the practicability of the proposed DMAR framework.