A robust deep feature learning approach for personalized residential electricity plan recommendation

The deregulation of the electricity market has led to the proliferation of diverse electricity retail plans, posing substantial information filtering challenges for residential users. In the meantime, the digitalization of electricity systems has driven the development of electricity plan recommendation services with advanced metering infrastructure (AMI) data to offer personalized decision support for all residential customers in selecting suitable electricity plans. This paper proposes a robust deep feature learning-based Electricity Plan Recommender System (RDFL-EPRS), designed to address the complexity and reliability of the multiplex features significantly influencing the recommendation performance for residential electricity plan selection. The system leverages both user-input and electricity plan data, employs multiple imputation on stacked denoising autoencoders to address and rectify missing or unreliable user-input features, and incorporates various deep learning techniques in the recommendation model to accommodate the diverse types of input features. This model learns intricate feature interactions and maps non-linear relationships with electricity plan ratings, ultimately generating a recommendation list for the Top-N most appropriate electricity plans. Extensive simulations validate RDFL-EPRS for significant improvements in missing/abnormal value imputation and electricity plan recommendations. Compared to state-of-the-art techniques, the proposed system provides more precise and more robust recommendations, supporting target users in making reliably informed decisions. © 2025 The Authors. Published by Elsevier B.V.