Machine learning for optimal net-zero energy consumption in smart buildings

The goal of the study is to offer a data-based layout utilizing reinforcement learning for optimizing the energy consumption (EC) for one smart home (SH) using solar photovoltaic (PV) systems, energy storage systems (ESS), and SH devices. This method differs from current data-driven optimization techniques for the home energy management (HEM) system in the following ways: i) The proposed robust scheme is solved using the Columnand-Constraint Generation (CCG) approach in order to plan EC for each controllable device, along with the ESS charge and discharge, and ii) A deep neural network (DNN) predicts indoor temperature which affects EC of the air conditioner (AC). Through the integration of the CCG algorithm with the DNN scheme, the developed algorithm decreases the user energy cost while maintaining the desired level of satisfaction and efficiency features of the device. Simulated homes include a PV system, an AC, a washer machine, and an ESS using time-ofuse pricing which are all modeled by their digital twin model in a net-zero scheme. According to the outcomes, the suggested algorithm reduces energy costs by 12% compared to the current optimization approach. The proposed smart home system integrates error-handling measures to address uncertainties. It incorporates error margins, adaptive learning for model updates, fallback mechanisms, real-time sensor validation, and scenariobased optimization to enhance robustness in the face of inaccurate temperature predictions or unexpected events.