A Deep Architecture using Multiple TabNets for Environmental Prediction and Control in Smart Home

Abstract

A deep neural network model is proposed in this research for indoor environmental prediction and control in the smart home. It attempts to benefit directly from human experience by making use of imitation learning, a paradigm that is closely related to reinforcement learning. In imitation learning, the agent learns from real human experience. The research uses the state-of-the-art deep attentive tabular network architecture, which is an extension of deep neural networks. The tabular network, which is designed specifically to handle tabular data, is able to outperform all other machine learning algorithms in current use. The proposed model incorporates four such tabular networks. Promising results described here demonstrate how current developments in machine learning can be adopted effectively in HEMS related applications.