Bayesian Optimization-Based Hyper-parameter-Tuned Neural Network Regression for Smart Home Energy Consumption Modelling Using Weather Information

This paper presents an illustrative study on the predictive analysis of household energy consumption patterns in the context of hybrid energy-powered smart homes having wide variety of appliances. Leveraging the proliferation of smart sensors and the integration of the Internet of Things (IoTs), our research focuses on the integration of weather information in predicting the user behaviour for different household appliances and the total energy consumption. By employing nonlinear regression modelling using artificial neural networks (ANNs), we develop black-box predictive models aimed at predicting total energy usage given the user behaviour patterns and weather information in order seamlessly design an optimized smart home energy system. We use the adjusted R2 or the coefficient of determination as the performance metric in deciding the best fitted model. In such data-driven modelling, the regression methods need to be further fine-tuned over a wide choice of hyper-parameters which significantly affect the accuracy, speed, and consistency of the machine learning (ML) models. We here benchmark the hyper-parameter optimized model with a simpler hyper-parameter optimized linear model to show the efficacy of the nonlinear model over the linear counterpart. The reported research endeavours to advance the paradigm of sustainable urban living by leveraging data-driven ML models for hybrid energy-powered smart homes, with the goal of fostering energy-efficient and ecologically responsible urban environments.