Wasserstein generative adversarial networks-based photovoltaic uncertainty in a smart home energy management system including battery storage devices

Abstract

Rooftop photovoltaic (PV) power generation uncertainty is one of the prominent challenges in smart homes. Home Energy Management (HEM) systems are essential for appliance and Energy Storage System (ESS) scheduling in these homes, enabling efficient usage of the installed PV panels' power. In this context, effective solar power scenario generation is crucial for HEM load and ESS scheduling with the objective of electricity bill cost reduction. This paper proposes a two-step approach, where a machine learning technique, Wasserstein Generative Adversarial Networks (WGANs), is used for PV scenario generation. Then, the generated scenarios are used as input for the HEM system scheduler to achieve the goal of cost minimization. The generated solar energy scenarios are considered in a single household case study to test the presented method's effectiveness. The WGAN scenarios are evaluated using different metrics and are compared with the scenarios generated by Monte Carlo simulation. The results prove that WGANs generate realistic solar scenarios, which are then used as input to a Mixed Integer Linear Programming (MILP) problem aiming for electricity bill minimization. A 41.5% bill reduction is achieved in the presented case study after scheduling both the load and ESS, with PV fluctuations taken into account, compared to the case where no scheduling, PV, or ESS are considered.