ON INTELLIGENT CONTROL OF RESIDENTIAL ENERGY SYSTEMS

Ninad K Gaikwad

Residential houses are increasingly adopting rooftop photovoltaic (PV) systems, battery energy storage, and electrically driven heating, ventilation, and air-conditioning (HVAC) systems, collectively introducing significant operational flexibility at the grid edge. When coordinated intelligently, this flexibility can be exploited by advanced control strategies to improve both economic efficiency and energy resiliency. Under normal on-grid conditions, the controllability of HVAC systems and the scheduling flexibility of PV and battery resources can be leveraged to reduce household energy costs by responding to time-varying electricity prices, and other grid signals. In contrast, the growing occurrence of extreme weather events has led to more frequent and prolonged power outages, during which this same flexibility becomes critical for maintaining residential energy resiliency. During such off-grid conditions, a house must remain habitable by sustaining thermal comfort and supplying critical loads despite limited energy resources and the high startup power requirements of air-conditioning systems. Rooftop PV systems combined with battery storage provide a clean and practical means to support both objectives; however, managing these resources effectively requires intelligent decision-making capable of balancing thermal comfort, critical load support, and energy availability. Therefore, there is a strong need for control frameworks that operate across on-grid and off-grid conditions, for individual houses as well as larger residential communities with heterogeneous distributed energy resources (DERs). This thesis develops a unified framework for designing and benchmarking intelligent controllers for residential energy systems at both single-house and community levels, operating under on-grid and off-grid conditions. The framework begins with the development of Model Predictive Control (MPC) formulations rooted in optimization principles, which act as intelligent controllers that manage PV and battery resources while maintaining thermal comfort and critical load support. Building on these formulations, equivalent reinforcement learning (RL)-based controllers are developed, in which their states, actions, and reward structures are inspired by the MPC formulations. The intelligent controllers are evaluated against two benchmarks: a baseline controller with no intelligence and a rule-based controller that incorporates limited logic-based decision rules but lacks optimization capability. These comparisons are carried out across multiple scenarios involving different community and DER configurations to assess performance in both on-grid and off-grid settings. To enable systematic testing and benchmarking of these controllers, we develop a scalable multi-house simulator that supports both single-house and community-level studies. The simulator allows flexible configuration of houses with different DER setups—such as PV + Battery, Battery-only, PV-only, or no DER—and integrates realistic weather and load data from the National Solar Radiation Database (NSRDB) and the Pecan Street dataset. It provides full control over simulation parameters, operating modes, and data pipelines, enabling closed-loop evaluations of MPC-based controllers and sample-based training and closed-loop evaluations of RL-based controllers within a unified environment. This framework facilitates fair and reproducible comparisons between optimization-based and learning-based approaches under diverse on-grid and off-grid scenarios. This work contributes a standardized and extensible testbed that integrates a high-fidelity simulator, data pipelines, and a suite of baseline controllers spanning MPC-, RL-, and logic-based approaches, capable of operating across diverse community, DER, and on-grid/off-grid scenarios. The proposed testbed enables reproducible benchmarking and provides a robust foundation for other researchers to develop, evaluate, and advance intelligent home energy management controllers, while also offering principled intelligent control design insights based on MPC formulations and MPC-inspired RL architectures under diverse scenarios.

ON INTELLIGENT CONTROL OF RESIDENTIAL ENERGY SYSTEMS

Files and links (1)

Abstract

Metrics

Details