Surrogate-Assisted Adaptive Evolutionary Algorithm for Efficient HVAC Control in Home Energy Management Systems

The residential energy sector has seen one of the greatest recent increases in energy prices. With energy bills expected to keep rising, homeowners need efficient systems that can reduce both energy costs and overall consumption. As Heating, Ventilation, and Air Conditioning (HVAC) systems constitute the largest energy consumers in residential buildings, optimizing their control is essential. However, traditional control methods are fundamentally reactive and often fail to provide a comprehensive home energy management strategy that jointly coordinates HVAC operation with onsite Photovoltaic (PV) generation, Energy Storage Systems (ESS), and grid interactions (including selling energy back to the grid), which can lead to sub-optimal energy consumption. This paper proposes a novel Online Surrogate-Assisted Adaptive Differential Evolution (SA-ADE) algorithm integrated within a Model Predictive Control (MPC) framework for holistic Home Energy Management Systems (HEMS) that reduces energy costs while maintaining occupant comfort. The holistic framework co-optimizes HVAC operations with onsite Photovoltaic (PV) generation and Energy Storage Systems (ESS). The framework integrates a hybrid machine learning approach, using Gaussian Processes (GP) and Stochastic Long Short-Term Memory (StochLSTM), to accurately model indoor temperature dynamics and forecast HVAC loads. To reduce the computational complexity of solving the proposed Mixed-Integer Nonlinear Programming (MINLP) problem, we develop a XGBoost-based surrogate model for SA-ADE. This surrogate approximates the fitness function, reducing computational runtime by over 90% while maintaining high optimization fidelity. Extensive high-fidelity simulations using EnergyPlus across five distinct climate zones demonstrate that the proposed approach achieves average energy savings of 6.56% and reduces occupant temperature comfort violations by 61.66% compared to state-of-the-art baselines.