Nonlinear Moving Horizon Estimation and Model Predictive Control for Buildings with Unknown HVAC Dynamics

TL;DR

This paper introduces a novel nonlinear moving horizon estimation and model predictive control framework for HVAC systems in buildings, utilizing minimal data and combining physics-based and neural network models for improved energy efficiency and occupant comfort.

Contribution

It develops an integrated approach using RC models and neural networks for online system identification and control, enabling scalable, data-efficient HVAC management without detailed building models.

Findings

Successful simulation validation of the proposed control system.

Achieved energy savings and occupant comfort improvements.

System operates effectively with only BMS data and minimal storage.

Abstract

We present a solution for modeling and online identification for heating, ventilation, and air conditioning (HVAC) control in buildings. Our approach comprises: (a) a resistance-capacitance (RC) model based on first order energy balance for deriving the zone temperature dynamics, and (b) a neural network for modeling HVAC dynamics. State estimation and model identification are simultaneously performed using nonlinear moving horizon estimation (MHE) with physical constraints for system states. We leverage the identified model in model predictive control (MPC) for occupant comfort satisfaction and HVAC energy savings and verify the approach using simulations. Our system relies only on building management system data, does not require extensive data storage, and does not require a detailed building model. This can significantly aid the large scale adoption of MPC for future occupant-centric control of grid-interactive buildings.