Distributed model predictive control via Proximal Jacobian ADMM for building control applications

Abstract

This paper investigates a distributed model predictive control (DMPC) framework for building control applications. The proposed framework is general in that it can be easily customized to solve the dynamic optimization problem for a broad class of multi-zone buildings with relatively complex HVAC systems. The Proximal Jacobian alternating direction method of multipliers (ADMM), a recent variant of the traditional Gauss-Seidel sequential ADMM is employed and adopted to solve the centralized optimization problem, which ultimately leads to an agent-based parallel updating scheme with guaranteed convergence. A case study on the HVAC energy optimization of a multi-zone building is presented to show the effectiveness of the proposed method.