Model Predictive Control, Optimal Control, Distributed Methods, Energy Efficiency, Thermal Comfort
Model based predictive control (MPC) is increasingly being seen as an attractive approach in controlling building HVAC systems. One advantage of the MPC approach is the ability to integrate weather forecast, occupancy information and utility price variations in determining the optimal HVAC operation. However, application to largescale building HVAC systems is limited by the large number of controllable variables to be optimized at every time instance. This paper explores techniques to reduce the computational complexity arising in applying MPC to the control of large-scale buildings. We formulate the task of optimal control as a distributed optimization problem within the MPC framework. A distributed optimization approach alleviates computational costs by simultaneously solving reduced dimensional optimization problems at the subsystem level and integrating the resulting solutions to obtain a global control law. Additional computational efficiency can be achieved by utilizing the occupancy and utility price profiles to restrict the control laws to a piecewise constant function. Alternatively, under certain assumptions, the optimal control laws can be found analytically using a dynamic programming based approach without resorting to numerical optimization routines leading to massive computational savings. Initial results of simulations on case studies are presented to compare the proposed algorithms.