An Optimization-Based Approach for Facility Energy Management with Uncertainties

Abstract

Effective energy management for facilities is becoming increasingly important in view of rising energy costs, the government mandate on reduction of energy consumption, and human comfort requirements. This paper presents a daily energy management formulation and the corresponding solution methodology for HVAC systems. The problem is to minimize the energy and demand costs through control of HVAC units while satisfying human comfort, system dynamics, load limit constraints, and other requirements. The problem is difficult in view of the facts that the system is nonlinear, time-varying, building-dependent, and uncertain and that the direct control of a large number of HVAC components is difficult. In this paper, HVAC setpoints are control variables developed on top of a direct digital control (DDC) system. A method that combines Lagrangian relaxation, neural networks, stochastic dynamic programming, and heuristics is developed to predict system dynamics and uncontrollable load and to optimize the setpoints. Numerical testing and prototype implementation results show that our method can effectively reduce total costs, manage uncertainties, and shed the load; is computationally efficient; and is significantly better than existing methods.