MPC-trained ANFIS for control of MicroCSP integrated into a building HVAC system

Mohamed Toub, Mohammed V University of Rabat
Mahdi Shahbakhti, Michigan Technological University
Rush D. Robinett III, Michigan Technological University
Ghassane Aniba, Mohammed V University of Rabat

©2019 AACC. Publisher’s version of record:


This paper presents the design of an easily implementable rule-based controller that can minimize the electrical energy consumption of a building heating, ventilation, and air-conditioning (HVAC) system integrated with a microscale concentrated solar power (MicroCSP) system. A model predictive control (MPC) scheme is developed to optimize Mi-croCSP electrical and thermal energy flows for HVAC use in a building. Despite its attractiveness regarding energy savings and thermal comfort satisfaction, MPC requires high computational resources and can not be easily implemented on the common low-cost HVAC controllers available in the market. To cope with these issues, two MPC-trained adaptive neuro-fuzzy inference system (ANFIS) models are designed to control the building HVAC with MicroCSP. Simulation results exploiting real operation data from an office building at Michigan Technological University and our newly purchased MicroCSP are presented. It is shown that the resulting controller can reproduce the MPC reasoning and performance while being simpler and much more computationally efficient.