Experimental Study of the Model Predictive Control for a Residential Split Air Conditioner

Abstract

Buildings consume 60% of the total building energy consumption for space heating and cooling purposes, and even more with inefficient and outdated air-conditioners (AC) - which are most common in developing countries. Although there are already conventional control strategies like ON/OFF control and PID in practice, their performance worsens once the building’s complexity increases and they cannot cope-up with the stochastic nature of the inside and outside environment. So one of the most efficient and preferred control strategies in complex systems like a building would be model predictive control (MPC). This study uses MPC to predict and control a 1.5 tonne inverter-based air conditioner. A cascaded control strategy - including various control methods, i.e., hysteresis, bang-bang, MPC, and an adaptive MPC, is implemented over the inbuilt thermostat-based control. MPC minimizes the setpoint tracking error and optimizes real-time thermal comfort by generating optimal temperature setpoints. An automated hardware setup is developed using Arduino Mega-2560, Raspberry Pi 3 B+ model, calibrated temperature, and humidity sensors that set the optimal setpoints to the air-conditioner via an infrared remote controller. Both MPC and adaptive MPC are able to maintain thermal comfort strictly within the desirable range, i.e., ±0.5 PMV throughout the experimental day, with thermal comfort hours of 24 hrs. and 23.67 hrs., respectively. On the contrary, with regular thermostat-based control, the thermal comfort is not maintained at any manually set temperature setpoint. With an adaptive MPC, energy saving of more than 7.06% is achieved as compared to the air-conditioner’s regular operation, running at a constant 24∘C - that is taken as the baseline for this study.