Dead-band vs. machine-learning control systems: Analysis of control benefits and energy efficiency

Abstract

In residential and commercial buildings, thermostat controllers have been typically utilized to maintain room temperature near desired set-point. Recently, advanced computing and statistical technologies, such as Fuzzy Inference System (FIS) and Artificial Neural Network (ANN) algorithms, were introduced to complement the control performance for optimal energy use in building thermal systems. However, most schemes were developed to control fuel use or fan motor speed in a plant or a system, and showed some disadvantages to immediately respond to sensitive changes in thermal demands for a zone scaled level.This paper introduces heating energy models capable of controlling the amount of supply air and its temperature simultaneously, and the FIS and ANN algorithms are developed to control the optimal supply air conditions for a heating season. Both the FIS and ANN models are compared to thermostat controllers with 4-step dead-band setups from normal to sensitive levels. The sum of errors, caused by the difference between desired set-point and controlled room temperatures, and the amount of energy supply are used to define control precision and energy efficiency of the control models. From the simulation results, the machine-learning based ANN controller averagely reduces control errors by 88% and mitigates increases in energy consumption by 2% in comparison with thermostat on/off controllers. The control system can be effective when various sensitive settings are required as a type of buildings and rooms without an excessive increase in energy use.