Self-learning fuzzy air handling system controller

Abstract

Modem air-conditioning commonly employ the 'central all-air system' and the variable air volume (VAV) system, in particular, is widely used everywhere around the world for energy conservation. Proportional-integral-differential (PID) control for air handling units (AHUs) is simple and straightforward but static fuzzy control is very often more robust, more energy efficient and faster in responding to changes due to executing expert knowledge. However, AHU control is highly dynamic where the system characteristics vary continuously. The static fuzzy rules are quite general in nature. They cannot be effective and optimal for all operating conditions because they are based on linguistic rules suggested by human experts. A new self-learning fuzzy controller has been designed in which the control policy is adaptable to changes in the control process and the environment. The controller can therefore always be operating at its optimal settings. The air handling system is modelled continuously by an artificial neural network. Although the process is quite intensive computationally, no system model needs to be assumed beforehand. Based on the minimisation of a performance indicator addresses both set-point errors and energy consumption, the self-learning ability continuously updates the defuzzification parameters of the controller. Computer simulation has revealed that the new self-learning fuzzy controller can achieve an even faster response rate and better energy consumption profile than those of the static fuzzy controller.