Lam H. N.
Year:
1995
Bibliographic info:
Building Simulation, Madison, USA, 1995, p. 151-157

Conventional control schernes for air condítioning systems are in lack of the capabilities to adapt to a changing enviroranent and to optimize against given criteria. In this paper, a methodology is presented, which employs a classifier system with genetic algorithm to enable an air-conditioning controller to learn from its own experience the best control strategy against a given performance evaluation scherne. Solar insolation and outdoor air temperature are chosen to be the input environmental variables of the classifier system which utilizes a simple genetic algorithm to formulate the optimal control rule based on Fanger's thermal comfort index of predicted mean vote (PMV). A split-unit air conditioning system with a variable-speed compressor is used for experimental testing of the genetics-based rule learning system. Preliminary results obtained are encouraging.