Yazdan Daneshvar - Department of civil engineering, Qazvin branch, Islamic Azad University, Qazvin, Iran.
Majid Sabzehparvar - Department of industrial engineering collage of engineering, karaj branch, Islamic Azad University, Karaj. Iran.
Seyed Amir Hossein Hashemi - Department of civil engineering, Qazvin branch, Islamic Azad University, Qazvin, Iran.

In this study, artificial neural networks, artificial neural network combination with genetic algorithm and neural network combination with Kalman filter were used to optimally model and control a real air conditioning system. Using the above methods, the system is first trained and after verifying the modeling accuracy, the capability of this modeling to predict the future conditions of the system is investigated. In addition to the subsystems investigated in both heating and cooling phases by mass and energy equations in Simulink simulated by Matlab software, the results of this section are finally compared with the optimal modeling results. The most important advantage of artificial neural network modeling over mass and energy equation modeling approaches is that it captures all the uncertainties and nonlinear properties of the air conditioning system due to the use of real data for modeling. It takes. Therefore, this method can optimize energy consumption in air conditioners by predicting the future conditions of the system and by precisely adjusting the time of turning on and off the main energy consuming equipment. The most important achievement of this research is more accurate and realistic modeling of the nonlinear air conditioning system.Comparing the methods used in the research for simulation methods using mass and energy equations, modeling using Bayesian trained neural network, artificial neural network modeling using MLP, modeling using neural network and genetic algorithm, modeling Using neural network and Kalman filter, the square error is equal to ۰.۰۰۶, ۰.۱۸, ۰.۰۵۶, ۰.۱۴۵۶ and more than ۰.۵, respectively.

HVAC control systems, Artificial Intelligence, Extended Kalman-filter, Genetic Algorithm, artificial ‎neural networks