Neural Estimator Automatic Fluorescent Daylight Control System

Abstract

The daylight control system represents an electric light system used in office or design laboratory applications. The system tries to maintains constant the illuminance level on the working plane even the daylight contribution is variable. From other point of view the daylight control system is the lighting system that compensates the daylight variation in a room (office, design laboratory). The importance of this type of lighting system is that it satisfies the following requirements: user visual comfort and electrical energy savings. Considering these requirements the lighting system has to be implemented such an automatic control system with negative feedback. The behavior of the automatic lighting system will depend mainly on the controller behavior. In the present paper, a feed-forward artificial neural network (FANN) was chosen to control the lighting process using the Control by Estimation Iterative Algorithm. Due to the control strategy for a stable behavior of the automatic lighting control system without or with acceptable overshoot (regarding the control system step response) the learning rate of the FANN needs to have very small values and in a short range. To remove this shortcoming in present paper is proposed a modified learning error which allows the learning rate to have a wider range of values for which the automatic lighting control system has a good behavior. Also, is proposed a new way that the user can modify the speed reaction of the automatic control system regarding the daylight changes.