Prediction of the artificial illuminanceusing neural networks

This study deals with calculation of the illuminance preferences in offices. For this purpose, a neural network model is formed that predicts the artificial illuminance preferences of the office workers by taking 'daylight illuminance level', 'time of day' and 'initial level of illuminance as input data. A four-layered feed-forward back-propagation neural network having five neurons in each hidden layer is used. The data used in this study are taken from the experiments recorded in three identical real cell-offices at the 2nd floor of Philips Design and Application Centre in Eindhoven during 1992-1993. For 1351 data, the difference between the experimental and neural network-based predicted illuminance levels are compared and it is seen that the probability of absolute errors below 150 lux is 65%. Similarly, the probabilities of absolute errors below 300 lux and 450 lux are 80% and 88%, respectively. These results are encouraging for designing lighting control systems.