Construction Of A Psychophysical Scale Of Visual Comfort of Lighting Based On A Neural Network: preparation Of The Experiment

Abstract

One of the important questions in lighting engineering is to determine the sensation of discomfort from lighting installations. There is no unified psychophysical scale for assessing the visual comfort of lighting (VCL) at any arbitrary distribution of luminance in space. This article considers a mathematical model of the scale based on a neural network (NN) as an ‘expert’ that trained to determine the comfort of perception of lighting depending on the light source’s luminance and background. The experimental data obtained at the lighting engineering department of the National Research University “MPEI” were used to train the NN. The experiment results presented in this article are consistent with the numerical scale for estimating the VCL proposed by Lakiesch and Holladay. A new model allows predicting the sensation of VCL with an accuracy of up to 70 %. This work allows formulating criteria for NN’s input and output parameters to choose a metric for evaluating NN’s performance, such as the confusion matrix, ROC curves, and a metric, such as the probability distribution for each sensation depending on the input parameters. It clearly follows, that amount of initial data is not allowed to make a final conclusion. One more experiment required considering the algorithm used to calibrate the experimental installation, instructions for observers, and the obtained results processing.