Multidimensional Data-Driven Life Prediction Method for White LEDs Based on BP-NN and Improved-Adaboost Algorithm

Abstract

In view of the light-emitting diode (LED) and its life prediction, despite its currently wide use, IES TM-21-11 parametric life prediction method is incapable to extrapolate under multidimensional conditions (include working environmental conditions). This paper presents a multidimensional back propagation-neural network (BP-NN) based life prediction method which considers different driving currents and ambient temperatures. In this method, parameters such as temperature, electric current, initial chromaticity coordinates and initial luminous flux serve as the inputs, while life serves as the output. Since the traditional NN can easily get trapped in local minima and be affected by low precision, the BP-NN method is improved using Adaboost algorithm. The expected life predicted by the improved method is compared with that of traditional BP-NN and IES TM-21-11. The LED lamps of different power grades are compared for verification purposes. The results show that when predicting LED's lifetime, the improved method reduces the average relative error by on average by 54% compared with traditional BP-NN. However, the improved method takes 63.6% longer to operate, which requires the users to choose an appropriate model in accordance with particular operating conditions.