根据LED可靠性与相关参数的映射关系,建立拓扑结构为6-12-1的BP神经网络。以实测白光LED芯片的理想因子、结温、色温漂移等参数为输入量,以寿命为输出量,计算模型精度。研究结果表明,该模型有良好的外推能力及鲁棒性,可在短时间内成功预测LED寿命,神经网络训练结果相关系数为99.8%,检验组误差小于3%。

We proposed the 6-12-1 topology BP neural network according to the LED reliability and relevant element. The ideal factor, junction temperature, color temperature drift of the white LED chip and so on were measured as the input and the life as output to calculate the precision of the model. The model shows a good ability of extrapolation and robustness, and can predict the life of the LED in a short time. The linear correlation of ANN reaches 99.8%, and the inspection group error is less than 3%.