【摘要】 根据LED芯片光谱的质心波长和半高全宽，构建了采用人工神经网络预测LED结温的方法。采用LED热阻结构分析系统测量不同衬底温度、不同电流下的结温，同时采用光谱仪测量相对光谱分布，选择GaN基芯片发出的蓝色光谱分布为研究对象，计算其质心波长和半高全宽，将质心波长、半高全宽和驱动电流作为输入参数，结温作为输出参数，构建人工神经网络。最后用该模型预测LED发光时的实际结温。研究表明，在采用大量数据学习后，本方法精度可以达到1.23℃，满足实际工程需求。此外，本方法不需要接触LED光源，准确性较高，具有明显的技术优势。更多还原

【Abstract】 A new non-contact approach to predict LED junction temperature using artificial neural network based on centriod wavelength and full width at half maximum(FWHM) of the LED chip was presented. LED junction temperature under different substrate temperatures and different currents was measured by using LED thermal resistance analysis system,and the corresponding spectral distribution was measured by a spectrometer. The blue spectral distribution of the LED chip was selected as the research object,from which centroid wavelength and FWHM were calculated. Then,these two parameters and the driving current were taken as input parameters,the junction temperature was taken as output parameters to construct an artificial neural network. Finally,the model was used to predict the actual junction temperature of LED. By learning with a large amount of measured data,the measurement error of this new network reached 1.23℃. This method demonstrates significant technical advantages of non-contact and high accuracy.更多还原