基于TrAdaBoost算法的近红外光谱模型传递研究

随着近红外光谱检测仪种类的增多,不同仪器间的校正模型存在无法共享问题,可利用模型传递解决.以食用油为研究对象,在主机上建立油酸质量比的极限学习机校正模型,利用迁移学习中的TrAdaBoost算法把主机模型传递到从机上,探讨标准化样品数量对模型传递效果的影响,并与直接标准化算法、缺损数据重构算法和极限学习机自编码器的模型...

Near Infrared Spectroscopy Calibration Transfer Based on TrAdaBoost Algorithm

With more and more types of near infrared spectroscopy detectors, the inability to share calibration models between different instruments has become the main problem that limits its application, and calibration transfer has become the key to solve this problem. Taking edible oil as the research object, the extreme learning machine model of its acid value on the master instrument was established. And the TrAdaBoost algorithm in transfer learning was used to transfer the master model to the slave model, and the dependence of calibration transfer on the number of standardization samples was explored. It was also compared with the direct standardized, missing data recovery and transfer via extreme learning machine auto encoder method. The results showed that the predictive power of the slave samples after the TrAdaBoost calibration transfer algorithm was most effective and very close to the predictive value of the master sample-master model. The R2 of the validation set was increased from 0.489 to 0.892, the root mean square error of prediction (RMSEP) was reduced from 4.824mg/g to 0.267mg/g. Specifically, the model effect was almost independent of the number of standardized samples. The next degree of effect was the transfer via extreme learning machine auto-encoder method algorithm (TEAM), the missing data recovery algorithm (MDR) and direct standardized algorithm (DS) in decreasing order, respectively. It was shown that the TrAdaBoost can be effectively applied to the calibration transfer, and it can realize the communication between different spectroscopic instruments, which provided an idea for the calibration transfer.