关键词: AdaBoost, 神经网络, 软测量, 集成算法, 增量学习

Abstract: To make up for the shortages of existing updating methods of soft sensor modeling, a new improved AdaBoost algorithm available to increment learnability was proposed by combining the increment learning with AdaBoost algorithm for ensemble learning. Then, an ensemble BP network was formed by integrating the new improved AdaBoost with BP neural network so as to develop a soft sensor model which will not only improve the accuracy by using single BP network but also ensure the updating ability for increment learning. A soft sensor model of molten steel composition in a 60-ton LF (ladle furnace) in Fu-Steel was developed according to the new method proposed, and its prediction result showed that it can meet the requirements for production.