Study On The Incremental Learning Applied In The Intelligent Curing Systems Based On Electronic Nose

As one of the important research direction in machine learning, coming with the age of big data, incremental learning is considered as an important and effective solution to solve the problem of large data since it does not need to rely on all samples,only use and preserve the characteristics, concepts learned former or representative samples, can effectively achieve the sample reduction and learning new knowledge,has a unique advantage in dealing with memory limitation and repeated learning.Smell is one of the important characteristics of the material, E-nose(electronic nose) technology is a new nondestructive smell detection technology based on the bionic olfaction principle. Compare to the traditional chemical and optical detection method, the e-nose method is rapid, accurate, objective and convenient, has been a hot spot in field of bionic olfaction research.With the arrival of the "industry 4.0" times, intelligent has become the main development tendency of the future industry. While the intelligent research around intelligent baking currently is less and the baking process mainly relies on the human experience and automation of machine, a huge gap exists between the real unmanned intelligent automatic baking.As an importance part of the tobacco industry, the tobacco curing process which has the problem of low level intelligence and automation has been studied in this thesis in order to realize the automatic and intelligent curing of tobacco. Based on the self-developed curing control platform, using e-nose, camera and other collecting device to gather the information of tobacco including smell, image and moisture,exploring the adjusting rules of the tobacco features while curing, combining the curer's regulation towards the curing curve, a machine learning network was constructed to predict the curing parameters according the collected tobacco features.Furthermore, incremental learning methods have been studied to cope with the complex curing environment for various kinds of tobacco.As the predicting artificial neural network for tobacco is a kind of regression network, the incremental learning about support vector regression machine becomes the main content of this thesis. After analyzing the weaknesses of current incremental learning methods and characteristic of tobacco curing adding learning samples, the SOINN(Self-organizing incremental neural network) network was used to get thetopology node information and WSVR(Weighted support vector regression machine)was used to train the network in the intelligent curing system. This method can preserve the original data and weaken the bad influence of the general incremental model in incremental learning. Meanwhile, the problem of SOINN, such as data concept drifts has been considered, and the corresponding processing method was proposed. The algorithm was tested under artificial sine function data set, UCI data set and the measured data set.Finally, based on the existing intelligent curing control platform, an actual operation procedure of intelligent tobacco curing was designed, which provides reference for realization of intelligent curing.