Research on Soft-sensing Method of Water Content of Fixed Tea Based on Improved Clustering

Abstract

The water content of tea leaves is an important process parameter of fixation of tea, existing measurement method for water content of tea is not suitable for on line measurement because of the limits of measurement speed, sample handling and cost. So, the water content of fixation leaves can only be evaluated by sense organ and be controlled by setting process parameters according the experience. In the process of removing tea activity,in order to predict the water content by the process parameter of fixation, the process of fixation  in a cylinder is analyzed and the parameters having effect on the water content of tea is found in this paper.And 350 groups of production scene data including the water content of tea leaves and the process parameters of fixation were collected, while the 300 groups of them are used to cluster with K-means algorithm, linear model is built in each class with recursive least squares. Taking the remaining 50 groups of data as test data, to calculate the water content of tea leaves with the linear model of the class in which the clustering center is nearest to the test data in distance, the result is compared with the measured value. The tests indicate that the MSE of the soft-sensing Modeling is less than 0.02 and the maximum error is 0.0463.