Abstract:  Moisture is the medium of various reactions inside the leaves during the processing of green tea, and it is an important factor affecting the quality of green tea. This studymeasured and recorded the moisture changes in each process of the tea processing process, and then studied and analyzed its dynamic changes. The results showed that the water loss process of fresh tea leaves was mainly concentrated in the process of fixation and drying. The process of spreading green water loss was less and the process was soft. The moisture was redistributed during the regaining process, and the tea was shaped by the rolling process, after/during whichthe moisture change was not obvious. Select the main dehydration processes in the two processing processes of fixation and drying, fix the drum speed, take temperature, initial moisture content and working time as input, and time moisture content as output, using BP neural network algorithm and support vector machine (SVM) algorithm to establish a moisture content prediction model for green tea fixation and drying. The established model was used for prediction, and the actual measured moisture content was compared and analyzed. The results showed that the R2 of the BP and SVM prediction models for the fixation were 0.99901 and 0.99932, respectively, while the R2 of the drying process were 0.99729 and 0.99786. The prediction model built by the SVM algorithm performed better and the prediction was more accurate. Compared with the drying process, the model had higher accuracy and better effect in the prediction of the moisture content of the fixation with higher moisture content and more obvious changes.

Key words: green tea processing, moisture content, BP, SVM, prediction model

摘要： 水分是绿茶加工过程中叶片内部各种反应的介质，是影响绿茶品质的重要因素。通过测定记录茶叶加工过程各个工序的水分变化，研究分析其动态变化规律。结果表明：茶鲜叶失水过程主要集中在杀青和烘干工序，摊青失水较少且过程柔和，回潮过程水分重新分布，揉捻过程进行茶叶塑形，水分变化均不明显。选择杀青和烘干这两个加工过程中的主要失水工序，固定滚筒转速，以温度、初始含水率以及工作时间作为输入，时刻含水率作为输出，利用BP神经网络算法与支持向量机（SVM）算法，建立绿茶加工杀青和烘干过程含水率预测模型。采用建立的模型进行预测，与实际测量的含水率进行对比分析和误差分析，结果表明对于杀青过程BP和SVM预测模型的R2分别为0.999 01和0.999 32，烘干过程R2分别为0.997 29和0.997 86，即SVM算法建立的预测模型表现更好，预测更加精确，并且相比烘干环节，模型对含水率更高、变化过程更明显的杀青工序的含水率预测精度更高，效果更好。

关键词: 绿茶加工, 含水率, BP, SVM, 预测模型