基于主元神经网络的片烟松散回潮控制系统建模与优化  

Modeling and optimization of tobacco strips loosening and conditioning control system based on principal component neural network

作　　者：文广球 江婷 伍先 谢华坤 邓红伟 周成林 WEN Guangqiu;JIANG Ting;WU Xian;XIE Huakun;DENG Hongwei;ZHOU Chenglin(Lingling Cigarette Factory of China Tobacco Hunan Industrial Co.,Ltd.,Yongzhou 425006,Hunan,China;First State Technology(Hangzhou)Co.,Ltd.,Hangzhou 311202,China)

机构地区：[1]湖南中烟工业有限责任公司零陵卷烟厂,湖南省永州市425006 [2]首域科技(杭州)有限公司,杭州市311202

出　　处：《烟草科技》2024年第7期104-112,共9页Tobacco Science & Technology

摘　　要：为解决松散回潮工序传统PID控制模式存在出口含水率波动大、系统控制延时等问题,基于BP神经网络和主元回归方法提出一种主元神经网络出口含水率预测模型;以预测偏差最小为目标,建立基于遗传算法的松散回潮加水量寻优模型和实时反馈加水量修正机制,得到模型参数自适应的最优加水量。以湖南中烟工业有限责任公司零陵卷烟厂的“芙蓉王(硬)”牌卷烟配方原料为对象进行测试,结果表明:①预测模型预测的出口含水率与实际出口含水率平均绝对误差为0.14%,能较好地预测出口含水率变化趋势;②与传统PID控制模式相比,采用主元神经网络控制模式后,出口含水率过程能力指数均值提升65%,标准偏差均值降低37%。该技术可为提升松散回潮出口含水率控制精度和稳定性提供支持。As a solution to the large fluctuation of moisture content in output tobacco and time delay associated with the traditional PID control mode of loosening and conditioning procedure,a principal component neural network prediction model for moisture content in output tobacco was proposed based on BP neural network and principal component regression method.To minimize the prediction error,a genetic algorithm-based optimization model for water addition during loosening and conditioning and a real-time feedback mechanism for correcting the water addition were established to obtain the optimal water addition with self-adaptive model parameters.The prediction model was tested on blended tobacco material for“Furongwang(Hard)”brand cigarette in Lingling Cigarette Factory of China Tobacco Hunan Industrial Limited Corporation.The results showed that:1)The average absolute error between the predicted and actual moisture contents was 0.14%,the variation trend of moisture content in output tobacco could be well predicted.2)Compared with the traditional PID control mode,the principal component neural network control model increased the mean process capability index of moisture content in output tobacco by 65%and decreased the mean standard deviation by 37%.This technology is helpful for promotion of control accuracy and stability of moisture content in output tobacco from the loosening and conditioning procedure.

关 键 词：松散回潮 片烟 出口含水率 主元神经网络 遗传算法 预测控制 

分 类 号：TS432[农业科学—烟草工业]