毛精纺前纺工艺参数重要性的BP网络定量评价法.doc

毛精纺前纺工艺参数重要性的BP网络定量评价法 刘 贵，于伟东1,2 东华大学 纺织材料与技术实验室,上海　201620; 武汉科技学院 纺织与材料学院,湖北 武汉　430073) 摘 要（小五黑体）在BP神经网络建模技术的基础上,提出利用神经网络输入层与输出层之间的网络权值及其分布来求各输入参数重要程度的方法。将采集到的毛精纺企业前纺工艺参数运用BP神经网络分别建立了粗纱CV值和粗纱单重的预测模型。结果表明：所建模型的平均相对误差都低于3%；采用样本数据验证,其预报值与实测值间的相关系数都高于0.95。对所建模型的网络权重进行提取,分别计算出13个输入参数对粗纱CV值和粗纱单重的重要性,挖掘出显著而有效的参数。经对比认为，BP网络法比多元回归显著性分析(MRSA)更为精准，可用于对实际生产加工的预报和控制。（小五宋体） 关键词（小五黑体）毛精纺；前纺工艺参数；模型；BP神经网络；定量评价法（小五宋体） 中图分类号（小五黑体）： TS 131.9（小五宋体） 文献标志码（小五黑体）： Quantitative evaluation method for the significance of worsted fore-spinning parameters based on BP neural network LIU Gui1，1,2（五号） (1.Textile Materials andTechnology Laboratory Donghua University, Shanghai　201620,China; 2.Department of Textiles and Materials, Wuhan University of Science and Engineering, Wuhan, Hubei　430073,China） Abstract Based on BP neural network model technology, a new approach was developed and applied to appraise the input parameters′significant degree through the weightiness and its distribution between the input and output layer. Usingthe fore-spinningworking procedure data gathered fromtheworsted textiles enterprise, the roving unevenness and weight prediction models were established respectively. The results indicated that the models′mean relative errors are all less than 3%; the correlation coefficientR2between the prediction value and the actual are all more than 0·95. Using the weightiness extracted fromthe established models, the 13 input parameters′significance to the roving unevenness and weight were calculated respectively, and the remarkable and effective parameters are excavated out. Meanwhile contrasting to the multivariate regression significance analysis (MRSA), the BP neural network method is more exact than MRSA and can be used in the forecast and control of the actual produce and manufacture (小五) Key words Double glow; Artificial neural network; Prediction model (小五) （正文五号宋体,单倍行距）对精毛纺织厂而言,前纺工序作为整个加工过程的第一环节,其加工质量对后道工序将产生直接影响。根据实际生产经验,细纱的条干不匀率和细纱机的断头率相对于末道粗纱的质