基于实数编码遗传算法的神经网络成本预测模型及其应用.pdf

第 21 卷第3 期 控制耀论与应用 Vol.21 No.3 2004 年6 月 Control 文章编号: 1000 - 8152(拙)4)03 倍。423 - 04 基于实数编码遗传算法的神经网络成本预测模型及其应用 刘 威，李小平，毛慧欧，柴天佑 (东北大学自动化研究小心，辽宁沈阳 11α泊的 摘要:在生产过程中，影响产品成本的因素多而复杂，回素之间相互影响，存在桐合现象，因此准确预测成本是 个熏要又难以解决的问扭.通过遗传算法(Genetic Algoritlun) 与误盖反向传播(自ror Back Propagatíon) 神娃网络相 结合，提出了用实数编阔的自适应变异遗传算法训练神经网络权重的混合算法，避免了传统神经网络易陷入局部 极小的缺点.以矩阵形式表示产品成本组成，建立了产品成本组成模型，以此为基础建立了考虑成本因素之间互相 影响的神经网络产品成本预测模型，并成功应用于某钢铁企业产品成本的预测，提高了预测精度. 关键词:成本预测~ ;遗传算法;神经网络;实数编码 中图分类号:τ刊 文献标识码:A Neural network cost prediction model based on real幽 coded genetic algorithm and its application LlU Wei , LI Xiao翩 ping ，如1AO Hui-ou , CHAI Tian-you (Re能耐h Center of Automation ,North倒stem University. Shenyang Liaoning 11αlO4 .α甘na) Abstract: In production process , many complex factors which influence cost affect 创蚀 。也.er and the coupling phe幡 nomenon exis衍， 80 it is im阴阳lt and difficu1t to pr回ict 阳 C08t. By combining genetic a1goritlun with error back propagation neural network , a hybrid a1goritlun that trained neura1 network weight by real心创ed adaptive mutation genetic a1goritlun is 伊b sented. and it overcomes the disadvantage that 阳ditional neural network is 臼syωfa11 into 1∞a1 minima.ηle product cost cα孙 position is expre随时 by matrix ，加防咽uct cost composition mαlel is established ,on 也巳 basis of the mα制， the prc咖ct cost pre- diction model based on neur百1 network is established. and the interactions 刷nong cost factors are taken into aωoun1. Fur也ermore ， the model is sucωssfully applied to cost prediction in some iron and steel enterpr啊，and 仰伊时iction p阳ision is improved. Keywo时S: cost 阴叫iction; genetic algori出m; neur丑I network; real-c