基于反向传播神经网络的注塑模具用零件报价模型  

A Study on Injection Mould Part Quotation Model Based on Back Propagation Neural Network

作　　者：柳伟[1] 杨超[1] LIU Wei YANG Chao(National Engineering Research Center of Die ＆Mold CAD, Shanghai Jiao Tong University, Shanghai 200030, Chin)

机构地区：[1]上海交通大学模具CAD国家工程研究中心,上海200030

出　　处：《上海交通大学学报》2017年第10期1207-1213,共7页Journal of Shanghai Jiaotong University

摘　　要：针对诸多通用报价方法在注塑模具用零件价格评估过程中存在的效率与精度等问题,提出了基于反向传播(Back Propagation,BP)神经网络的多专家报价模型.利用遗传算法优化BP网络权值和阈值抵御随机参数引起的局部最优解,同时对报价误差的高斯分布规律进行分析以确定理想的拓扑结构.通过多专家模型给出的多个报价,提出了价格评估算法筛选合理报价,避免单一模型报价的随机性.经过验证,该方法有较高的准确性和效率,报价平均误差约为6.90%.此报价模型降低了传统报价的难度,并提高了其稳定性.General part quotations for injection mould often hold shortcomings such as low efficiency and poor accuracy. Aiming at this issue, a multi-expert quotation framework is proposed based on back propagation （BP） neural network （NN） in this paper. Weights and thresholds of the network are optimized by using generic algorithm which can avoid to fall into a local optimum solution. Meanwhile, an ideal topological structure of NN is decided by studying the Gaussian distribution of errors. A rotational quotation is thereby achieved after synthetical consideration of the evaluating results by synthesizing multiple experts. It is proved that the method greatly eliminates the randomness of a single quotation. Experiments exhibit a preferable performance, in which the average error of quotations is about 6.90 %. This scheme reduces the difficulty of traditional quotation and enhances its robustness.

关 键 词：注塑模具用零件 价格评估 多专家模型 神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]