基于循环神经网络的物料采购需求智能化挖掘方法  

An Intelligent Mining Method of Material Purchasing Demand Based on Cyclic Neural Network

作　　者：李咏凯 连莲 刘译聪 赵朋 LI Yongkai;LIAN Lian;LIU Yicong;ZHAO Peng(Material Branch of State Grid Jibei Electric Power Co.,Ltd.,Beijing 100075,China)

机构地区：[1]国网冀北电力有限公司物资分公司,北京100075

出　　处：《微型电脑应用》2022年第12期111-114,共4页Microcomputer Applications

摘　　要：为了最大程度保留有价值信息,提升企业对物料采购价格需求挖掘能力,降低挖掘误差,提出基于循环神经网络的物料采购需求智能化挖掘方法。利用小波分析方法预处理初始需求序列,多尺度分解初始序列,获取若干高、低频序列;利用相空间重构多尺度序列高、低频序列,将重构序列引入循环神经网络模型;经模型训练重构数据后,构建物料采购需求智能化挖掘模型,实现物料采购需求智能化挖掘。试验结果表明,该方法能够有效保留初始序列中的价值信息,训练误差和测试误差平均值为3.02%和2.61%,绝对误差和相对误差平均值为0.58%和3.82%,具有较好的实际应用价值。In order to retain valuable information to the greatest extent, improve the ability of mining material purchase price and reduce mining error, an intelligent material purchase demand mining method based on RNN is studied. The wavelet analysis method is used to preprocess the initial demand series and decompose the initial series to obtain some high and low frequency series, and the phase space is used to reconstruct the high and low frequency series. Experimental results show that the method can effectively retain the value information in the initial sequence. The mean values of training error and test error are 3.02% and 2.61%, and the mean values of absolute error and relative error are 0.58% and 3.82%, respectively. It has better practical value.

关 键 词：循环神经网络 物料采购 采购需求 智能化挖掘 小波分解 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]