基于变结构算法的神经网络预测方法及其应用  

Neural network prediction based on variable structure algorithms and its application

作　　者：刘铁男[1] 赵秀明[1] 

机构地区：[1]大庆石油学院电气信息工程学院,黑龙江大庆163318

出　　处：《大庆石油学院学报》2007年第1期91-93,共3页Journal of Daqing Petroleum Institute

基　　金：黑龙江省自然科学基金项目(TF2005-26);黑龙江省教育厅科研项目(10541014)

摘　　要：针对油田系统常规的建模和动态预测方法的局限性,采用多层动态前向网络作为油田系统的数学模型.研究基于变结构算法的多层动态前向网络预测模型与预测技术;分析梯度算法的不足,提出改进方案;给出单变量多层前向网络预测模型和相应的变结构梯度法,并且把该模型和算法推广到多变量系统,改进了油田系统的建模和预测方法.新方案在油田产量预测应用中颇有成效.In order to eliminate limitations of conventional modeling and dynamic prediction method,based on variable structure algorithms,multilayer dynamic forward networks are considered as the models of oilfield systems,and the prediction models and technique of multiplayer dynamic forward networks are studied.The deficiency of gradient learning algorithm is analyzed.An improvement scheme is given,so the algorithm performance is improved.Thus the method of modeling and prediction for oil field is renewed.In the application of the new scheme,excellent result has been obtained,which proves that the new scheme is very effective and useful.

关 键 词：多层动态前向网络 变结构算法 产量预测 多变量 非线性系统 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]