摘要
神经网络预测的优点是预测指标与影响因素指标的历史拟合好,微分模拟预测的特点则是预测过程中更加注重预测指标的自身变化趋势。两者有机结合,首先将微分模拟得到的油田产量与其影响因素的输入输出关系视为时变系统,再把BP神经网络引入到微分模拟参数识别中,建立具有时变特征的功能模拟预测新方法。该预测模型中的参数随时间变化,具有自适应性;能在神经网络训练过程中通过变学习率的方式解决其与微分模拟胶合过程可能出现的不收敛问题;对中长远预测有更好的效果。最后将这一新方法应用于国内某油田的产量预测中,经软件计算,预测结果与实际的吻合程度明显高于其它几种预测方法。
One advantage of neural network forecast is the good historical matching between forecast indices and influence factors indices,while the differential simulation forecast pays more attention to the change trend of forecast indices.In this paper,these tow methods are organically combined.At first,the input-output relation between oilfield output and their influence factors is viewed as a time-varying system,then the BP neural network is introduced to parameter identification of differential simulation to obtain a new forecast method of functional simulation based on time-varying system.This new forecast model owns good self-adaptability since its parameters change with time.Moreover,it has better effect in mid-long term forecast because the non-convergence problem appeared in the coupling process between it and differential simulation can be overcome in the training process of neural network by variable learning rate.In the end,a practical output forecast case in a certain oilfield in China is given.The computational results show that the forecast is in good agreement with the reality,even much better than the results obtained by other forecast methods.
出处
《西南石油大学学报（自然科学版）》
CAS
CSCD
北大核心
2008年第4期181-184,共4页
Journal of Southwest Petroleum University(Science & Technology Edition)
基金
“油气藏地质及开发工程”国家重点实验室开放基金(PLN0702)资助.
关键词
油田产量
预测
功能模拟
时变系统
oilfield output
forecast
functional forecast
time-varying system
作者简介
刘志斌（1962-），男（汉族），四川武胜人，教授，博士生导师，主要从事石油工程计算技术研究。
