模糊神经网络油气预测

Oil and gas prediction using fuzzy neural network

大庆石油管理局物探公司研究所, 大庆 163357

Research Institute Geophysical Exploration Company,Daqing Petroleum Administration, Daqing 163357

本文在分析研究BP网络油气预测存在问题的基础上, 提出了一种模糊神经网络。这种网络与BP网络相似, 它是将模糊的概念结合于网络之中, 使神经网络可以处理结构化的知识, 亦即由专家给出的规则, 从而提高了油气预测结果的可信度。同时, 在网络训练时采用同伦学习算法, 大大提高了网络训练的收敛速度, 避免了用梯度下降法训练网络所产生的局部收敛现象。模糊神经网络已在大庆探区多个高分辨率区块进行了油气预测的实际应用, 取得了较好的效果。

Based on the analysis and study of problems existing in oil and gas prediction using BP network, we propose a kind of fuzzy neural network. The network is similar to the BP network. It combines the con-cept of fuzziness with the network, and make the neural network able to process the structural knowl-edge, i. e. the regulation given by specialists. So the credibility of the oil and gas prediction result is im-proved. Meanwhile, the homtopy learning algorithm is used in netwok training, which greatly increas-es the convergence speed of networ training and avoids the local convergnce phenomenon produced bytraining network using the gradient method. The fuzzy neural network has been used to predict oil andgas in several high-resolution blocks in the Daqing area, and a good result is obtained.