高分辨率非线性反演方法及应用研究
|
摘要
文章所述的ANFIS是将模糊技术与神经网络技术相结合而形成的一种高分辨率非线性反演方法 ,是集模糊逻辑推理的强大结构性知识表达能力与神经网络的强大自学能力于一体的新技术。ANFIS是由可更改的前提和结论两部分组成 ,采用 (GD +LSE)混合算法自适应地进行调节 ,该算法减少了原始纯反向传播算法的收缩空间的维数 ,因而收敛速度非常快。这种高分辨率非线性反演方法兼有二者的优点 ,对原始模型没有依赖性。在实际反演中 ,首先由井点出发构造测井资料与井旁地震道的非线性映射关系 ,并根据地下介质在横向上的变化特征来更新非线性映射关系 ,并在地质约束下进行自适应外推 ,其反演过程是总体反演过程 ,以实现高分辨率反演 ,获得高分辨率的储层参数剖面。实践证明 ,应用ANFIS可建立储层参数与地震信息之间的较精确关系 ,所获得的储层参数剖面具有精度和分辨率高的特点 ,是提高储层预测效果的一种有效手段
The adaptive neural fuzzy inference system(ANFIS)mentioned in the paper is a high resolution nonlinear inversion method formed by unifing the fuzzy technique with the neural network technique and it is also a new technique of incorporating the strong structural knowledge expressing ability of fuzzy logical inference with the strong teaching oneself ability of neural network as a whole. The ANFIS is composed of two parts, i.e. alterable prerequisites and conclusions and can be adaptively adjusted by applying(GD+LSE)mixed algorithm. The dimensionality of contraction space of the original net backpropagation algorithm may be reduced by use of the mixed algorithm, so that the convergence rate will be very fast. This high resolution nonlinear inversion method is of the advantages of the both, having no any dependence on the initial model. In actual operation, the nonlinear mapping relation between log data and well vicinal seismic traces was set up at the well point firstly, then renewed according to the characters of the lateral change in underground mediums and then extrapolated adaptively under the control of geological conditions. This process is a total inversion process so as to realize the high resolution inversion and to obtain a high resolution reservoir parameter section. Through practice it was proved that a precise relation between reservoir parameters and seismic message may be set up by use of the ANFIS and the reservoir parameter section achieved is of the characters of high accuracy and high resolution, which has a vast range of application prospects, being an effective means of improving reservoir prediction.
The adaptive neural fuzzy inference system(ANFIS)mentioned in the paper is a high resolution nonlinear inversion method formed by unifing the fuzzy technique with the neural network technique and it is also a new technique of incorporating the strong structural knowledge expressing ability of fuzzy logical inference with the strong teaching oneself ability of neural network as a whole. The ANFIS is composed of two parts, i.e. alterable prerequisites and conclusions and can be adaptively adjusted by applying(GD+LSE)mixed algorithm. The dimensionality of contraction space of the original net backpropagation algorithm may be reduced by use of the mixed algorithm, so that the convergence rate will be very fast. This high resolution nonlinear inversion method is of the advantages of the both, having no any dependence on the initial model. In actual operation, the nonlinear mapping relation between log data and well vicinal seismic traces was set up at the well point firstly, then renewed according to the characters of the lateral change in underground mediums and then extrapolated adaptively under the control of geological conditions. This process is a total inversion process so as to realize the high resolution inversion and to obtain a high resolution reservoir parameter section. Through practice it was proved that a precise relation between reservoir parameters and seismic message may be set up by use of the ANFIS and the reservoir parameter section achieved is of the characters of high accuracy and high resolution, which has a vast range of application prospects, being an effective means of improving reservoir prediction.
引文
1 李正文等.沉积盆地有效储集层综合识别技术.成都:四川科学技术出版社,2003
2 赵振宇等.模糊理论和神经网络的基础与应用.北京:清华大学出版社,1996
3 张贤达.现代信号处理.北京:清华大学出版社,1995
4 JangJSR .ANFIS :Adaptive etwork basedfuzzyinferencesys tems.IEEETransactiononSystems,ManandCybernetics,1993;23(3):665～685
5 MartinezRD .Complexreservoircharacterizationbymultiparame terconstrainedinversion,SEG EAEGResearchWorkShopRes ervoirGeophysics,1988
6 HaganMT .Neuralnetworkdesign.PWSpublishingcompany,1996
7 AndersonJA .Anintroductiontoneuralnetworks.Cambridge,MA :MITpress,1995
2 赵振宇等.模糊理论和神经网络的基础与应用.北京:清华大学出版社,1996
3 张贤达.现代信号处理.北京:清华大学出版社,1995
4 JangJSR .ANFIS :Adaptive etwork basedfuzzyinferencesys tems.IEEETransactiononSystems,ManandCybernetics,1993;23(3):665～685
5 MartinezRD .Complexreservoircharacterizationbymultiparame terconstrainedinversion,SEG EAEGResearchWorkShopRes ervoirGeophysics,1988
6 HaganMT .Neuralnetworkdesign.PWSpublishingcompany,1996
7 AndersonJA .Anintroductiontoneuralnetworks.Cambridge,MA :MITpress,1995
版权所有：© 2023 中国地质图书馆 中国地质调查局地学文献中心