概率神经网络在丽水——椒江凹陷月桂峰组沉积微相识别中的应用

由于海上钻井取芯较少,所以东海陆架盆地丽水—椒江凹陷古新统月桂峰组地层沉积微相识别存在局限.运用概率神经网络对研究区进行沉积微相识别.首先,通过地震相-沉积相响应分析和测井曲线主成分分析,发现研究区地震相和沉积相之间存在耦合对应关系,因此选择地震相作为概率神经网络输入项中的范畴自变量参数,同时提取出能对沉积微相区分较好的自然伽马、自然电位、声波时差、密度测井、补偿中子、井径测井曲线值作为概率神经网络输入项的数值自变量；然后,选用2 199个学习样本对神经网络进行训练,经过65次试验,搜索出变量的最佳平滑因子,建立研究区20种沉积微相类型的判别模式；最后,利用建立的神经网络对研究区沉积微相进行识别.结果表明：跟岩芯分析的结果对比,运用概率神经网络识别的结果准确率达到90％以上,该方法应用于未取芯井区域沉积微相的识别具有可行性.

Application of Probabilistic Neural Network to Sedimentary Microfacies Recognition of Yueguifeng Formation in Lishui-Jiaojiang Sag

Sedimentary microfacies recognition of Yueguifeng Formation in Lishui-Jiaojiang sag of East China Sea Shelf Basin is circumscribed because of the shortage of offshore drilling well core. Probabilistic neural network （PNN） was used for sedimentary microfacies recognition of Yueguifeng Formation. Firstly, based on the response analysis of seismic facies-sedimentary facies and principal component analysis of logging curve, relationship between seismic facies and sedimentary facies was coupling and corresponding, so that seismic facies was selected as the category independent variable from the input item parameters of PNN, meanwhile, natural gamma, self potential, acoustic, density logging, compensated neutron and caliper logging curves, which were helpful to distinguish different sedimentary microfacies, were taken as the numerical independent variables from the input item parameters of PNN; secondly, 2 199 learningsamples were used to train PNN, the optimal smoothing factors of variables were searched after 65 tests, and the discrimination model for 20 kinds of sedimentary microfaeies in study area was established; finally, the PNN established was used to identify sedimentary microfacies. The results show that compared with the result of well core analysis, the accuracy rate of sedimentary microfacies identified by PNN is more than 90%, so that it is feasible to identify sedimentary microfacies from the non-cored area by PNN.