油气识别的模糊聚类与遗传神经网络技术

IDENTIFICATION OF THE OIL AND GAS BASED ON FUZZY CLUSTERING AND GENETIC-NEURAL NETWORK TECHNIQUES

为了解决遗传BP神经网络在储层油气识别中存在的问题,采用改进的遗传算法优化了RBF网络的连接权值及结构,不仅解决了神经网络易陷入局部最优的问题,而且提高了网络的泛化性能.针对储层性质差别大会影响油气识别精度的问题,给出基于马氏距离的模糊聚类方法,对原样本空间按储层性质聚类得到了新的样本空间,并以常规测井和录井资料作为网络的输入参数进行了油气识别.通过样本的聚类处理,提高了遗传神经网络映射的精度.

In order to solve the problem in the identification of the oil and gas reservoirs, the improved genetic al- gorithm is adopted to optimize the connection power and structure, thus the problem of the partially optimized neu- tral network can be resloved, and moreover the generalized capability of the network has been enhanced. In the light of the problem of influencing the identification resulted from the sharp contrast among the reservoir properties, on the basis of Markov Distance, the fuzzy clustering method is presented, and then the new sample space is ob- tained by the clustering of the original sample space by means of the reservoir properties. At the same time, taking the conventional well logging and mud logging data as the input parameters of the network, the oil and gas identifi- cation is conducted. With the help of the sample clustering process, the reflecting precision of the genetic-neutral network is enhanced.