基于GMDH-SVR致密气开发过程参数优化

本文基于非常规油气的致密气开发流程图,利用K-means将4291口气井根据单位压降产气量聚为三类。通过改进GMDH算法对训练集和测试集样本规模比例的完备性和样本次序的随机性进行研究以增强算法鲁棒性。以单位压降产气量作为评价参数对三类气井开发效能主要相关的11个参数进行特征提取并统计各参数作为主成分的概率,选择累计概率80%及以上最少参数作为SVR模型的输入变量,单位压降产气量作为输出变量,选择RBF 作为核函数。分别使三类井各一个样本井的压裂参数上下浮动20%,应用SVR模型统计相应单位压降产气量变化范围是-18.08%,13.42%]、-2.34%,5.39%]、-16.10%,15.21%],分析不同压裂参数组合对单位压降产气量的影响趋势,确定工程实践最优压裂参数,提高气井效能。

Development Process Parameter Optimization of Tight Gas

Based on development flow chart of tight gas which is one type of unconventional gas, the paper uses K-means to cluster 4291 gas wells into three types. We improve GMDH method in two ways: 1.we complete sample's quantitative proportion of train set and test set. 2.we decrease the impacts of sample sequence on method's results. The paper sets gas production per unit pressure drop as evaluation parameter, analyzes eleven parameters which are related to the production efficiency of gas wells and counts the probabilities of parameters that are selected as principle components. We select the fewest parameters whose probabilities are added up to more than 80% as input variables and evaluation parameter as output variable for SVR method, whose kernel function is RBF. We apply one sample well for each type of wells to SVR method, count relevant range of gas production of unit pressure drop-18.08%,13.42%]、-2.34%,5.39%]、-16.10%,15.21%] by changing volumes of construction parameters at the range of ±20% and analyze production efficiency of gas wells under different volumes and combinations of construction parameters.