论文部分内容阅读
近年来我国石油产量跟不上需求,供需矛盾进一步凸显,导致石油的对外依存度已经连续几年超过警戒线,为了缓解供需矛盾,石油的增储上产是一种有效措施,但精确地识别石油储层成为增储上产的一大难题,而特征选择是精确识别石油储层的有效保障.本文提出了一种增强型自适应差分演化算法,即ESADE算法,在算法中使用了双种群的概念,构造了一个简单的双层差分演化,并且在算法的选择操作中加入模拟退火的思想;接着将ESADE算法作为特征选择的搜索策略,将ReliefF算法、BIF算法、FCBF算法及随机抽选特征算法作为评价准则库,SOM神经网络算法、模糊C均值算法、K均值算法和K近邻算法作为分类器库,得到了一种基于ESADE的特征选择算法.然后将此算法应用于某油田oil81、oil82、oil83、oil84和oil85五口井的测井数据集上进行石油储层的油层、差油层、水层和干层的分类识别,并与未进行特征选择直接进行分类的结果进行比较及相同分类正确率下不同分类算法组合及不同属性选择的比较.实验结果表明与SOM神经网络算法、模糊C均值算法、K均值算法及K近邻算法这四种分类算法相比,基于ESADE的特征选择算法能在利用较少属性的同时提高分类准确率,并能够提供不同的属性和分类算法的最优组合方案.
In recent years, China’s oil production can not keep up with the demand, and the contradiction between supply and demand has become even more conspicuous. As a result, the dependence on oil has exceeded the warning line for several years in a row. In order to ease the contradiction between supply and demand, oil production and storage increase is an effective measure, Oil reservoirs have become a major challenge for increasing production and increasing production, and feature selection is an effective guarantee for accurate identification of oil reservoirs.This paper presents an enhanced adaptive differential evolution algorithm, namely ESADE algorithm, which uses two populations , A simple two-layer differential evolution is constructed, and the idea of simulated annealing is added to the selection operation of the algorithm. Secondly, the ESADE algorithm is chosen as the search strategy for feature selection. The ReliefF algorithm, the BIF algorithm, the FCBF algorithm and the random selection As an evaluation criterion library, SOM neural network algorithm, fuzzy C-means algorithm, K-means algorithm and K-nearest neighbor algorithm are used as classifier library, a feature selection algorithm based on ESADE is obtained.Then the algorithm is applied to a field oil81, oil82, oil83, oil84 and oil85 well logging data set for oil reservoir oil layer, poor oil layer, water layer and dry layer classification knowledge And compared with the results of direct classification without feature selection and the combination of different classification algorithms and different attribute selection under the same classification accuracy rate.The experimental results show that compared with SOM neural network algorithm, fuzzy C-means algorithm, K-means algorithm, Compared with these four classification algorithms, the ESADE-based feature selection algorithm can improve the classification accuracy while using less attributes and can provide the optimal combination of different attribute and classification algorithms.