Research On Automatic Recognition Of Sandstone Correlation Layer Of Completion

Depth correction for well completion is a crucial aspect in the process of perforation data calculation. Particularly, the identification of sandstone correlation layer is the core in the depth correction of sandstone layer. The reasonableness of sandstone correlation layer selecting directly influences the accuracy of the calibration results and construction data.Currently, identifying the sandstone correlation layer is confronted with the following problems:firstly, the decision procedure cannot get rid of the artificial pattern, which affects greatly both the efficiency and automation process; secondly, artificial discrimination mainly depends on the experience of experts. Due to the subjectivity in understanding the curve shape and the indefiniteness and non-integrality of experiences, there are certain ambiguities and errors to the results, which make the regular definition method inapplicable; thirdly, the missed detections of sandstone correlation layers have unfavorable effects on the results even the proportion of them is low in the whole layers. Some effective classification methods are needed for unbalanced data.Faced with these problems, this paper presents a new approach with a combination of pattern classification approaches and geological knowledge on the base of the thorough research in identification method of sandstone correlation layer. The main contents can be stated as follows:1. In order to describe the abnormal shape similarity for the gamma ray curves, an extraction method based on the feature of logging curve shape similarity is proposed under the analysis of similarity measure for a variety of time series. Combined with the visual experience, the curve trend and the similarity of fluctuation are comprehensive analysed, which can be proved to be accurate and efficient by comparing with the judgment results of various methods and can be one of the characteristic parameters in the samples of sandstone correlation layers.2. For the purpose of detecting the curve’s mutated depth, an extraction method based on the mutation feature of logging curve with optimized moving T-test is designed. By introducing moving T-test in mutation analysis of well-logging curves and processing optimally of the weighted average method for multiple window length detection values, the restriction of the moving window length can be offset towards the algorithm effect.3. According to the analysis of the Adaboost algorithm and the problems of the Adaboost algorithm applied to the unbalanced data, the improved Adaboost algorithm based on unbalanced classification is put forward. Three kinds of samples are attached great importance in improving the algorithm, i.e.,positive error samples, negative error samples and noise samples,adjusting the sample distribution with the cost and error’s control on the weight update of the three kinds of samples,so as to realize the algorithm’s focus to positive error.Thr ough the experiments in the standard data sets and actual correlation layer data sets, the improved algorithm significant boosts in identifying rare class correctly, which can be finally applied to calculating the depth data of perforation combined with the actual situation.