A SVM-based method for abnormity detection of log curves

Rapid and accurate abnormity detection of log curves is critical in the quality control for logging industry. Traditional methods based on manual detection have been proven to be ineffective and unreliable. A machine learning method based on Support Vector Machine (SVM) is proposed in this paper to address this problem. The SVM classifiers are established according to the suspicious sections selected from log curves and the detected results given by experts. A genetic algorithm (GA) is introduced for optimization of parameters. With GA and SVM fusions, the optimal models for classifiers are determined to detect abnormity sections. Experimental results of China XiangJiang Oilfield show that an accuracy of 95% is achieved for suspicious straight sections and 96% is achieved for suspicious bouncing sections, which has proven the feasibility of this method.