Research On Waveform Classification Method Based On Post-Stack 3D Seismic Data

The result of 3D seismic waveform classification can be used to predict the underground oil and gas storage situation which is a very important step to oil exploration. This paper firstly discusses the background and technology-related applications waveform classification, fully summarize the various classification algorithms presented in the previous preprocessing, feature extraction and classification algorithm selection and final strengths and weaknesses in the existing methods proposed algorithm, finally process a new three-dimensional seismic signal classification based on the specific work which including the following four aspects.1. A lot of noise is bring into seismic signal during the acquisition process, to solve this problem, this paper presents the seismic signal preprocessing based on fitting method. Comparison of the different characteristics of the fitting algorithm, this paper finally chooses the Chebyshev polynomial fitting method, which retains the original information of the signal.2. Seismic horizons resolution is one of the key steps of the seismic signal processing. We studied the singularity detection method for singularity analysis and seismic waveforms, based on the current horizon interpretation position to adjust the horizons to its nearest singular position. Comparing the original seismic horizons and the result of horizons error correction found that this method is good for the error corrected which occurred during horizons resolution.3. According to the characteristic of the three-dimensional post-stack seismic signals, we propose a new feature extraction algorithm using artificial immune algorithm, compared to the traditional method this method can not only extract the excellent features, but also has a very strong robustness in noisy environments strong environment, can still extract good signal characteristics, and maintain a good information of the original signal, to provide a reliable guarantee for the follow-up to the correct classification of the signal. Meanwhile, the method in the process of extracting the feature brings the effect of dimensionality reduction.4. The general algorithm is support vector machines(SVM) and decision tree algorithms in supervised classification. Support vector machines in a supervised classification of areas occupy a lot of advantages, but this method has its inherent shortcomings, This method is prone to over-fitting.To solve these problems, this paper proposes a new method for three-dimensional seismic signal supervised classification based on random forest. The method add over fitting detection when doing model selection, not only considering the accuracy of the model on the training samples, but also consider the complexity of the model impact on the final classification result.We finally run the proposed method and the traditional classification method for the actual data are classified, and compare their classification results, find that the proposed classification method has good increase in space complexity and time complexity, and the final classification method is also very good.