The Application Of Fuzzy Multilevel Window Length TFPF In Noise Attenuation For Seismic Exploration Data

Seismic exploration is now the main method for oil and natural gas detection.The seismic exploration data we collect usually contains strong random noise due tothe constraints of a variety of complex conditions. It decreases signal-to-noise ratioand resolution of the data, and affects the accuracy and reliability of seismicexploration research and the overall quality of seismic exploration record. Therefore,improving the signal-to-noise ratio of seismic exploration data is the primary task ofthe processing work.TFPF, which is short for Time-Frequency Peak Filtering, is an effective methodfor attenuating random noise in seismic exploration data. It has the advantages offew constraint conditions and strong adaptability to seismic data with lowsignal-to-noise ratio. The principle of the method is based on time-frequencyanalysis, and signals mixed with random noise are coded as instantaneous frequency.Considering the character that analytic signal instantaneous distribution concentratesalong its instantaneous frequency, we can estimate instantaneous frequency throughthe peak of time-frequency distribution. If the valid signal is linear, no biasedestimation can be obtained. For the nonlinearity of seismic signal, PseudoWigner-Ville distribution (PWVD) is used, which makes the instantaneous frequencyin the window be partial linear, therefore the selection of window length is crucial.When applying TFPF to process seismic exploration data, the demands ofattenuating random noise and retaining signals are conflicting, the traditional way ofselecting a fixed window length can hardly make a balance of noise attenuating andsignal retaining. In order to overcome the restriction of fixed window length,Fuzzy C-Means(FCM) Clustering and TFPF are combined on the basis ofvarying-window-length thought，and fuzzy multilevel window length TFPF methodis proposed. This method can consider the signal-noise-mixed situation in seismic exploration data, the spatial correlation character of events, time domain and spatialinformation comprehensively, and use FCM to divide the unprocessed seismic datainto two parts. When most components of the part are signals, it is judged as signalpart, the same with the judgment of the noise part. Then do multilevel windowlength TFPF, use short window for the signal part, long for the noise one, andcombine the two parts as the final result. Compared with the single window lengthfiltering, multilevel window length TFPF can meet the different demands of signaland noise, multilevel window length has better result of noise reduction, meanwhiledetails of the signals are well retained. Compared with point by pointvarying-window length TFPF, it has avoided the shortcomings of the existingvarying-window length TFPF method such as the large computationally expensiveand signal distortion.In order to verify the validity and practicability of multilevel window lengthTFPF method, it is applied to theoretical model experiments and real seismic recordprocessing, which include single event record, common-shot-point simulation recordand real seismic record, and make comparison and analysis according to the timedomain waveform and spectrum. The experiment result shows that compared withthe traditional single window length TFPF, the method can divide signal part andnoise part reasonably, and create favorable conditions for choosing time windowlength flexibly. Because of the combination of advantages of long and short windowlength, both the preservation of the signals and reduction of noises have obtainedgood results.