Researches On Recognition Of Well Logs Based On The Chaotic Time Series Analysis

Logging is measuring the parameters of earth layers in electronics, acoustics and radioactivity by sending the cables into the wells underground using modern electrical apparatus. Logfacies of an oil-layer is a set of logging response which can distinguish from other kinds of oil-layers. Logfacies recognition is a primary way to understand the sedimentary structures of oil layers correctly. As the development of oil field becomes larger and larger in scale, the manual recognizing operation cannot satisfy the need of the oil field of the development. The automatic interpretation of oil layers by computer becomes more and more interested.Well-logs are the main bases of discrimination of sedimentary facies of oil layers. Doing this task needs rich experiences, so well-log facies recognition is a special domain of pattern recognition that needs large amount of geological knowledge. The purpose of this thesis is to develop some novel methods and algorithms for solving the practical issues. Based on the chaotic time series analysis theories and methods, this paper carried out following main tasks:(1) Chaos identification of well-log time series. By analyzing the reasons of chaos formation in well-logs, it was presented for the first time that chaos exists in well-logs. And surrogate data method was used for proving that well-logs contain chaos indeed. The obtained conclusion provides a premise for the application of chaotic time series analysis in well-log facies recognition.(2) Chaotic feature extraction of well-logs. Based on the obtained conclusion of part 1, two different chaotic features of well-logs, correlation dimension and maximal Lyapunov exponents, were extracted. Calculation of the correlation dimensions shows that different oil layer groups have clearly different fractal dimension. These results can be used in predicting the main oil layer group in the early period. An improved method to estimate maximal Lyapunov exponents for well-logs was presented. This algorithm has many advantages suchas easy calculation and good precision. At the same time, experiment results show that all the oil layer groups' maximal Lyapunov exponents are larger than zero. It is testified again that the conclusion of part 1 is correct.(3) Smoothing and feature extraction of oil layer well-logs. In this paper, a type of asymmetric Gaussian function fitted by nonlinear least square method is used to data smoothing of oil layer well-logs. After studying the shapes of oil layer well-logs, a special algorithm suitable for these time series was presented. At the same time, asymmetric Gaussian model parameters can be used directly as feature values for the classification of well-log shapes. The recognition results are satisfactory. At last, an improved asymmetric Gaussian function was presented for better fitting for the shapes of well-logs of oil layer.(4) Well-log fades classification algorithm using Gaussian mixture models. Other than traditional well-log fades recognition, a new well-log classification approach was given. This method is based on reconstructing the chaotic attractors of the nonlinear dynamics in the phase spaces. The modeling of the different attractors was done using Gaussian mixture models. The algorithm only needs the numbers of mixtures, the well-logs samples of oil layers and their class labels as input during learning process. The experiment results for 4 kinds of main sandstones are satisfactory. This shows that the chaotic modeling is very promising for complex pattern recognition.It can be seen that this thesis enriches the application domains of chaotic time series analysis and provides new ways for well-log representation and recognition. The paper has strong engineering background, novel thinking and nice future.