| At present,many commonly used methods for logging are mathematical statistics,multi-variables analysis,artifical intelligence,computer graph ,& image processing,degree of fuzzy grey incidence,fuzzy cluster analysis,fuzzy comprehensive evaluation,grey cluster analysis and so on.And these methods were programmed at the time.Due to the complexity of beds,the accuracy ratio for logging is 60-75%.In this thesis,some mathematical methods.for example time series analysis,disciminant analysis and exponential smoothing model,are used for logging interpretation.By example analysis,the methods proposed in this thesis was shown effective and applicable to engineering.As an important embranchment of probability and statistics,time series analysis has been regarded as a powerful and potential tool for prediction and controle has been developed rapidly in recent years.In this paper,main work is as follows:(1) Four kinds of AR(p) models are introduced,which include parameters estimation,rank confirmation and forecasting. The model is composed of parameters estimation by the general least squares and the improved least squares based on linear model theory for multi-dimensional AR(p) model,rank confirmation by best criterion functions ,which includes Final Prediction Error Criterion, A-Information Criterion and B-Information Criterion ,the routine F-test,the Fast F-test and a improved F-test,and forecasting by direct forecasting formula and recursive forecasting formula.(2) The combined prediction approaches of AR(p) model & exponential smoothing and its application.Paper[45]&[46] proposed a new method for forecasting the grade of copper matter based on the collected data from the factory and it established the dynamic Auto-Regressive and Exponential Smooth model by the system identification. Paper[50] proposed a new method for forecasting the oil production, it established the combined prediction based on one-dimensional dynamic Auto-Regreesive and Grey system model. Paper[51] proposed a new method for forecasting the oil production, it established the combined prediction based on multi-dimensional time series model and neural network model.Based on these,this thesis proposed a new combined prediction model based on the one-dimensional Fuzzy AR(p) model & exponential smoothing model which had better forecasting accuracy ratio than the combined prediction model based on the one-dimensional AR(p) model & exponential smoothing model.And for multi-factors forecasting,this thesis proposed two kinds of combined prediction model which include thecombined prediction model of multi-dimensional AR(p) model & exponential smoothing model and the combined prediction model of multi-dimensional Fuzzy AR(p) model & exponential smoothing model.Finally four kinds of combined prediction model were contrasted by examples.(3) AR(p) model & discriminant analysis and its application for the identification of oil & water beds. A new method is proposed for the identification of oil & water beds,which is AR(p) model & discriminant analysis.The AR(p) model of four logging curves AC,GR,SP and RT of 47 oil beds and 17 water beds were established.Then oil & water beds were identified by discriminant functions.Through an example analysis,the accuracy ratio of identification of 47 oil beds and 17 water beds which based on four kinds of AR(p) model were contrasted.Finally 15 unkown layeres were identified.(4) Software system for logging interpretation based on time series AR(p) model by the applied mathematical software MATLAB 6.5.Software system for logging interpretation were programmed by the applied mathematical software MATLAB 6.5 based on the operating system Windows98,2000 and WindowsME/NT/XP.It main functions are layer identification based on four kinds of AR(p) model & discriminant analysis and combined prediction based on four kinds of AR(p) model & exponential smoothing model.It includes.data management,graph management,one-dimensional AR(p) modeling, multi-dimensional AR(p) modelin... |
PDF Full Text Request