摘要
For lack of robust theoretical guidance in the respect of genetic mechanism,inspection accuracy and interpretation model,and without any reliable evaluation models and standards,the traditional interpretation techniques of mud logging data produces results with poor reliability.At present,drilling targets are getting more diversified,and fluid identification becomes challenging for complex reservoirs,deep reservoirs,thin interbedded reservoirs and unconventional reservoirs.In view of this,based on mathematical statistics technologies,the natural gas component from the mud logging chromatographic analysis was first taken as the basic data set;the different response abilities were then analyzed of Euclidean distance and Mahalanobis distance to multiple variables in variable space.Additionally,the different abilities were discussed of spatial distance-based fluid identification models in identifying gas and water layers.Finally,a new identification model was built and applied in the fluid identification of the Upper Triassic Xujiahe reservoirs in the western Sichuan Basin,achieving a coincidence rate of above 83%.In combination with the characteristics of this new model,an optimized model was further built and also compared with the Euclidean model and Mahalanobis model.The optimized model was more powerful in fluid identification and the accuracy of prediction results reached up to 97.22%.Meanwhile,a mathematics-based evaluation method of traditional mud logging interpretation was proposed and a comparative analysis of the optimized model and traditional model was also performed.