Study On Oil-Based Mud Electrical Imaging Logging Data Processing Method Based On Machine Learning

With the deepening of oil and gas exploration,the electrical imaging logging technology plays an important role in logging interpretation.Due to the high temperature and high pressure of deep reservoirs,as well as the water-sensitivity effect of shale reservoir,logging personnel have gradually used high temperature and high pressure oil-based mud instead of water-based mud,but because of the high resistivity of oil based mud,it also brings new challenges to the electrical imaging logging technology.In this paper,the processing method of electro-imaging logging data in oil-based mud environment is studied in order to eliminate the influence of high resistance mud and mud cake in formation parameter measurement.The main research contents and conclusions are divided into the following parts:The first part establishes the numerical simulation model of oil-based mud electrical imaging logging tool.Based on the working principle of oil-based electrical imaging logging instrument,a numerical simulation model was established by COMSOL simulation software.Firstly,the geometric model was established,and material properties,boundary conditions and physical field conditions were set up to verify the effectiveness of the model and collect measurement signals.The second part studies the methods to eliminate the effect of mud.The equivalent circuit model of the low-resistivity formation is established,and the curves of the calculated value of the formation resistivity with the true value is analyzed.The results show that the calculated value of the formation resistivity reverses at high resistivity,which is caused by the capacitive coupling effect of the formation,Therefore,the equivalent circuit model considering the capacitance coupling effect of the formation is re-established.A quick calculation model of formation resistivity,formation permittivity and standoff was established by using the dual-frequency correction method.The curves of the calculated values of the three parameters with different parameters was analyzed.The results show that the apparent resistivity by the fast calculation model can basically reflect the change of formation resistivity quantitatively,and the calculation process is only affected by the standoff.It is not affected by other parameters,and the influence of high resistance mud on formation resistivity measurement is eliminated to a large extent.The calculated value of standoff under this model can only roughly reflect the change of true value,and the calculated value is not affected by other parameters.The relative permittivity of the formation calculated by this method is similar to that of the plate gap,and can’t quantitatively characterize the change of the formation permittivity,and the calculated value is affected by the mud permittivity and standoff.The third part studies the quantitative expression of the three parameters.It was found that dual-frequency correction method could only roughly reflect the changes of each parameter,and could not accurately characterize the three-parameter value.Therefore,BP neural network and SVR(support vector machine for regression)were introduced for three-parameter inversion study.By studying the influence of different transfer functions on the accuracy of inversion,the BP neural network is constructed by selecting appropriate transfer functions and network parameters.The training set and the test set are randomly selected,and the three parameters are inverted for several times to verify the effectiveness of the algorithm and parameter selection.The relationship between the predicted value and the true value of each inversion is analyzed.The results show that the BP neural network can predict the three parameters well,the inversion accuracy is above 90%,and the inversion speed is fast.Also,the appropriate kernel function is selected to establish the SVR algorithm.The three parameters are inverted several times based on SVR,and the inversion results are compared and verified.The results show that SVR can well predict the formation resistivity and standoff,and the inversion accuracy rate is above 95%.However,the inversion effect of formation permittivity is not good,which can only be achieved under specific samples,and SVR is only suitable for training data sets with small sample size.The inversion method based on BP neural network and SVR solves the problem that the dual-frequency correction method cannot represent the three parameters quantitatively,and provides strong support for the processing of oilbase mud electrical imaging logging data.In the fourth part,three parameter inversion based on dispersion characteristics is studied.Based on Maxwell-Garnett dispersion model and SMD dispersion model,the dispersion characteristics of formation permittivity are studied.After determining the values of formation permittivity at multiple frequencies,BP neural network and SVR are used to invert the three parameters.The inversion results are good.However,when the relationship between formation permittivity and frequency cannot be accurately fitted,the inversion results of formation permittivity at low frequencies are poor.