Применение методов машинного обучения для прогноза или замещения недостающих каротажных данных
Nine machine learning methods (ANN, ANFIS, ELM, FM, SVM, GPR, RF, RT, k-NN) are compared using the example of predicting acoustic logging data. With machine learning, the solution to the regression problem can be used not only for predicting geophysical fields, but also for filing in missing data. The constructed curve T(Р) of the P-wave interval time can be considered as a forecasted result, if acoustic logging is expected later; if additional acoustic logging is not possible, then the synthetic curve T(Р) replaces the log-derived one for further interpretation. The RF method is shown to provide the best test results.