Modifying a Model for Predicting Rheological Behavior of Drilling Fluid Using Artificial Neural Network

Due to the influence of many parameters on the drilling fluid, the precise determination of the rheological behavior of the fluid is important. Therefore, the elimination of imprecise experimental methods which have been done based on trial and error, and using logical mathematical intelligence methods such as artificial neural networks instead of previous methods are needed. In the present study, for predicting the rheological properties of drilling mud, including Plastic viscosity, funnel viscosity, and yield point, the data from four wells of the field of X which contains 240 rows of data (4080 data) to test and 23 rows (391 data points) to test the model have been used. This includes 14 types of parameters in a fluid, depth, and temperature formation (a total of 17 parameters). By using artificial neural networks, the structures for predicting the neural network, and the rheological properties of drilling mud were built, and finally, three separate optimization models for the plastic viscosity (PV), funnel viscosity (FV), and the yield point (YP) were designed that in all three models, the network had two layers, 17 inputs and one output in last layer. Moreover, the number of neurons in the hidden layer, 16 neurons for the PV model, 19 neurons for FV model, and 19 neuron for YP model were determined. Finally, the coefficients of determination of these models, which were tested, were R2 PV = 0.99, R2 YP = 0.98, and R2 FV=0.97 indicating a high compliance test results with reality.