Experimental Investigation and Prediction of Pour Point Using Artificial Intelligence

Pour point as an important physical property of crude oil is a measure of its low temperature fluidity. The accurate determination of this property is of significance as the temperature decrease below the pour point of the crude oil causes severe production and transportation problems. In this study, for the first time, two types of artificial neural networks (ANNs), including multilayer perceptron (MLP) and radial basis function (RBF), were proposed to predict the pour point. First, the MLP network was modeled and evaluated using different methods. To this end, the optimal number of the input parameters and the best activation function were examined. The results showed that the best predictive MLP network model is constructed using two parameters of wax content and cloud point and SoftMax-SoftMax activation function. The regression of training, validation, and testing datasets of the constructed MLP were 79.5%, 74.1%, and 76.8%, respectively. To validate the constructed MLP-ANN, some experiments were conducted on an oil sample. The dataset obtained from the experiments was then used to predict the pour point. The absolute error of 0.94 °C indicated the great performance of the MLP-ANN in predicting the pour point. Finally, the prediction performance of the MLP-ANN was compared to the RBF-ANN. The results showed the higher predictive accuracy of the MLP-ANN in comparison with the RBF-ANN. Based on the obtained results, the proposed MLP-ANN can be used with confidence in lieu of the expensive and time-consuming laboratory measurements to determine the pour point of crude oils.