A general regression neural network model offers reliable prediction of CO2 minimum miscibility pressure

Abstract

This study introduces a general regression neural network (GRNN) model consisting of a one-pass learning algorithm with a parallel structure for estimating the minimum miscibility pressure (MMP) of crude oil as a function of crude oil composition and temperature. The GRNN model was trained with 91 samples and was successfully validated with a blind testing data set of 22 samples. The MMP for six of these data samples was experimentally measured at the Petroleum Fluid Research Centre at Kuwait University. The remaining data consisted of experimental MMP data collected from the literature. The GRNN model was used to estimate the MMP from the training data set with an average absolute error of 0.2 %. The GRNN model was used to predict the MMP for the blind test data set with an average absolute error of 3.3 %. The precision of the introduced model and models in the literature was evaluated by comparing the predicted MMP values with the measured MMP values and using training and testing data sets. The GRNN model significantly outperformed the prominent models that have been published in the literature and commonly used for estimating MMP. The use of the GRNN model was reliable over a large range of crude oil compositions, impurities, and temperature conditions. The GRNN model provides a cost-effective alternative for estimating the MMP, which is commonly, measured using experimental displacement procedures that are costly and time consuming. The results provided in this study support the use of artificial neural networks for predicting the MMP of CO 2 .

Comparison of Machine Learning Methods for a Diabetes Prediction Information System

“…Simplicity of implementation and high ability to generalize are the main advantages of the AdaBoost algorithm. Among the limitations of the algorithm should be noted the need for no noise in the data, which can lead to retraining [9]. In addition, an important role in the efficiency of this algorithm is played by the dimension of the sample for training [8].…”

An Approach towards Increasing Prediction Accuracy for the Recovery of Missing IoT Data based on the GRNN-SGTM Ensemble

“…In general, the approach to solving the task using GRNN is not new, with refinements and modifications of neural networks of this type [37][38][39][40][41][42][43] seeming promising, taking into consideration their advantages over neural networks of other types. These advantages can be represented as follows [36,43] Considering the velocity performance of modern computers, as well as the ability to apply cluster technologies to solve tasks using this type of neural network on separate clusters, the main desirable disadvantage of GRNN networks to be minimized is significant operating errors, which provides a basis for the research described in this paper.…”

“…To analyze some basic features of the GRNN algorithm [34,37], let us consider a determined set of observations for a particular phenomenon or object. Each observation contains a vector of independent variables x and a dependent component −y.…”

Generalized regression neural network association with terahertz spectroscopy for quantitative analysis of benzoic acid additive in wheat flour

“…Artificial neural networks (ANN) have the potential to deal with the nonlinear problem in the THz spectra analysis [13,14]. However, it should be pointed out that general regression neural networks (GRNN) were adopted instead of frequently employed back-propagation neural networks (BPNN), because GRNN has the following advantages: single-pass learning so no back propagation is required, high accuracy in the estimation for uses of Gaussian function, and it can handle noises in the inputs [15,16]. BPNN requires selecting training parameters and defining network architectures in the process of training.…”