POWER CONTROL AND OPTIMIZATION: Proceedings of the 7th Global Conference on Power Control and Optimization
978-983-44483-63
Functional Networks as A Best-Model Selector in Computational Intelligence Hybrid Models for Petroleum Reservoir Characterization
Fatai Anifowose, Tarek Helmy, Abdulazeez Abdulraheem
PCO Conf-Proc 2008 (2013), - PDF
Abstract. Functional Networks (FN) has been known as a computational intelligence (CI) technique with excellent functional approximation capability. When the input data is supplied to the networks, the output is determined by a set of neurons, which are defined by a function. FN is an advancement of the Artificial Neural Network technology. Feature selection is an important part of the model building process where the dimensions of input spaces are reduced for better performance. The feature selection process has been used in petroleum reservoir modeling but the non-linear selection capability of FN has not been studied. In this paper, we used the model selection segment of the learning algorithm of FN to select the dominant input parameters for the prediction of porosity and permeability of petroleum reservoirs using hybrid CI paradigm. The effect of the FN-based feature selection process on the improved performance of a Type-2 Fuzzy Logic-Support Vector Machine (T2F-SVM) hybrid model was investigated. T2FL was used to extract rules directly from the best models selected by the FN algorithm and the output was then passed to SVM for the final prediction process. The results of this were then compared with that of T2F-SVM hybrid without the FN component. The results showed that the FN-T2F-SVM performed better with higher performance indices than the T2F-SVM, hence suggesting the feature-selection capability of the learning algorithm of FN.
© 2013 PCO based on American Institute of Physics