Aggregation and modeling using computational intelligence techniques

In waterflood management, there exists several models to describe a petroleum reservoir for predicting the future production rates using scheduled injection rates. Most of them have the ability to estimate how much the injectors impact some specific producers, namely, the interwell connectivities between the injectors and the producers. Knowing these values not only reduces the cost of water injection, but can also increase the oil production.;In the first part of this thesis, we construct four different models for the interaction between a group of injectors and a producer and then dynamically estimate the parameters of these models, along with the interwell connectivities using an Iterated Extended Kalman Filter (IEKF) and Smoother (EKS). We then use the Weighted Average (WA) and Generalized Choquet Integral (GCI) to aggregate the estimated interwell connectivities. These two aggregation functions are optimized to minimize mean-square errors in future forecasted production rates. This is done by using Quantum Particle Swarm Optimization (QPSO) to search for the optimal set of weights which are required by both aggregation methods. Several experiments are conducted to show the improved average performance of our approach on a set of data from a real reservoir, and the performances of the above two aggregation methods are also compared and analyzed.;A similarity measure between fuzzy sets is a very important concept in fuzzy set theory. There have been a lot of different similarity measures proposed in the literature, for both T1 FSs and IT2 FSs. The second part of this thesis presents theoretical studies that were performed for the most advanced fuzzy logic sets that are currently under research-general type-2 fuzzy sets. In our study, based on the alpha-plane representation for a general type-2 (GT2) FS, the similarity measure is generalized to such T2 FSs. Some examples that demonstrate how to compute the similarity measures for different T2 FSs are given.;Next, the third part of this thesis proposes a new method--the HM method--to model words by normal IT2 FSs, using data intervals that are collected from a group of subjects. The HM method uses the same bad data processing, outlier processing and tolerance limit processing to pre-process the data intervals, as is used in the Enhanced Interval Approach (EIA); it then uses a new confidence-interval-based reasonable interval test to keep only those data intervals that share a common interval. In the Fuzzy Set Part, the common overlap is first determined for a group of data intervals; the IT2 FS model for a word is then determined from the remaining data intervals that exclude the overlap. The HM method has a new way to establish if a word should be modeled as a left shoulder, interior or right shoulder IT2 FS. The resulting IT2 FSs have both normal lower and upper membership functions (MFs), which makes them unique among IT2 FS word models. We also compare the IT2 model obtained from the HM method with those from the EIA using Jaccard's similarity measure.;The fourth and fifth part of this dissertation, a generalization of the Linguistic Weighted Average (LWA), the Linguistic Weighted Power Mean (LWPM), is studied. Based on the LWPM, a new type of fuzzy statistic, namely, the Linguistic Weighted Standard Deviation, is proposed. In classical statistics, the first- and second-order statistics, i.e., the mean and standard deviation, are the most important ones. In this thesis, we extend the definition of the standard deviation and makes it possible to compute the standard deviation when data contains not only numbers, but also words. The generalized standard deviation is called the Linguistic Weighted Standard Deviation (LWSD). The LWSD is viewed as a special case of the LWPM when the parameter r in the LWPM is set to be 2. Two numerical examples that utilize the new LWSD are presented: one is synthetic where all the data are generated randomly, and the other is a practical decision making problem. These examples demonstrate that the LWSD can provide extra information to a decision maker when only uncertain input data (words) are available. We believe that the concept of the LWSD will certainly play an important role in many future applications. In the final part of this dissertation, an IT2 version of the LWPM is defined, and an application of the Perceptual Computer (Per-C) to evaluate learning outcomes using the IT2 LWPM is also presented.;The final part of this dissertation draws conclusions, and provides some suggestions for future research.