Currently, sleep staging is accomplished is by clinical polysomnography (PSG). By extracting features from different combinations of electroencephalogram (EEG), electrooculogram (EOG) and electromyogram (EMG) signals, this study uses neural network to perform sleep staging. A whole night and complete sleep stage contains wake stage, rapid eye movement (REM) stage, stage 1, stage 2, and slow wave sleep (SWS) stage. This project focuses on the classification of wake stage and light-sleep (stage 1 and 2). These three stages are classified by a two-step process. At first, wake stage and light-sleep are divided into two parts. Second, light sleep is divided into stage 1 and stage 2. For a fixed number of channels, this work identifies the best combination of signal channels. In addition, by simultaneously considering the Neighboring epochs Rule classifier, this work also introduces an empirical rule to improve the classification accuracy. Among the tested databases which contain two Medicine center and sixteen sets of different signal channels, the best results are obtained from the group of patients with the low average RDI value. They include the group that has a mean 15% SWS and the group that uses CPAP. As a whole, the combinative features of four channels are better results of classification. For our best results, the sensitivity and PPV of wake and stage 2 varies from 85%~88%, and those of stage 1 are respective 44.84% and 53.61%. And the total classification of sleep staging is 84.59%. Apparently, the research has satisfactory results on sleep staging. Keywords: Sleep Medicine, Sleep stage, Neural Networks

In many practical applications, it is well known that data collected inevitably contain one or more anomalous outliers; that is, observations that are well separated from the majority or bulk of the data, or in some fashion deviate from the general pattern of the data. The occurrence of outliers may be due to misplaced decimal points, recording errors, transmission errors, or equipment failure. These outliers can lead to erroneous parameter estimation and consequently affect the correctness and accuracy of the model inference. In order to solve these problems, three robust fuzzy neural networks (FNNs) will be proposed in this dissertation. This provides alternative learning machines when faced with general nonlinear learning problems. Our emphasis will be put particularly on the robustness of these learning machines against outliers. Though we consider only FNNs in this study, the extension of our approach to other neural networks, such as artificial neural networks and radial basis function networks, is straightforward. In the first part of the dissertation, M-estimators, where M stands for maximum likelihood, frequently used in robust regression for linear parametric regression problems will be generalized to nonparametric Maximum Likelihood Fuzzy Neural Networks (MFNNs) for nonlinear regression problems. Simple weight updating rules based on gradient descent and iteratively reweighted least squares (IRLS) will be derived. In the second part of the dissertation, least trimmed squares estimators, abbreviated as LTS-estimators, frequently used in robust (or resistant) regression for linear parametric regression problems will be generalized to nonparametric least trimmed squares fuzzy neural networks, abbreviated as LTS-FNNs, for nonlinear regression problems. Again, simple weight updating rules based on gradient descent and iteratively reweighted least squares (IRLS) algorithms will be provided. In the last part of the dissertation, by combining the easy interpretability of the parametric models and the flexibility of the nonparametric models, semiparametric fuzzy neural networks (semiparametric FNNs) and semiparametric Wilcoxon fuzzy neural networks (semiparametric WFNNs) will be proposed. The corresponding learning rules are based on the backfitting procedure which is frequently used in semiparametric regression.

Water production is a challenge in production operations because it is generally costly to produce, treat, and it can hamper hydrocarbon production. This is especially true for gas wells in unconventional reservoirs like shale because the relatively low gas rates increase the economic impact of water handling costs. Therefore, we have considered the following questions regarding water production from shale gas wells: (1) What is the effect of water production on gas production? (2) What are the different water producing mechanisms? and (3) What is the water production potential of a new well in a given gas shale province. The first question was answered by reviewing relevant literature, highlighting observed deficiencies in previous approaches, and making recommendations for future work. The second question was answered using a spreadsheet based Water-Gas-Ratio analysis tool while the third question was investigated by using artificial neural networks (ANN) to decipher the relationship between completion, fracturing, and water production data. We will consequently use the defined relationship to predict the average water production for a new well drilled in the Barnett Shale. This study also derived additional insight into the production trends in the Barnett shale using standard statistical methods. The following conclusions were reached at the end of the study: 1) The observation that water production does not have long term deleterious effect on gas production from fractured wells in tight gas sands cannot be directly extended to fractured wells in gas shales because the two reservoir types do not have analogous production mechanisms. 2) Based on average operating conditions of well in the Barnett Shale, liquid loading was found to be an important phenomenon; especially for vertical wells. 3) A neural network was successfully used to predict average water production potential from a well drilled in the Barnett shale. Similar methodology can be used to predict average gas production potential. Results from this work can be utilized to mitigate risk of water problems in new Barnett Shale wells and predict water issues in other shale plays. Engineers will be provided a tool to predict potential for water production in new wells.

The Temperature Sensitivity Analysis of Power System Load Demand with Neural Networks Chih-Hung Chen* Chao-Shun Chen** Institute of Electrical Engineering National Sun Yat-Sen University Kaohsiung, Taiwan, R.O.C. ABSTRACT The analysis of customer load characteristic plays the fundamental role of power system operation. Based on the load survey study, the load pattern of each customer class is derived to achieve more effective load forecast for system planning to reduce the risk of system capacity shortage. For the load survey study, a stratified sampling method has been used to select the proper size of customers for meter installation to collect the customer power consumption. By the way, the customer load patterns derived can represent the load behavior of whole customer population. The standardized daily load pattern of each customer class has been solved with the mean per-unit method of customer load. According to the total power consumption by all customers within the same class and considering the corresponding daily load pattern, the daily load profile of the customer class is then determined. The standard daily load pattern of each customer class and total power consumption within the territory of service districts of Taipower system are integrated to construct Taipower system daily load profile. The temperature sensitivity analysis of customer power consumption is performed for each customer class by applying neural networks. The proposed method has been used to investigate the change of power consumption due to temperature rise for each district and Taipower system. For the districts with high ratio of the air conditioner loading, the increase of power consumption is in proportion to the temperature. It is concluded that the research of temperature sensitivity on power consumption can support power system operation and better capacity planning of power system in the future. *Author **Advisor

This paper is to investigate the impact of temperature sensitivity to the load profiles of power system by artificial neural networks (ANN). The load survey study is performed to derive the typical load patterns of the residential, commercial, and industrial customers respectively. By executing the training process of customer power consumption and temperature, the ANN model is created to derive the temperature sensitivity of power consumption for each customer class, which is then used to solve the impact of temperature rise to system power profiles. According to the system load composition and temperature sensitivity of power consumption by each customer class, the hourly increase of system power loading due to 1¢J temperature rise is solved. To study the temperature effect to the system reliability, the ¡§IEEE Reliability Test System¡¨ is selected as test system for power system reliability analysis. Based on the temperature sensitivity of power consumption for each customer class and load composition of each load bus. The power demand is updated with the temperature rise. The temperature sensitivity of commercial customers is very significant because of the high air conditioner loading. When the system load composition is most composed of commercial customers, the power demand are due to temperature rise will have very critical impact to system reliability. On the other hand, the tempearture rise will have less impact of reliability analysis for the system which serves high percentage of industrial customers. It is concluded that the research of temperature sensitivity on power consumption can provide important information for system reliability analysis. Better substation planning and system capacity expansion can be obtained to meet system reliability criterion by taking into account the temperature effect to system loading.

In the near future, multimedia will form the majority of Internet traffic and the most popular standard used to transport and view video is MPEG. The MPEG media content data is in the form of a time-series representing frame/VOP sizes. This time-series is extremely noisy and analysis shows that it has very long-range time dependency making it even harder to predict than any typical time-series. This work is an effort to develop multi-step-ahead predictors for the moving averages of frame/VOP sizes in MPEG-coded video streams. In this work, both linear and non-linear system identification tools are used to solve the prediction problem, and their performance is compared. Linear modeling is done using Auto-Regressive Exogenous (ARX) models and for non linear modeling, Artificial Neural Networks (ANN) are employed. The different ANN architectures used in this work are Feed-forward Multi-Layer Perceptron (FMLP) and Recurrent Multi-Layer Perceptron (RMLP). Recent researches by Adas (October 1998), Yoo (March 2002) and Bhattacharya et al. (August 2003) have shown that the multi-step-ahead prediction of individual frames is very inaccurate. Therefore, for this work, we predict the moving average of the frame/VOP sizes instead of individual frame/VOPs. Several multi-step-ahead predictors are developed using the aforementioned linear and non-linear tools for two/four/six/ten-step-ahead predictions of the moving average of the frame/VOP size time-series of MPEG coded video source traffic. The capability to predict future frame/VOP sizes and hence the bit rates will enable more effective bandwidth allocation mechanism, assisting in the development of advanced source control schemes needed to control multimedia traffic over wide area networks, such as the Internet.

In this thesis, we study the least trimmed squares artificial neural networks (LTS-ANNs), which are generalization of the least trimmed squares (LTS) estimators frequently used in robust linear parametric regression problems to nonparametric artificial neural networks (ANNs) used for nonlinear regression problems. Two training algorithms are proposed in this thesis. The first algorithm is the incremental gradient descent algorithm. In order to speed up the convergence, the second training algorithm is proposed based on recursive least squares (RLS). Three illustrative examples are provided to test the performances of robustness against outliers for the classical ANNs and the LTS-ANNs. Simulation results show that upon proper selection of the trimming constant of the learning machines, LTS-ANNs are quite robust against outliers compared with the classical ANNs.

<p>Today one of the greatest aims within the area of bioinformatics is to gain a complete understanding of the functionality of genes and the systems behind gene regulation. Regulatory relationships among genes seem to be of a complex nature since transcriptional control is the result of complex networks interpreting a variety of inputs. It is therefore essential to develop analytical tools detecting complex genetic relationships.</p><p>This project examines the possibility of the data mining technique artificial neural network (ANN) detecting regulatory relationships between genes. As an initial step for finding regulatory relationships with the help of ANN the goal of this project is to train an ANN to predict the expression of an individual gene. The genes predicted are the nuclear receptor PPAR-g and the insulin receptor. Predictions of the two target genes respectively were made using different datasets of gene expression data as input for the ANN. The results of the predictions of PPAR-g indicate that it is not possible to predict the expression of PPAR-g under the circumstances for this experiment. The results of the predictions of the insulin receptor indicate that it is not possible to discard using ANN for predicting the gene expression of an individual gene.</p>

Thesis (Ph. D.)--Texas A&M University, 2004. / Vita. Includes bibliographical references (p. 171-177).

Thesis (Ph. D.)--Lehigh University, 2000. / Includes vita. Includes bibliographical references (leaves 138-147).