Emin Germen

Case studies on the use of neural networks in eutrophication modeling

Abstract Artificial neural networks are becoming more and more common to be used in development of prediction models for complex systems as the theory behind them develops and the processing power of computers increase. A three layer Levenberg–Marquardt feedforward learning algorithm was used to model the eutrophication process in three water bodies of Turkey (Keban Dam Reservoir, Mogan and Eymir Lakes). Despite the very complex and peculiar nature of Keban Dam, a relatively good correlation (correlation coefficient between 0.60 and 0.75) was observed between the measured and calculated values. For Mogan and Eymir, which are much smaller and more homogenous lakes compared to Keban Dam Reservoir, correlation values as high as 0.95 were achieved between the measured and calculated values. Neural network models were able to model non-linear behavior in eutrophication process reasonably well and could successfully estimate some extreme values from validation and test data sets which were not used in training the neural network.

Self organizing map (SOM) approach for classification of power quality events

In this work, Self Organizing Map (SOM) is used in order to classify the types of defections in electrical systems, known as Power Quality (PQ) events. The features for classifications are extracted from real time voltage waveform within a sliding time window and a signature vector is formed. The signature vector consists of different types of features such as local wavelet transform extrema at various decomposition levels, spectral harmonic ratios and local extrema of higher order statistical parameters. Before the classification, the clustering has been achieved using SOM in order to define codebook vectors, then LVQ3 (Learning Vector Quantizer) algorithm is applied to find exact classification borders. The k-means algorithm with Davies-Boulding clustering index method is applied to figure out the classification regions. Here it has been observed that, successful classification of two major PQ event types corresponding to arcing faults and motor start-up events for different load conditions has been achieved.

Network traffic classification with Self Organizing Maps

Anomaly detection in network traffic is one of the most challenging topics in the study of computer science and networking. This paper introduces a classification method for analyzing network traffic behavior. In order to distinguish the normal traffic with well-known anomalies such as port scanning and DOS attacks, Self Organizing Maps (SOMs), one of the well- known artificial neural network architecture, is used. The measurement of traffic is performed by using Simple Network Management Protocol (SNMP). In this work, it is proposed a SOM-based classifier to discriminate three types of network traffic as port scanning, heavy-download and the rests. It is worth to mention that impressively satisfactory results have been obtained. The method has also been enhanced to obtain better results by trying to find trajectories on the map with sliding the input vectors in time and developed an alarm mechanism. Here it is possible to detect whether consecutive trajectories are hit by one of the classes or not. The success rate of the system is approximate to certain.

Subspace-Based Identification of Acoustic Noise Spectra in Induction Motors

In this paper, we study the identification of acoustic noise spectra in induction motors by using a recently developed frequency-domain cross-power spectrum estimation algorithm. This algorithm is a noniterative high-resolution spectral estimator. In a test rig, from multiple experiments sound data are collected by an array of five-microphones placed hemispherically around motors in a reverberant and noisy room. In order to explore the issue of assembly micromisalignments, each motor is removed from the test rig and then replaced, after which the experiment is then repeated. The identification algorithm is used to detect changes in acoustic noise spectra of induction motors due to mechanical and electrical faults most frequently encountered in industry. Not only the autopower spectra of the individual microphones, but also the cross-power spectra of the microphone pairs are estimated. As a byproduct, it is demonstrated that one microphone is sufficient to identify noise spectra. The estimated acoustic spectra, or more compactly statistics extracted from them, can be used in the development of preventive maintenance programs for induction motors in service.

Identification of acoustic spectra for fault detection in induction motors

In this paper, we study fault detection problem for induction motors by using a recently developed cross-power spectral density estimation algorithm from sound measurements. In a test rig, from multiple experiments the sound data were collected by an array of five-microphones placed hemispherically around motors in a reverberant and noisy room. After an experiment was performed, each motor was removed from the test rig and was reinstalled for the next experiment to verify the consistency of the experimental procedure. The mechanical and electrical faults frequently encountered in induction motors were isolated by the identification algorithm, which is a non-iterative high resolution spectral estimator. The estimated acoustic spectra, or more compactly statistics extracted from them, can be used in the development of preventive maintenance programs for induction motors in service.

A New Approach to Estimate RED Parameters Using Global Congestion Notification

In communication networks, congestion avoidance in routers is one of the hottest topics. In this work, a new queue management approach is proposed on the RED (Random Early Detection) algorithm by monitoring the global congestion situation of an autonomous system. In order to observe the congestion situation of the system, a traffic is generated between routers and a centralized unit. IP routers are arranged in order to send information packets regarding current output queue levels to the centralized unit which helps to produce a global picture of congestion. The autonomous routers update their RED parameters according to the congestion notification of the control unit. In this work the benchmarks of the proposed algorithm have been investigated in an OPNET model and improvements on recent queue management technologies which are Drop Tail and RED have been introduced.

Self organizing map (SOM) approach for classification of mechanical faults in induction motors

In this work, Self Organizing Map (SOM) is used in order to detect and classify the broken rotor bars and misalignment type mechanical faults that often occur in induction motors which are widely used in industry. The feature vector samples are extracted from the sampled line current of motors with fault and healthy one. These samples are the poles of the AR model which is obtained from the spectrum of sampled line current. The waveforms are obtained from four different 3 hp test motors. Two of them have different number of broken rotor bars, one test motor has misalignment problem and the last one is the healthy motor. Broken rotor bar and misalignment faults are successfully classified and distinguished from the healthy motor using SOM classification with the feature vectors. It is also worth to mention that discrimination of different number of broken rotor bars has been achieved.

Anomaly Detection with Self-Organizing Maps and Effects of Principal Component Analysis on Feature Vectors

Network anomaly detection is the problem of scrutinizing of unauthorized use of computer systems over a network. In literature there are plenty different methods produced for detecting network anomalies and the process of anomaly detection is one of the major topics that computer science is working on. In this work, a classification method is introduced to perform this discrimination based on Self Organizing Network (SOM) classifier. Also, rather than proving well-known abilities of SOM on classification, our main concern in this work was investigating effects of Principal Component Analysis on quality of feature vectors. In order to signify the power of success, KDD Cup 1999 dataset is used. KDD Cup dataset is a common benchmark for evaluation of intrusion detection techniques. The dataset consists of several components and here, it is used ‘10% corrected’ test dataset. Since the feature vectors obtained from the dataset have prominent impact of success on the method, the usage of PCA and a method of choosing reliable components are introduced. At the end it is mentioned that the success of decision by the proposed method has been improved. In order to clarify this improvement, a detailed comparison of changing number of principal components on the success of decision mechanism is given.

Improving the resultant quality of kohonen's self organizing map using stiffness factor

The performance of Self Organizing Map (SOM) is always influenced by learn methods. The resultant quality of the topological formation of the SOM is also highly dependent onto the learning rate and the neighborhood function. In literature, there are plenty of studies to find a proper method to improve the quality of SOM. However, a new term “stiffness factor” has been proposed and was used in SOM training in this paper. The effect of the stiffness factor has also been tested with a real-world problem and got positive influence.

A SELF ORGANIZING MAP BASED APPROACH FOR CONGESTION AVOIDANCE IN AUTONOMOUS IP NETWORKS

This work presents a Self Organizing Map (SOM) based queue management approach against congestion in autonomous Internet Protocol (IP) networks. The new queue management approach is proposed with consideration to the pros and cons of two well-known queue management algorithms: Random Early Detection (RED) and Drop Tail (DT). At the beginning of this study, RED and DT are compared by observing their effects on two important indicators of congestion: end- to-end delay and delay variation. This comparison reveals that the performances of RED and DT vary according to the level of global congestion: under low congestion conditions, when packet losses caused by congestion are unlikely, DT outperforms RED; while under high congestion, RED is superior to DT. The SOM based ap- proach takes into account the variations in the global congestion levels and makes decisions to optimise congestion avoidance. A centralized observation unit is designed for monitoring global congestion levels in autonomous IP networks. A traffic flow is generated between each router and the observation unit so as to follow the changes in the global congestion level. For this purpose, IP routers are specialized to send packets carrying queue length information to the observation unit. A SOM based decision mechanism is used by the observation unit, to make predictions on the future congestion behavior of the network and inform the routers. Routers use this information to update their congestion avoidance behavior, as their ability to update their RED parameters is enhanced by the congestion notications sent by the observation unit. In this work, multiple simulations are undertaken in order to test the performance of the proposed SOM-based method. A considerable improve- ment is observed from the point of view of end-to-end delays and delay variations, by comparison with DT and RED as used in recent IP networks.