Design and application of neural networks and intelligent learning systems

Advances in computing technologies have opened up the way for designing and developing intelligent learning systems that are able to solve complex real-world problems. In this aspect, computational intelligence-based techniques, which include neural computing, evolutionary computing, fuzzy computing, and other data-based computing methods, are useful in undertaking different tasks that are difficult to tackle using conventional approaches. In this special issue, a total of ten papers, based on extended papers from the 12th International Conference on Knowledge-Based and Intelligent Information & Engineering Systems (KES2008) as well as from other submissions, are presented. The papers address how neural network-based systems as well as other intelligent learning systems can be applied to solve practical problems in a variety of domains. A summary of each paper is as follows. Neural-network-based controllers have been used in many power electronics circuits. In the first paper, a B-spline network is proposed to function as a controller for power electric systems. The B-spline network is suitable for real-time implementation owing to its linear nature and local weight-updating procedure. The B-spline network controller is designed and analyzed using a frequency domain stability model. The design process of the controller is simple and straightforward, which is an important consideration in industrial applications. Applicability of the controller to power converter in a UPS is demonstrated, and the results show that the proposed B-spline network controller is able to achieve low steady-state error with fast error convergence. Induction motors are commonly used in modern electric drives. They require non-linear control systems as owing to variability of the motor parameters under different conditions. In the second paper, a study of the indirect rotor field oriented control system of an induction motor including deviations in the stator resistance is described. An approach based on an adaptive model reference system is used to identify the rotor time constant, and a neural network model is used to produce the estimated rotor speed. The difference between the actual and the estimated rotor speed is utilized for manual tuning and automatic stator resistance tuning based on the fuzzy logic principles. The results obtained show the effectiveness of the proposed approach. Feature selection is a key success factor for neural network applications. In the third paper, the circle segments method is used to provide visualization of the relationship between the input features and target outputs, and to select important features. Based on the selected features, the multi-layer perceptron network is used for function approximation and pattern classification. The efficacy of the proposed approach is evaluated empirically. A performance comparison with the response surface