1 A MODIFIED FUZZY ARTMAP ARCHITECTURE FOR POWER SYSTEM DYNAMIC STABILITY ASSESSMENT SHAHRAM JAVADI Electrical Engineering Department Azad University - Research & Science Center IRAN NASER SADATI Electrical Engineering Department Sharif University Of Technology IRAN E-mail: sadati@ee.sharif.ac.ir MEHDI EHSAN Electrical Engineering Department Sharif University Of Technology IRAN E-mail: ehsan@ee.sharif.ac.ir Abstract: In this article, a new approach for power system dynamic stability assessment has been presented. The proposed approach is based on standard Fuzzy ARTMAP discussed in our previous paper [1] and its modified version called PROBART (Probability ART). Besides Fuzzy ARTMAP, Neural architecture with probability Fuzzy ARTMAP is considered as a modified Fuzzy ARTMAP Network for noisy mapping tasks. Like other ART-based systems, Fuzzy ARTMAP has advantages over feedforward Networks and is especially suited for classification- type problems. In this paper it is used to approximate the noisy continuous mapping. Results show that properties for noisy mapping problems, One particular feature, match tracking, is found to cause over-learning of the data. A modified variant is proposed. Without match tracking which stores the probability information in the map field, this information is subsequently used to compute output estimates. The modified Fuzzy ARTMAP variant namely PROBART, is found to outperform the Fuzzy ARTMAP in the mapping task. Keywords: Power Systems, Dynamic Stability, Fuzzy ARTMAP Neural Network, PROBART 1. Introduction Application of artificial Neural Networks to the solution of engineering problems has received significant attention during recent years. This is mainly due to the fact that neurocomputing is not merely an extension of conventional artificial intelligence. It can learn in a manner similar to the human brain. Conventional computers are programmed with an algorithm while humans learn. Although expert systems can take the knowledge acquired by a human expert and automate the reasoning process, they still cannot learn anything on their own. On the other hand, Neural Networks distribute knowledge in form of patterns in a network of processing elements instead of locating each element of knowledge in a single artificial memory. A significant result is that similar patterns allow the network to generalize. These networks can be exposed to new situations and derive workable solutions based upon prior experience. Power systems have grown in both size and complexity, and dynamic characteristics of the system vary as well. Even a change in the system loading, generation schedule Network-Interconnection, and/or type of system protection may also give completely different stability outcome of a system for the same disturbances. An important task in power system operation is to decide whether the system is safe, critically or unsafely. If the system is safe, it is further of interest to know how much is safe. One of the methods for evaluating that power system is safe or not, is eigenvalue evaluation. There are many methods for determination of eigenvalues of those systems such as analytic methods [2]-[5]. Because of growing in both size and complexity of the power systems and requiring to accuracy and fast computation, analytical methods don’t have good efficiency. Instead, The Neural Networks have a good performance in this regard. A number of artificial Neural Networks (ANN) are developed so far such as multilayer Perceptron [6] and Radial basis Function (RBF) Networks [7]. Because of gradient descent method, which is used in these networks, these traditional Neural Networks suffer from divergency. In addition, these networks cannot learn new knowledge in real time during their operation.