The 4 th International Power Engineering and Optimization Conf. (PEOCO2010), Shah Alam, Selangor, MALAYSIA: 23-24 June 2010 Abstract— Voltage instability has recently become a challenging problem for many power system operators. This phenomenon has been reported to be responsible for severe low voltage condition leading to major blackouts. This paper presents the application of Artificial Immune Systems (AIS) for online voltage stability evaluation that could be used as early warning system to the power system operator so that necessary action could be taken in order to avoid the occurrence of voltage collapse. Key features of the proposed method are the implementation of clonal selection principle that has the capability in performing pattern recognition task. The proposed technique was tested on the IEEE 30 bus power system and the results shows that fast performance with accurate evaluation for voltage stability condition has been obtained. Index Terms— Artificial Immune Systems, Pattern Recognition, Machine Learning, Voltage Stability. I. INTRODUCTION oltage stability is the ability of a power system to maintain steadily acceptable bus voltage at each node under normal operating conditions, after load increases, following system configuration changes or when the system is being subjected to a disturbance. The progressive and uncontrollable drop in voltage eventually results in a wide spread voltage collapse. The phenomenon of voltage instability is attributed to the power system operation at its maximum transmissible power limit, shortage of reactive power resources and inadequacy of reactive power compensation tools [1-3]. Problems related to voltage stability issues have attracted greater concern among the power system engineers since power systems nowadays have evolved through continuing growth in interconnection and operating in highly stressed condition[4]. Therefore voltage stability evaluation becomes part of the power system operation routine and has been treated by a wide spectrum of computational approaches with major goal of determining the voltage stability condition of the system and its margin from instability condition [5]. There are 1 S. I. Suliman is with the Faculty of Electrical Engineering, Universiti Teknologi Mara, Malaysia . Phone: 603-55435051; fax: 603-55435077 2 T. K. Abdul Rahman is with the Faculty of Electrical Engineering, Universiti Teknologi Mara, Malaysia . Phone: 603-55435051; fax: 603-55435077; e-mail: khawa@ salam.uitm.edu.my several methods to estimate or predict voltage instability using indexes. They are based on multiple load flow solution such as the Voltage Instability Proximity Index (VIPI) and Voltage Collapse Proximity Indicator (VCPI). Evaluation techniques based on load flow full of or reduced Jacobian analysis such as singular value decomposition, eigenvalue calculation, sensitivity factors and modal analysis are time consuming for a large power system [6-8]. More robust and efficient index for determining the set of low-voltage solutions closest to the operatable solution is also used. This index utilizes the structure of the power flow equations to determine good initial guesses for these solutions, and is generally enough to determine more than just a single low-voltage solution [9-10]. A fast method to evaluate voltage stability condition is desirable since rapid action and accurate decision is needed so that the occurrence of voltage collapse could be avoided. Nowadays, power system control and operation has made use of intelligent systems in determining system condition for the on line monitoring system. This is due to their capability in giving fast decision or evaluation with acceptable accuracy. The applications of intelligent systems have been reported to solve power system problems since early 1980’s [3]. Expert System (ES) and Artificial Neural Network (ANN) are the most common intelligent systems employed in solving various problems in power system operation and planning [3]. The learning capability of ANN has been exploited for power system security evaluation as reported in references 4 to 7. For example, a multi-layer feed-forward perceptron ANN was implemented for predicting power system dynamic stability as described in reference 4. This method utilized the most critical eigenvalue obtained from the S-matrix as the indicator or index to power system dynamic stability and taken to be the output of the ANN. The relationship between the index (output) and the input quantities (nodal active and reactive power, nodal voltage magnitudes and angles) are represented as a three-layer feed-forward neural network. The neural network was constructed with the back propagation algorithm. On the hand, a supervised clustering algorithm of neural network was used to assess the dynamic stability as mentioned in reference 5. The algorithm implemented the adaptive threshold valued to the traditional clustering method and used the supervised output as the convergence constraints. The technique has able to reduce the training time of the neural network. In reference 6, ANN was used to predict the frequency of the centre of inertia during the first ten seconds of a dynamic process after the occurrence of generator outage. The evaluation from the developed ANN would help the Artificial Immune System Based Machine Learning for Voltage Stability Prediction in Power System S. I. Suliman 1 , T. K. Abdul Rahman 2 , Senior Member,IEEE, V 978-1-4244-7128-7/10/$26.00 ©2010 IEEE 53