|
Business Systems and Knowledge Modelling Lab |
 |
Research ProjectIntelligent Decision Support Systems for Emergency/Critical SituationsResearch Goal Real-time alarm processing and fault diagnosis requires a response time of 100ms to assist an operator (who suffers from information overload) in a power system control centre to isolate power network (consisting of 110 parameters) faults in an emergency situation. The purpose of this project is to design and implement a parallel and distributed multi-agent system which will process real time alarm data from 16 substations linked with Thomastown Terminal Power Station & isolate faults within 100ms based on 1500 alarm events/minute. The previous work (which achieved response time of 30secs[1]) is based on collaboration with State Electricity Commission (now the Victorian Power eXchange (VPX)). Funding Acknowledgements Problem Space A typical power system consists of three levels of control, namely system level control or EMS (Energy Management System) control, grid or regional level control, and distribution level control. Alarm processing systems are used at all these three levels which can give some indication of the complexity involved. The EMS systems are not only responsible for the overview of the power system grid but also the generation of power and interconnection to other power systems. The grid or regional systems are responsible for transmission of power between generation and distribution systems. Finally, the distribution based systems are concerned with maintenance and delivery of power to the customers from a distribution substation. An alarm is a structured signal from the computerized supervisory control and data acquisition (SCADA) system which is used in a power system control centre at system, regional or distribution level. Alarm processing thus relies on status information or measurements gathered at a large number of points distributed throughout the system. The advances in computer and telecommunications technology in the last three decades have made it possible for a large SCADA system to scan 20,000 to 50,000 points every few seconds. Such a SCADA system will have the ability of displaying 500/1000 or even more alarm messages per minute (Kirschen et al. 1988; Munneke and Dillon 1989; Kirschen and Wollenberg 1992). These days three or more operators are needed to oversee such a complex system which scans dozens of terminal stations and substations spanning hundreds of miles. The enormity of the system and rate at which messages are displayed on the monitor increases the complexity of the problem multi-fold. This complexity leaves a operator in the control centre who has to analyse these messages in a highly constrained time frame suffering from high stress and cognitive overload. Under normal circumstances, the operators carry out routine actions and adjustments to optimize the security and economics of the power system. Uncontrollable events such as sudden load fluctuations, equipment failures and atmospheric perturbations can propel the system from a stable and secure state to an insecure and unstable state. The operator in these circumstances has to take immediate action in order to pull back the system into an acceptable state. Failure to respond quickly, can lead to catastrophic circumstances like the state of the system may continue to deteriorate with some loads getting disconnected or in extreme cases, the entire system may collapse, leading to a blackout for hours. An approximate estimate of number of alarms which could be triggered under such or similar events in a regional control center is (Durocher 1990):
Such an enormous rate of alarm messages makes a quick response from the operator difficult. First, by suggesting a catastrophe (because of large number of messages) , it may increase the level of stress beyond the threshold level at which the performance of the operator drops down sharply. Having overcome the element of surprise, the operator must sift through a large number of messages to find the cause of the problem. This may even involve scanning through the design manuals or drawings as the operator finds it difficult to process the large number of messages on their own. A significant amount of time can be wasted in this search which might be crucial to prevent a deterioration of the situation. Finally, an operator working under stress and with an over abundance of data may easily be misled as to the true nature of the problem. Demonstrations Team Members Publications [2] R. Khosla and T. Dillon , Neuro-Expert Systems for Power System Problems(expanded version)', in International Journal of Engineering Intelligent Systems , CRL Publishing, U.K., Vol. 2, No. 1, March 1994, pp. 71-78 [3] R. Khosla, `Knowledge Learning systems Techniques Utilizing Neuro-systems
and [4] R. Khosla and T. Dillon, 'Neuro-Expert System Applications in Power
Systems', a [5] R. Khosla and T. Dillon, 'GENUES Architecture and Application', a
chapter in a [6] R. Khosla and T. Dillon, 'A Symbolic-Connectionist Model for Time
Critical and [7] R. Khosla and T. Dillon, 'An Integrated Neuro-Expert System Model
for Real- [8] R. Khosla and T. Dillon , 'Neuro-Expert Architecture with application in a Power System Control Centre', in Fourth IEEE conference on Tools with Artificial Intelligence , Arlington, Virginia, U.S.A., November, 1992 , pp.471-72. [9] R. Khosla and T. Dillon , 'A Cognitive Architecture for Problem Solving with Symbolic-Connectionist Constituents', in European Conference on Cognitive Science in Industry , Luxembourg, Sept. 1994, pp.369-82. [10] R. Khosla and T. Dillon,' Distributed Symbolic-Subsymbolic Agent Architecture for Configuring Power Network Faults', in International Conference on Multi-Agent Systems, ICMAS’95, San Francisco, USA, pp. 451. June, 1995, [11] R. Khosla and T. Dillon , 'Learning Knowledge and Strategy of a Generic Neuro- Expert System Model', in AAAI (American Association of Artificial Intelligence) International Symposium on Integrating Knowledge and Neural Heuristics , Florida, USA, May, 1994., pp. 114-21. [12] R. Khosla and T. Dillon , 'A Distributed Real-Time Alarm Processing System with Symbolic-Connectionist Computation', in Proc. of the IEEE Workshop in Real Time Applications, Washington D.C., July, 1994., pp. 104-9. [13] R. Khosla and T. Dillon , 'Abstraction - a Key to Integration of Artificial Neural Networks and Expert Systems', in Workshop on Integration Technology for Real- Time Intelligent Control Systems(IRTICS'93) - ESPRIT Project , Madrid, Spain, Oct. 1993, pp. 18-1-18-13 [14] R. Khosla and T. Dillon , 'Applying An Integrated Neuro-Expert System Model in a Real-Time Alarm Processing System', in SPIE proceedings on Applications of Artificial Intelligence XI: Knowledge Based Systems in Aerospace and Industry , Orlando, Florida, U.S.A, April 1993, pp. 68-79. [15] R. Khosla and T. Dillon, 'Combined Symbolic-Artificial Neural Net Alarm Processing System'- (deals with power system aspects of alarm processing), in Eleventh Power Systems Computation Conference (PSCC), Avignon, France, Sept. 1993, pp. 259-266. [16] R. Khosla and T. Dillon,'Enabling Technology for Diagnostic Applications,
in [17] R. Khosla. and Q. Li.,"Multi-Layered Multi-Agent Architecture
with Fuzzy [18] R. Khosla and T. Dillon, 'A Neuro-Expert System Approach to Power
System
Content Approved by: Director of BII-BSKM Lab
Page maintained by: Business Web Developer Last Updated: 23 January, 2007 |