| Abstract: | Thermal units must be maintained periodically as prescribed by the electric utility industry law. As time to execute maintenance works increases with thermal unit capacity, maintenance scheduling has a great influence on the reliability and economy of a power system. In the recent amendment in the law, three inspection rankings have been introduced and scheduling over several consecutive years becomes mandatory, thus making maintenance scheduling extremely difficult. Reflecting a recent stringent supply capability, the emphasis is laid on security rather than a minimum operating cost, having been the primary objective in determining the schedules. Therefore, this study aims to level the spinning reserve at each period under study in the maintenance scheduling while taking into consideration all the amendments in the law. Although rigorous methods such as integer programming and branch and bound method can solve small scale problems, large size problems are beyond these techniques due to an exponential explosion in the number of possible combinations. The prime objective of this paper is to investigate the capability of the Hopfield neural network (HNN) in solving the newly formulated maintenance scheduling problem. The scheduling problem has been mapped on the HNN with slight problem relaxations to facilitate the implementation. A small scheduling problem that determines the maintenance schedules of 3 generators over 3 years (divided to 78 periods) has been solved by the neural network simulator. It has been made clear from simulation results that the proposed approach is very promising in handling a complicated combinatorial optimization problem. |
