The quality of power supply is more and more important in today's industry. The distribution network has to keep up with the growing consumption. This requires increasing effort in design and maintenance to provide quality power supply.
In many cases the reason of supply disturbance is caused by faults on medium voltage networks. Recognizing the problems, then locating and repairing it is often a complicated task. With decision support by intelligent systems much improvement can be achieved.
This work focuses on one aspect of this problem: Throughout system restoration many switching has to be done (eg. to separate a certain part of the network for trials). Meanwhile as many consumer as possible have to be energized with the shortest interruption. In a small network with few switches and tie lines this doesn't look like a challenging task, but in a more complex network it poses difficulties.
A new genetic algorithm based method is introduced in this article to address the issues mentioned above, which can determine the necessary switching operations.
The objective is to supply the maximum number of consumers, with few switching operations and with short interruption. Switches can be locked in case that certain switches can not be operated, or are required to stay in opened or closed position. Some restriction has to be considered: Over-current on lines and transformers has to be avoided and the voltage level need to be in the regulation defined threshold. For this reason a load-flow calculation is necessary, which is preferably fast and simple, because it needs to be calculated many times. Fortunately the parameters needed for the load-flow are available at the utilities here in digital format. With the help of real time measurements on the HV/MV transformers, consumption can be estimated. Only this and the position of the switches has to be online available at starting of the algorithm (these are also available today). Thus new equipment installations on the network are not necessary. Of course if extra information can be used, it improves the quality of the results.
The solving process is separated in two parts. Thereby the complexity of the problem is reduced to a more manageable level. Firstly the desirable network topology is determined, secondly the order of the switching operations is decided.
The first part is a multi-objective constrained problem, for which a genetic algorithm based on NSGA-II is developed.
The second part is a single-objective constrained problem. For which a genetic algorithm with dynamic stochastic ranking is developed.
Through testing it is observed that the method provides good solutions reliably.
The results can be applied to
computer assisted decision making,
network design (eg. Remote controllable switch placement),
training simulators (eg. At staff trainings, we can have objective indicator of their performance in random situations).