Advanced hybrid genetic algorithm for short-term generation scheduling

This paper is concerned with the development of hybrid techniques for short-term generation scheduling in which the basic element is a genetic algorithm (GA). Particular emphasis is placed on investigating hybrid approaches, involving combining a GA with an artificial neural network (ANN) and/or a priority list (PL), which are significantly more cost effective than the techniques available hitherto. First, a prescheduling of the generating units using the ANN is presented. Since the solution attained via the ANN may not be feasible for the entire scheduling period, the GA (or the GA with a PL) is subsequently employed to adjust the prescheduling. To take load-forecast uncertainty into account and to make the schedule feasible and cost effective, a typical ±5% load forecast error is incorporated into the ANN training data. The proposed techniques are shown to be efficient and flexible based on the simulations performed on the IEEE reliability test systems. In addition, the proposed hybrid techniques are shown to have higher potential to enhance performance and expedite the solution, overcoming the problems associated with the traditional GA-based technique