A memetic algorithm for dynamic economic load dispatch optimization

The dynamic economic load dispatch (DELD) problem is an extension of the conventional static load dispatch problem in the context of electrical power generation. In the static case, the problem is to optimize the settings for each unit in a generating station so as to supply sufficient power to meet a given overall predicted demand for minimal cost. In the dynamic version of the problem, predicted demand exists for each of a number of successive periods (e.g. 24 hourly periods), and the static version of the problem is to be solved for each period. Until now, the DELD has been treated as a series of static problems. In this paper, we take a memetic algorithm (MA) that has recently provided superior results on some benchmark problems for the static ELD, and we now adapt it for the dynamic case, and investigate a simple dynamic optimization approach to this where the final population of a previous period is used to intialise the population for the next period. This is compared with two baselines, in which (i) the static problems are solved independently, and (ii) the static problems are solved together, treated as a single multi-part problem with suitably adjusted constraints. We evaluate our methods on two benchmark cases of the DELD for which published results exist, and we show that the basic dynamic optimization approach, using our MA, has superior performance to both the baseline approaches and to other approaches published in the literature so far.