Capacity generation expansion problems have traditionally been represented with low time resolution models due to their high computational cost, very often using blocks of hours with similar demand. However, the current transformation of the power system with the new generation and consumption technologies, the flexibility and reserve requirements, and the expected new behavioral consumption patterns, requires more complex and detailed models with higher time resolution to provide accurate investment decisions and allow for closer analyses. In particular, these challenges require chronological hourly models with constraints linking all the years of the planning horizon, compromising in most cases the computational feasibility. This paper presents a new approach to synthetize a reduced representative time period for capacity expansion problems, for being used in detailed chronological hourly models, while keeping them computationally feasible. The representative period is synthetized by selecting, with a genetic algorithm, those real days that minimizes the distance between the duration curves of a set of relevant variables (such as demand, renewable generation, ramps, etc.) computed for the original and for the representative periods. Results show that investments decisions with the representative period are very similar to those obtained with the full planning horizon, while computational times are strongly reduced.
Keywords: Generation expansion planning, hourly dispatch models, genetic algorithms.
Publicado: junio 2018.