Abstract: The wet-dry features of annual runoff depend on the size and time-history distribution characteristics of runoff itself.In view of this,we put forward a LBA-PP model of wet-dry classification of annual runoff by searching the optimum projection direction using bat algorithm (LBA) improved with a Lévy flight strategy in association with projection pursuit (PP) model.We also construct a particle swarm optimization (PSO) algorithm PP model for comparison,with the annual runoff at Xiyang station in Yunnan Province as a case study.Results show that the LBA algorithm is superior to PSO algorithm,and is of good convergence accuracy,robust performance and global optimization ability.Using LBA algorithm to find the best projection direction of PP model not only improves the classification accuracy of the PP model,but also provides a new way and method for the selection of the PP model. In the LBA-PP model,the annual runoff is considered,and the time history information is distributed.The classification results are more scientific and objective than those of conventional method.

Key words: annual runoff classification, bat algorithm, projection pursuit model, parameter optimization, Lévy flight strategy