Completion of hydrological data using neural networks, for reassessment of reservoirs in Vietnam

摘要

In this paper, a new approach to generating input learning patterns (LPs), which can handle incomplete data records and tolerate unreliable data quality, is introduced. Based on an architecture of feed-forward neural networks (FNNs), its application for infilling the gaps of monthly rainfall data of Tan-cuong Station (Nui-coc Reservoir) and monthly runoff data of Ke-go Station (Ke-go Reservoir), in Vietnam, shows better results compared with the simple average method (SAM) combining the outputs of different linear regression (LRG) models.