Abstract
Precipitation is a kind of dynamic variable with multiple time scales, and has complex nonlinear characteristics. Although deep learning model has strong learning ability for nonlinear data. However, due to the particularity of precipitation, especially hourly precipitation, the distribution of data is not uniform, and the characteristics of small numerical precipitation are not obvious. It is difficult for the model to capture the characteristics of precipitation. Therefore, this paper proposes a preprocessing method for logarithmic transformation of hourly precipitation and then normalization. After logarithmic transformation, the distribution of hourly precipitation data is more uniform, and the data characteristics are more obvious, especially the characteristics of small value precipitation, which is more conducive to deep model learning. Through experimental comparison, it is found that the logarithmic transformation of hourly precipitation data has achieved better experimental results in various deep learning models.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have