Deep learning interpretability-based water level flow relation jacking influence separation method

Abstract

The invention discloses a deep learning interpretability-based water-level flow relation jacking influence separation method. The method comprises the following steps of: collecting multi-source element data influencing a water-level flow relation of a hydrometric station; selecting a deep learning model for simulating the water level and flow relationship of the hydrological station; preprocessing multi-source element data and constructing a water level flow simulation regression model, adopting a selected deep learning model to fit a mapping relation between the multi-source elements and the water level of the hydrologic station, and constructing a regression model of water level flow simulation of the current hydrologic station; and performing precision evaluation and interpretability analysis on the water level flow simulation regression model, evaluating the precision and robustness of the model, and finally separating the jacking degrees of different jacking influence factors on the water level of the hydrometric station based on an improved deep learning interpretability method. On the basis of a deep learning method and an improved interpretability technology, the influence separation between hydrological station water level flow simulation under the influence of multi-source elements and the jacking elements of the water level flow relation of different hydrological stations is realized.

Images