A deep learning-based water diversion pipeline working condition prediction method and system

Abstract

The application discloses a water diversion pipeline working condition prediction method and system based on deep learning, and relates to the technical field of deep learning.The method first acquires fluid pressure and pipe wall strain data in a water hammer impact window of a water diversion pipeline to construct a sequence; extracts a trajectory sequence through phase space noise reduction and topological projection reconstruction; performs a shape extreme value scanning and path backtracking operation to acquire a dynamic time warping alignment feature sequence, and calculates a wave speed misplacement compensation sequence and a fluid-solid regular residual sequence; then, the residual sequence is subjected to a dispersion measurement calculation to extract a distortion dispersion feature; finally, each feature sequence and a compensation parameter are input into a pre-trained fluid-solid synchronous evolution memory prediction network to output a working condition prediction result.The application removes wave speed time lag interference through physical constraint regularization, amplifies weak pathological evolution in combination with multi-dimensional dispersion measurement, overcomes the defect that a pure data-driven model is prone to fitting noise, and significantly improves the precision and robustness of pipeline hidden disease identification in a strong noise background.

Images