Channel large-range water depth rapid prediction method compatible with curved river reach

The invention discloses a channel wide-range water depth rapid prediction method compatible with a curved reach, and the method comprises the steps: 1, building a water depth field represented by time and space functions based on a historical water depth data set; the representation of the water depth field is established in a form of being decomposed into a plurality of orthogonal modes; establishing historical water depth data represented based on a water depth space mode and a corresponding time coefficient; 2, taking historical water depth data represented by a time coefficient matrix as a sample set, taking a time coefficient as a prediction target, taking hydrological data and river channel characteristic data as input, and completing prediction output through a neural network; and 3, sequentially carrying out reverse normalization inversion processing on the output time coefficients to obtain the predicted water depth. According to the rapid prediction method for the water depth of the channel, the neu