Mountain torrent early warning method and system based on cross-source interpretable deep learning model

The invention relates to the technical field of mountain torrent early warning, and particularly discloses a mountain torrent early warning method and system based on a cross-source interpretability deep learning model, and the method comprises the steps: obtaining mountain torrent parameters and environment parameters in an early warning region, and calculating the correlation degree of the mountain torrent parameters and the environment parameters; environment parameters are selected and combined according to the correlation degree, and corresponding mountain torrent parameters are connected to serve as samples of the early warning area; counting samples of different early warning areas, training a neural network model, and synchronously constructing an interpretation content library; and when the neural network model is applied, determining supplementary content of an output result according to the explanation content library. According to the method, other possible inputs of the same output can be obtaine