The key techniques of building a real-time forecast model for debris flow disaster using neural network (NN) method are explained in detail in this paper, including the determination of neural nodes at the input layer, the output layer and the implicit layer, the construction of knowledge source and the initial weight values and so on. The neural network-based real-time forecast model for debris flow disaster is built using the rainfall parameters of 40 historical debris flow disasters as training data, which included multiple rainfall factors such as the rainfall of the day disaster happening, the rainfalls of 15 days before the disaster, the maximal rainfall intensity of one hour and ten minutes. Based on the torrent classification and hazard zone mapping of the study region, combined with the rainfall monitoring in the rainy season and real-time weather forecast models, the NN-based early-warning system for debris flow disaster ran well. In this system, GIS technique, advanced international software and hardware were applied, which made performance of the system steady and its applicability wide. It can forecast some most important indices, the probability, the critical rainfall, the warning rainfall, and the refuge rainfall of debris flow occurring, and reduce the direct disserve in the debris flow disasters through the real-time monitoring of rainfall or local weather forecast. As it was a visual information system, we could monitor the variation of the torrent types and hazardous zones, and the torrent management through it, so it could serve the local management and decision-making on the debris flow disaster warning and prevention.
展开▼
