A New Disaster Monitor and Forecast System Based on RBF Neural Networks

A new disaster monitor and forecast system based on RBF neural networks is proposed. This disaster forecast system consists of disaster spatial monitor subsystem that is pre-trained by off-line learning algorithms and disaster time forecast subsystem developed by online learning algorithms. The disaster spatial monitor subsystem aims to detect trend of the objective behavior, once the unstable condition is detected, the real time series will be collected and used to forecast disaster by the disaster time forecast subsystem. Using real time data on the eve of disaster, which contain more information about the disaster, this system can largely improve the pertinence and guarantee the accuracy of disaster forecast. To illustrate the feasibility and effectiveness, this system is applied to the landslide forecast. Simulation results show that the spatial monitor subsystem can effectively forecast the stability of the various monitoring points and the time forecast subsystem established by online learning algorithms has better forecast precision as compared with the subsystem established by the Orthogonal Least Squares algorithm. The results also illustrate that the system holds high precise if real time date are sufficient and has a broad application prospects.