Rapid identification of rainstorm disaster risks based on an artificial intelligence technology using the 2DPCA method

摘要

An artificial intelligence technology with the two-dimensional principal component analysis (2DPCA) method is introduced into the early identification of rainstorm risks. The characteristic subspace for historical rainstorm events can be constructed through the 2DPCA method. When identifying an ongoing rainfall process, the most similar rainstorm event can be found through comparing the features of the ongoing rainfall to the events in the characteristic subspace. Taking the most similar historical rainstorm event found as a reference, the possible duration and intensity of the ongoing rainfall process can be estimated, thus achieving early identification of rainstorm risks. Two groups of validation experiments are performed based on a database including 116 rainstorm events observed in Shenzhen metropolis, China. The validation shows that although the ratio of historical events to validation events impacts the performance of identification, the identified historical events are generally similar to the ongoing rainfall processes in terms of rainfall duration, range of influence, magnitude, and maximum single-station rainfall.