【摘要】 台风致灾因子、承灾体和灾情之间是一个复杂的非线性动力系统,准确高效地提取重要指标对台风灾情等级进行预评估,是防灾救灾工作的重要依据。采用主成分分析、等距特征映射和信息熵特征提取的承灾体关键指标,和致灾源作为输入神经元,灾情等级作为输出神经元,建立台风灾情概率神经网络预评估模型。结果表明,基于等距映射非线性特征提取的概率神经网络预评估模型的准确率达到90%。更多还原

【Abstract】 Typhoon hazard,between hazard bearing body and the disaster is a complex nonlinear dynamical system; accurately and efficiently extract the important indicators for the pre-assessment of typhoon disaster grade is an important basis for disaster prevention and relief work. In this paper,we apply principal component analysis,isometric feature mapping and entropy to extract key indicators of hazard bearing body,with hazard source as the input neurons,and disaster grade as output neurons,establishing probabilistic neural network pre-assessment model in typhoon disaster. The results show that the accuracy of probabilistic neural network pre-assessment model based on the non-linear feature extraction isometric feature mapping reaches 90%,the model has a satisfactory level of accuracy and generalization ability,provide a new way for natural disaster risk assessment,having certain reference value.更多还原