Abstract:The dynamic reserve in meizoseismal area is influenced by many factors and embody nonlinear characteristics, which is difficult to estimate. Under this circumstance, a coupling model between particle swarm optimization, extreme learning machine and AdaBoost algorithm(PSO-ELM_AdaBoost) is proposed based on principal component analysis(PCA). Taking 60 groups of debris flow samples from the Wenchuan meizoseismal area, this paper selects six influencing factors including: the gully’s area, relative height difference, the main channel length, the distance from seismogenic fault zone, the mean gradient and the total amount of material source based on the source formation and starting mode of the debris flow. First, Pearson Correlation Coefficient(PCC), Grey Relational Grade(GRG) and Maximal Information Coefficient(MIC) were used to analyze the sensitivity and response of the impact factors, so as to verify the rationality of the selection factors. Secoudly, the sample data are dimensionality reduced based on PCA to avoid information redundancy. Then, the PSO-ELM_AdaBoost coupling model is used to train and predict the processed sample data and compare the results with the estimated values of BP, SVM, ELM and PSO-ELM models. Finally, in order to verify the suitability of the model, A test sample set, which is obtained by extracting a sample from each sub-research area and three samples from other meizoseismal area, use the PSO-ELM_AdaBoost coupling model to calculate. The results show that the coupling model has higher accuracy and better versatility and stability compared with traditional calculation model and other neural network models. It can provide valuable reference for the design of debris flow prevention and control.