Abstract: Taking a slope as a unit, the regional vulnerability assessment based on potential disaster intensity is one of the important problems to be solved urgently. In this paper, the city of Xiangxiang in Hunan is selected as the research area. On the basis of susceptibility regionalization with the weighted information value method, the elevation, slope height, slope, slope direction and monthly average rainfall of the highest prone points of slope units are extract one by one as the characteristic parameters, which are put into the BP neural network, PSO-BP neural network, random forest and support vector machine model, respectively. A landslide volume prediction model based on BP neural network algorithm optimized by PSO is constructed through training and precision test comparison. A comprehensive vulnerability evaluation model is established with disaster volume as disaster intensity index and building density, population density and property density as vulnerability indexes. Regional vulnerability evaluation based on potential disaster intensity is carried out for the study area. The divisions of high-vulnerable areas (1.5% of the total area), medium-vulnerable areas (28.5% of the total area) and low-vulnerable areas (70% of the total area) are completed, which realize the organic combination of the disaster intensity of the disaster-causing body and the vulnerability of the disaster-bearing body in the process of regional vulnerability evaluation, and enhance the objectivity and scientific nature of the evaluation.