Exploring the spatio-temporal changes of poverty and identifying the factors that cause poverty can provide reference for the formulation and implementation of poverty alleviation policies.Poverty is caused by many factors. Geographically Weighted Regression model (GWR) can analyze the spatial differences in the influence of various factors on poverty,but there is a strong correlation between the factors causing poverty,which leadsto multicollinearity. Principal Component-based Geographic Weighted Regression method (PCA-GWR) is usedin this paper by combining the natural, economic and social attributes toanalyze the characteristics of the spatial pattern of poverty.In order to explore the spatio-temporal changes of poverty, this paper analyzes the temporal and spatial patterns of village-level poverty incidence from 2013 to 2017. Spatial autocorrelation analysis was performed using global Moran's I index and local G coefficientrespectively.Selecting Yongxin County of Jiangxi Province as the research area, the results show that: (1) There is a high correlation between independent variables affecting poverty. When these variables are put together in GWR model, the multicollinearity problem is easy to occur, and the results of GWRanalysis are not reliable. In order to eliminate the multicollinearity problem, Principal Component Analysis (PCA) was performed on the variables that were significantly correlated with the dependent variables. Three principal components were extracted by principal component analysis, including self-development ability of rural subjects, topographic and vegetation index. The Variance Inflation Factors（VIF）value of the variable in the PCA-GWR model is significantly lower than that in the GWR model. The PCA-GWR model effectively solves the multicollinearity problem in the GWR model. (2) The result of PCA-GWR found that the poverty in Yongxin County is the result of the combination of natural factors such as topographic factors and vegetation distribution and the self-development ability of rural subjects such as low-education, lack of labor, disease. And the effects of these factors presented different spatial patterns. This can provide a reference for the formulation of government poverty alleviation policies. (3) From 2013 to 2017, the incidence of poverty in Yongxin County decreased from 11.27% to 0.97%, showing a downward trend year by year, and the poverty gap between villages decreased year by year. The incidence of poverty from 2013 to 2015 was high in the west and low in the east. The overall value in 2016 and 2017 was low. (4) From the perspective of spatial correlation: on the whole, the spatial correlation between 2013 and 2016 is positive, and it is randomly distributed in 2017; Locally, the distribution of cold and hot spots did not change much from 2013 to 2016, the cold spots were distributed in the middle, and the hot spots were concentrated in the southwest. In 2017, hot spots are distributed in the south, and cold spots are scattered in the north.