Electric vehicle charging behavior is the basis for accurately predicting charging demand and achieving smart charging. Considering the multi-peak distribution characteristics of the charging behavior time data. An adaptive diffusion kernel density estimation model (ADKDE) based on the diffusion equation is used. Firstly, the Gaussian kernel function is converted into a linear diffusion process using the diffusion heat equation, and the asymptotic mean integrated squared error (AMISE) is used to select the adaptive optimal bandwidth for the diffusion kernel function in order to fit the multi-peak data distribution better. The validation is based on the charging order data of a district in Beijing in 2021. The results show that the adaptive diffusion kernel density estimation model can best track the changes of peaks and troughs of real samples, and has more accurate fitting results for one-dimensional and high-dimensional data.