The invention discloses a shared electric vehicle station user demand prediction method. According to the method, a station is selected as a circle center, two data acquisition subsystems are established by taking a buffer area within a range of square and round one kilometer as a research object, wherein one system acquires characteristic data such as population economic characteristics, employment, entertainment, medical service, traffic modes and road networks in the buffer area, and the other system acquires parking space number data of a station, and travel data of the shared electric vehicle in the continuous time; deep mining and fusion are performed on the two pieces of data to form input data, and finally a whole set of shared electric vehicle station user demand prediction technology is established based on semi-parametric spatial geographic weighting Poisson regression; The technology can provide user demand change prediction support for the design of a berth number change adjustment scheme in the existing site layout, the influence of site surrounding built-up environment change on site user demands and the like.