The invention relates to a UTE-LSTM-SA-based
community user charging load prediction method, an electronic device and a medium, and the method comprises the steps: obtaining multi-
source data, the multi-
source data comprises
community user behavior data, environment data and power data, the
community user behavior data comprises commuting distance,
travel time and user type labels, and the environment data comprises the environment data and the power data; the environment data comprises temperature data and
weather data, and the power data comprises historical charging load and time-of-use
electricity price; constructing multi-dimensional features based on the multi-
source data, wherein the multi-dimensional features comprise commuting association features, time scene features, external influence features and abnormal value correction features; constructing a community user charging load prediction model based on a UTE-LSTM-SA prediction framework and the multi-dimensional features; and solving the community user charging load prediction model based on constraint conditions to obtain a community user charging load prediction result. According to the method, the commuting distance, the environment temperature and the time-of-use
electricity price characteristics are fused, the charging load in the next 24 hours is predicted through the customized
deep learning model, and the prediction precision is improved.