Short-term power load prediction method based on distributed online learning

The invention relates to a short-term power load prediction method based on distributed online learning, which comprises an offline stage and an online stage, and in the offline stage, at each meteorological sensor end, after meteorological measurement data is preprocessed, training data under the sensor is formed by combining corresponding time and load values. And then performing offline regression learning by using an extreme learning machine to obtain a power load prediction model and a corresponding model weight coefficient under the meteorological data. In the online stage, meteorological data are preprocessed at each sensor end to form measurement data fingerprints, the measurement data fingerprints are substituted into the corresponding power load prediction model, and a final power load estimation value is obtained by using a weighted summation method. And at the same time, parameter updating is performed on the power load prediction model by using online data. The method has the advantages of being