Multi-source-Load Prediction Based on Multi-task Learning

As wind power and photovoltaic power generation account for an increasing proportion of the energy supply, the forecasting of the output of wind and solar power plants has become crucial. Meanwhile, accurate prediction of power load is also important for grid operation and energy dispatch. Therefore, in this paper, a joint prediction method considering the spatial correlation characteristics of source-load is proposed based on multi-task learning and convolutional gated loop unit. First, the regional source-load correlation characteristics are analyzed. Then, a multi-task learning method using dynamic weight averaging technology is introduced, and reasonable weights are assigned to a single task to realize the joint prediction of wind power plant field groups as well as the joint prediction of source-load. The results of the algorithm show that the accuracy of wind power and load prediction is improved compared with the single field station prediction method.