In the scenario of joint power forecasting for multiple sites and multiple types of sources and loads

traditional time-series forecasting models are often difficult to efficiently extract spatiotemporal correlations and inherent relationships among different types of sources and loads

resulting in insufficient forecasting accuracy. To address this issue

this paper proposes an ultra-short-term source-load joint forecasting method that integrates an improved graph convolutional neural network (GCN) with a group cross-gating mechanism. First

an ultra-short-term joint power forecasting model based on the improved GCN is constructed to fully extract the spatiotemporal correlation features of source (wind

solar) and load (electricity

heat) power. Then

a group cross-gating mechanism is designed and incorporated into the improved GCN

thus the cross-modulation of information between different types of sources and loads is realized

effectively utilizing their inherent relationships. This significantly improves the accuracy of ultra-short-term joint power forecasting for multiple sites and multiple types of sources and loads. Finally

based on historical data of wind power

solar power

electric load

and heat load from a region in northern China

comparative experiments are conducted. The results verify the effectiveness and superiority of the proposed method in terms of prediction accuracy.