In power systems， in order to implement on-line optimal dispatching，preventative control，security assessment and potential transmitting capacity decision-making， it is fundamental and crucial to grasp the load variation regularity of each node. Based on previous researches， this paper makes a further study and analysis on adaptive dynamic ultra-short term forecasting used for multi-node active and reactive load variation regularity. This paper proposes a load data hierarchical and partitioned processing method， establishes a formula to reflect their mutual restraint and relation， creates a model to describe transmission system multi-node load dynamic characteristic on the basis of top layer forecasting using recursive least square support vector machines（RLS-SVM） algorithm， and constructs an ultra-short term load forecasting overall frame of adaptive dynamic model. The application in an actual power system control center of Shandong Province has been verified with satisfactory results.