An online multiple-model modeling method based on spatial partition tree and lazy learning is suggested for complex nonlinear system. The new method establishes the optimum local model of the system based on lazy learning algorithm, which is on the basis of divide-and-conquer principle and input-output data. As working points changing, multiple local models were built to realize the exact modeling for the global system. To select local neighborhoods of the query points, a hierarchical searching strategy based on spatial partition tree is present, and as a result, the real-time performance of the modeling is improved. Simulation results showed the effectiveness of the proposed method.
