The nonlinear blind source separation algorithm based on kernel function has been widely applied in blind signal progress. However the learning rate of traditional kernel function method is fixed. If the learning rate is unsuitable, the aogorithm would be convergenting difficult or would never convergent. So combined with simulated anneling aogoruth ,an adaptive nonlinear blind source separation based on kernel function was proposed. The result of simulation and experiment indicate that the improved algorithm can improve the convergence performance and the effect of BSS. The improved algorithm has better ability of noise reduction and feature extraction than fixed rate nonlinear blind source separation.