A novel adaptive filter algorithm based on DFP technique

Applying a variable convergence factor technique, we derive an adaptive filter based on Davidon-Fletcher-Powell (DFP) algorithm and present update recursion of the inverse autocorrelation matrix estimation. Under MATLAB 7.0, DFP algorithm is implemented in parameter identification and short-term traffic flow prediction. Simulation results obtained demonstrate that divergence may exist in application to parameter identification because of inappropriate parameters´ selection for LMS and RLS algorithm that must predefine parameters algorithms. However, DFP algorithm can always guarantee its stability and convergence characteristics. Applications to short term traffic flow prediction show that DFP algorithm is well capable of reflecting change tendency and regularity of traffic flow series and is characteristic of higher prediction accuracy.