Non-parametric regression forecasting method is widely used in short-term traffic flow forecasting. To improve
its searching speed and parameters optimization, the ??-dimension(KD) tree is used as the structure of the pattern database,
which can effectively increase the searching efficiency and allow newly-recognized traffic flow pattern to insert into the
pattern database dynamically. Then, the genetic algorithm is used to optimize parameters setting. Experiment results show
that better parameters are found by using the algorithm. These results are valuable for the short-term traffic flow forecasting
system.