基于DFP的自适应算法及其在短时交通流预测中的应用

An Effective Adaptive Algorithm Based on DFP Method and Its Application to Short-Term Traffic Flow Prediction

研究了一种基于可变收敛因子的Davidon-Fletcher-Powell(DFP)自适应算法,给出算法中自相关逆矩阵估计的递归更新公式。对DFP算法中参数τ(n)的作用及算法的计算复杂度进行了分析。当分别输入正弦信号和高斯白噪声时,对不同滤波器阶数的τ(n)随样例个数变化情况进行了仿真,并将DFP算法分别应用于3 min和5 min短时交通流预测。结果表明:τ(n)最终将趋于稳态值0.5,DFP算法能够较好地反映交通流量变化的趋势和规律,预测精度较高。

By applying a variable convergence factor on the basis of a posteriori error assumption,we study an adaptive filter algorithm based on Davidon-Fletcher-Powell method and present the update recursion equation of estimate of inverse of auto-correlation matrix.The effects of the parameter τ（n） and computational complexity of the DFP algorithm are presented.Under MATLAB 7.0 environment,when the input signals are sine curve waves and white Gaussian noise,respectively,how τ（n） changes with the sample numbers is simulated under different filter orders.Moreover,DFP algorithm is implemented in 3-minute and 5-minute short-term traffic flow prediction.Simulation results and their analysis demonstrate preliminarily that： τ（n） eventually tends to steady-state value 0.5 and the proposed DFP algorithm is well capable of reflecting change tendency and regularity of short-term traffic flow and presents high-precision prediction.