基于滑动窗宽的非参数变宽直方图方法研究

Variable-Width Histogram Based on Slide Bandwidth Nonparametric Density Estimation

针对等宽直方图难以定性确定分组数的问题,提出一种基于滑动窗宽非参数概率估计的变宽直方图算法。该算法通过对等宽直方图的区间进行合并或划分操作,使直方图上部轮廓线与参数概率密度函数曲线尽量相吻合,并以此确定直方图的分组数。在概率密度函数未知的情况下,采用滑动窗宽非参数概率密度估计算法预先估计出概率密度函数。滑动窗宽非参数概率密度估计是一种改进算法,将固定窗宽算法和变窗宽算法相结合,实现窗宽根据样本的分布情况,在不同的估计点自动调整窗宽的取值。最后仿真实验表明,该直方图算法是有效的。

The variable-width histogram based on slide bandwidth nonparametric density estimation is proposed for equal-width histogram is difficult to set group number. Through combining or dividing opera- tion for equal-width histogram, the contour line of variable-width histogram and the probability distribu- tion function（PDF） of parameter may inosculate. On the condition of unknown PDF, the method of slide bandwidth nonparametric density estimation can be adopted. This method is an improved algorithm to combine the fixed bandwidth algorithm and the slide bandwidth algorithm, and then the bandwidth can be auto-regulative for different estimation dot according to the distribution of sample. The practical process of the variable-width histogram based on slide bandwidth nonparametric density estimation is provided and the results of comparative experiment indicate that the new algorithm is effective.