| 直接加权优化辨识的最小概率设计 |
 点此下载全文 |
| 引用本文：王建宏,许莺,毛少杰, 徐波.直接加权优化辨识的最小概率设计[J].计算技术与自动化,2015,(3):21-25 |
| 摘要点击次数: 2354 |
| 全文下载次数: 52 |
|
|
| 中文摘要:研究非线性系统辨识的一种新方法-直接加权优化辨识，对于该方法中未知权重值求取，采用统计学习理论中的最小概率准则作为逼近的误差准则函数。最小概率策略选择为最小化估计误差边界应小于某指定门限值的概率，此估计误差边界来源于有限个数的数据点。将最小概率准则转化为一个最大化问题，对于此最大优化问题，通过代数运算来求解此最大化问题以得到未知权重值的显式表达式。权重估计值有类似核估计的渐近收敛特性，且为独立未知噪声方差值的显式形式。最后用仿真算例验证本文方法的有效性。 |
| 中文关键词:非线性系统辨识 直接加权优化辨识 最小概率设计 统计学习 |
| |
| A Minimal Probability Design in Direct Weight Optimization Identification |
|
|
| Abstract:This paper studied a new nonlinear system identification method-direct weight optimization identification. To solve the unknown weight values, the minimal probability criterion from statistic learning theory was used to be the approximated criterion function. The minimal estimation error bound must be less than one certain probability value of some specified threshold value in the whole minimal probability method. This estimation error bound is derived from limited number of data points. Then the minimal probability criterion can be converted into a maximize problem through some changes. The explicit expressions about these unknown weights, which exist in the maximize problem, can be obtained through some algebra operators. These weight estimation values are similar to the kernel estimation and t independent of the unknown noise variance, because they all have same asymptotic property. Finally, the efficiency of the proposed strategy can be confirmed by the simulation example results. |
| keywords:nonlinear system identification direct weight optimization identification minimal probability design statistic learning |
| 查看全文 查看/发表评论 下载pdf阅读器 |
|
|
|
位访问者,今日访问
次