限值响应超平面模式识别的模型及计算方法

Hyperplane recognition model and methodology for estimation of response values within limits

用模式识别的观点，探讨了以本征值分解法研究响应值在窄的上下限内的限值控制问题．通过变换坐标表象（ｒｅｐｒｅｓｅｎｔａｔｉｏｎ），可以将一个超平面嵌入响应值符合控制要求的输入模式的薄分布层中，并划分出一定的模糊分界面±ｄ’，使之与响应值的限定范围±ｄ相对应．子空间中的超平面由协方差矩阵的一组本任解确定，且最小本征值所对应的本征矢为超平面的法矢．从拟合特定输入模式的分布出发，建立输入－输出之间的一般联系，减少了在模型建立和修正过程中对系统正常运行的干扰．

From the view point of pattern recognition, response values within narrow limits are studied by use of eigenvalue decomposition method. By way of change of coordinate representation, a hyperplane may be laid in a thin layer, within which responce values accord with control demands. A narrow ragion |y'|≤d' can be approximately seperated by a hyperplane, which corresponds to the response values within the limits of ±d. The hyperplane in subspace is determined by a at of eigen solutions of the covariance matrix,with the least order eigenvector as its normal vector. Relations between input and output are established by fitting the distribution of input pattern and thus disturbances to the normal metion of system are decreased.