基于虚假最近邻点GT准则的化工模型变量选择

摘要

To against traditional variable selection method could not obtain the input variables, which could always reasonably explain the output, because the chemical industrial modeling is nonlinearity and the method belongs to unsupervised learning. This paper put forward a method for complex nonlinear chemical industrial modeling based on the false nearest neighbors and Gamma test( GT). Firstly, it inspired by the false nearest neighbors, thinking about only one variable every time though setting it zero, could search all variables simply. Secondly, it calculated the Gamma statistics for every variable using the Gamma test when it' s zero and not zero. Finally, it obtained the sensitivity of output to input to carryout the selection of variables. The example of linearity and nonlinearity were given to validate the method. Finally, confirming the input variables of BP artificial neural networks for the hydrocyanic acid by the method, the modeling is high precision. Therefore, it provides a new-method for the variable selection of the complex nonlinear chemical industrial modeling.%针对传统变量选择方法对复杂非线性化工模型进行变量选择时,由于缺乏输出变量的有效监督,导致所选择输入变量不能有效解释输出变量的问题,提出基于虚假最近邻点Gamma检验(Gamma test,GT)准则的变量选择方法.首先借鉴虚假最近邻点法,实现对所有变量的全面搜索；再采用能够在输出变量监督下进行非线性系统噪声估计的GT准则,计算各输入变量置零前后数据噪声的伽马统计量,得到输出变量对各输入变量的敏感度,以此为依据进行变量选择.使用线性、非线性模型验证了该方法的有效性.最后对氢氰酸复杂非线性化工过程建模进行变量选择,结果表明合理的变量选择有效地提高了模型精度.