基于改进GMDH的科技成果中试转化系统模型构建

基于科技成果中试转化运行原理，搭建中试系统机理模型，利用改进GMDH构建相应数学模型。改进GMDH将含有回归项线性相关检测的快速递归算法（FRA）代替传统最小二乘（OLS）进行GMDH多层建模的部分多项式求解，使用初始变量添加法避免有效神经元过早丢失。以改进的GMDH对上海市中试转化统计数据进行建模及仿真实验，结果表明，所建模型比传统GMDH模型精度高，泛化能力强。

Technological Achievements Pilot Conversion Model Building Based on the Improved GMDH

On the basis of analysis of the pilot process operation principle, this paper builds a pilot conversion mechanistic model and its mathematical model through using the improved batch data processing algorithm （GMDH） ; This paper replaces the traditional least squares algorithm with the fast recursive algorithm （FRA） containing a linear correlation detection to calculate the coefficients of the polynomial in the process of the muhilayer modeling, and designs the initial variables to avoid active neurons losing early. The improved algorithm is applied to system modeling based on the Shanghai pilot conversion statistical data, the simulation results show that the accuracy and generalization capability of the built model are better than the traditional model.