Improvement of GMDH Parametric Modeling and Its Application to Coal Price Research
Zengguang Li, Jing Wang, Lu Ningyun, Fei Dong · Systems Engineering · 2012
An improved GMDH modeling algorithm is presented to overcome the shortcomings of the existing GMDH algorithms in the constructing of partial polynomials and the explicit output expression of GMDH network.The input variables(i.e.,the first layer nodes) are always included in the candidate pool of the partial polynomials for determining the construcution of the higher layers,which can prevent the loss of useful information,simplify the model structure and improve the model's performance.Furthermore,the weight coefficient method or the symbolic expression method is used to obtain the explicit output expression of GMDH network directly.The improved GMDH algorithm is then applied to the case study of coal price in China.The results show that the obtained GMDH model can reflect the actual coal price model structure with high fitting precision.Meanwhile,it can predict accurately the changing tendency of coal price,indicating that the model has good trend extrapolation ability.