CAJ | 학술논문

将GM(1,1)预测模型与广义回归神经网络(GRNN)相融合,构建一种兼具两者优点、互补型的灰色广义回归神经网络(GGRNN).以1985-2007年度广西木薯鲜薯总产量为数据样本,采用GGRNN模型进行广西木薯产量预测研究.研究结果表明,GGRNN训练期平均拟合指数、预测期平均拟合指数分别为0.99和0.93,分别比GM(1,1)模型高0.09和0.04.该组合模型在拟合精度和预测精度方面均优于单一的GM(1,1)预测模型,并具有自学习能力、非线性映射能力以及适应性强等优点,为木薯产量预测的定量化和智能化提供了一条有效途径.
장GM(1,1)예측모형여엄의회귀신경망락(GRNN)상융합,구건일충겸구량자우점、호보형적회색엄의회귀신경망락(GGRNN).이1985-2007년도엄서목서선서총산량위수거양본,채용GGRNN모형진행엄서목서산량예측연구.연구결과표명,GGRNN훈련기평균의합지수、예측기평균의합지수분별위0.99화0.93,분별비GM(1,1)모형고0.09화0.04.해조합모형재의합정도화예측정도방면균우우단일적GM(1,1)예측모형,병구유자학습능력、비선성영사능력이급괄응성강등우점,위목서산량예측적정양화화지능화제공료일조유효도경.
Gray general regression neural network (GGRNN) is constructed by combining gray model [GM (1, 1)] and general regression neural network (GRNN) with their advantages and a complementary for each other. In the present study, the use of GGRNN was illustrated by the yield prediction of cassava. The total yield of cassava in Guangxi during 1985-2007 was used as data samples to predict the yield of cassava during 2004-2007 by GGRNN model. The results showed that the average FI in training time and predicting time of GGRNN was 0.99 and 0.93, with an increasing of 0.09 and 0.04 as compared to GM(1,1), respectively. The fitting and predicting precision of GGRRN were better than that of GM(1,1), and GGRRN had advantages on convenience of calculation, nonlinear mapping ability and wide suitability, etc. So it would provide an effective method on quantitative and intelligent prediction of yield of cassava.