论文部分内容阅读
根据灰色新息优先利用思想,定义新的累加生成,与灰色离散模型结合,构建出新息优先累加生成的灰色离散模型(NIPDGM(1,1))。在四种误差准则下,给出了参数优化方法。进一步利用数值模拟,研究NIPDGM(1,1)模型在不同误差最小化下对信息的重视程度,分析表明在序列累加生成过程中,四种优化形式对信息的重视较为一致。在实证部分,以高速公路软土路基沉降以及江苏省能源消费问题为例,分析NIPDGM(1,1)模型的建模精度,结果表明:在NIPDGM(1,1)实证模型中,不同误差优化方式对信息的重视程度与数值实验结论相符;与GM(1,1,t2)、反向累加GOM(1,1)、倒数累加GRM(1,1)、GM(1,1)、DGM(1,1)、无偏GM幂模型相比,NIPDGM(1,1)对路基沉降的建模精度更优;与RBF神经网络、灰色累加生成RBF神经网络(GRBF)、支持向量机(SVM)、灰色累加生成支持向量机(GSVM)相比,NIPDGM(1,1)对能源消费的模拟误差大些,但预测误差更小,表明新模型具有更好的泛化能力。
According to the idea of prioritizing gray interest, the new cumulative generation is defined and combined with the gray discrete model to construct the gray discrete model (NIPDGM (1,1)) which is generated by priority accumulation. Under the four error criteria, the method of parameter optimization is given. Furthermore, numerical simulation is used to study the importance of information in NIPDGM (1,1) model under different error minimization. The analysis shows that the four optimization forms have the same emphasis on information in the sequence accumulation. In the empirical part, the modeling precision of NIPDGM (1,1) model is analyzed with the settlement of expressway soft soil subgrade and energy consumption in Jiangsu Province as an example. The results show that in the NIPDGM (1,1) empirical model, different error optimization (1,1), GOM (1,1) in the reverse direction, GRM (1,1), GM (1,1), DGM (1,1) Compared with unbiased GMM model, NIPDGM (1,1) has better modeling accuracy for subgrade settlement. With RBF neural network, RBF neural network (GRBF) and support vector machine (SVM) Compared with GSVM, NIPDGM (1, 1) has a higher simulation error of energy consumption but a smaller prediction error, indicating that the new model has better generalization ability.