首页|ANN模型与分段线性插值及回归模型的比较及应用

ANN模型与分段线性插值及回归模型的比较及应用

扫码查看
对ANN模型、分段线性插值模型和非线性回归模型从原理上进行了比较,ANN模型易于构建各影响因素与因变量间复杂关系,非线性回归模型和分段线性插值模型可以将自变量与因变量间的关系通过表达式直观表达。以荆江三口分流量与枝城流量的关系为应用算例,采用相关系数、纳什效率系数、均方根误差和平均绝对误差等4个评价指标对3个模型的拟合精度和误差大小进行了比较。结果表明:3个模型均可应用于模拟枝城流量与荆江三口分流量的关系,但3个模型的计算值与实际值间的误差大小存在差异,从4个评价指标综合来看,ANN模型计算值与实测值的误差最小,分段线性插值模型次之,回归模型计算精度相对较低。
Comparison and application of ANN model,piecewise linear interpolation model and nonlinear regression model
The ANN model,the piecewise linear interpolation model and the nonlinear regression model are compared in principle.The ANN model is easy to construct the complex relationships between the various in-fluencing factors and dependent variables.The nonlinear regression model and the piecewise linear interpola-tion model can intuitively express the relationship between the independent and dependent variables through expressions.Taking the relationship between the diversion flow rate through the three outlets of Jingjiang River and the flow rate at Zhicheng station as an application example,the correlation coefficient,Nash efficiency co-efficient,root mean square error and mean absolute error are used to compare the prediction accuracy and error of the three models.The results show that all the three models can be applied to simulate the relationship be-tween flow rates through the three diversion outlets and at the main channel of Jingjiang River.However,based on the four evaluation indexes,the error between the measured value and calculated value by the ANN model is the smallest,followed by the piecewise linear interpolation model,and the calculation accuracy of the regression model is relatively low.

ANN modelnonlinear regression modelpiecewise linear interpolation modelJinjiang Riverthe three diversion outlets

赵伟、毛继新、关见朝、吴兴华

展开 >

中国水利水电科学研究院流域水循环模拟与调控国家重点实验室,北京 100048

中国长江三峡集团有限公司长江生态环境工程研究中心,北京 100038

ANN模型 非线性回归模型 分段线性插值模型 荆江河段 三口分流

中国水利水电科学研究院基本科研项目中国长江三峡集团有限公司项目

SE0199A102021201903144

2024

泥沙研究
中国水利学会

泥沙研究

CSTPCD北大核心
影响因子:0.817
ISSN:0468-155X
年,卷(期):2024.49(4)
  • 8