摘要
为合理选取溪洛渡水电站岩体力学参数,运用偏最小二乘回归对神经网络输入数据进行处理,提取了对系统具有最佳解释能力的新综合变量,较好地克服了各因素间的多重线性相关性问题,解决了由于输入数据的严重相关性造成的神经网络模型不稳定及收敛速度慢的问题。结果表明,与单一方法相比,结合方法简化了网络结构,增强了网络稳定性。这一研究能为优化设计提供可靠的依据。
In order to select the mechanical parameters of rock mass reasonably for the construction of Xiluodu Hydropower Station, the input data of neural network are processed with the partial least-squares regression, from which new aggregate variables with the best explanatory ability to the system are abstracted. In this way, the multi-linear correlation among all the factors is overcomed and then the problem from the instability and slow convergent velocity of the neural network model due to the serious correlation of input data is solved as well. The result shows that compared with the single method, the combined method can not only simplify the structure of the network, but also can enhance the stability of it. Generally, this study can provide a reliable basis for the optimization of the design concerned.
出处
《水利水电技术》
CSCD
北大核心
2008年第5期20-22,39,共4页
Water Resources and Hydropower Engineering
基金
2005年度河南省高等学校创新人才培养工程
2005年度河南省高校杰出科研人才创新工程项目(HAIPURT,2005KYCX015)
华北水利水电学院青年基金资助(HSQJ2008008)
关键词
小脑神经网络
偏最小二乘回归
力学参数
溪洛渡水电站
CMAC neural network
partial least-squares regression
mechanical parameter
Xiluodu Hydropower Station