论文部分内容阅读
针对基于统计的大坝安全监测预报模型中,多效应量间和多影响因子间都存在互相关性,且效应量与影响因子又呈现出复杂的非线性动力系统特征,从而导致预报模型可信度降低的问题,提出了优化方案,首先对多效应量和影响因子采用基于主成分提取的关联分析,实现去相关和空间降维,并按关联性次序将变换后的正交基作为模型输入因子,建立改进的BP神经网络回归,利用人工粒子群算法搜索网络的最优参数,从而获得预报模型。经与逐步回归、简单BP神经网络回归比较验证,实例表明本预报模型具有收敛快、鲁棒性强和预报精度较优等特点,兼有大坝性态分析评估辅助意义,具有一定的工程实用性。
For the statistical dam safety monitoring and forecasting model, there exists cross-correlation between multi-effect variables and multiple influence factors, and the effects and influencing factors show complex characteristics of nonlinear dynamic system, which leads to the credibility of the forecast model Degree of reduction, put forward an optimization scheme. Firstly, the correlation analysis based on principal component extraction is applied to the multi-effects and influence factors to achieve decorrelation and space-reduction, and the transformed orthogonal basis is input as the model according to the relevancy order Factor to establish improved BP neural network regression, using artificial particle swarm optimization algorithm to search the optimal parameters of the network to obtain the forecasting model. The results show that the forecasting model has the characteristics of fast convergence, strong robustness and good forecasting accuracy. It is also helpful for the analysis and evaluation of the dam properties and has certain engineering practicability .