基于主成分分析和RBF神经网络的融雪期积雪深度模拟

Snow depth simulation in the snowmelt period based on principal component analysis and RBF neural network

积雪深度作为估算雪水当量的重要参数,是融雪径流计算及雪融性水灾预警、监测和评估的重要判别因子。融雪期积雪深度的变化过程是一个复杂的非线性系统,传统的预测方法难以精确表述其变化规律。文中采用主成分分析法和RBF神经网络相结合的方法建立融雪期积雪深度模拟数学模型,首先采用主成分分析法对模拟因子进行降维处理,以减少各模拟因子之间的多重共线性,其次采用逼近能力、分类能力和学习速度均优于BP神经网络的径向基(RBF)神经网络建立模型进行融雪期积雪深度变化的模拟,并针对特定流域进行实例分析。结果表明:使用基于主成分分析的RBF神经网络对融雪期积雪深度的模拟是可行的,其模拟成果可为今后该地区积雪深度的预测提供一定的理论支持,并可为该地区进行融雪径流及洪水研究提供合理的预报建议。

Snow depth is not only an important parameter to estimate snow water equivalent,but also an important distinguishing factor to calculate snowmelt runoff,and flood warning,monitoring and evaluation.The snow depth in snowmelt period is a complex nonlinear system.The traditional forecasting methods is difficult to accurately describe the change rule.In this paper we used the method of principal component analysis and RBF neural network to establish combining the snowmelt snow depth simulation mathematical model.Firstly the principal component analysis was adopted to reduce the dimension of the simulation factor,Secondly by using RBF to simulate the variation of snow depth snowmelt,and carried on the example analysis in a particular river.The results show that using principal component analysis and RBF neural network to simulate snow depth is feasible.The simulation results of the snow depth can be predicted to provide some theoretical support,and provide reasonable suggestions for forecasting snowmelt run off and flood for this area.