基于PSO-SVR-ANN的丹江口水库秋汛期月尺度径流预报模型

Monthly runoff forecast model of Danjiangkou Reservoir in autumn flood season based on PSO-SVR-ANN

丹江口水库位于汉江上游,是南水北调中线工程水源地,因受华西秋雨影响经常发生较大的秋汛过程。目前应用于丹江口水库径流预报的方法普遍存在预报精度不高和稳定性不强等缺点。针对上述问题,构建了PSO-SVR中长期预报模型,利用粒子群优化算法(PSO)寻找支持向量回归机(SVR)的惩罚系数C、不敏感系数ε以及高斯径向基核函数的gamma参数,在此基础上利用分层感知器人工神经网络(ANN)分析了SVR预报误差规律和特征,建立了PSO-SVR-ANN耦合模型,实现了径流预报的自纠正。结果显示,PSO-SVR-ANN秋汛期预报模型平均相对误差较小,均在10%左右;合格率较高,均处于80%以上水平。实验结果表明,PSO-SVR-ANN模型与PSO-SVR相比,预报精度更高,稳定性更强,可信度更高,具有一定的实用价值,为相关研究提供了参考。

Danjiangkou Reservoir,located in the upper reach of the Hanjiang River,is the water source of the Middle Route of the South-to-North Water Transfer Project.At present,the Danjiangkou Reservoir runoff forecast generally has the disadvantages of low prediction accuracyand poor stability.In order to solve the above problems,a PSO-SVR medium-and long-term runoff prediction model is developed.The particle swarm optimization algorithm(PSO)is used to optimize the penalty coefficient C,insensitive coefficientεand gamma parameters of Gaussian radial basis function kernel function of support vector regression(SVR).On its basis,hierarchical perceptron artificial neural network(ANN)is used to analyze the prediction error of SVR,and the coupled PSO-SVR-ANN model is established to realize the self-correction of the runoff forecast.The results show that the average relative error of PSO-SVR-ANN model for the autumn flood season is as small as around 10%,with the qualified rate all higher than 80%.In addition,compared with PSO-SVR,PSO-SVR-ANN model has a higher prediction accuracy,stronger stability and higher credibility with practical importance,which provides useful instructions for related research.