基于小波去噪和FA-SVM的中长期径流预报

Mid-long term runoff forecasting based on wavelet de-noising and FA-SVM

为优化支持向量机(Support Vector Machine,SVM)模型参数,提高中长期径流预报精度,建立了基于FA-SVM的中长期径流预报模型。该模型以样本中训练期均方差最小为目标函数,利用萤火虫算法(Firefly Algorithm,FA)对支持向量机主要参数(惩罚系数C、核函数参数g和不敏感损失系数ε)进行了优化。以岷江上游的紫坪铺水库为例,运用小波去噪法对各月径流序列进行数据预处理后,利用FA-SVM模型与BP神经网络模型进行了中长期径流预报。结果表明:①运用小波阈值法能够较好地滤除各月径流序列的系统噪声和测量噪声;②FA-SVM模型中长期径流预测效果较好,预报精度等级均在丙级以上;③FA-SVM模型的预报效果优于BP神经网络模型。

To optimize parameters of support vector machine(SVM)model and enhance mid-long term runoff forecasting accuracy,we established a mid-long term runoff forecasting model based on FA-SVM.In this work,the minimum mean square deviation of the training period data was taken as the objective function and Firefly Algorithm(FA)was used to optimize three main parameters(i.e.,penalty coefficient C,kernel parameter g and insensitive loss coefficientε)of SVM model.Also,we adopted wavelet de-noising method to de-noise monthly runoff data.Then,both FA-SVM model and BP neural network model were used to forecast mid-long term runoff in Zipingpu Reservoir in the upper reaches of the Minjiang River.The results showed:①Wavelet de-noising method could be used to filter the system noise and measurement noise of monthly runoff data.②The FA-SVM model worked well for mid-long term runoff forecasting,and its forecast accuracy grade was above grade C.③The comprehensive runoff forecasting effectiveness of FA-SVM model was superior to that of BP neural network model.