基于改进布谷鸟搜索算法优化的SVM月径流预测

SVM Monthly Runoff Prediction Based on Improved Cuckoo Search Algorithm Optimization

为提高水文预测预报精度,提出基于Mittag-Leffler、Pareto、Cauchy3种重尾分布改进的布谷鸟搜索算法(MLCS、PCS、CCS)优化的支持向量机(SVM)月径流预测模型。选取6个典型测试函数对MLCS、PCS和CCS算法进行仿真验证,并与基本布谷鸟搜索算法(CS)的仿真结果进行比较。利用MLCS、PCS、CCS、CS算法优化SVM关键参数,建立MLCS-SVM、PCS-SVM、CCS-SVM、CS-SVM模型对云南省姑老河站枯水期月径流进行研究,并利用实例前40年和后13年资料对各模型进行训练和预测。结果表明,MLCS、PCS、CCS算法寻优能力优于标准CS算法,具有较好的寻优精度和全局搜索能力。MLCS-SVM、PCS-SVM、CCS-SVM模型对实例1-3月月径流预测的平均相对误差分别在4.89%~4.94%、6.87%~7.07%、6.87%~7.09%之间,预测精度较CS-SVM模型分别提高了34.5%、8.30%、23.6%以上,具有较好预测精度和泛化能力,表明MLCS、PCS、CCS算法能有效优化SVM相关参数。模型及方法可为水文预测预报及其他相关预测研究提供参考。

In order to improve the accuracy of hydrological forecast,this paper proposes a support vector machine(SVM)monthly runoff prediction model optimized by Mittag-Leffler,Pareto,and Cauchy improved cuckoo search algorithms(MLCS,PCS,CCS).Six typical test functions are selected to simulate and verify the MLCS,PCS and CCS algorithms,and compared with the simulation results of the basic cuckoo search algorithm(CS).MLCS,PCS,CCS,and CS algorithms are used to optimize the key parameters of the SVM,and MLCS-SVM,PCS-SVM,CCS-SVM,and CS-SVM models were established to study the monthly runoff during the dry season at the Gulaohe Station in Yunnan Province.And the data for the next 13 years are used to train and predict each model.The results show that MLCS,PCS and CCS algorithms have better optimization capabilities than standard CS algorithms,and have better optimization accuracy and global search capabilities.The average relative errors of the MLCS-SVM,PCS-SVM,and CCS-SVM models for the monthly runoff forecast of the examples from January to March are between 4.89%~4.94%,6.87%~7.07%,and 6.87%~7.09%,respectively.The model has improved by 34.5%,8.30%,and 23.6%,respectively,and has better prediction accuracy and generalization ability.It shows that MLCS,PCS,and CCS algorithms can effectively optimize SVM related parameters.Models and methods can provide references for the hydrological forecasting and other related forecasting studies.