论文部分内容阅读
在针对支持向量机(SVM)最佳算法参数难以确定的缺陷以及基本布谷鸟搜索(CS)算法局部搜索能力弱、寻优精度低等不足,通过在迭代过程中对鸟窝位置进行改进,提出基于高斯扰动的布谷鸟搜索算法(GCS)。利用GCS算法优化选择SVM惩罚因子C和核函数参数g,构建GCS-SVM年径流预测模型,并与基本CS算法、粒子群优化(PSO)算法和遗传优化(GA)算法优化选择SVM学习参数的CS-SVM、PSO-SVM和GA-SVM模型作为对比,以云南省西洋街站年径流预测为例进行实例研究,利用实例前30a和后17a资料分别对模型进行训练和预测。结果表明:1GCS-SVM模型对实例预测样本预测的平均相对误差绝对值和最大相对误差绝对值分别为3.49%、7.88%(5次平均),精度优于CS-SVM、PSO-SVM和GA-SVM模型,表明GCSSVM模型具有较高的预测精度。2GCS算法全局寻优能力强、收敛速度快、调节参数少,利用GCS算法优化SVM学习参数有利于提高SVM模型的预测精度和泛化能力。
In view of the defect that the parameters of SVM are difficult to be determined and the weakness of local search and the poor precision of the basic cuckoo search (CS) algorithm, we propose the improvement of the position of the bird’s nest in the iterative process Cuckoo Search Algorithm Based on Gaussian Dispersion (GCS). GCS-SVM annual runoff forecasting model was constructed by GCS algorithm optimization and selection of SVM penalty factor C and kernel function parameter g, and optimized SVM learning parameters with basic CS algorithm, particle swarm optimization (PSO) algorithm and genetic algorithm (GA) CS-SVM, PSO-SVM and GA-SVM as an example, the case study of annual runoff in Xiyang Street Station in Yunnan Province was taken as an example. The models were trained and predicted using the data of 30 years before and after 17 years respectively. The results show that the absolute value of relative average absolute error and maximum relative error of 1GCS-SVM model are 3.49%, 7.88% (5 averages) respectively, and the precision is better than that of CS-SVM, PSO-SVM and GA- SVM model, indicating that the GCSSVM model has a high prediction accuracy. The global optimization ability of 2GCS algorithm is strong, the convergence speed is fast and the adjustment parameters are few. Using GCS algorithm to optimize SVM learning parameters is helpful to improve the prediction accuracy and generalization ability of SVM model.