论文部分内容阅读
鉴于支持向量机(SVM)最佳算法参数难以确定及模拟退火算法(SA)、遗传算法(GA)在实际应用中存在的不足,将SA算法与GA算法优点结合起来,提出模拟退火遗传算法(SAGA)。利用SAGA算法搜索SVM学习参数,构建SAGA-SVM预测模型,并与基于SA算法、GA算法搜索SVM学习参数的SA-SVM、GA-SVM模型作对比,以云南省某水文站枯水期3—4月月径流预测为例进行实例研究,利用实例前28年和后6年资料对模型进行训练和预测。结果表明:(1)SAGA算法兼顾了SA、GA算法二者的优点,有效地抑制了SA、GA算法早熟收敛现象,提高了算法在解空间中的探索能力和效率;(2)SAGA-SVM模型对实例后6年枯水期3、4月月均径流预测的平均相对误差绝对值分别为5.98%、3.36%,精度优于SA-SVM、GA-SVM模型,表明SAGA-SVM模型具有较高的预测精度和泛化能力。
In view of the difficulty of determining the best algorithm parameters of SVM and the difficulty of the SA algorithm and the GA, there are some shortcomings in the practical application. Combining the advantages of the SA algorithm and the GA algorithm, SAGA). SAGA algorithm was used to search SVM learning parameters to construct SAGA-SVM prediction model. Compared with SA-SVM and GA-SVM based on SA algorithm and GA algorithm to search SVM learning parameters, Monthly runoff forecasting as an example case study, using the example of the first 28 years and 6 years after the data to train and predict the model. The results show that: (1) SAGA algorithm takes into account the advantages of both SA and GA algorithms, effectively restraining premature convergence of SA and GA algorithms, and improving the exploration ability and efficiency of the algorithm in solution space; (2) SAGA-SVM The average absolute relative errors of monthly average runoff prediction in March and April in the past six years after the model were 5.98% and 3.36% respectively, which were better than SA-SVM and GA-SVM models, indicating that the SAGA-SVM model had higher Predictive accuracy and generalization ability.