【摘要】 支持向量机在径流中长期预报的应用中,普遍采用网格搜索法率定其参数,存在耗时较长、参数选取不当而导致预报精度低等问题,针对该问题提出了一种基于遗传算法的支持向量机模型,该模型结合遗传算法收敛速度快的特点对支持向量机参数进行优化选择,实现参数的全局自动化选取。应用乌江流域某电站的径流预报结果显示,相对于基于网格搜索参数寻优的支持向量机模型及神经网络模型,基于遗传算法参数寻优的支持向量机模型预报精度更高,泛化能力更强。更多还原

【Abstract】 In the predication of mid-term and long-term run-off by using the support vector machine( SVM),the gridresearch method is commonly applied for parameter calibration. This method is time-consuming and often causes low accuracy in predication if the parameters were calibrated unwisely. To solve these problems,a SVM model based on genetic algorithm was proposed in this paper. This model,which adopted the rapid convergence rate of the genetic algorithm,realized optimized parameter calibration,and made it possible for overall automatic parameter calibration. A case study of a hydropower station on Wujiang River was presented to verify the feasibility of the proposed method. The comparison of the forecast results between the proposed method and the SVM method based on grid-research and neural network algorithm indicates that the proposed method possesses higher forecasting accuracy and stronger generalization ability.更多还原