差分进化优化参数的LSSVM中长期径流预测

针对LSSVM预测模型参数难以确定的问题,利用差分进化(DE)算法的收敛速度快和全局优化能力,优化LSSVM模型的惩罚因子和核函数参数,避免了人为选择参数的盲目性。将优化后的LSSVM模型应用于中长期径流预测问题。选取黄河三门峡站1919年至1992年径流量实测数据进行分析和训练,对1993年至2002年的年径流量进行预测,并与BP神经网络和SVM模型进行比较。研究结果表明,该模型具有较高的预测精度。

Parameters Selection for LSSVM based on Differential Evolution to Mid-long Term Runoff Prediction

To solve the problem of the uncertain parameters of LSSVM,the learning algorithm of least squares supports vector machines forecasting model optimized by differential evolution is proposed.Two parameters of LSSVM model study by differential evolution algorithm’s abilities of the fast convergence and global optimization are optimized.It can escape from the blindness of man-made choice.The proposed model is applied to the mid-long term runoff forecasting problem.Actual data from 1919 to 1992 of Sanmen station in Huanghe area is taken as the sample data to be analyzed.The results show that the mean relative error of the proposed method is only 5.39%,which is less than those of BP and SVM model by 3.53% and 2.23%,respectively.