基于小波支持向量机的径流预测性能优化分析

发布时间：2018-06-23 09:33

本文选题：年径流预测 + 小波分解　；参考：《水力发电学报》2017年10期

【摘要】：中长期径流预测是水库调度的重要前提和难点问题。在数据驱动预测模型已有研究基础上,提出了基于小波分解的参数优化支持向量机(WD-SVM-PSO)预测模型,实现了对历史径流过程的分频预处理、分类训练、参数优化及交叉验证,从样本数据、模型参数、训练机制三方面对预测模型性能进行优化。采用淮河流域响洪甸水库1959—2014年径流过程进行模型验证,结果表明:WD-SVM-PSO模型预测合格率为93%,且具有良好的泛化性能,有效规避了过拟合现象;进一步通过对照试验仿真,定量揭示了耦合预测模型三方面要素所起的作用大小依次为:样本数据预处理训练模型模型参数。该结论可为分析和完善数据驱动径流预测模型、提高径流预测精度和可靠性提供参考借鉴。
[Abstract]:Long-term runoff prediction is an important prerequisite and difficult problem for reservoir operation. Based on the existing research of data-driven prediction model, a wavelet decomposition-based parameter optimization support vector machine (WD-SVM-PSO) prediction model is proposed, which realizes the preprocessing, classification training, parameter optimization and cross-validation of historical runoff process. The prediction model performance is optimized from three aspects: sample data, model parameters and training mechanism. The runoff process of Xianghongdian Reservoir in Huaihe River Basin from 1959 to 2014 is verified. The results show that the qualified rate of the prediction of the WD-SVM-PSO model is 933, and it has good generalization performance, which effectively avoids the over-fitting phenomenon. It is revealed quantitatively that the effect of the three elements of the coupled prediction model is as follows: sample data preprocessing training model parameters. The conclusion can be used for reference to analyze and improve the data-driven runoff prediction model and improve the accuracy and reliability of runoff prediction.
【作者单位】：安徽农业大学水利工程系;合肥工业大学水资源与环境系统工程研究所;华北电力大学可再生能源学院;
【基金】：国家自然科学基金(51509001) 安徽省自然科学基金(1608085QE112) 国家重点研发计划项目(2016YFC0401305;2016YFC0402208) 安徽省高校优秀青年人才支持计划项目(gxyq ZD2017019)
【分类号】：TV121;TV697

【相似文献】