支持向量回归在贮灰坝渗流监测中的应用

鉴于支持向量机在机器学习方面表现出的良好性能，提出了基于支持向量回归(SVR)算法的贮灰坝渗流监测模型。采用基于平行网格搜索的交叉验证法选择模型参数，避免了参数选择的盲目性、随意性，提高了预测精度。实例分析表明，该渗流监测模型与传统的神经网络（反向传播（BP）网络、径向基核函数（RBF）网络）模型相比，具有预测精度高、泛化能力强等优点，能够快速、准确地预测出指定位置的测压管水位，对贮灰坝水头预报和电厂的安全生产具有实用价值。

Application of SVR to Seepage Monitoring of Ash Storage Dam

Considering the favorable performance of support vector machines (SVM) in the respect of machine learning, a seepage-monitoring model based on support vector regression (SVR) is put forward. The cross-validation method via parallel grid search is used to avoid blindness and randomness in model parameter selection, and the prediction precision is improved. The comparison with traditional artificial neural network models (BP and RBF) shows that the proposed model has higher precision and generalization ability. The model can forecast fleetly and exactly the piezometric level of any appointed place. It is valuable for the water head forecast of ash storage dams and the safe operation of power plants.