关键词: 径流预测, 核主成分分析, 支持向量机, 粒子群优化

Abstract: To increase accuracy and stability of the runoff forecast model, KPCA-PSO-SVM based runoff forecast method is researched. On the basis of analyzing the influential factors of runoff forecast, the kernel principal component analysis (KPCA) is employed to extract main features from runoff influential factors and the principal component is used as the input of support vector machine (SVM), and the runoff forecast model is built. In the model, the parameter is optimized by the particle swarm optimization (PSO) algorithm. Finally, the model is applied to forecast the runoff at the Yamadu station on Ili River and the result shows that during the training and validation period, the average relative error are 0.77% and 7.64% respectively. By comparison with other forecast models, the fitting and generalization of this model are the best, thus it can be used as an optional method in long term runoff forecast.

Key words: runoff forecast, kernel principal component analysis, support vector machine, particle swarm optimization