目前应用于丹江口水库年径流预报的方法主要为物理统计和人工神经网络(ANN) 等方法, 但这些方法普遍存在预报精度不高和稳定性不强等缺点。选择回归支持向量机(SVR) 模型应用于丹江口水库年径流预报, 针对惩罚系数 C、核参数σ和不敏感损失系数ε三个参数在实际赋值过程中存在计算量大、难以得到最优值等问题, 将粒子群优化算法(PSO) 加入到 SVR 模型中, 建立 PSO-SVR 模型, 实现了参数的自动优选。结果表明, PSO-SVR 模型较之 SVR 模型, 提高了预报精度; 较之 ANN 模型, 稳定性更强, 可信度更高。该模型具有较好的应用价值, 可为南水北调中线工程调度方案制定提供一定的参考依据。

At present, the methods of annual runoff forecast for Danjiangkou reservoir mainly include physical statistical approach and artificial neural network (ANN) . However, these methods have the disadvantages of low accuracy and low stability. In this paper, we applied the regression support vecto rmachine (SVR) model to the annual runoff forecast for Danjiangkou Reservoir. Considering that the penalty coefficient C, the kernel parameterσ, and the insensitive loss coefficient ε all require a large amount of calculation and it is difficult to obtain their optimal value in the actual assignment process, we added the particle swarm optimization (PSO) algorithm to the SVR model and established a PSO-SVR model to realize the automatic optimization of parameters. The results showed that the PSO-SVR model has higher prediction accuracy compared with the SVR model, and has better stability and reliability than the ANN model. The model has a good application value, and can provide some reference for the development of the operation scheme of the middle route of the South-to-North Water Transfer Project.

“十三五”国家重点研发计划子课题“梯级水库群影响下流域水文循环演变规律研究”(2016YFC0402201-01) ; 中国水利水电科学 研究院基本科研业务费专项项目(WR0145B212017)