基于CEEMD-LSSVM-NNBR模型中长期入库径流模拟

为提高中长期径流模拟精度,提出使用完备总体经验模态分解(CEEMD)产生更低噪信号作为模拟模型输入。应用CEEMD方法对径流序列作信号分解,时间序列被分解为若干子序列,每一子序列通过最小二乘支持向量机(LSSVM)模型分别模拟,之后将每个子序列模拟结果重构以获得最终径流模拟结果。以石头峡水库入库径流模拟为例,试验结果表明,与LSSVM模型相比,CEEMD-LSSVM模型可提高水库汛期入库径流模拟精度。对于整体序列和汛期序列,其纳什效率系数、平均相对误差和均方根误差模拟效果均明显提高;但在枯水期模拟精度不理想,主要由于枯水期径流量数值波动较小、序列平缓所致。因此,CEEMD方法对于汛期径流序列分解更具优势,对枯水期径流序列分解效果有待提高。将适用于枯水期径流模拟的最近邻抽样回归模型(NNBR)与CEEMD-LSSVM结合成组合模型CEEMD-LSSVM-NNBR,可用于全年入库径流模拟。

Middle and long-term runoff simulation based on CEEMD-LSSVM-NNBR model

In order to improve the accuracy of middle and long-term runoff simulation, a complete ensemble empirical mode decomposition(CEEMD) was used to produce lower noise signals as model input. The signal decomposition of runoff series was carried out by CEEMD and the entire time series was decomposed into several sub-series. Each sub-series was simulated by the least square support vector machine(LSSVM) model, and later the simulation results of each sub-series were reconstructed to obtain flow simulation results. Taking the Shitouxia Reservoir inflow runoff simulation as an example,the results showed that the CEEMD-LSSVM model could improve the accuracy of inflow runoff simulation in flood season compared with the LSSVM model. For the whole series and flood season series, the Nash efficiency coefficient, mean absolute relative error and root mean square error were obviously improved but the accuracy was lower in dry season. Therefore, the CEEMD method was better in decomposing runoff series in flood season, but needed to be improved in dry season. The nearest neighbor sampling regression model(NNBR) and CEEMD-LSSVM were combined into CEEMD-LSSVM-NNBR model to simulate the inflow runoff.