基于小波分解的日径流逐步回归预测模型

Stepwise regression model for daily runoff prediction based on wavelet decomposition

本文以预测水文站的上游水文站的日径流序列为依据，利用小波分解和重构得到预测水文站及上游水文站的日径流序列在1-4尺度下的概貌分量，然后以各站的原始径流序列及其在不同尺度下的概貌分量为候选预报因子，建立了径流逐步回归多步预测模型。计算实例表明，由于引入了上游水文站的径流序列并提取了各站径流序列的不同尺度下的概貌分量，本文提出的基于小波分解的日径流逐步回归预测模型的预测精度高于小波网络模型和多元自回归模型，能对非凌汛期未来1～3d以及凌汛期1～7d的日均流量进行预测，可为制定水电站未来的发电计划提供科学的依据。

A stepwise regression model for runoff prediction based on wavelet decomposition is proposed The daily runoff time series of the hydrological stations in the upstream of the hydrological stations under consideration are introduced into the model. The general components of the daily runoff time series of both hydrological stations at timescale 1 -4 can be obtained by using the wavelet decomposition and reconstruction. Taking the original daily runoff time series and their general components as candidate independent variables, the stepwise regression models for daily runoff multi-step prediction can be established. A case study shows that the p model is better than the auto-regression model, and is able to predict the daily runoff in 1 - 3 days during non ice-jam period and 1 - 7 days during ice-jam period with acceptable accuracy.