| 摘要: |
| 【目的】探讨熵谱模型在月径流预报中的应用效果以及训练期长度对模型预报精度的影响，为熵谱模型在径流预报中的应用提供参考。【方法】月径流预报依据黑河莺落峡站月径流资料，采用伯格熵（BESA）和构造熵（CESA）2种熵谱模型进行月径流预报，并用平均相对误差（RE）、均方根误差（RMSE）、相关系数（R）和纳西效率系数（NSE）对模型预报精度进行评价。【结果】训练期长度过短会使模型阶数偏低，模型无法做出准确的预测；训练期长度过长会使模型阶数偏高，此时训练期和验证期精度反而略微下降；适中的训练期长度能够使模型的训练期和验证期精度均相对较高且稳定。对于黑河莺落峡站，BESA模型的最佳训练期长度为13年，CESA模型的最佳训练期长度为19年，CESA模型的训练期拟合精度和验证期预报精度均高于BESA模型，同时CESA模型在汛期预报精度相对较高，而BESA模型在非汛期预报精度相对较高。【结论】BESA和CESA 2种模型都可用于月径流预报，但需要合理选择训练期长度，使模型阶数适中且稳定，以提高预报精度和可靠性。 |
| 关键词: 月径流预报 熵谱分析 伯格熵 构造熵 时间序列分析 |
| DOI： |
| 分类号: |
| 基金项目:国家自然科学基金项目（91425302） |
|
| Monthly streamflow forecasting based on entropy spectral theory |
|
ZHOU Zhenghong,SU Xiaoling
|
| Abstract: |
| 【Objective】This study discussed the application of entropy spectral model in monthly streamflow forecasting and the effect of lengths of calibration period on model performance to provide reference for the application of entropy spectral model in streamflow forecasting.【Method】Burg entropy and configurational entropy spectral analysis model were introduced for monthly streamflow forecasting at Yingluoxia station in Heihe river basin.Relative error (RE),root mean square error (RMSE),correlation coefficient (R),and Nash Sutcliffe efficiency coefficient (NSE) were used to evaluate the model performance.【Result】Shorter training period resulted in lower model order and the model cannot forecast accurately.Longer training period resulted in higher model order and lower accuracy in the training period and verification period.Moderate length of training period resulted in high and stable accuracy in training period and verification period.At Yingluoxia station in Heihe river basin,the best calibration period length of BESA model and CESA model were 13 years and 19 years,respectively.The accuracy of the CESA model was higher than that of the BESA model.In verification period,the CESA model had higher forecast accuracy in the flood season,while BESA model had higher forecast accuracy in non-flood period.【Conclusion】Both BESA and CESA models can be used for monthly streamflow forecasting.However,the length of calibration period needs to be reasonably selected so that the model order can be moderate and stable to improve forecast accuracy and reliability. |
| Key words: monthly streamflow forecasting entropy spectral analysis Burg entropy Configurational entropy time series analysis |
