BRR-SVR月降水量预测优化模型

受多种因素影响,水文时间序列具有非平稳性。研究时间序列的传统模型如ARMA对数据的平稳性有较高要求,不适用于非平稳水文时间序列的研究。近年来,机器学习算法越来越多地被应用于研究水文过程,本文将支持向量机回归(SVR)和贝叶斯岭回归(BRR)应用于月降水量的预测。运用小波变换对降水数据进行分解和重构,然后对各子序列进行相空间重构,运用校验数据从SVR和BRR中选取每个子序列上精度更高的模型,构建耦合支持向量机回归和贝叶斯岭回归的BRR-SVR优化模型,并与单一的BRR模型和SVR模型加以对比。以北京站、南京站和太湖流域7个雨量站为例,采用确定系数、平均绝对百分比误差和平均绝对误差3项指标评估各模型的预测性能,以相对误差图探讨三类模型之间的差异,计算结果验证优化模型的有效性。

BRR-SVR optimization model for monthly precipitation prediction

Affected by many factors, the hydrological time series is non-stationary. Traditional models of time series,such as ARAM,require data to be stable and not suitable for forecasting hydrological time series. In recent years, machine learning algorithms are increasingly used to study hydrological processes. In this paper, Support Vector Regression (SVR) and Bayesian Ridge Regression (BRR) are applied to the prediction of monthly precipitation. Wavelet transform is used to decompose and reconstruct precipitation data,and then phase space reconstruction is carried out for each sub-sequence. The model with higher precision on each sub-sequence is selected from SVR and BRR with verification data, so as to construct the BRR-SVR optimization model and compare it with BRR and SVR models. Taking Beijing Station, Nanjing Station and seven rain stations in Taihu River Basin as examples, the prediction performance of each model is evaluated by coefficient of determination, mean absolute percentage error and mean absolute error, and the relative error graph is used to reveal the differences among the three models.The calculation results verify the validity of the optimization model for Nanjing Station and Taihu River Basin.