基于小波支持向量机特征分类的日径流组合预测——以宜昌三峡水库为例

Daily Runoff Combination Prediction Based on Wavelet Support Vector Machine Feature Classification——Taking the Three Gorges Reservoir in Yichang as an Example

河流径流预测作为水库调度和发电的重要前提,其预测精度直接影响水利工程的综合效益。基于径流历史数据,针对其波动和随机性提出一种小波分析-支持向量机(SVM)特征分类组合预测模型。该模型首先利用小波分解提取原始径流序列的高低频能量谱作为SVM样本标记,并对原始序列进行特征分类,分为"平稳型"和"突变型"序列,对应不同类型序列的小波近似信号和细节信号分别采用自回归和滑动平均模型(ARMA)和BP神经网络模型进行预测,再重构各序列预测结果。最后采用平均绝对百分比误差(MAPE)、均方根误差(RMSE)、希尔不等式系数(TIC)作为模型评价指标。结果表明:在3个评价指标下,所提模型都优于ARMA和BP神经网络模型,并具有更好预测稳定性。

As an important prerequisite for reservoir operation and power generation,the prediction accuracy of river runoff has a direct impact on the comprehensive benefits of water conservancy projects. Based on the historical data of runoff,this paper proposes a wavelet analysis support vector machine（ SVM） feature classification combined forecasting model for its volatility and randomness. Firstly,the wavelet decomposition is used to extract the high and low frequency energy spectrum of the original runoff sequence as the SVM sample mark,and the original sequence is classified by feature,dividing into stationary and abrupt sequences,the wavelet approximation signals and the detail signals,corresponding to different types of sequences,are predicted by auto-regressive moving average model（ ARMA） and BP neural network model respectively,then the prediction results of each sequence are reconstructed. Finally,the Mean Absolute Percentage Error（ MAPE）,the Root Mean Square Error（ RMSE） and the Theil Inequality Coefficient（ TIC） are used as the evaluation indexes of the model.The results show that： under the 3 evaluation indexes,the proposed model is better than the ARMA and BP neural network models,and it has better prediction stability.