基于FPSTWD算法与时间序列支持向量机的河流径流量预报

Prediction of river runoff based on FPSTWD algorithm and time series support vector machine

为及时掌握河道径流量变化趋势,为下游水库防洪调度提供依据,提出了基于时间序列的最小二乘支持向量机河道径流量实时预测模型。采用特征点分段时间弯曲距离算法对实时采集的时间序列数据进行分段与相似度计算,以缩减规模的子序列数据集对LSSVR模型进行训练优化,实现多个LSSVR子模型建模,将预测数据序列与LSSVR子模型的相似度匹配,自适应地选取最佳的子模型作为预测模型。应用该模型对某河径流量进行实时预测,模型评价指标中最大相对误差、平均相对误差绝对值和均方根误差分别为9.08%、3.25%与303m3。研究结果表明,该模型具有较好的预测性能,能够满足河道径流量预测的实际需求,并为下游水库防洪调度与水资源管理提供了重要参考。

In order to grasp the change trend of river runoff,provide the basic information for the flood control of reservoir.A least squares support vector machine time series of river runoff forecasting model is proposed.Using feature point segmented time warping distance algorithm on the real-time data of time sequence and similarity calculation,the sequence data reduction scale set for training and the optimization of the LSSVR model,the LSSVR sub model,the forecasting data sequence similarity with the LSSVR model to adaptively select the best matching,sub model as the predictive model.Application of the model of a river runoff forecasting,the absolute value of the maximum relative error,average relative error of the model evaluation index and the root mean square error of 9.08%,3.25% and 303 m^3 respectively.The results show that,this model has better prediction performance,which can satisfy the actual demand of river runoff prediction,and provide an important reference for flood control reservoir and water resources management.