水文时间序列解析-集成预测模型研究

Study on Prediction Model Based on Segregation and Aggregation of Hydrologic Time Series

为了克服实际工作中常规预测模型的弊端,本文提出了水文序列解析-集成预测模型(Prediction Model based on Segregation and Aggregation of Hydrological Time Series,PMSAHTS),通过分离水文序列中的趋势信号和周期信号得到消除了人类活动影响的序列纯随机信号,然后通过随机因子预测预报方法(如BP神经网络)使用这些随机信号进行训练和仿真预测,将预测结果与趋势、周期预测结果重新集成,得到水文序列的预测值。将该模型应用到和田子项目区进行年内月平均蒸发量的预测,结果表明,PMSAHTS模型达到了水文情报预报规范的合格要求,可以用于实际预测。

To overcome the shortcomings in conventional forecast methods, a new Prediction Model based on Segregation and Aggregation of Hydrological Time Series （PMSAHTS） was put forward. Impacts of human activities on hydrological data sequences were firstly eliminated through segregation of trend and period signals in the data sequences. Secondly, the remaining random sequences were used as inputs to train BP Neutral Network, and then the trained network was used to predict random sequences in the future. Finally, the predicted random sequences were aggregated with the prediction results of trend and period terms. Thus the predicted hydrological sequences were obtained. To demonstrate this model, PMSAHTS was applied to predict the annual month-average evaporation in the Hotan Sub-project Area. It was shown by the results, among all comparisons of predicted values with measured ones, 62.5% of then have a prediction relative error less than 20%, which suggests that the PMSAHTS was qualified for hydrological prediction in practice.