Study on Pr ediction Model Based on Segr egation and Aggr egation of Hydrologic Time Ser ies
水文时间序列解析-集成预测模型研究

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.