当前位置: X-MOL 学术Stoch. Environ. Res. Risk Assess. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Outflow sediment concentration forecasting by integrating machine learning approaches and time series analysis in reservoir desilting operation
Stochastic Environmental Research and Risk Assessment ( IF 4.2 ) Pub Date : 2020-04-20 , DOI: 10.1007/s00477-020-01802-3
Ming-Jui Chang , Gwo-Fong Lin , Fong-Zuo Lee , Yi-Cheng Wang , Peng-An Chen , Ming-Chang Wu , Jihn-Sung Lai

During typhoons, outflow sediment concentration at dam outlets should be accurately forecast for increasing the efficiency of turbidity current venting in reservoir desilting operations. However, forecasting hourly outflow sediment concentration is difficult because of complex physical processes and high temporal variability. This study proposes an outflow sediment concentration forecasting model (SOSVMAR) that integrates self-organizing map (SOM), support vector machine (SVM) and autoregressive method (AR) to overcome the poor representation of outflow sediment concentration extremes. SOSVMAR primarily uses SOM to extract valuable data and to identify the most salient features that are regarded as reprocessed input of SVM for precision improvement. After data extraction, AR is used for real-time correcting the model forecasts. An application in the Shihmen reservoir, which is the most important multi-purpose reservoir in northern Taiwan, was conducted to demonstrate the forecasting performance of the proposed model. Due to the 34.65% of storage capacity has been lost in the Shihmen reservoir, it is an urgent task to increase the efficiency of desiltation. In this study, SOSVMAR, SOSVM integrated with SOM and SVM without AR and traditional SVM are compared. Results indicate that the SOSVMAR outperforms SOSVM and SVM by accurately forecasting the maximum outflow sediment concentration and long lead time forecasting. By comparing the root mean square error of the results from SOSVM with SOSVMAR, SOSVMAR significantly improving percentage at the power plant intake and bottom outlet for long lead time forecasting (3 h lead time) is 40% and 35%, respectively. SOSVMAR can provide accurate forecasts because it uses the reprocess model and real-time correction. Thus, the proposed model can be applied to provide useful forecasting information for reservoir sediment management during desilting operations in a reservoir.

更新日期:2020-04-22
down
wechat
bug