Streamflow forecasting using multivariate relevance vector machine algorithm | |
| Author | Thi Reindar Tin Tun |
| Call Number | AIT Thesis no.WM-17-19 |
| Subject(s) | Streamflow--Forecasting Multivariate analysis |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Water Engineering and Management in School of Engineering and Technology |
| Publisher | Asian Institute of Technology |
| Series Statement | Thesis ; no. WM-17-19 |
| Abstract | One of the Artificial Intelligence (AI) techniques, multivariate relevance vector model (MVRVM) was applied for predicting streamflow volume sat four gauging stations, namely Y20, Y6, Y4 and Y17 of the Yom River Basin in Thailand. Daily discharge and daily water level values from 4 gauging stations and daily observed rainfall values from 10 rainfall stations play as an input data in the 75-25 partition. The available training dataset for gauging stations and rainfall stations are from 1/1/1991 to 31/12/2007 (16 years, 15290 data points) and the testing dataset are from 1/1/2008 to 31/12/2014 (6 years, 5939 points). To qualify the forecasts generated by the model, the statistics used for the selection of the model are the number of RVs, Correlation Coefficient (R2), and prediction efficiency (PE). Different kernel width and many different iterations of the training data were run to achieve the most stable accurate results based on the classical partition and “Heavy tail” kernel function. The results show that among those 4 gauging stations and 10 rainfall stations, there have been found out the highest number of R2value is 0.97only at gauging station Y6 and its upstream rainfall station code 400062 by using the width of 10and also which has the highest prediction efficiency value is 0.93, combine with the lesser number of RVs (33). The model only utilizes (33 RVs) from the full data set (7254 observations) that was used for training. In term of low, medium and high flows, the MVRVM model is performing well by showing that the maximum number of R2is 0.91and maximum prediction efficiency PE is 0.82at the width of 10 using 5448 data points. This study stated the promising results of up to two, three, five, seven days ahead predictions in streamflow forecasting and the optimal model for low, medium and high flows in Yom River basin by utilizing multivariate relevance vector model (MVRVM) model which is very useful in improving water management and decision making capability as well as an essential factor for flood mitigation due to their high prediction accuracy, robustness and easiness to master in limited amount of time. |
| Year | 2017 |
| Corresponding Series Added Entry | Asian Institute of Technology. Thesis ; no. WM-17-19 |
| Type | Thesis |
| School | School of Engineering and Technology |
| Department | Department of Civil and Infrastucture Engineering (DCIE) |
| Academic Program/FoS | Water Engineering and Management (WM) |
| Chairperson(s) | Andriyas, Sanyogita |
| Examination Committee(s) | Shrestha, Sangam;Duc Hoang Nguyen |
| Scholarship Donor(s) | Thai Pipe Scholarship |
| Degree | Thesis (M. Eng.) - Asian Institute of Technology, 2017 |