The reservoir inflow prediction of the Bhumibol and Sirikit dams using machine learning techniques
Issued Date
2022
Copyright Date
2022
Resource Type
Language
eng
File Type
application/pdf
Access Rights
open access
Rights
ผลงานนี้เป็นลิขสิทธิ์ของมหาวิทยาลัยมหิดล ขอสงวนไว้สำหรับเพื่อการศึกษาเท่านั้น ต้องอ้างอิงแหล่งที่มา ห้ามดัดแปลงเนื้อหา และห้ามนำไปใช้เพื่อการค้า
Rights Holder(s)
Mahidol University
Suggested Citation
Pheeranat Dornpunya (2022). The reservoir inflow prediction of the Bhumibol and Sirikit dams using machine learning techniques. Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/114310
Title
The reservoir inflow prediction of the Bhumibol and Sirikit dams using machine learning techniques
Author(s)
Advisor(s)
Abstract
The recent research has been devoted to the machine learning (ML) technique to enhance the predictive performance of hydrological prediction models for a successful planning and water resources management. This study aims at applying ML for reservoir inflow prediction of the Bhumibol (BB) and Sirikit (SK) dams which are two major dams supplying water in the Greater Chao Phraya River Basin (GCPYRB) in the central region of Thailand. Two types of the reservoir inflow prediction models of BB and SK dams
(1) univariate model and (2) multivariate model for both daily and monthly prediction were accordingly established using the XGBoost and LSTM algorithm. The XGBoost was used to predict the reservoir inflow of univariate prediction models. The LSTM algorithm was used to predict the reservoir inflow of multivariate prediction models. Model training and testing were implemented using observed inflow and climate data during 2000–2020 which were specified as key prediction inputs. It is found from the validation results that the XGBoost model can present more reliable and robust prediction results especially for the daily prediction model with the highest R 2 of 0.8854 and 0.8362 for BB and SK dams, respectively. The LSTM is well–suited for the prediction models with multiple outputs. It can give more trustworthy and robust prediction results for both BB and SK dams, particularly for the daily prediction model, which produces the acceptable R 2 values of 0.8729 and 0.7800 for BB and SK dams, respectively. However, it is apparent that the predictive results obtained by the univariate prediction models are higher than those obtained multivariate prediction models for both daily and monthly models. In addition, to apply the model results for reservoir operation practice, three scenarios of the daily predictive inflows generated in the different time periods using the best prediction models, were used as key input in the multi–reservoir reoperation model. The main finding indicates that utilizing daily predictive inflows of these three distinct scenarios for reservoir operation in GCPYRB can produce acceptable levels of reliability of 90.71%, 89.44%, and 90.70%, respectively, which are definitely close to that received for the current operation using observed inflows. IMPLICATION OF THE THESIS. The prediction models developed using ML technique for BB and SK dams can aid in predicting the reservoir inflow which is one of the important components for the achievement of dam–reservoir operation in the central region of Thailand. The precise and accurate prediction of both short–term and long–term time horizon of the reservoir inflows can lead to the proper management of water resources in coping with the anticipated hydrological events.
(1) univariate model and (2) multivariate model for both daily and monthly prediction were accordingly established using the XGBoost and LSTM algorithm. The XGBoost was used to predict the reservoir inflow of univariate prediction models. The LSTM algorithm was used to predict the reservoir inflow of multivariate prediction models. Model training and testing were implemented using observed inflow and climate data during 2000–2020 which were specified as key prediction inputs. It is found from the validation results that the XGBoost model can present more reliable and robust prediction results especially for the daily prediction model with the highest R 2 of 0.8854 and 0.8362 for BB and SK dams, respectively. The LSTM is well–suited for the prediction models with multiple outputs. It can give more trustworthy and robust prediction results for both BB and SK dams, particularly for the daily prediction model, which produces the acceptable R 2 values of 0.8729 and 0.7800 for BB and SK dams, respectively. However, it is apparent that the predictive results obtained by the univariate prediction models are higher than those obtained multivariate prediction models for both daily and monthly models. In addition, to apply the model results for reservoir operation practice, three scenarios of the daily predictive inflows generated in the different time periods using the best prediction models, were used as key input in the multi–reservoir reoperation model. The main finding indicates that utilizing daily predictive inflows of these three distinct scenarios for reservoir operation in GCPYRB can produce acceptable levels of reliability of 90.71%, 89.44%, and 90.70%, respectively, which are definitely close to that received for the current operation using observed inflows. IMPLICATION OF THE THESIS. The prediction models developed using ML technique for BB and SK dams can aid in predicting the reservoir inflow which is one of the important components for the achievement of dam–reservoir operation in the central region of Thailand. The precise and accurate prediction of both short–term and long–term time horizon of the reservoir inflows can lead to the proper management of water resources in coping with the anticipated hydrological events.
Degree Name
Master of Engineering
Degree Level
Master's degree
Degree Department
Faculty of Engineering
Degree Discipline
Environmental and Water Resources Engineering
Degree Grantor(s)
Mahidol University