An intelligent based decision-making modeling of reservoir operation for improved water management | |
| Author | Panuwat Pinthong |
| Call Number | AIT Diss. no.WM-08-04 |
| Subject(s) | Reservoirs--Management--Decision making Water resources development--Decision making |
| Note | A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Water Engineering and Management, School of Engineering and Technology |
| Publisher | Asian Institute of Technology |
| Series Statement | Dissertation ; no. WM-08-04 |
| Abstract | A decision-making procedure for reservoir operation and management plays an important role in balancing demand and supply for optimal social, economic and environmental benefits. An improved decision-making procedure enhances the operational efficiency of a multipurpose reservoir system. Development of such proves to be a challenging task. Present studies reveal a new adoption of an integrated modeling approach in artificial intelligence techniques for improved reservoir operation and management. The modeling approach applied a genetic algorithm to search for the optimal input combination of a neurofuzzy system. The optimal model structure was modified using the selection index (SI) criterion expressed as the weighted combination of normalized values of root mean square error (RMSE) and maximum absolute percentage of error (MAPE). The hybrid learning algorithm combines the gradient descent and the least-square methods to train the genetic-based neurofuzzy network by adjusting the parameters of the neurofuzzy system. The proposed genetic algorithm based neurofuzzy modeling was used for the inflow forecasting and decision-making models under the integrated modeling framework development that combines reservoir simulation and performance evaluation modules. In the development of the reservoir inflow forecasting model, a Thiessen polygon was used to delineate the boundary of the selected rainfall stations. The point rainfall depth in each station was then transformed into the rainfall volume covering the sub-catchment area according to the polygon. The traveling time between two consecutive rainfall volume generating points was considered as the lag time. Thus the streamflow volume at the outlet of the catchment area can be determined by taking the cumulative rainfall volume in each sub-catchment area then subtracting the losses. The autocorrelation analysis was initially used to determine the correlation between the candidate of rainfall volume input from the selected station and streamflow volume of the forecasting station. Then the highly correlated rainfall volume inputs from the autocorrelation analysis and the streamflow data with the lag time were used for cross-correlation analysis to determine the input data of the model. The decision-making model was linked to the reservoir simulation module and the performance evaluation module. The reservoir simulation model applied the water balance concept to simulate the physical system for reservoir operation. The engine for deriving reservoir release was in the decision-making model, which was developed based on the proposed genetic algorithm based neurofuzzy computing. The decision variables were reservoir storage and inflow, diversion flow, sideflow and water demand. The results obtained from the reservoir operation were evaluated by estimating the performance of the model based on three indicators namely reliability, vulnerability and resiliency in meeting the water demand and minimizing downstream flooding. The reliability was defined using time based satisfaction of demand. The vulnerability and resiliency on the other hand, were defined based on the magnitude of failure to meet the target values and the maximum duration of failure, respectively. The total water demand and the maximum full channel capacity were used as criteria for water supply and flood control, respectively. The applicability of the proposed modeling approach is demonstrated through its application on the Pasak Jolasid Reservoir, Pasak River Basin, Thailand. As a result of the reservoir inflow forecasting model, five input variables were selected: rainfall data at Lomsak, Nongphai, Wichianburi and Buachum Stations, and reservoir inflow data at the Pasak Jolasid Station. The reservoir inflow was forecasted using 1 to 7 days and 15 days ahead where high flow was found to be more accurate than low flow. The performances of the forecasting model in training and testing periods were between 89% to 98% and 65% to 92% respectively. As a result of a decision-making based reservoir operation model, the performance evaluation indicated that the water release predicted by the genetic-neurofuzzy model gave higher reliability for water supply and flood protection compared to the actual operation, the releases based on simulation following the current rule curve, and the predicted releases based on other approaches such as the fuzzy rule-based model and the neurofuzzy model. Also the predicted releases based on the newly developed approach resulted in the lowest amount of deficit and spill. Using the inflow forecasting data, the average annual water deficit was 282.5 MCM, while the deficit of water demand in the reservoir operation model using the observed inflow data was 176.1 MCM. The average annual spills produced by the reservoir operation model using the data of inflow forecasting and observed inflow were 1,684 MCM and 1596.3 MCM, respectively. Genetic algorithms were found to be immensely beneficial in reducing the time taken in model calibration, and an integrated genetic-neurofuzzy model can easily be used in the application because it required fewer data while providing high accuracy. The modeling methodology based on the genetic-neurofuzzy modeling approach indicated that the developed decision-making model would assist in improving operational performance of the Pasak Jolasid Reservoir. The reservoir operating policy is determined based on the requirement of meeting the downstream water demand and the flood control. The computed releases based on the proposed methodology met the requirement with higher level of satisfaction compared to the releases determined using the other techniques. Furthermore, consideration of the storage stage, the reservoir inflow, the sideflow, the diversion flow from the other basin, and the water demand as input variables in defining the optimal architecture of the genetic-neurofuzzy model provided an improved representation of the system operation. The optimal scheme of the reservoir operation was translated into a practical reservoir operational guideline which can assist operators in making decisions on how to manage the system effectively. The proposed methodology can be extended for defining the operational scheme of a multi-reservoir system (e.g., linking between the Pasak and Chao Phraya River Basins). |
| Year | 2009 |
| Corresponding Series Added Entry | Asian Institute of Technology. Dissertation ; no. WM-08-04 |
| Type | Dissertation |
| School | School of Engineering and Technology |
| Department | Department of Civil and Infrastucture Engineering (DCIE) |
| Academic Program/FoS | Water Engineering and Management (WM) |
| Chairperson(s) | Babel, Mukand S.; |
| Examination Committee(s) | Kojiri, Toshiharu;Sutat Weesakul;Manukid Parnichkun;Gupta, Ashim Das; |
| Scholarship Donor(s) | Commission on Higher Education, Ministry of Education; |
| Degree | Thesis (Ph.D.) - Asian Institute of Technology, 2009 |