In the deterministic optimization method, a historical series of inflow to the reservoir through the operation time is considered and the release volume from the reservoir is optimized to provide downstream consumption in these conditions. The disadvantage of such models is that the optimal solutions cannot be generalized to other possible inflows to the reservoir and if the inflow to the reservoirs changes, the optimal solutions will no longer work and the system must be operated in the form of an optimizer algorithm. Thus, the main objective of this research is to integrate the support vector machine model with the NSGA-II optimization algorithm for optimal real-time operation from the system.
This study utilizes an integration of the NSGA-II multi-objective algorithm and WEAP simulator model so that the first objective is to maximize the reliability of providing the needs in front of the second goal, i.e., to minimize water table drawdown at the end of the operation time. The dam role curve or the amount of released volume from the reservoir is optimized to supply downstream uses in these conditions. However, in certain optimizations, the optimal solutions cannot be generalized to other possible inputs to the reservoir, and if the inflow to the reservoirs changes, the obtained optimal solutions are no longer efficient and the system must be re-optimized in the form of an optimizer algorithm. Therefore, to solve this problem, a new method is extended on the basis of the combination of the support vector machine and NSGA-II algorithm for optimal real-time operation of the system. In this case, after completing the algorithm and extracting the optimal variables, there will be a relationship between the monthly inflows to the reservoir, water storage volume in the reservoir, reservoir volume changes and downstream needs (as independent parameters) and the optimal release rate variable (as a dependent parameter). This means that in each future simulation period, by determining the first four parameters at the beginning of each month, the optimal real-time release volume will be determined.
The results demonstrate that the average error rate of optimal rules derived from support vector machines is less than 2.5% compared to the output of the NSGA-II algorithm in the verification step, which indicates the efficiency of this method in predicting the optimal pattern of the dam role curve in real-time. In this structure, based on the inflow to the reservoir, the volume of water storage in the reservoir and changes in the reservoir storage (at the beginning of the month) and the downstream demands of the current month, the optimal release amount can be achieved in real-time. Therefore, the developed support vector machine has the ability to update the rule curve of the dam based on the new data if the input flow to the dam changes and provide the possibility of operating the system in real time. In this structure, unlike the common structure of deterministic optimization, if the inflow changes, it is no need to re-optimize to understand the optimal coefficients, however by utilizing the relationship obtained from the support machine method it is possible to predict the release volume in real-time on the basis of the inflow to the reservoir (measured at the first day of the month), the volume of water storage in the reservoir (measured at the first day of the month) and changes in the reservoir storage and downstream needs in the current month.
In general, the results showed that the SVM-NSGA-II developed model has good capability and efficiency in solving complex and completely nonlinear problems and providing optimal solutions based on 24 answers on the Pareto optimal front. Among these solutions, according to the valuation of the objective functions, the solution that had the lowest groundwater drop function and the highest percentage of supplying the demands was selected as the superior answer. The results of applying the optimal release values from the dam (optimal role curve of the dam) showed that considering the optimal policy, the percentage of providing and reliability of supplying the most demands is appropriate and acceptable. The developed SVM-NSGA-II model has the ability to provide optimal operation policies based on new data of the inflow to the dam in a way that allows us to optimally manage the system in real-time.