Hybrid SVM-CIPSO methods for optimal operation of reservoir considering unknown future condition

Abstract In this paper, new hybrid methods have been proposed to solve reservoir operation optimization problem for uncertain water inflows condition by equipping improved particle swarm optimization (IPSO) algorithm with support vector machine (SVM) method. The constrained version of IPSO algorithm (CIPSO) has been used here to improve the efficiency of the IPSO algorithm. In CIPSO algorithm, the problem constraints have been explicitly satisfied leading to smaller search space and finally smaller computational cost. Two approaches have been considered to propose the hybrid methods. In the first approach, named SVM-CIPSO1, water inflows into the dam reservoir have been predicted using SVM model and these predicted values have been used to solve reservoir operation optimization problem using CIPSO algorithm. However, in the second approach, named SVM-CIPSO2, at first, the CIPSO algorithm has been applied to solve reservoir operation optimization problem using the historical data and finally the optimal water release values have been used as input and output data to create a SVM model for predicting optimal water release from reservoir for the future condition. For comparison purpose, the ANN model has been also used to predict the water inflow or release values for the future condition and the standard form of IPSO algorithm has been also used to solve the optimization problem. Here, to evaluate the proposed approaches, the optimal water release values form Zayandehroud dam reservoir have been obtained using proposed methods and the reliability, resiliency, vulnerability and sustainability indexes have been computed. Comparison of the results indicates the capability of the proposed methods to predict the optimal water release values for future condition with acceptable accuracy. In other words, the RMSE values of SVM model for test, validation and training processes are 9.7104 (23. 56196), 11.2553 (42.69093), and 7.9556 (47.9346) MCM, respectively, which are obtained using second (first) approach. In addition, the best reliability, resiliency, vulnerability and sustainability index values are 51.95% (45.45%), 45.95% (38.10%), 3.539 (0.0041) MCM and 62.02% (55.74%), respectively, which are obtained using SVM-CIPSO2 (SVM-CIPSO1) method.