عنوان مقاله [English]
Basin-scale river basin simulation models are vital tools in assessing and quantifying large-scale water resource systems performances under different scenarios. However, simulation models are not capable of doing multiperiod
optimization for optimal planning and management purposes. Linking river basin simulation models to multiperiod optimization algorithms could be a remedy to such shortcoming.
Nevertheless, large-scale river basin models, when used as the simulation model of an optimization algorithm, may face difficulties, in terms of the needed time for combined optimization-simulation models to run. The more hydrologic and socio-economic processes to be addressed in the simulation model, the more the computational burden of the combined model would be.
Meta-modeling is one of the useful approaches to dealing with this difficulty, where a fast-running approximate model, called meta-model, replaces the exact simulation model. This study presents application of the particle swarm optimization (PSO) algorithm, linked to the MODSIM decision support system (DSS), as a river basin simulation model, resulting in the integrated PSO-MODSIM optimization-simulation model, for solving basin-scale water allocation problems. The developed model was applied to the problem of optimal water resource development plans, as well as optimal water allocation of the Atrak River Basin, as a real case study in north-east Iran. Where there are serious competitions between upstream and downstream provinces of the basin in
utilization of available water resources. The objective function of the model, which consists of design variables; i.e., reservoir capacities, and operational; i.e., priority numbers of reservoir target levels, was calibrated
based on improving water allocation conditions compared to a benchmark scenario, in which development projects were not implemented. Since the integrated model was time consuming, support vector machines (SVMs) were used as a meta-model to develop the PSO-MODSIM$sim$ SVM model, in which an SVM-based
surrogate model replaces MODSIM. The performance of the PSO-MODSIM, PSO-MODSIM$sim$ SVM and another model, with a meta-model of artificial neural network (ANNs) type, were analyzed and compared through their application to optimal water allocation and planning water resource development projects in the basin. Before using SVMs as a meta-model, we first showed how well they perform in approximating benchmark ultidimensional mathematical functions. Then, basin-scale optimal water allocation problems were solved using surrogate optimization techniques. Both SVM and ANN models were able to approximate and represent the MODSIM DSS reasonably. However, the results show that SVM-based surrogate optimization has performed satisfactorily in terms of both the quality of the solutions and the saving in computational burden compared to the model employing ANN as the meta-model.