Simulation–optimization framework for multi-season hybrid stochastic models

A novel simulation–optimization framework is proposed that enables the automation of the hybrid stochastic modeling process for synthetic generation of multi-season streamflows. This framework aims to minimize the drudgery, judgment and subjectivity involved in the selection of the most appropriate hybrid stochastic model. It consists of a multi-objective optimization model as the driver and the hybrid multi-season stochastic streamflow generation model, hybrid matched block boostrap (HMABB) as the simulation engine. For the estimation of the hybrid model parameters, the proposed framework employs objective functions that aim to minimize the overall errors in the preservation of storage capacities at various demand levels, unlike the traditional approaches that are simulation based. Moreover this framework yields a number of competent hybrid stochastic models in a single run of the simulation–optimization framework. The efficacy of the proposed simulation–optimization framework is brought out through application to two monthly streamflow data sets from USA of varying sample sizes that exhibit multi-modality and a complex dependence structure. The results show that the hybrid models obtained from the proposed framework are able to preserve the statistical characteristics as well as the storage characteristics better than the simulation based HMABB model, while minimizing the manual effort and the subjectivity involved in the modeling process. The proposed framework can be easily extended to model multi-site multi-season streamflow data.