Research On Multi-objective Optimization Model For Calibration Problem Of Xin'anjiang Model

Hydrological model is a generalization of complex hydrological processes in the real world,simulating a series of hydrological phenomena such as rainfall,confluence and evaporation.The study of hydrological model is one of the important research fields in hydrology and water resources,and also an important tool to simulate hydrological processes and understand hydrological laws.It is an irreplaceable role in the calculation of precipitation,surface entrapment and infiltration,evaporation,groundwater and so on.It is gradually applied to flood prediction,water resources management and reservoir scheduling and other fields.According to the description of the hydrological model process,it can be divided into two categories: the conceptual hydrological model and the hydrological model based on the physical mechanism.Each hydrological model has many parameters that representing the hydrological process,which are difficult to measure from the actual project.But its value has a very important effect on the prediction of hydrological model flood,which leads to the calibration of the hydrological model parameters.In the field of hydrological model calibration,different objective functions are constructed to evaluate the different characteristics of hydrology.Some objectives focus on the fitting of the whole process line of floods,others focus on the fitting of large diameter values or the fitting of small diameter values.Therefore,the selection of the objective function is critical to the optimization of the hydrological model parameters.At the same time,the application of hydrological model parameter calibration in practical engineering shows that single target optimization only reflect one aspect of the hydrological process,and the practical engineering needs to consider more.Therefore,the multi-objective optimization model of hydrological model parameter calibration problem came into being.At present,the research on the calibration of hydrological model parameters mainly focuses on the improvement of hydrological model structure or optimization algorithm.Most of them are single target optimization,and the choice of the objective function is more subjective,there is no one principle can refer to.Therefore,in this paper,the error function of the calibration of hydrological model parameters is analyzed,makes the selection of the objective function has a reference principle.Then uses the multi-objective evolutionary algorithm to optimize the objective function,and then uses the idea of ensemble learning to improve the hydrological model flood predictive accuracy and stability.The major work of this paper include:1.Error analysis of hydrological model.Selects the conceptual hydrological model: Xin'anjiang model,choose orthogonal design method to design experiment.On the basis of the experimental results,it is analyzed whether these error functions are suitable as the objective function of the parameter calibration problem of Xin'anjiang model.At the same time,the correlation between the objective functions is analyzed by using the further processing of the experimental results.The research work has certain reference value for the selection of objective function of parameter calibration of hydrological model,and provides a reference basis for selecting objective function of multi-objective optimization algorithm.At the same time the research method can also be applied to other hydrological models.2.Application of Pareto-based multi-objective evolutionary algorithm and ensemble learning in calibration of hydrological model parameters of Xin'anjiang.The predicted and the measured runoff of single field flood fit well,but the generalization ability still has room for improvement.The Pareto solution set by the multi-objective evolutionary algorithm can meet the requirement of ensemble learning.The multi-objective evolutionary algorithm NSGA-II is used to get the Pareto solution of the objective functions,and the partial of the Pareto solution is selected as based learners.Then,the regression prediction method is used to integrate the results of the based learners,the experimental results show that the combination of multi-objective evolutionary algorithm and ensemble learning can improve the generalization ability and prediction accuracy of Xin'anjiang model.