Research On Runoff Forecasting Method Based On Multi-model Forecasting Information Fusion And Model Parameter Optimization

With the rapid development of China's water conservancy construction and the large-scale reservoirs being put into operation successively,the upper reaches of the Yangtze River presents a situation of large-scale reservoir group joint operation.The power generation efficiency,flood control and drought resistance capacity,and ecological guarantee capacity of the river basin have been improved significantly.At the same time,higher requirements have also been put forward for the safe and economical operation of hydropower energy systems.As one of the core contents in the field of hydrology,runoff forecasting is closely related to the efficient use of water resources and the economic operation of hydropower stations.Accurate and reliable runoff forecast results are prerequisites for safe operation of cascade reservoirs in the river basin.In order to ensure the safe operation of hydropower and energy systems,it is imperative to establish a complete runoff forecast system to provide high-precision runoff forecast information for flood control decisions of cascade reservoir groups in the river basin.Therefore,this paper takes the upper reaches of the Yangtze River as research object,explore techniques and methods for improving the accuracy of Mid-term and short-term runoff forecasting.The primary research substance and achievements are as follows:(1)Focus on the issue of improving the accuracy of long-term runoff forecasting in the catchment,based on the multiple linear regression model,BP neural network model and Elman neural network model,a multi-model combined forecasting study was developed.This paper proposes a combined forecasting model based on information entropy weighting method and a combined forecasting model based on neural network.The results show that compared with the single forecasting model,the two combined forecasting models can improve the forecasting accuracy significantly.The forecasting effect of the combined forecasting model based on the neural network is optimal,it can provide data support for formulating mid-and long-term power generation dispatching regulations in catchments.(2)Aiming at the problem of hydrological model parameter optimization affecting the forecasting effect,based on the Xin'anjiang model,the three elements of the flood process,namely the relative errors of total runoff volume,certainty coefficient and relative peak flood error,are used as evaluation indicators to establish single-objective evaluation index parameter optimization model and multi-objective evaluation index parameter optimization model,The SCE-UA single-objective parameter optimization algorithm and the MOSCDE multi-objective parameter optimization algorithm were introduced to optimize the parameters of the above model.Using the above model to carry out flood forecasting from xiluodu~xiangjiaba,The results show that The multi-objective parameter optimization model has a significantly better forecasting effect than the single-objective parameter optimization model.(3)In order to obtain the optimal solution from the solution set generated by the multi-objective algorithm,the fuzzy optimization criterion is selected,and the coefficient of variation theory is used to determine the weight,and a non-inferior solution optimization method is proposed.The hydrologic model parameter optimization rate of Xiluodu ~ Xiangjiaba interval is defined as an example to verify the effectiveness of the proposed non-inferior solution optimization method.This method can quickly give the optimal parameter combination,greatly reduce the workload,and improve the multi-target the practicality of the algorithm is conducive to the promotion and application of multi-objective algorithms and has a certain theoretical value.(4)Using the three-target MOSCDE algorithm to optimize the parameters of the Xin'anjiang model,Tank model,and TOPMODEL model,we can calculate the results of the runoff of the basin.Using a segmented Muskingum method to establish a confluence model and obtain the calculation results of the flood course evolution.In addition,serial and parallel methods were introduced to correct the results and improve the overall flood forecasting effect in the basin.The results show that the serial and parallel correction method is effective and the accuracy of the short-term runoff forecast achieved is in full compliance with the requirements and has engineering application value.