Study On Prediction Of Runoff And Sedimentation And Multi-objective Optimal Operation Of Reservoir

Sustainable development of the economy and society puts forward more and more demands for hydraulic engineering. And how to conduct multi-objective optimal operation of reservoirs to achieve the unification of short-term and long-term benefit and economic and social benefits becomes a significant research subject. In this study, some topics involved in multi-objective optimal operation of reservoirs are investigated and the conclusions are as follows:(1) Two generalization models of natural watercourses, single channel mode and channel-beach mode, are presented to generalize the typical main sections of natural watercourses. And the flow evolution is simulated using the Saint-Venant equations set and the expansion Saint-Venant equations set with side flow which are resolved by Preissmann implicit difference method. A case study shows that the simulated flow evolution process is consistent with the actual process. Moreover, the improved BP neural network with genetic algorithm is employed to predict the propagation time of flow and the corresponding network model are established. A case study shows the neural network models are suitable for prediction of propagation time of flow.(2) Daily runoff prediction is investigated and a stepwise regression model of daily runoff prediction based on wavelet decomposition is proposed. The daily runoff time series at the hydrological stations upstream of the hydrological station under consideration are introduced into the former model. And then, the general components of the daily runoff time series at the hydrological stations under consideration and upstream at different timescales are obtained using the wavelet decomposition and reconstruction. Taking the original daily runoff time series and their general components as candidate independent variables, the stepwise regression models of daily runoff multi-step prediction are established. A case study shows that the stepwise regression model of daily runoff prediction based on wavelet decomposition is superior to auto-regression model, and is able to predict the daily runoff in 1~3 days in non-freeze-up period and 1~7 days in freeze-up period with accepted accuracy.(3) Prediction of sediment concentration is investigated. Taking Toudaoguai hydrological station as an example, the factors affecting the sediment concentration in the non-freeze-up and freeze-up runoff models are analyzed and the neural network models of sediment concentration prediction are established. A case study shows that, compared with the multiple regression models, the neural network models are able to simulate the nonlinear relationships between the sediment concentration and its factors much better and to provide certain reference for water and sediment regulation and reservoir operation.(4) Reservoir sedimentation is investigated and a compound predicting model of reservoir sedimentation is proposed in this study. First, the 1-D coupled model of sediment transportation is established by combining the 1-D unsteady transportation model of non-uniform suspended load and the 1-D model of unsteady density current siltation. And then, the solutions of the 1-D coupled model of sediment transportation are studied, and the fictitious flow method is introduced to forecast the evolution and the arriving time of the density current. The study of two cases indicates that the 1-D coupled model of sediment transportation can simulate the processes of reservoir sedimentation, the evolution and arriving time of the density current very well. Finally, the compound predicting model of reservoir sedimentation is established by combining the 1-D coupled model of sediment transportation and the BP networks. A case study indicates that the compound model is more than two hundred and fifty times as efficient as the 1-D coupled model of sediment transportation and can predict reservoir sedimentation with acceptable accuracy.(5) A multi-objective optimal operation model of water-sedimentation-power in reservoirs is established. Moreover, the inertia weight self-adjusting mechanism and Pareto-optimal set are introduced into the particle swarm optimization and an improved multi-objective particle swarm optimization (IMOPSO) is established. And then, the IMOPSO is employed to solve the optimal model and obtain the Pareto-optimal front. A case study shows that the former method is able to obtain the Pareto-optimal front with fine distribution properties. The Pareto-optimal front can reflect the results of each optimized plan intuitively and provide more effective support for the operating staff. Compared with NSGA-II, IMOPSO has close global optimization capability and is very suitable for multi-objective optimization problems.