Study On Watershed Hydro-meteorological Characteristics Analysis And Runoff Nonlinear Integrated Forecasting

Watershed hydro-meteorological characteristics analysis and runoff comprehensive forecasting have long been two of the most important research subjectcs in the field of hydrology,and are important foundations for the optimal planning and management of water resources as well as the sustainable development of human society.Owing to the unique climatic conditions and geographical features,the spatial-temporal distribution of water resources in China is severely inhomogeneous.What's more,due to the strong influence of global warming and human activities,the river water cycle is undergoing profound changes,which intensifies the spatiotemporal variation and inhomogeneity of water resources distribution.In recent decades,the construction of large water conservancy projects and the implementation of inter-basin water transfer projects changes the physical mechanisms and manifestations of the watershed natural hydrological cycle,which results in the change of the formation and evolution law of hydro-meteorological processes.These factors put forward higher requirements for watershed hydro-meteorological characteristics analysis and runoff comprehensive forecasting.Focus on the key scientific and technical problems faced by the evolution law of the natural water cycle process and the water resources optimal allocation and utilization in changing environment,this paper analyzed the spatio-temporal evolution of hydrometeorological elements and studied the state of art methods of nonlinear comprehensive forecasting.Relevant achievements of this paper can promote the development of hydrological analysis and forecasting in engineering application and is important for the safe and economic operation of hydropower stations as well as the rational allocation and efficient use of water resources.The following are four of the main research contents and innovations of this study:(1)Based on long-term hydrometeorological observation data,this study analyzed the seasonal and inter-annual variations of the hydrometeorological elements including temperature,rainfall and runoff of the past 37 years of the Jinsha River Basin.The novel MASH(moving average over shifting horizon)method and the Mann-Kendall trend test,the linear regression test and the Sen's slope estimation were introduced to analyze the seasonal and interannual variations of the meteohydrological processes.The correlations between the meteorological and hydrological elements were further analyzed to reveal the complex response mechanism of the runoff process to the meteorological elements.The results of the experienments showed that that MASH method can eliminate the influence of noise and outliers on the data and reveal the long-term trends of hydrometeorological processes.The internal elements of the natural water cycle in the Jinsha River Basin have shown an increasing trend and the tendencies of changes are not evenly distributed throughout the year.The meteorological elements and the runoff processes are closely related,which indicates that the climate change have a certain impact on the watershed runoff processes.(2)Conceptual rainfall-runoff modelling is a widely-used approach for rainfall-runoff simulation in streamflow forecasting.The objective of this paper was to introduce an extended non-dominated sorting genetic algorithm-II(NSGA-II)for the automatic calibration of a hydrologic model.The orthogonal design based initialization technique was exploited to produce a more uniformly-distributed initial population.At the same time,a chaotic crossover operator as well as a chaotic mutation operator were presented to avoid trapping into local minima and to obtain high quality solutions.Finally,a multi-criteria decision-making(MCDM)approach combing Shannon entropy weighting method and an improved technique for order preference by similarity to ideal solution(ITOPSIS)based on projection was introduced to prioritize the Pareto optimal solutions and select the comprehensive optimal solution as a follow-up step.Hydrological data from two river basins named the Leaf and Muma River basins were exploited to test the ability of the orthogonal chaotic NSGA-II(OCNSGA-II)for solving the multi-objective HYMOD(MOHYMOD)problem.The results demonstrated that the OCNSGA-II can obtain betterdistributed Pareto optimal front and thus can be exploited as an effective alternative approach for the multi-objective automatic calibration of hydrologic models.(3)Due to the strong nonlinearity and non-stationarity of the medium and long-term streamflow time series in the upper reaches of Yangtze River,this paper firstly identified the chaotic characteristics of the monthly runoff time series and deduced the optimal delay time and embedding dimension using chaos theory.The reconstructed phase space matrix was exploited as the input matrix and an improved Adaboost.RT algorithm based on adaptive dynamic threshold was introduced to train several extreme learning machine weak learners for monthly streamflow forcasting.Secondly,the empirical wavelet transform(EWT)was employed to eliminate the redundant noises by discasding the decomposed mode with the highest frequency.The input weights and hidden biases of the artificial neural network were optimized using the multi-verse optimizer(MVO)algorithm for annual runoff forcasting.Results obtained from this study indicated that the proposed hybrid models can capture the nonlinear characteristics of the medium and long-term streamflow time series and thus provideing more accurate forecasting results.(4)Due to the deterministic forecasting of streamflow can only provode a single point value of the targeted variable of the future time,a streamflow interval prediction method based on multi-objective kernel extreme learning machine was proposed.The double-output kernel extreme learning machine was exploited to predict the upper and lower bounds of the possible forcasting value of streamflow.The probability that the runoff observations may fall into the prediction interval has also been given.The orthogonal chaotic NSGA-II algorithm was used to optimize the hidden layer output weights of the kernel extreme learning machine to obtain streamflow prediction interval with high quality.Results have shown that the multi-objective interval prediction model can overcome the disadvantages of the single-objective interval prediction model to select the optimal penalty coefficient.At the same time,the multi-objective interval prediction model can obtain prediction intervals with different confidence levels at the same time.Compared with the interval prediciton model based on multi-objective neural network,the one based on multi-objective kernel extreme learning machine has better stability and generalization ability.The orthogonal initialization technique and chaotic genetic strategy can improve the distribution and convergence performance of the NSGA-II algorithm.A multi-objective interval prediction model with stronger global convergence ability was obtained and thus improving the reliability and validity of streamflow prediction intervals.