Research On Grey Wolf Optimal Least Squares Support Vector Machine Model Based On Variational Modal Decomposition And Its Application In Runoff Forecasting

Since the Industrial Revolution,humans' influence on the climate has become greater and greater,and the ability to build reservoirs and dams has become stronger.The formation of runoff is also more and more affected by human activities,resulting in more complex and non-stationary runoff.It's getting more and more obvious.Therefore,runoff forecasting,which is of great significance in water resources utilization and flood prevention and disaster relief,has gradually become a difficult and hot research topic in related fields.For this reason,in view of the difficulties in runoff prediction,this paper establishes a gray wolf optimization least squares support vector machine prediction model based on variational modal decomposition to predict the monthly runoff of the Meixian section of the Weihe River,hoping to provide it for the runoff forecasting work in the middle reaches of the Weihe River Certain reference value.The main tasks of my paper are as follows:(1)This article takes the 600 months runoff collected by the Weijiabao Hydrological Station in Meixian Section of the Weihe River as the research object.First,statistics on the change characteristics of the runoff in this section are collected.The runoff changes in this section are relatively uniform over the years and unevenly distributed during the year.,70% of the runoff is concentrated between May and October.From the 600-month time scale,the monthly runoff has obvious periodicity.(2)This paper establishes a gray wolf optimized least squares support vector machine prediction model,and conducts simulation experiments on monthly runoff samples.The paper gives a detailed description of the gray wolf optimization algorithm and least squares support vector machine.Aiming at the difficult problem of general least square support vector machine hyperparameter selection,the gray wolf optimization algorithm is used to penalize the least square support vector machine with the penalty factor C and kernel The parameter ? is optimized to find.Establish a gray wolf optimized least square support vector machine prediction model,select 300,400,and 500 samples from the 600-month runoff distribution for training,and compare with the particle swarm optimization model and the genetic optimization algorithm optimized model.The results show that the gray wolf optimization model converges faster.The 50-iteration running time of the gray wolf optimization model is 0.0328 seconds faster than the particle swarm and 0.4279 seconds faster than the genetic optimization model.(3)This paper combines the variational modal decomposition algorithm into the gray wolf optimization least square support vector machine model,and conducts simulation experiments.The thesis introduces the theory of variational modal decomposition algorithm,the establishment of variational function and the reconstruction conditions.Variational modal decomposition decomposes the original runoff data into a series of characteristic components with strong regularity,and then optimizes them with gray wolf The least squares support vector machine model respectively predicts the characteristic components,and finally reconstructs the prediction results of each characteristic component to obtain the final prediction results of the runoff forecast model.The gray wolf optimization least squares support vector machine model of variational modal decomposition was used to correct the prediction,and the prediction results were compared with the undecomposed gray wolf optimization model and the gray wolf optimization model of empirical mode decomposition.The results show that the prediction results of the decomposed optimization model are significantly better than those of the undecomposed optimization model.The determination coefficient of the prediction results of the variational modal decomposition optimization model is 0.016 higher than that of the empirical mode decomposition optimization model,and the average absolute error is less totaled 0.0028411 billion cubic meters.