Coupling And Optimization Of Hydrological Time Series Forecasting Models

The Modaoxi is a representative small and medium-sized mountain drainage area which is located in the upstream of the Yangtze River.In this paper,it have collected rainfall data and runoff data in the drainage area,these data were measured by the hydrological station(Longtan Hydrological Station Yulong,Jiannan,Moudao and Longju Rainfall Station)year after year which are established by our country.This paper uses a variety of different mathematical methods to analyze the characteristics of these data,and then building a mathematical model to forecast the hydrological time series.Then the Wavelet Analysis coupled with Artificial Neural Network to establish WNN model,and based on the global optimization characteristics of Genetic Algorithms,the threshold,weight,and time scale factors of the WNN model were optimized,and finally established an optimized WA-GA-ANN model.Longtan Station is the only hydrological station in the Modaoxi River Basin and controlling with the largest rainfall collection area.The analysis of the rainfall runoff characteristics of the station hopes to provide scientific guidance for flood control and disaster mitigation of water resources in the Modaoxi.The whole-basin rainfall prediction model was established,hoping to contribute to the mid and long term hydrological forecast of the basin.The main conclusions of this paper are as follows:(1)This study uses Trend Regression method,Mann-Kendall Rank Correlation method,and Moving Average method to identify the trend term of the series;Time Series Cumulative Correlation Curve method and Ordered Clustering method to identify the jump term of the annual runoff sequence at Longjiao Station;Fourier Analysis,Maximum Entropy Spectrum Analysis,and Wavelet Analysis are used for period identification.After analyzing the rainfall sequences of Longjiao Station from 1959 to1990 and Longtan Station from 2001 to 2010,it was found that the rainfall in the control basin of the station has a periodicity of 2a scale,and the overall rainfall shows an increasing trend,but in 1963?Rainfall tended to decrease between 1966 and1982-1988.Due to the impact of human activities,runoff produced abrupt changes in2001.(2)The Wavelet Analysis are discussed in detail from the aspects of wavelet denoising,decomposition layers,and periodic analysis.The wavelet de-noising threshold was optimized by using the Stein SURE method and the Entropy Criterion Threshold Selection method.These two threshold selection methods were applied to the rain sequence de-noising at Yulong Station.It was found that the peak value of the sequence after de-noising was significantly reduced,that is,the sequence The system error is reduced.At the same time,a method for determining the number of wavelet decomposition layers for white noise detection is proposed.The length of the rain sequence at Yulong Station is 192,and the optimal number of wavelet decomposition layers is 2.This is consistent with the maximum decomposition scale obtained by the empirical formula.(3)Taking the rain sequence of Yulong Station as an example,the paper uses Autoregression,Fuzzy Analysis,and Gray System Analysis to establish prediction models respectively,compare the residual and relative errors of the original and simulated sequences,and find that the emerging models are accurate,high efficiency and strong operability.It also explored the model principles of BP,RBF,and GRNN,and based on the rainfall data of Jiannan,Moudao,Longju,and the rainfall runoff data of Longjiao station,a prediction model was established to simulate the daily maximum water level of Longjiao station.The eigenvalues of the GRNN simulation sequence are closer to the measured sequence.(4)The original series is de-noised with Entropy Criterion,and the white noise detection method is used to determine the number of decomposition layers,and then input to the WNN model.Taking the monthly rainfall sequence and daily rainfall sequence of Yulong Station from 2001 to 2016 as examples,the influence of the time series scale on the prediction results of the coupled model was studied.It was found that the longer the sequence and the smaller the time scale,the more accurate the prediction results.Taking the monthly average rainfall,water level,and flow sequence of Yulong Station from 2001 to 2016 as an example,the effect of sequence complexity on the prediction accuracy of the coupled model was explored.The more complicated the sequence itself,the lower the prediction accuracy.Finally,in this paper,a set of genetic algorithm is used to optimize the parameters of the wavelet neural network model based on wavelet neural network weight and threshold setting.