Research On Runoff Prediction Based On Intelligent Algorithm

Runoff is an important component in the field of hydrology,an important condition for constituting regional industrial and agricultural water supply,and a constraint factor for the scale of socio-economic development.Reliable runoff prediction is of great significance for water resources management,reservoir scheduling and reducing the loss of life and property caused by flood and drought disasters,and then influenced by climate change,human activities and geographical features,the formation and development of runoff presents a high degree of The complexity,nonlinearity and volatility of runoff bring challenges to the accurate prediction of runoff.How to solve the problems of low accuracy,poor reliability and weak stability of current runoff prediction methods by intelligent algorithms,and build a prediction model with good prediction performance and strong robustness has become the focus of current research.In this paper,the daily runoff data and water level data of some hydrological stations in the middle and lower reaches of the Yellow River basin in Henan Province from 2002 to 2020 are used as the research objects.Recurrent Unit(GRU),Broad Learning(BL),Least Squares Support Vector Machine(LSSVM),Quantum Particle Swarm Optimization(QPSO)to build different types of runoff prediction models and conduct comparative analysis to provide new methods and new ideas for runoff prediction.The research contents and conclusions are as follows:(1)The ADF(Augmented Dickey-Fuller test,ADF)test,PP(Phillips & Perron test,PP)test,Mann-Kendall,Pettitt test were used to analyze the smoothness and abrupt variability of runoff from four important hydrological stations in the middle and lower reaches of the Yellow River to reveal the runoff The results show that the runoff at the four hydrological stations is non-stationary and has abrupt changes,which provides a reliable theoretical basis for the subsequent runoff prediction.(2)In the single runoff prediction model,four single runoff prediction models,GRU,BL,LSTM,and LSSVM,are constructed,among which the GRU model based on deep neural network and BL based on width learning can better simulate the change trend of runoff,and compared with the regression model LSSVM,the GRU and BL models have higher prediction accuracy and credibility,and the cross-sectional comparison of different The prediction results of different hydrological stations can be found that all four machine learning models have better prediction results in hydrological stations with less volatility of runoff series,which indicates that reducing the non-stationarity of runoff series can improve the prediction accuracy of runoff.(3)In the combined "decomposition-synthesis" model,we propose an improved runoff prediction model combining Variational Mode Decomposition(VMD)and recurrent neural network GRU,and firstly,from the perspective of similarity and proximity,we use a waveform matching method based on gray The results show that,the combined model of modal decomposition has better prediction accuracy and credibility than the single model,which indicates that the VMD signal decomposition method can obtain the variation patterns embedded in the runoff series,reduce the non-smoothness of the series,and effectively improve the accuracy,credibility,stability and robustness of runoff prediction.(4)Construct a "decomposition-optimization-synthesis" runoff combination model.The runoff prediction model TVF-EMD-IQPSO-OS-BL is proposed based on Time-Varying Filters Empirical Mode Decomposition(TVF-EMD)and improve Quantum Particle Swarm Optimiztaion(IQPSO)optimized BL,which uses the idea of regularization to solve the problem of BL easily falling into overfitting.The model uses the idea of online sequence to improve BL,so that BL can update the model online without retraining according to the change of data,and uses IQPSO to optimize BL to solve the shortage of artificially set model parameters to choose the optimal parameters.The results show that the TVF-EMD can effectively avoid the modal aliasing problem and better preserve the time-varying characteristics of the signal.Compared with the TVF-EMD-SVM,the prediction accuracy and confidence of TVF-EMD-IQPSO-OS-BL are improved by at least 43.3% and 13.2%,and the forecast level reaches the qualified level of hydrological specifications.