Research On River Water Level Flow Forecasting Method Based On Hybrid BP Neural Network

Global hydrological change has an important and far-reaching impact on terrestrial ecosystems,and the simulation as well as prediction of hydrology are both important parts of studying hydrological change.Accurate prediction of hydrological factors such as water level and river flow plays an important role in water resources planning,reservoir scheduling and operation,shipping management and flood prevention.Based on the monitoring data of four hydrological stations in Yichang,Zhicheng,Luoshan and Hankou in the middle reaches of the Yangtze River,the intrinsic relationship between water level and river flow is explored and revealed in this paper.In addition,the method for predicting water level and flow is established based on hybrid BP neural network,so as to provide theoretical support and technical platform for hydrological prediction and warning.The main work and achievements of this paper are as follows:(1)In the case of vacancies and anomalies in the time series data of water level and flow of the four stations in the middle reaches of the Yangtze River from 2010 to 2013,the advantages and disadvantages of several commonly used data processing methods are compared and analyzed.And the central metric filling method is determined to process the data of vacancies and anomalies,so as to ensure the completeness and credibility of the data as well as the accuracy of the prediction model.(2)According to the complex rope curve relationship between water level and flow,the VMD-BP model is constructed for simulating the relationship between water level and flow and then for predicting the flow.Variational modal decomposition(VMD)decomposes the sequence data into several combinations reflecting the original sequence,which can reduce the fluctuation of the data of water level and flow.Taking the advantages of BP neural network in non-linear fitting into account,the two methods are combined to predict the flow,so as to improve the prediction accuracy.The model is then validated by the discharge data of Luoshan hydrological station,and the prediction results are compared with those of single-input BP neural network and multiinput BP neural network.The results show that,the overall prediction error of the VMD-BP method is only 1.41%,and the prediction effect of this method on lowfrequency components is excellent.(3)Aiming at the hysteresis relationship between upstream and downstream water levels,based on the spatio-temporal sequence model,combined with distributed lag model(DLM),VMD model and BP neural network,the DLM-VMD-BP model is constructed to predict water levels.The water level data of four stations in the middle reaches of the Yangtze River are used for case analysis,the prediction results are compared with regression model,single input BP neural network and multi-input BP neural network.The results present that,DLM-VMD-BP model has a good effect in water level forecasting of upstream and downstream,and it is significantly useful for water level forecasting,early warning,and rational optimal allocation of hydrological stations.(4)By integrating the VMD-BP river flow forecasting model and the DLM-VMDBP upstream and downstream water level forecasting model,and by using ArcGIS Engine secondary development and MySQL database technology,the system of water level and river flow forecasting as well as early warning in the middle reaches of the Yangtze River is developed and constructed.The system realizes four functional modules,namely information query management,information analysis,forecasting and early warning,and system management,which provides an information management platform for hydrological forecasting and prediction in the middle reaches of the Yangtze River.