Study On Runoff Prediction Of Goaf With Special Underlying Surface Based On Improved BP Neural Network Model

Floods are characterized by unstable inter-annual variation and easy to be affected by special topography.Due to the lack of flood diversion,detention and passing capacity in general river basins,the flood peak discharge caused by the flood not only destroys the living environment of the surrounding residents,brings economic losses,but also poses a huge hidden danger to the lives and health of the residents.Therefore,it is very important to forecast the flood process in the basin.Due to the existence of large-scale coal mining activities in Shanxi Province,the underlying surface conditions of the river basin have been changed,thus affecting the flood formation process.At the present stage,many hydrological experts have carried out researches on the flood formation process under the special underlying surface conditions.However,due to the complex topography of the special underlying surface conditions in the goaf,the mechanism system of runoff generation and confluence in the basin has not been completely established,and the relevant rainfall and runoff data in many areas are not perfect.At present,there is no single model that can solve all these problems.In this paper,the upper reaches of Fenhe River in Shanxi Province,where coal mining activities are frequent,are taken as the research object,and the rainfall and runoff data in the basin controlled by Jingle Hydrological Station are selected for analysis and prediction.Of mined-out area before first use of rainfall and runoff data,based on the MATLAB software to establish a suitable BP neural network model,secondly by genetic algorithm(GA)and particle swarm optimization algorithm(PSO)to optimize BP neural network model,get a GA-BP,PSO and BP model,by running the model to get the predicted result analysis,and then using the mined-out area rainfall,runoff data after the appearance of the original model of parameter debugging,end up with is suitable for the mined-out area special underlying surface conditions of flood forecast model.The results are as follows:(1)Of mined-out area before using the data of BP and GA-BP and PSO and BP neural network model for prediction of 1990 years ago,the mined-out area special underlying surface conditions to form the session before the flood forecast effect is good,the percent of pass is 66.67%,66.67% and 80%,respectively,can play the role of guiding practice,and the other a few flood prediction effect is not ideal.Investigate its reason,after 1990,due to a large number of coal mining,the formation of different size of the goaf.(2)After 1990,in view of the situation that the prediction value of the field flood in the goaf appeared too large error,BP,PSO-BP and GA-BP neural network prediction models were adjusted to re-predict the field flood in the control basin of Jingle Hydrological Station under the special underlying surface condition of the goaf,and the predicted value of the field flood was obtained.With the improved model,the qualified rate of flood runoff prediction under the special underlying surface of goaf is increased to 58.33%,66.67% and 66.67%.(3)The PSO-BP neural network model >GA-BP neural network model >BP neural network model is used to predict the runoff prediction effect of the large flood under the condition of the special underlying surface of the goaf.The GA-BP neural network model >PSO-BP neural network model >BP neural network model is used to predict the runoff prediction effect of small and medium-sized flood under the condition of special underlying surface of goaf.(4)The research results show that the runoff prediction model suitable for the special topography is established according to the special underlying surface conditions of the goaf,and the prediction effect is very significant,which has an important guiding significance for the practical application.