Research On The Movement Of River Bed Pebble Based On LSTM Neural Network

Bed load movement generally exists in the transitional section between natural mountain rivers and plain rivers.However,due to river bed pulsation,river bed sand composition,uncertainty of river bed surface morphology,bed surface location,and upstream section recharge conditions in these areas,it is very important.Large uncertainty,therefore,the movement of the load pebble has great uncertainty in time and space.In order to prevent natural disasters related to the movement of bedload pebbles,as well as for the purpose of dredging waterways,constructing water conservancy facilities,treating river channels,and predicting river evolution,etc.The information of study the movement process of bedload pebbles in time and space and their position at each moment is necessary.In this paper,multi-fusion sensor technology is used to obtain the state information during the movement of the pebble,and on this basis,the LSTM neural network is used for trajectory prediction.The main work and conclusions are as follows:1.In order to obtain the data set,use MEMS multi-fusion sensors and 3D printing technology to conduct data collection for anhydrous experiments and generalized water tank experiments.First,the rationality of the collected data was verified under anhydrous conditions,and then various physical parameters of the model pebbles were obtained through a tank experiment.After the physical parameters of the model pebbles were obtained,the quaternion method and strapdown attitude navigation principle were used to solve the data.Calculate and construct a data set;2.On the basis of the original structure of the LSTM neural network,add the Attention mechanism model,and combine the Kalman filter to propose a new composite model;3.After the data set is obtained through design experiments,test with the proposed composite model to evaluate the performance of the proposed model.Experimental results show that the method proposed in this paper is superior to the single LSTM and LSTM?Attention models in terms of accuracy and trajectory fitting.