Research On Forecasting System Of Qiantang River Tidal Bore Based On Deep Belief Network

Tidal water of Qiantang river is the one of the most spectacular natural landscapes in the world,it brings people a visual feast,at the same time,create security risks to the masses on both sides of Qiantang River.Since people do not timely understand the characteristics of Qiantang River tidal bore,there are many people were drowned in Qiantang River every year.Therefore,how to construct the effective tidal level prediction model is very important to the tidal Protection work of the Qiantang river.In view of the shallow network can not dig the original relationship between the Qiantang River tidal data and other data,which affect the reconstruction accuracy of the predicted data,becauce of monolayer feature without hierarchical structure obtained via shallow network model learning.For that reason,a method for predicting the tidal level of the Qiantang River based on deep belief network was presented in the paper.The main work is as follows:(1)Research on the principle of Qiantang River tidal bore.the previous Qiantang River bore prediction models are built on the basis of historical tidal data on single.In view of the above mentioned factors,this article has studied the characteristics of Qiantang River tidal bore.According to the causes of the generation of Qiantang River tidal bore,the velocity variation characteristics of tidal bore,the sound data of tidal bore,this paper were selected the "day" as data,the average tidal bore velocity,the sound data of tidal bore,tidal level historical data and tidal level data of downstream sites as a data for building a model from 5 aspects combined with the realities of Qiantang River hydrological stations.It can extract the information furtherly between the Qiantang River bore characteristics.(2)A method for predicting the tidal level of the Qiantang River based on deep learningwas presented in the paper.According to the characteristics of the Qiantang River tidal bore,this paper build the prediction model of Qiantang River tidal bore combined with the deep belief network extracting the advantages in data.Firstly the multiple limited Boltzmann machine(RBM)was used to extract the Qiantang River data,and then the BP neural network was to reconstruct the prediction data.According to the comparation of the prediction results of the BP neural network and the support vector machine,the results showed that the prediction error of Qiantang prediction model based on the deep belief network decreased.(3)Model performance optimization.This paper does not directly use lunar "day" as data,but build the sine function which can simulate the trend of the Qiantang River tidal bore level andreduce the pressure of training model.At the same time,the impulse item is added in the training to avoid getting the model into local optimal.The experiment proves that these improvements have optimized the model.