Research On Prediction Method Of Mooring Ship Motion Responses In Open Wharf Based On Big Data

There are many factors relevant to mooring ship in open wharf,I made a research on them and predict the motion response with the artificial neural network and build a model on the Hadoop platform.Many factors affect the mooring ship,such as the power factor,the wind,the waves,the flow,and also constraints cable and fenders,as well as the characteristics of the ship itself.In all of these,the waves are most important,and there are many relevant researches,mainly in wave height,period and the angle of incidence on three factors.The larger the wave height,wave represents more energy inherent;the larger periodic wave,the ability of the wave to communicate is stronger,and if the frequency of the wave frequency and the ship itself is similar or identical resonance,will cause greater losses;different wave incident angle,to varying degrees on the ship six degrees of freedom of movement.The method of data preprocessing is data mining,to determine the overall correlation between variables,determine the approximate input variables,and then filter the error value,outliers and missing values,make the appropriate treatment;calculate Pearson correlation coefficients calculated from the data,and ultimately determined the input variables by the correlation.Introduces development and architecture of Hadoop,a Big Data platform,studied principle and structure of the main components--HDFS and MapReduce;deep studied about the topology of BP neural network,the learning process,and put forward improvement.In the learning process of BP neural network,the gradient descent method usually used to update the connection weights between neurons,for this proposed dual learning rate approach,setting the second learning rate was some multiple of the first learning rate,the bigger learning can accelerate the convergence rate,the smaller one helps improving accuracy;and in the choice of the learning rate,using adaptive learning approach,combined with the dual learning rate.Simulation result shows that the algorithm can effectively accelerate convergence.Build a model of MapReduce framework and transplant BP neural network algorithm on Hadoop platform;tested in pseudo-distributed,compared to non-Hadoop BP neural network,there is a slightly improvement of training speed in larger data sample.