The Research Of Multi-dimensional Visibility On Ocean Based On Machine Learning

Atmospheric visibility is one of the important meteorological indexes of sea,which has great influence on marine resources development,military activities,navigation and Marine meteorological research.At present,the research on visibility mainly focuses on the prediction of single dimension.The analysis of the experimental results is relatively lopsided.Therefore if the machine learning method can be used to carry out a multi-dimensional statistical analysis of visibility in the limited data,which is the change of visibility can be understood more comprehensively.Therefore,this paper conducts a multi-dimensional research on visibility on ocean based on machine learning.The main work and conclusions are as follows:(1)Through the theoretical analysis of BP(Back propagation)neural network model and LMS(Least mean aquare),the LMS-BP(Least mean square-back propagation)neural network time dimension prediction model of sea visibility was proposed,which solved the over-fitting and under-fitting situations of BP neural network prediction model due to the large training sample and high correlation.At the same time,through the statistical analysis of multi-dimensional meteorological and pollutant data for training modeling,the short-time visibility prediction with high accuracy can be achieved.Compared with the existing neural network prediction model,the accuracy is improved by 10%.Through the research methods of this article analyzed the distribution of pollutants,experiments show that the pollution of land only has certain influence to the coastline and near sea area than middle and far sea area.(2)An improved random forest spatial interpolation algorithm for visibility is proposed.The main idea is to use Gaussian hybrid clustering model to cluster the sample points with different visibility levels,and then carry out the spatial interpolation prediction of the random forest algorithm,which improves the interpolation accuracy of the random forest spatial interpolation algorithm.Experiments show that the proposed improved random forest visibility spatial interpolation algorithm has higher interpolation accuracy than the IDW,the Co-kringing and random forest spatial interpolation algorithm,and which compared with the random forest spatial interpolation algorithm on the interpolation precision improved by 15%.At the same time,experimental analysis of spatial distribution prediction in three aspects of different seasons,different visibility levels and different weather conditions,and the experimental results show that the visibility on ocean spatial interpolation precision have higher interpolation accuracy in winter,afternoon,sunny and visibility is that higher than 10 km.At the same time,the spatial distribution of visibility in coastal and near sea area varies greatly with time,and the visibility is poor in coastline and near sea areas than middle and far sea area.