Research On The Key Problems Of Marine Engineering With Machine Learning

In the context of intelligent marine engineering,we focus on the key problems of information extraction,trends ananlysis,thermocline recognition and crowd-evacuation model for emergency.We introduce machine learning models to analysis marine engineering data and the main researches are as follows.1.With the fast development of marine engineering and the boom of hotspots on marine research,the number of scientific manuscripts increases explosively and it becomes extremely difficult to extract information from a larger number of papers and search for research hot topics.We collected 300,000 abstracts from 14 marine engineering journals(2010-2019).Using LDA model and Phrase LDA model,the hot topics and cutting-edge technologies are captured.Combined with the clustering algorithm we predicted the hot topics and trends in marine engineering.Meanwhile,we also verified some of the important research trends.2.From the aforementioned hotspots analysis,we found that the research on physics property of ocean plays a key role in marine engineering research.A good example in case is that thermal layer is an important phenomenon and is of great value for marine-climate prediction and underwater communication.However,the coverage of the marine data is not comprehensive and meticulous.Moreover,it is also lack of label.World Ocean Atlas 2013(WOA13)is one of the few gloable marine-climate data.And the original data of WOA13 is large and noisy.We first preprocess the data into a raster map.Then,Generative Adversarial Networks(GAN)is used to extend the preprocessed data.At last,we combined the deep learning model ResNet and information entropy to recognize the thermal layer.The experimental results show that the recognition results are better than the original data.It is also indicated that GAN network can improve the marine data mining.3.From the aforementioned hotspots analysis,we find that risk control is becoming one of the hotspots of this research area.With the change of global climate,the probability of ships being affected by extreme weather increases.With high density of population and limited export,sea-vehicles are prone to congestion,stampede and other incidents.Therefore,the establishment of appropriate scenario model and crowd model can provide more reasonable evacuation guidance and improve the survivability in emmergency.We introduced convolution neural network,Faster R-CNN and other deep learning methods for data preprocessing,object tracking,projection coordinate conversion,and social force behavior simulation,and the trajectories of crowd are predicted.