A Dynamic Marine Oil Spill Prediction Model Based on Deep Learning

Water pollution resulting from shipborne oil spills has caused tremendous damage to local marine ecology and has led to huge property losses. By establishing a multi-angle monitoring and early-warning mechanism, the oil spill's trajectory to a large extent can be effectively predicted, and thereby the accompanying economic losses can be reduced. In this study, real-time processing is performed on an oil spill monitoring video frame to analyze the characteristics of the spilled oil, such as the edge contour features, diffusion rate, centroid, area, etc. An initial system model and an oil spill behavior monitoring model were established based on the previously mentioned characteristics. A long short-term memory network in a recurrent neural network was introduced to deal with the memory information to obtain the connection between features and influencing factors. The spatial variable distribution of a dynamic grid reference system, as a substitution of the traditional original data sequence, was used as the system input. The result shows that the model has good stability and can provide a reliable interactive prediction; the result is especially significant given the vigorous exploitation of petroleum resources and the rapid development of maritime transportation in today's global economy. The auxiliary oil spill early-warning system investigated in this paper provides a scientific basis for targeted strategic oil spill emergency planning.