Research Into Machine Learning Methods For Optimal Marine Aquaculture Layout

A wide variety of marine life provided people chemical energy, mineral energy,medical energy and space resources. Therefore, studying marine economy is thenecessary road for sustainable development. The density and distribution of marinelife in the water will have a certain impact on the tide of the seawater. As a result, theenhancement of marine life according to local conditions is the premise of obtaining agood economic benefit and it is very important for people to study the optimalaquaculture layout.As a new machine learning method that was developed on the basis of thestatistical learning theory, support vector machine has solved many questions forlearning methods such as the small number of samples, nonlinear, over-learning, highdimension, and so on. It has a good application in the field of prediction, classificationand regression. On the other side, due to the relatively small observed data inOceanography, too many influent factors and the complexity of physical processes, alarge amount of data are difficult to extract and accurately used. Based on all thequestions, the thesis proposed a method to combine support vector machine andhydrodynamic to solve the marine aquaculture layout problem.Focusing on the particular breeding sea Sanggou Bay, with little observation datain this case, the thesis studied how to make full use of the data provided by historicaldocument, combine with hydrodynamic model, use statistical learning method and diguseful information from the limited data. First, the resistance generated by the marineaquaculture raft frame is added to the hydrodynamic model. Then the model isimproved through optimizing the drag coefficient of the raft frame. After that, thepaper used SVM to obtain the change of the flow in the whole bay. Because of the non-fixity and diversity, the thesis proposed a new method thatcombines multi-output support vector machine with the POM model to solve theactual situation. Combined with the hydrodynamic model and with a large number ofexperiments, we can conclude that the multi-output support vector machine not onlysaved the time, but also produced more accurate results. It can provide a large amountof data for oceanographic research, as well as scientific evidence on the optimallayout for improving marine environment.