Study On Fuzzy RBFNN Model For Forecasting Fishery Habitat Suitability Index

As an integral part of our fisheries industry, marine fisheries alsooccupy an important position in the national economy. As the watersworldwide overfishing and serious environmental pollution, the currentbottom fish resources have been severely damaged, while cephalopodsand pelagic fish production was not affected. Thus, the tuna fisheries,especially on oceanic tuna fisheries for the fishing objects are muchfavored by the country’s fisheries, many developed countries competingfishing. Bigeye tuna, also known as fat tuna, is one kind of migratory fish.Bigeye tuna are in large species, with more common in cluster activities.Bigeye tuna are mainly distributed in the Atlantic, Indian and Pacifictropical and subtropical waters. Tuna longline fishery dates back to18thcentury Japan, fishing in1907. After the updated fishing operation, thefishing area was extended to more than50sea miles away from the shorewaters. To mid-twentieth century, based on improved fishing vessels,equipment innovations, fishing and other fishing technology hasimproved the rapid expansion of fishing boats. In the next, Korea, ChinaTaiwan, Indonesia and other countries have entered the Indian Ocean tunato catch large fish.With the long-term sea fishing operations and fishing practices,aquaculture workers in China have accumulated a lot of marine catch data. But suffering without processing data analysis, they decided many fishingwaters delineation and determine the best fishing time mostly based ontheir experience. These situations led to our long-standing blind fishing,and higher cost of offshore fishing. If you can know the case of fisheriesin advance and do theoretical analysis to predict, the analysis can providestrong support for the offshore fishing industry. Thus, taking into accountthe actual needs of aquaculture workers, it is very important to carry outfishing tuna fishing situation prediction research.This thesis summarizes the existing research conducted by someaquaculture fish species and points out some problems.First, multiple regression analysis is often used as a traditional fishingprediction model theoretical tool, but the researchers ignored the inherentcharacteristics of marine data, that between the dependent variable andthe dependent variable is not simply divided into independent, and mutualinfluence is always exist. The primary assumption of multiple regressionanalysis is that the independence between the dependent variable.Secondly, the marine environment is more complex factors, the existingfishing conditions most predictive models contained fewer environmentalfactors, only temperature and sea level data. This is bound to have anegative impact on fishing situation prediction accuracy of the model.Now, it is needed to consider some of the factors which have a significantimpact factor. Second, at present most of the fishing situation prediction model is a relational model that just takes into accounts the marineenvironmental factors and production relations. The field of specialistexpertise available for the prediction model can provide more guidancefor modeling, but usage is not high. Finally, the advantages of radial basisfunction neural network has be verified by many scholars, but as for itsblack box nature of neural network research, mostly stay in the expertsystem or other system simulation is used. Few researchers try to digitself from the neural network. The development of the neural network islong, with complete theory. However, most researchers now use themodel simulation experiments simulate the assumption, which greatlylimits the ability of neural networks applied to the actual. So, it is worthyto study how to apply the real neural network theory to meet realproblems. The majority study of this paper is to collect relative data fromthe Indian Ocean, according to the number of tuna hooks, production dataand the marine environment factor data (water temperature, chlorophyllconcentration and sea surface height) and the experience of fisheriesdomain expertise, using radial basis function neural network to learningas well as to extract fishing knowledge. And analysis of the Indian Oceantuna habitat index, and ultimately get a better fishing situation predictionmodel. The forecast model system can be used to guide the productionand fisheries fishing activities, explore and discover the distribution offishing grounds. The main contents of the following aspects: (1) Comparative study of knowledge discovery algorithms. Existingknowledge mining algorithms include decision trees, support vectormachine, association rules, K-Means clustering, genetic algorithms, fuzzyclassification and clustering, rough classification and rule induction,neural networks and other methods. Considering the advantages anddisadvantages of existing methods and specialty of marine fisheries data,the RBFNN is chosen to dig the knowledge of fish habitats, to providedata support for further analysis and forecast of catch data.(2) Data collection. Collection of the Indian Ocean MarineEnvironmental factor data, including sea surface temperature, deep seatemperature, sea surface height, chlorophyll concentration, and theproduction of tuna longline data from January1999to December2009(5°N-5°S，40°E-80°E). Spatial resolution of SST and SSL data is0.25°×0.25°, while the output data is5°×5°average range. Therefore,before data analysis, the marine environment and habitat index factor datashould be matched, to ensure the same standards-based measurement. Weuse bilinear regression method to process data, in order to improve dataavailability and reliability.(3) Improve RBFNN Performance. RBF neural network learning methodfocuses on the recognition of the basis function centers and width. Thisarticle uses fuzzy clustering algorithm combined with the width fornew-type radial basis function neural network model which can be used to predict.(4) Use intelligent search method to obtain implicit knowledge ofRBFNN. An important drawback of existing neural network is difficult toshow the implicit network processing. This study is aim to describe thecontained rules of neural networks. Mix the views of different experts inthe field, to determine the discrete sample data analysis and create newdataset for rules extraction by Harmony Search algorithms.This is the first time the proposed method to dig the rules ofknowledge. All the rules could be applied to the fishing situationprediction system. The model obtained valuable knowledge of fisheriesindustry. Fish farms can analyze the situation and improve the predictionaccuracy. This study provides a new way of thinking.