Research On Framework And Key Technologies Of Intelligent Fisheries Big Data Platform

With the development of Internet of Things,cloud computing and big data technology,the amount of fishery data in China is rapidly accumulating and expanding,which is both an opportunity and a challenge for the fishery industry.Through investigation,the existing fishery information platform provides relatively perfect data retrieval,data display and other services,but the various fishery data are isolated from each other,ignoring the relationship between the data and the actual value,far from meeting the maximum demand for the use of fishery data value.Therefore,in order to better analyze the relationship between large-scale fishery data,make full use of the value of fishery data resources,and solve the problems of relatively backward fishery informationization,low utilization rate of data analysis and easy formation of data islands in China,this paper designs a large-scale intelligent fishery data platform based on Hadoop distributed architecture.The platform stores massive fishery production data,and generates corresponding data analysis model through large data related technology and neural network algorithm.It analyses and predicts aquaculture production,aquaculture water quality analysis and evaluation,fishery disasters and other important fields in fishery.Finally,it will be sent to the government,fishery-related units,fishery-related enterprises,scientific researchers and fishery practitioners.Staff and others provide the required information services.Based on Hadoop big data technology and the characteristics of fishery production environment and industry,this paper designs a four-tier architecture for intelligent fishery big data platform.Aiming at the problems of complex aquaculture environment,many factors affecting aquaculture production and unpredictability in fishery,this paper uses BP neural network with strong non-linear mapping ability as data analysis method to analyze the number of aquaculture factors.According to the complex non-linear relationship between them,the corresponding prediction model of aquaculture production is put forward;aiming at the problem that the input parameters of traditional BP neural network are difficult to choose,the optimal featurecombination of neural network input is obtained by the method of choosing priority features by classification of decision tree algorithm;aiming at the problem that the number of hidden layer nodes in BP neural network is difficult to determine,this paper combines "dichotomy".Segmentation algorithm "quickly determines the number of hidden layer nodes in the neural network,realizes the optimization of the traditional BP neural network algorithm,improves the convergence speed of the network and reduces the error of the network output without affecting the accuracy of the network output.Using MapReduce distributed programming model,the parallel design of BP neural network algorithm is carried out,and the parallel algorithm of BP neural network based on MapReduce is constructed to meet the needs of massive fishery data processing.Taking the prediction of red tide grade in fishery disasters as an example,the detailed design and result analysis of the algorithm are carried out.Finally,the intelligent fishery data platform with high efficiency and fault tolerance is constructed by combining the key technologies of large data with the neural network algorithm.