Deep Learning Supported FPSO Security Situation Classification

Waters in China are rich in oil and gas resources. FPSO(Floating Production Storage and Offloading) has become a very popular solution in exploitation activities of ocean oil and gas resources in Bohai Sea and South China Sea. In recent years, many mooring accidents happened with the increase in tonnage of FPSO and the complexity of marine environments. This paper presents a novel multi-sensor feature learning model using DBN(deep belief network) to avoid a number of FPSO mooring accidents.The rest of this paper is presented as follows. First, presents the structure of FPSO single-point mooring system and analysis the influence factors of FPSO system, and some related work about accident monitoring of FPSO with existing techniques. Then, presents the Deep Learning algorithm and RBM(restriction Boltzmann machine) as a basic composition unit of DBN, and some good practice studies of the DBN algorithm in various areas.Then, this paper details the proposed the feature learning and classification model based on historical data. The developed FPSO state monitoring model based on DBN can be structured in three consecutive stages:(1) collecting sensory dataset for DBN training and testing, and defining the classification standards,(2) training the DBN with the training datasets and developing DBN-based feature learning and classification models,(3) validating DBN-based classification models with testing sensory dataset. At last, the security situation monitoring approach for FPSO using DBN-based classification technique is compared with two existing machine learning techniques which are introduced briefly in this paper. The experiment demonstrates the superiority of the proposed model by taking the data of ground truth as case studies.