Study On Intrusion Monitoring And Early Warning In Offshore Oil And Gas Exploration Area

As the exploitation of offshore oil and gas resources in China tends to mature,the future oil and gas development area will gradually focus on the deep and far sea.However,the latest offshore exploration and exploitation operation was invaded and strongly disturbed by third-party ships,and even affected the progress of exploitation operation for a time.In this thesis,the neural network is used to support the rapid response of ships in the exploration area and improve the possible response-ability of the third party in the following aspects:(1)Aiming at the problem that the time window length of ship trajectory prediction is not clear,a standard process of constructing the historical ship intrusion emergency process into a time-series network model is proposed.By extracting the relationship between each subject and the relationship between subjects in the event,and attaching time attributes to the relationship,a network model reflecting the evolution of the relationship between subjects over time is formed.The global importance evaluation method is used to calculate the importance of each subject in the network,with an accuracy of 70%.The key nodes in the process are successfully identified,and it is determined that the length of the ship trajectory prediction time window should be 15 minutes or more.(2)Aiming at the problem that the trajectory prediction accuracy of third-party ships in the operation sea area does not meet the user’s demand,this thesis proposes a trajectory prediction model based on Adaboost-LSTM.By comparing the deviation of prediction results obtained by different numbers of predictors,six weak predictors are selected to build an integrated framework.By optimizing the input data,adjusting the weight of each weak predictor,and taking the real navigation data of ships in the limited sea area in recent years as the data set for training.The accuracy of the improved model is 40%higher than that of the single LSTM model,and the trajectory prediction time window can reach 24 minutes.(3)In view of the difficulty of driving away third-party invading ships in the region,a multi–Guardianship to multi invading ship driving model based on deep reinforcement learning is proposed in this thesis.By constructing the mathematical model of ship motion and optimizing the absolute position parameters of the ship in the input data as the relative position parameters,the corresponding optimization measures for each step of the thirdparty ship are realized.The case results show that compared with the random action selection strategy without rules to follow,the method proposed in this thesis can reduce the total fuel consumption by more than 30%,and control the stability of the displacement effect of the best performance and the worst performance within the fluctuation range of 10%.