Research On Safety Decision Technology Of Full-ocean-depth Underwater Vehicles Based On Bayesian Network

With the continuous development of human society,the non-renewable resources on land are gradually scarce,and it is imperative for human beings to comprehensively and thoroughly develop and utilize the ocean.As a high-efficiency underwater working platform,the unmanned underwater vehicle(AUV)plays an irreplaceable role in marine development.With the expansion of the scope of AUV applications and the increase in operation tasks,AUV systems have become more and more complicated,coupled with the harsh and complicated marine environment,various failures will inevitably occur.By intelligently monitoring the AUV's movement status and formulating reasonable decision-making information based on the state of the AUV,the next work plan of the AUV is determined,which is of great significance to the safety of the AUV.This article focuses on the safety decision-making technology of the full ocean depth AUV,and mainly conducts the following research work:Research on overall scheme of the AUV safety decision system for the whole sea.According to the specific processes of the whole sea deep AUV in the deep sea environment,the research and analysis of the main dangerous states that AUV may face and the decision-making scheme adopted.According to the characteristics of the full-sea-depth AUV,a Bayesian three-layer decision-making network framework structure suitable for the subject is studied,the network construction process is designed,and the overall scheme and work flow of the full-sea-depth AUV safety decision system are established.Research on the safety decision-making reasoning model of AUV based on Bayesian network.According to the Bayesian network design process,the node attribute information of each layer of the Bayesian network is studied.Aiming at the problem of Bayesian network structure learning,this paper studies the network structure learning through the scoring search algorithm.Aiming at the problem of missing prior probabilities of some node variables in Bayesian networks,a method combining sample statistics and expert experience is used to complete the missing probability information.Netica software is used to establish a network model of the full sea depth AUV safety decision system,and the effectiveness of the Bayesian network established for the full sea depth AUV safety decision is verified by simulation analysis.Research on the improved Bayesian network inference model.Aiming at the problem of attribute redundancy in traditional Bayesian networks,a method of attribute reduction based on rough sets and upper and lower approximation sets is studied,and the rationality of the reduction is verified by Pearson product moment correlation coefficient and two-sided T test.Aiming at the problem that the weights of node variables that have a significant impact on the diagnosis decision result are not reflected in the Bayesian network,a method based on word frequency-inverse document frequency weighting factor(TF-IDF)algorithm and multiple linear regression model for attribute weighting And verify the rationality of attribute weighting through node sensitivity analysis.The safety decision reasoning simulation comparison of the Bayesian network before and after the improvement is performed by Netica software to verify the effectiveness of the improved Bayesian network.Experimental research on the improved Bayesian network model security decision system.In order to verify the rationality of the safety decision system designed and improved in this paper for the assessment of dangerous states,a simulation experiment of a pool when a single dangerous state occurs and a simulation experiment of a pool when multiple dangerous states occur simultaneously are performed.In order to verify the effectiveness of the safety decision-making system in the actual marine environment,according to the specific requirements of the project,a 1500-meter sea trial in the South China Sea was conducted for the abnormal water leakage signals and deep and altitude abnormal signals during AUV operation.