Research On Collision Avoidance Model In Marine Traffic Flow Simulation

With the rapid development of the shipping industry and the advancement of shipbuilding technology,the marine traffic is becoming increasingly busy and the ship operations are more complicated,which greatly increases the risk of marine traffic accidents.As an important technical means to study marine traffic problems,marine traffic flow simulation has guiding significance for solving problems such as marine traffic planning and marine traffic flow evaluation.Collision avoidance model is one of the core of marine traffic flow simulation research.Traditional traffic flow simulation technology has problems such as low fidelity or poor timeliness.It is necessary to introduce new methods to improve the model.Based on the analysis of the modeling accuracy and practicability of the existing marine traffic flow simulation,this paper proposes a ship collision avoidance model based on Reinforcement Learning.The main research includes the following aspects:(1)Collecting AIS data of ships in real sea areas,decoding and pre-processing data,using ArcGIS to reproduce the marine traffic flow on the map,and using Hadoop technology to screen the trajectory data of the encounter situation,choosing three typical encounter situations among many encounter situation.(2)Selecting the Deep Q-Leaming algorithm in Reinforcement Learning to model the collision avoidance model.Through the navigation simulator,the three encounter situations are simulated under the same latitude and longitude,the same navigation state and the same wind flow,and the actual simulation map is intercepted as the comparison reference of the simulation results.(3)Training the collision avoidance model,through the convolutional neural network to train the Q-Leaming model by using the navigation trajectory data,and stop training when the error between the netwrork target Q value and the real Q value meets the preset accuracy or reaches the maximum training number.The output model completes the learning of the collision avoidance strategy.(4)Designing the simulation experiment,input the initial state data of the three encountering situations into the collision avoidance model,simulating the results with Matlab tools,and compare the simulation results with the actual simulation.Then comparing the simulation effect of the collision avoidance model with the traditional sea traffic flow simulation effect,and analyze the simulation fidelity and timeliness.