The Research On Theory And Method Of Vessel Collision Avoidance Base On The Fuzzy BP Neural Network

The study of vessel collision risk index and decision-making system of vessel collision avoidance has always been at the center of concern for all sailors and navigators. A review of the literature shows that research in this field still faces a number of unresolved issues, including real collision risks for vessels.The purpose of studying vessel collision risk index is to provide a scientific basis for decision-making aimed at avoiding vessel collision. So far, navigators would only gather some raw data through observing the target vessel, such as distance, direction, and their changes, to decide whether there is a risk for their own vessel to collide with the target vessel. However, a further study and processing of the observational data may uncover additional, sometimes unexpected, valuable information.Artificial Neural Networks is a new artificial intelligence information processing system that simulates and extends the cognitive function of a human brain by using electron or photon elements to duplicate the structure and function of certain neural cells. A biological neuron is very slow compared with an electronic circuit (10^-3 vs. 10^-9 second). However, a human brain may accomplish many tasks at a speed by far the faster than an existing computer. This is mainly because of the huge parallel functional potential of the biological neural network, that is, all neural cells can work at the same time. A neural network based on connectivity mechanism is characterized by massive parallel functioning, massive connectivity, distributive storage, high non-linearity, high tolerance of errors, changeable structure, and non-precision of calculation. The neural network has the ability to self-learn, self-adapt, and self-organize, capable of dealing with most fuzzy and complex questions.BP neural network is currently the most widely used and theoretically supported neural network model. It uses the smooth activate function (also called activate function) and has one or more latent layers, with the neighboring layers fully connected by a weighting value. It is a feedforward neural network, which means that information under processing will move forward layer by layer. It learns the weighting value based on the difference between the ideal output and the actual output, and then modifies the weighting value via back propagation. The study of BP neural network has become a heated topic internationally, particularly its application in all kinds of control areas. Because of its unique function, BP neural network has been successfullyused in many ways, such as detecting and judging a work environment, recognizing a picture, deciding on the choice, and optimizing a system.In operating a vessel, how to deal with the encounter situation of vessels and how to choose the best collision avoidance strategy and timing is a long-standing topic for research study. This thesis attempts to use a neural network's unique functions of self-learning, self-adaptation, self-organization, and capability of dealing with non-linear problems to design or optimize a collision avoidance system.First, using Dc to denote the most representative argument for a judgment and Tcto indicate network input, the system will determine the basic risk index through learning expert examples, fully carrying out the self-learning, self-adaptation, and self-organization functions of the neural network. Based on the definitions of DCPA and TCPA, the thesis will define and distinguish between a spatial collision risk index and a temporal collision risk index. Both models are based on multiple factors and the idea of stimulus-reaction in order to determine a reasonable compound operator for each index.Second, the thesis attempts to design a rational BP neural network based on experiment and previous research, using signal processing and characteristic abstraction technique to directly obtain an input signal vector incorporating all major parameters of a vessel encounter situation. This is done by learning from the expert samples and training so that the neural network is adjusted to an optimal weighting value and threshold value demarcation number distribution. The study used the raw data obtained by one vessel about another vessel as the input for the BP neural network, and obtained satisfactory results through simulation. This trained network can determine several levels of collision risk index, with the number of levels and classification system being determined by the designer, according to a specific vessel encounter situation.Third, the thesis analyzes a fuzzy risk index determination system, and studies the potential factors affecting the various empirical parameters in the risk under the jurisdiction function. It optimizes the system with the optimal result of the self-learning of the BP network on the parameters affected by various environmental factors to make it more objective, precise, and automated. It quantitatively defines three vessel encounter situations provided by the collision avoidance rules and determines collision avoidance actions. Based on those collision avoidance rules and survey of the sailors, the study analyzed the data qualitatively and quantitatively. The results summarize the knowledge of sailors about maritime collision avoidance rules,...