A Study Of Waterway Traffic Flow Forecasting Model Based On Ais Data And Artificial Intelligence Algorithms

China’s recent development has resulted in a highly export-oriented and ocean-dependent economy.Its dependence on marine resources and space has been greatly increased.The country’s maritime rights and interests in the outer seas also need to be constantly maintained and expanded.As the most common tool for maritime transport and preservation of maritime rights or interests,ships are becoming larger and equipped with increasingly sophisticated technologies.Meanwhile,as the awareness and interest of many countries to such rights and intersts arise,the ship transportation is also increasing very quickly.The increasing demand for maritime trade has also strained ports,and resulted in traffic jams and accidents,which lead to other problems,including ecnomic losses and increased air pollution.The development of a reliable short-term traffic forecasting tool is therefore vitally important for the operation and management of ships in ports and water channels.Reliable forecast of the arrival and departure of ships plays an important role in developing these mechanisms and it is expected that water transportation system will benefit greatly from the use of such a technique.However,there are still many pending problems related to the prediction of ship traffic flow,such as inaccurate predictions and immature methodologies.These issues serve as the primary motivation of this thesis,which aims to develop an advanced ship traffic flow forecasting method using Automatic Identification System(AIS)data and artificial intelligence algorithms.In order to achieve the above goals,the following work has been accomplished in this thesis:The theories and parameters of ship traffic flow and corresponding forecasting models are first systematically summarized based on a review of existing studies from the literature.After a background introduction of how AIS ship data is carried out,the method for acquiring the AIS data employed is outlined.The data is then analyzed to obtain important parameters such as the characteristics of ship traffic flow,proportions of different ship lengths,and ship velocities.The accuracy of the acquired data is then validated through a cross-validation.Based on the results of the ship traffic flow analyses,it is found that Neural Network(NN)models fits the characteristics of the data and the purpose of this study.A Neural Network model was therefore developed for ship traffic flow forecasting.As the selection of the input variables of the Neural Network model require optimization and it is difficult to do so using traditional optimization techniques,a Genetic Algorithm was adopted to select the final input variables from a large of candidate input space.In order to validate the results of the proposed forecasting model,traditional ship traffic flow forecasting models were also implemented and the results were compared with those from the proposed NN model.The comparisons have proven that the proposed forecasting model is superior than the traditional ones,in terms of forecasting accuracy.The above models are to predict short-term ship traffic flows of one cross-section of water channel hence they cannot fully take the advantage of available data at multiple cross-sections which are at either upstream or downstream of the same water channel.In this regard,the forecasting models based on traffic flow data from multiple cross-section are developed.It is found that the multiple cross-section models(MCMs)have an improved accuracy when compared to single cross-sction models(SCMs).Finally,the MCMs are applied to forecast vessel traffic of three channel sections of the Yangtze River and it is found that such models are useful for short-term traffic forecasting at different spatial and temporal scales.