Vessel Anti-collision Forewarning based on Mean Impact Value and Random Forest

In order to explore the inherent laws of ship collision and determine the optimal timing of ship collision avoidance, a Vessel Anti-collision Forewarning method was proposed based on mean impact value (MIV) algorithm and random forest (RF) algorithm. First, the initial index system for vessel anti-collision forewarning was established from vessel static information, vessel dynamic information and environment information by combining macro and micro risk of collision. Second, MW feature selection algorithm was utilized to extract main factors from the initial index system. Finally, aiming at the shortcomings of the existing single classifier forecasting models, RF combined classifier was introduced to identify and forewarn the vessel collision risk situation at different encounter situation. Experimental results show that RF model has higher prediction accuracy, stability and generalization ability compared with BP neural network, support vector machine (SVM), classification and regression tree (CART) classifier model.