Research on the Prediction of Quay Crane Resource Hour based on Ensemble Learning

In Container terminals,a quay crane’s resource hour is affected by various complex nonlinear factors,and it is not easy to make a forecast quickly and accurately.Most ports adopt the empirical estimation method at present,and most of the studies assumed that accurate quay crane’s resource hour could be obtained in advance.Through the ensemble learning(EL)method,the influence factors and correlation of quay crane’s resources hour were analyzed based on a large amount of historical data.A multi-factor ensemble learning estimation model based quay crane’s resource hour was established.Through a numerical example,it is finally found that Adaboost algorithm has the best effect of prediction,with an error of 1.5%.Through the example analysis,it comes to a conclusion:the error is 131.86%estimated by the experience method.It will lead that subsequent shipping cannot be serviced as scheduled,increasing the equipment wait time and preparation time,and generating additional cost and energy consumption.In contrast,the error based Adaboost learning estimation method is 12.72%.So Adaboost has better performance.

Autonomous Vessel Scheduling

This study deals with an autonomous vessel scheduling problem when collaboration exists between port operators and an autonomous vessel company.A mixedinteger nonlinear programming model is developed,including decisions in assigning autonomous vessels to berths at each port and the optimal arrival time of each vessel at each port in an entire autonomous shipping network.This study aims to minimize the total cost of fuel consumption and the delay penalty of an autonomous vessel company.The nonlinear programming model is linearized and further solved using off-the-shelf solvers.Several experiments are conducted to test the effectiveness of the model and to draw insights for commercializing autonomous vessels.Results show that a company may speed up an autonomous vessel with short-distance voyage once fuel price decreases to gain additional benefits.

IMPROVING SOLVER PERFORMANCE THROUGH REDUNDANCY

It is well known that hierarchies of mathematical programming formulatlons with different numbers of variables and constraints have a considerable impact regarding the quality of solutions obtained once these formulations are fed to a commercial solver. In addition, even if dimensions are kept the same, changes in formulations may largely influence solvability and quality of results. This becomes evident especially if redundant constraints are used. We propose a related framework for information collection based on these constraints. We exemplify by means of a well-known combinatorial optimization problem from the knapsack problem family, i.e., the multidimensional multiple-choice knapsack problem （MMKP）. This incorporates a relationship of the MMKP to some generalized set partitioning problems. Moreover, we investigate an application in maritime shipping and logistics by means of the dynamic berth allocation problem （DBAP）, where optimal solutions are reached from the root node within the solver.