Creating a Forecasting Model for a Volatile Environment

Examensarbete för masterexamen
Camitz, Axel
Johansen, Mattias
The container terminal in the Port of Gothenburg, Skandia Container Terminal, is operated APM Terminals that is part of the Danish consortium A.P. Møller - Mærsk A/S. In the terminals hinterland operations, containers arrives and departs from the terminal by either rail-road or truck. For the rail-road, there is a sufficient system for controlling the volumes containers arriving to and departing from the terminal. No such system is in place when i comes to the truck traffic. Many container terminals in Sweden provide hauliers with slot times for when they can visit the terminal. However, APM Terminals, do not use such slot times. Instead, the haulers issue a visit code that gives them access to enter the terminal anytime during the upcoming two week. Since the visit codes are valid for such a long time, there is no sufficient indication of future volumes. To provide planners and management with predictions of future volumes, a forecast system needs to be developed. To forecast future demand in predictable environments are relatively easy and can be done by using simple models. For more volatile environments where the fluctuation of demand can seem irregular, the development of forecast methods are not as simple. The aim of this study was to develop a forecast model that could predict the daily volumes of trucks arriving at the Skandia Container Terminal. To produce the most accurate and usable forecast system the performance of different models were compared. The best performing time series model was determined by using the measures of RMSE, MAPE, MaxAE and MaxAPE. Parallel to the time series models, two causal models were developed, one simple linear regression model and one multiple linear regression model. The findings suggest that the time series model based on ARIMA yields a forecast with the highest predictive ability. Furthermore, it was seen that the operations of container freight has potential for digitization and more advanced method of forecasting. This study shows how more data and more advanced model could be used to potentially make models with higher predictive ability.
Forecasting , Time series forecasting , Machine Learning , LSTM , Container Operations , Hauliers , Freight
Arkitekt (konstruktör)
Geografisk plats
Byggnad (typ)
Teknik / material