Short-term prediction of outbound truck traffic from the exchange of information in logistics hubs: A case study for the port of Rotterdam
Transportation Research Part C: Emerging Technologies
Short-term traffic prediction is an important component of traffic management systems. Around logistics hubs such as seaports, truck flows can have a major impact on the surrounding motorways. Hence, their prediction is important to help manage traffic operations. However, The link between short-term dynamics of logistics activities and the generation of truck traffic has not yet been properly explored. This paper aims to develop a model that predicts short-term changes in truck volumes,
... ed from major container terminals in maritime ports. We develop, test, and demonstrate the model for the port of Rotterdam. Our input data are derived from exchanges of operational logistics messages between terminal operators, carriers and shippers, via the local Port Community System. We propose a feed-forward neural network to predict the next one hour of outbound truck traffic. To extract hidden features from the input data and select a model with appropriate features, we employ an evolutionary algorithm in accordance with the neural network model. Our model predicts outbound truck volumes with high accuracy. We formulate 2 scenarios to evaluate the forecasting abilities of the model. The model predicts lag and nonproportional responses of truck flows to changes in container turnover at terminals. The findings are relevant for traffic management agencies to help improve the efficiency and reliability of transport networks, in particular around major freight hubs. Transportation Research Part C 127 (2021) 103111 2 flows. A comprehensive view of the existing methods for long-term freight generation can be found in Holguín-Veras et al. (2014) . Additionally, only a limited number of data collection methods (surveys, demographics) have been tested for this problem. Furthermore, these methods mostly use site-related characteristics, i.e. the number of employees; area; and capacity, which results in significant errors in the case of a relatively high number of trucks e.g. seaport's terminal (Sarvareddy et al., 2005) . Most importantly, however, these methods do not address the dynamics of within-day variations of truck flows. Currently, therefore, traffic managers lack the tools to make accurate short-term predictions of truck flows. As truck demand is a consequence of the logistics operations of multiple actors, it is worthwhile to explore the possibility to predict flows based on the exchange of information between them. Since the early 2000s, several studies have emphasized the importance of information exchange in freight transport and logistics chains (Giannopoulos, 2004; Di Febbraro et al., 2016; Banister and Stead, 2004) . Cooperation between different logistics actors is important in intermodal hubs, such as seaport terminals, where a prompt exchange of information is needed for cooperative planning and execution of intermodal freight transport (Di Febbraro et al., 2016) . Due to the advances in information and communication technologies, many ports around the world now use different systems to benefit from information transmission. In particular, Port Community Systems (PCS) are there to ensure that everyone in the hinterland transport chain (e.g. Port authority, terminals, careers, depots) can easily exchange information. These systems are generally built upon an integrated central database where all the information from clients of the port and government agencies, like customs, come together. PCS could align the vessel arrival time and container discharge time to the truck's container pick up time. This information is mostly useful for optimizing terminals operations (Heilig and Voß, 2017) . PCS also has the potential to provide other additional information services. For example, road carriers could be offered planning information services, to give terminals and empty depots a prior notification of the trucks' arrival times. Terminal operators, in return, could manage these arrivals and speed up the loading and unloading process of the containers at the terminal gates. One of these functions, that we will explore here, is the provision of timely warnings about truck flow dynamics to traffic managers. The main objective of this paper is to propose an approach to predict the next hour's truck volume, based on PCS data and truck count data. The key contributions of this paper are as follows.