An artificial intelligence use case for smart port resource allocation

Ports and their ability to operate as a transit hub – smoothly and efficiently – is under public scrutiny. It has become clear that the congestion emerging in ports across the world is causing concern, not only among maritime and supply chain professionals but even for the world’s largest economy.

As much as there are constraints associated with the port and the maritime side, the hinterland flows of goods in and out of the port are causing woes too. Since the beginning of 2021 port congestion has been gradually building up at ports across the globe, in particular at the ports on the US West Coast.

Extended operating hours are seen as a way to enable the ports to handle more volume. This topic has even been discussed at the White House in light of the severe congestion at the West Coast ports.

Simulations empowered by artificial intelligence (AI) systems can prove the point, but even more importantly they can provide more insights which help to define the optimal opening hours per day based on predictions. In fact, the primary benefit of AI is the smarter allocation of resources.

Winds should be favourable for the implementation of AI-based approaches as many ports are choked and wish to become smarter in managing traffic flows for better utilisation of infrastructure.  This topic is right at the heart of the discussion around the transformation towards smart ports which requires large-scale digitalisation of businesses, governmental services, buildings, logistics and transport. 

In most cases, ships are handled on a first come first served basis, which creates enormous challenges for planners that need to design the operations that bring goods in and out of the port, ideally based on predictable situations. Several solutions have been discussed to overcome this operational challenge and optimise the utilisation of different infrastructure, like ships, trucks, ports, roads, rail tracks etc., involved in maritime supply chain networks.

Among the options are: Slot management systems and just-in-time operations; Expansion of buffer zones at the port; Increasing manpower in the ports, even bringing in national guards; and Flexible opening hours.

In this article we look at evidence of how new approaches empowered by artificial intelligence (AI) can help to flatten the curve of truck arrivals at the port. Based on a real-life case with data from the Port of Valencia, we reflect on how to better predict and synchronise the port-bound traffic flows.

Expanding the capabilities of the port

Traffic jams, queues and delays are part of daily life, particularly for professionals working in ports. The port is the space where unsynchronised port-bound flows from the hinterland meet unsynchronised cargo flows arriving from the seaside. The port is the buffer zone that has to absorb shocks caused from peaks at both ends. This logically requires flexibility through buffer zones or time reserves.

The challenge is real and significant. In the United States of America, it is estimated that at least seven of the nation’s 10 busiest ports by volume face congestion regularly. Ships are waiting near the shore for days, or even weeks, to get unloaded.

Ports are short of space to stack containers until they are picked up for loading onboard ships or on trucks for transport to their destinations on land. The trucks wait in line for hours – up to eight or nine hours in some cases – to pick up one single container. Chassis to move containers are in short supply.

While container rates exploded and other transport and logistics costs increased, the situation has caused delays for customers across the country that are lasting weeks. The situation has become a trade barrier with a significant negative impact on the economy.

Idling cargo ships in the Ports of Los Angeles and Long Beach also account for more than 100 tons of pollution per day, exceeding the emissions of the 6 million cars in the region – adding to the negative impact on both air quality and public health.

This port congestion that has gradually been building up from the beginning of 2021 reached its peak with 109 container ships at the ports of Los Angeles and Long Beach on January 9, 2022.

The port, transport and logistics industry has basically two options to get back to normal. It could increase the pool of equipment to have spare capacity at its disposal so that when demand surges or supply is disrupted, the extra ships, containers, vehicles, and trailers can provide the additionally required supply. This is wasteful, costly, and inherently inefficient. Alternatively, it could introduce methods to allow the existing capacity to handle a larger volume of cargo. In this paper, we focus on the latter option.

“There are 168 hours in a week. For the most part, our terminals are open less than half those hours. Without expanding terminals or building new facilities, we could handle still more cargo by utilising more of those hours,” says Mario Cordero, Executive Director of the Port of Long Beach.

AI can help to understand the impact of such measures and whether the curve can really be flattened and how. We have used real-life data from the Port of Valencia to test if this is the case.

AI study of Port of Valencia

A deeper study of the vessel and truck flows to Port of Valencia shows that, based on public vessel schedule information only, useful predictive models for truck traffic rates can be applied on varying time scales. The same study also demonstrates how data sharing of e.g., cargo exchange or advanced vessel scheduling information, helps to improve prediction accuracy and time range.

Based on the AI system that was developed, we considered cases where the port is assumed to operate continuously, i.e., trucks arriving at the port at any time and date. Figure 1 below shows the actual distribution of gate exit events by hour of the day, and the simulated distribution of exit events when no hourly restrictions are set.

In reality, the distribution would probably not be this uniform when the port is operated by flexible opening hours, but the outcome of the simulation indicates that there is capacity for distributing traffic more evenly.

This flattening of the curve, i.e., cutting the peaks of the utilisation the port’s surrounding infrastructure, requires data sharing and port integration with the seaside and landside communities and would significantly release the burden on the port, the cities close to ports and the hinterland infrastructure. This requires the consent of the pool of stakeholders involved, with truck drivers at the core. This might be challenging but the potential to reduce waiting time could act as an incentive.

Figure 2 shows actual and simulated daily numbers of exit events for the first half of 2019, and a hypothetical simulation where the port is operating continuously, i.e., truck arrivals are also allowed during Saturdays, Sundays, and holidays.

In the simulation, the standard deviation of the number of events per day – taking into account only the active days – is reduced by 30-40 % if continuous traffic is enabled. We would like to highlight that this analysis was performed on a subset of events which means that the traffic volumes are relatively small relative to the size of the port.

The study shows that AI systems can predict port container flows with high precision and that prediction systems can calculate different scenarios as alternative methods of operating, with different impacts on the fluidity of traffic flows highlighting the benefits of different specific adjustments.

The application examples are simplified but demonstrate how the considered modelling approach can be adapted to simulate effects of changes in operational parameters. Resource and traffic planners can utilise the traffic rate predictions to dynamically adjust resource allocations, e.g., to prepare in advance for days or weeks when peaks are predicted. The recommendation is thus not to expand opening hours per se, but to adopt flexibility when peak times are predicted.

The results of the above-described real-life study indicate the additional value that can be created through AI systems in a port environment. AI systems are based on trust and data sharing. Building trust hinges on the involvement of trustworthy parties as leading examples and their clear communication and subsequent actions.

The adoption of such AI systems relies on an acceptance among major actors that optimising cargo flows in the interconnected and interdependent self-organising world of maritime supply chain networks requires a holistic view and approach, with at least a minimum level of collaboration to ensure sufficient data sharing. Beyond the adoption of such AI tools, the major challenge will be the implementation of the necessary adjustments to current policies and practices.

Interestingly, the AI system explored in this article is also designed to enable predictions in cases where only publicly available data on vessel locations and port call schedules is available, with a trade off in prediction accuracy. However, and most importantly, dynamic operating times may require changes to current labour regulation, for example, which could require the consent of several stakeholder groups. Stakeholder management is therefore as important for digital transformation as technical capabilities.

Flexible hours for higher throughput

Global maritime supply chains have been disrupted. Shipping companies wish to be quickly served, trucks and their drivers do not want to wait in line for hours to pick up a single container, train operators do not wish to leave ports with empty wagons, and ports do not want to have their infrastructure and resources idle. AI systems enable the integration of different communities, like port, seaside and hinterland corridors, which helps to deal with disruptions through better synchronisation of cargo flows.

Synchronising cargo flows through a holistic approach to planning and management requires the availability of data through data sharing. Introducing dynamic scheduling for the operating hours of ports appears promising, but true optimisation of opening hours requires a certain level of digitalisation and data sharing across relevant channels and communities.

AI tools are immediately available and AI systems can quickly be built, while expanding capacity takes time. The shortage of semiconductors limits the number of new trucks that can be put into service. The lack of drivers persists or has even increased as working conditions degraded. Significant new ship capacity will only kick in 2023/2024. As such, more optimised infrastructure utilisation offers many benefits.

This article is another call to move beyond operations based on physical presence towards virtual coordination. Some initiatives focus on one aspect, like just-in-time arrival procedures, but all pieces of the puzzle need to be aligned and based on the same logic. AI powered predictions combined with slot management for visiting ships, trucks, and trains and their integration would provide the foundations for integrated and synchronised cargo flow management, which is in the interests of the entire maritime supply chain network community.

Dynamic opening hours is a straightforward approach. AI allows for peaks to be predicted and consequently for a dynamic approach to be applied to port operating hours, i.e., extended openings when peaks are predicted and reducing them when traffic levels are expected to be low. AI systems will help to understand the actual impact of extended and dynamically fixed operating hours.

The Port of Valencia study has shown that, through AI systems, the port may be able to serve more ships, manage more truck visits, and optimise its storage capacity while providing relief to road infrastructure during peak times.

Editor’s note: This is an edited version of the original article, which contains a full list of references and is downloadable here.