Applications of Artificial Intelligence in Port Management and Operations: A Prisma-Based Review

2.1. Port Management and Operations / Lučko upravljanje i poslovanje

Port management is a service-oriented activity that coordinates administrative, economic, and technical tools to achieve commercial objectives and meet sustainability requirements [9]. It involves governance of port activities through the interaction of public authorities and private actors, with management models shape by institutional arrangements [9]. Core functions include strategic planning, infrastructure development, and adaptive capacity to evolving global trade dynamics. Building on this foundation, port authorities are now redefining these traditional paradigms by embedding digital transformation into their governance frameworks. This shift toward smart ports represents a strategic evolution – from static, institutionally driven decision-making to dynamic, data-enabled systems that enhance managerial responsiveness, operational adaptability, and institutional transparency [7], [10].

Port operations comprise the integrated activities that facilitate efficient vessel handling and cargo throughput, ensuring seamless connectivity between maritime and land transport [3], [11]. The International Transport Forum categorises port operations into three principal phases – landside handling; berth assignment and turnaround; and risk and safety management [12]. Table 1 shows the key technologies and benefits of each phase, demonstrating AI as a core enabler. In addition, AI-enabled digital twins of equipment and operational workflows create live virtual replicas that integrate machine-learning analytics with real-time data streams, enabling non-intrusive “what-if” simulations for capacity planning, maintenance scheduling and emergency-response drills. Ports utilising these end-to-end digital-twin platforms report throughput gains up to 20%, operating-expenditure reductions of 15-20% and significant environmental benefits [13].

Table 1 Key Technologies & Benefits for each Port Operations phases

Tablica 1. Ključne tehnologije i prednosti za svaku fazu lučkog poslovanja

2.2. Use of AI in the Port Sector / Primjena umjetne inteligencije (UI) u lučkom sektoru

AI has transitioned from pilot trials to a foundational element of contemporary port operations, underpinning improvements in planning, maintenance, scheduling, safety and sustainability. Core AI methods – ML, computer vision, optimisation algorithms and digital twins – now permeate all major port functions [3], [7], [8].

A flagship application is the AI-powered digital twin. High levels of accuracy in throughput forecasting have been reported, with hybrid network-analysis models achieving over 90% predictive accuracy [1], while digital twin-driven safety management systems have demonstrated reductions of approximately 25% in incident-response times during large-scale port trials [2]. Major providers (IBM, Siemens, Dassault Systèmes) have rolled out such platforms at Rotterdam, Antwerp and Shanghai terminals, with broad standardisation expected by 2026 [15].

Predictive maintenance represents another critical use case: neural-network models trained on vibration, temperature and load data can forecast crane and equipment failures up to 48 hours in advance. Research has demonstrated 85% recall for crane-fault predictions – cutting unplanned downtime by 20% – and similar outcomes have been confirmed in Chinese and Belgian ports [3], [13]. Machine-learning techniques such as Long Short-Term Memory (LSTM) networks and reinforcement-learning agents now optimise berth assignment and vessel-arrival forecasts. Researchers observed a 15% average reduction in vessel–berth mismatches [8], and report that ETA-forecast errors at leading terminals in Singapore and Hamburg have dropped from over 90 minutes to under 30 minutes [11] [17].

On the landside, deep-learning OCR systems paired with RFID readers automate gate processing and yard management. Researchers achieved 98% container-ID recognition under varied lighting, reducing gate-throughput times by 40% and improving yard retrieval times by 12%; when streamed over private 5G into TOS, these innovations yield a 35% cut in dwell times and a 12% CO₂-emissions reduction [3], [8], [13], [14]. AI also underlies advanced safety and security measures: Bayesian networks and support-vector machines forecast collision risks with 80% precision [4], and AI-driven intrusion-detection systems – compliant with National Institute of Standards and Technoloy Cybersecurity Framework (NIST CSF) and International Maritime Organization (IMO) Cyber Risk guidelines – monitor Operational Technology (OT) and Information Technology (IT) networks in real-time to flag anomalies. Sustainability is further enhanced by ML-based energy-management platforms that dynamically balance load and schedule shore-power connections, delivering up to an 18% reduction in peak energy demand and a 10% decrease in CO₂ emissions [6]. Successful, scalable AI adoption depends on modular, micro-services architectures, open APIs and robust data-governance policies. Studies highlight the need for standardised interfaces and algorithmic transparency to enable secure, multi-party data sharing in Industry 4.0 ports [7], and propose an Artificial Intelligence of Things-blockchain (AIoT–blockchain) framework to guarantee data integrity and traceability across complex port ecosystems [10].

4.1. Bibliometric Data / Bibliometrijski podaci

Of the 30 selected articles, published in international journals between 2021 and 2025, a growing trend in annual scientific production was observed. The year 2024 stood out due to a marked increase in the scientific community’s interest in the application of AI in port operations and management (Graph 1). Although an increase in publications for 2025 was projected, it was necessary to emphasise that, considering that the current year was still underway, the continuation of this trend could not be confirmed.

Graph 1 Number of articles by year of publication

Grafikon 1. Broj članaka prema godini publikacije

Regarding the journals, the 30 analysed articles were distributed across 23 different journals (Table 2), revealing an interdisciplinary landscape of research. Among these journals, 9 were directly related to the maritime sector, addressing topics such as naval engineering, maritime transport, port policy and management, as well as environmental impacts on the sector, with prominent publications such as the Journal of Marine Science and Engineering, Frontiers in Marine Science, and Maritime Policy & Management. The remaining 15 journals belonged to various fields of knowledge, including logistics and transport (Transportation Research Part E), applied sciences (Applied Sciences), electrical engineering and intelligent transport systems (IEEE Potentials and IET Intelligent Transport Systems), as well as sustainability (Sustainability), operations research (International Transactions in Operational Research), and sensor technology (Sensors and Materials). This diversity highlighted the multidisciplinary nature of AI research applied to the port sector, demonstrating how the maritime field intersects with other areas of knowledge to develop innovative and sustainable solutions.

Table 2 Journals with relevant publication.

Tablica 2. Časopisi s relevantnim publikacijama

Regarding the methodological approach, empirical studies predominated, representing 63% of the sample, with 19 articles adopting this methodology, suggesting a strong trend towards the practical validation of AI applications in the port context (Appendix).

In the analysis of the keywords, a total of 122 occurrences were observed, of which 84 were distinct. The most frequent keywords were: AI (13 occurrences), ML (8 occurrences), Port Operations (5 occurrences), and Port Management (4 occurrences). The remaining keywords showed a more scattered distribution, with one or two occurrences each, reflecting the diversity of topics covered within the context of artificial intelligence applied to port management and operations. This pattern suggested that, although the main themes were centred around AI and its applications in the port sector, a wide range of subfields were being explored within the research.

Graph 2 Authors’ geographical origin

Grafikon 2. Zemljopisno podrijetlo autora

The analysis of the authors’ geographical origin revealed contributions from 23 different countries. As illustrated in Graph 2, China stood out with 6 publications, followed by Belgium, Canada, Chile, India, Greece, Poland, and Portugal, each contributing 2 publications. This geographical overview highlighted a significant concentration of research in Asia and Europe, particularly in China, South Korea, and India. The presence of countries from America and Africa, such as Canada, Chile, Egypt, and Morocco, demonstrated the global reach and international relevance of AI research in the port sector.

4.2. Classification and Prevalence of Artificial Intelligence Methods / Klasifikacija i zastupljenost metoda umjetne inteligencije

An analysis of the 30 selected studies revealed that ML is the most widely adopted AI technique in port-related applications, appearing in 57% of the reviewed articles. Within this category, supervised learning methods such as Artificial Neural Networks (ANN), Support Vector Machines (SVM), and decision trees are the most commonly applied.

ANN, Deep Learning (DL) methods, particularly Convolutional Neural Networks (CNN) and Long Short-Term Memory networks (LSTM), appear in 40% of the literature, primarily supporting visual recognition and time series forecasting.

Reinforcement Learning (RL) is used in 13% of the studies, especially in contexts requiring sequential decision-making, such as quay crane scheduling and vessel traffic coordination in constrained environments. Computer Vision (CV) techniques are reported in 17% of the publications, frequently in combination with DL models for real-time object detection and automated monitoring.

Less frequent, but still relevant, approaches include Metaheuristics (6%), such as Genetic Algorithms (GAs) and particle Swarm Optimisation (SO), typically applied in scheduling and optimisation problems. Predictive AI Systems and Decision Support Systems (AI based) (20%), focused on forecasting ETA, failures, and energy consumption. Digital Twin technologies (6%), employed for simulation and visualisation of port operations. Blockchain integrated with AI (6%), used for ensuring transparency and traceability in logistics. Natural Language Processing (NLP) (3%), mentioned in system integration and data interpretation contexts. Table 3 presents the classification and prevalence of AI methods in the reviewed studies.

Table 3 Classification and Prevalence of Artificial Intelligence Methods in Reviewed Studies

Tablica 3. Klasifikacija i zastupljenost metoda umjetne inteligencije u analiziranim studijama

When we relate the types of AI to port areas, the analysis of the articles reveals a broad but relatively concentrated application of AI technologies in port operations, with a particular focus on several key management areas. The most frequently addressed domains include berth allocation and scheduling, container handling and yard operations, maritime and landside traffic management, predictive maintenance, and environmental and energy monitoring.

In berth allocation and quay crane scheduling, the predominant techniques are supervised ML models [8], [18], [11], [22], [2], [30], [17], [7], especially ANN, regression models, decision trees, and ensemble methods such as Random Forest and eXtreme Gradient Boosting (XGBoost). These are employed for tasks such as predicting vessel ETAs, estimating waiting times, and optimizing resource allocation, contributing to improved operational flow and reduced congestion.

Container handling at the quay and in the yard is another area that significantly benefits from deep learning solutions, particularly CNN and LSTM [2], [17], [1], [4]. These techniques are applied to image recognition, anomaly detection, and forecasting of logistic flows. The integration of AI with CV, AIoT, and autonomous systems such as automated Guided Vehicles (AGVs) and drones has enabled greater automation of internal logistics processes, enhancing both safety and productivity [7], [20], [22].

In maritime and landside traffic management, including channel navigation, vessel entry and exit, and lorry movement within terminals, there is a growing application of reinforcement learning algorithms [5], [23], [28]. Models such as Q-Learning and Deep Q-Networks are used to generate optimal, real-time decisions under operational constraints and dynamic environmental conditions.

Predictive maintenance and energy management are supported by forecasting models based on ML and digital twins [11], [8], [28]. These technologies facilitate failure prediction, more efficient resource usage, and better planning of interventions. Simultaneously, LSTM and regression-based models are applied in forecasting energy consumption and greenhouse gas emissions, contributing to the environmental sustainability of port activities [10], [27], [31], [32].

Finally, environmental monitoring [25] and operational safety [2], [7], [10] have benefited from the combined use of CV, IoT sensors, and DL, allowing for real-time control of air and water quality, as well as risk indicators. In some cases, these applications are complemented by NLP techniques [22] for automated data extraction and the integration of administrative systems.

In summary, the type of AI adopted is closely related to the operational context and the nature of the decision being supported. While supervised ML techniques and neural networks are widely applied in predictive and classification tasks, reinforcement learning methods are more commonly found in sequential decision-making and dynamic control problems. The current applications are mostly concentrated in operational areas such as berth scheduling, container handling, and equipment maintenance. However, there remains significant potential for expansion into strategic planning, integrated supply chain management, and large-scale environmental performance analysis.

4.3. Operational Impact and Benefits / Operativni utjecaj i prednosti

The analysis of AI-related benefits and contributions underscored the growing breadth and diversity of AI applications within port operations and management. The contributions observed vary according to the type of AI employed, demonstrating that these technologies address a range of logistical, operational and strategic challenges. Overall, the application of AI has shown benefits in terms of efficiency, predictability, sustainability and safety.

Among the most frequently used technologies, ML stands out as the most widely adopted. This prevalence is largely due to its versatility and adaptability to different operational problems, such as ETA [8], [16], [18], [25], vessel turnaround times [3], [17], resource allocation and real-time decision support [10], [22]. ML has enabled more efficient management of maritime traffic and container flows, leading to shorter waiting times, improved operational flow and optimised use of port infrastructure.

Within the ML domain, ANNs have proven particularly effective in modelling complex, non-linear relationships. They have been successfully applied to accident prediction [4], [5], [33], crane allocation and equipment control [18], [29], supporting more robust and adaptive operational decisions. In parallel, DL techniques – especially CNNs and LSTMs – appears particularly in applications involving computer vision [20], [23] and time series forecasting [8]. These models have enabled automatic recognition of containers, licence plate reading, visual anomaly detection and pattern prediction, thereby facilitating process automation and reducing human error.

Reinforcement Learning (RL) plays a significant role in sequential decision-making scenarios, such as quay crane scheduling, one-way channel traffic management and dynamic resource allocation under uncertainty [5], [23]. RL-based models have demonstrated the ability to adapt in dynamic environments, offering more resilient and efficient solutions compared to traditional deterministic or heuristic methods.

CV technologies are often used in conjunction with DL models. They support automated surveillance, obstacle detection and visual inspection of containers, significantly improving operational safety and asset tracking [20], [21]. Metaheuristics remain an effective approach for solving combinatorial optimisation problems, such as berth allocation, vehicle fleet scheduling and task sequencing in terminals [18], [29].

Other relevant technologies include Systems and Decision Support Systems (AI based), which support fault prediction in critical equipment and improve maintenance planning [32], [34]. Digital twins are emerging as key tools for real-time simulation and monitoring, enabling proactive management through dynamic data and virtual representations of port operations [11].

Though less frequent, the integration of blockchain with AI contributes to enhancing traceability and transparency in port logistics chains, particularly in contexts that require interoperability among multiple stakeholders [10], [24]. Lastly, NLP has been applied to automatic data extraction and system integration tasks, assisting the digitalisation of administrative processes and communication across heterogeneous systems [22].

In summary, the review highlights that those different types of AI provide distinct and complementary benefits. Their application is contributing to the transformation of traditional ports into intelligent, resilient and sustainable infrastructures, effectively meeting current demands for operational efficiency, safety, emissions reduction and digital integration. This technological shift points towards a port future increasingly driven by data and automated decision-making, with AI playing a central role in supporting competitiveness and sustainability across the maritime-port sector.

From a comparative perspective, the reviewed studies indicate that Machine Learning (ML) techniques currently demonstrate the highest level of technological readiness and empirical validation for practical implementation in port environments, particularly in applications such as ETA prediction, vessel scheduling, predictive maintenance and energy management. These methods are supported by a substantial number of real-world case studies and operational datasets, reflecting their maturity and reliability. Reinforcement Learning (RL), while showing strong potential in dynamic optimisation problems such as quay crane scheduling and vessel traffic management, remains largely confined to simulation-based or pilot-scale implementations, indicating a lower level of immediate operational readiness. Digital Twin technologies, although still emerging, exhibit growing empirical validation in safety management and decision-support contexts, particularly in hazardous cargo handling and emergency response; however, their large-scale deployment is still constrained by high infrastructure and integration costs. Overall, ML-based systems appear to be the most mature and deployment-ready category, whereas RL and Digital Twins represent high-potential technologies that require further large-scale empirical validation for widespread operational adoption.

4.4. Key Barriers to the Adoption of Artificial Intelligence in Port Management and Operations / Ključne prepreke za usvajanje umjetne inteligencije u lučkom upravljanju i poslovanju

Regarding the challenges associated with the implementation of AI in port operations and management, it is possible to identify a number of recurring and critical issues that range from technical limitations to organisational and legal barriers. These challenges have been systematised and classified according to the frequency with which they appear in the reviewed literature and their relevance to the effective application of AI in this context.

The most frequently cited challenge concerns the quality, availability, and integration of data. Several studies [8], [18], [19], [25], [29] pointed out that the absence of reliable, up-to-date, and structured data significantly limits the ability of AI algorithms – particularly those based on ML and deep learning – to generate useful outcomes. Issues such as noisy data, lack of standardisation across terminals, fragmented databases, and the absence of real-time sensors hinder not only the training of models but also their validation and deployment in real-world environments.

Another widely reported obstacle relates to insufficient technological infrastructure [4], [10], [20], [22], [24]. Many ports, particularly smaller ones or those located in less developed regions, lack the connectivity, storage capacity, sensors, or computing platforms required to support AI-based solutions. This limitation is often compounded by the high investment costs associated with upgrading equipment, adopting solutions such as digital twins or computer vision, and maintaining up-to-date and secure AI systems.

The computational complexity of the models themselves also presents a significant barrier [1], [5], [8], [26], [34]. Numerous articles highlight that hybrid algorithms, real-time simulations, and advanced optimisation methods not only require substantial processing power but also demand highly qualified human resources to operate, fine-tune, and interpret the systems. The shortage of professionals with expertise in data science, AI, and systems engineering is, therefore, a transversal challenge across most ports [22], [23], [30].

Integration with existing port systems, many of which are legacy or outdated, constitutes another frequently noted impediment [4], [23], [24]. AI solutions are rarely immediately compatible with the operational platforms already in use, necessitating either the adaptation or replacement of such systems – a time-consuming, costly process that also involves operational risks.

From an organisational perspective, there is a notable resistance to change and a lack of digital culture among many port stakeholders [22], [23]. The adoption of AI requires not only technological innovation but also the reconfiguration of processes, team training, and strategic alignment between operators, authorities, and service providers. This issue is further aggravated by the limited technical knowledge among port managers, which hinders informed decision-making regarding technological investments and adoption.

Other challenges identified include: difficulties in the economic and operational validation of solutions [4], [11], legal uncertainty and a lack of clear regulation (particularly concerning data sharing, system interoperability, and cybersecurity), as well as ethical and social concerns related to privacy and workforce automation [10], [20], [22], [32], [35]. Additionally, the high variability of port environments makes it difficult to generalise models and scale solutions across different terminals [5], [25].

Another important challenge is related to data governance and interoperability are increasingly being addressed at an international policy level. Initiatives promoted by the IMO, notably the Maritime Single Window framework established under the Facilitation (FAL) Convention, aim to standardise and harmonise data exchange among maritime and port stakeholders, thereby supporting interoperability and the integration of AI-enabled systems [36]. Similarly, European Union policy initiatives on port and transport digitalisation – including the European Maritime Single Window environment and the Digital Transport and Logistics Forum – emphasise standardized data architectures, secure data sharing and interoperability as prerequisites for advanced digital and AI-based solutions in ports [37], [38]. These frameworks indicate that the future adoption of AI in port environments will depend not only on technological maturity but also on regulatory alignment and robust data-governance structures.

In summary, the implementation of AI in ports faces a combination of structural, technical, economic, and human challenges. Overcoming these barriers requires a coordinated and multidisciplinary approach, involving technology, training, regulation, and a clear organisational strategy, so that ports may effectively and sustainably harness the transformative potential of artificial intelligence.