Agent Based Risk Management & Operational Modelling of Ports

R. D. Colwill and S. L. Yeung

The continuous development of world-wide shipping, both in terms of volume and individual ship size places increasing pressure on navigation safety in port approaches and on landside infrastructure. Within this complex environment sits the pilots, operators, and drivers who are so crucial to the safe and efficient operation of port systems. The ability to identify and predict the distribution of risk and identify operational efficiency is a key component of a projects development. Increasing mechanical reliability ensures that human factors are now the most significant elements of risk within these systems, yet are frequently the most poorly replicated. This paper describes the analysis of port systems through an agent based simulation approach that allows the key issues of human decision making in response to the environment and perceived threats to be better represented, and provides a platform for more accurate planning. The paper reviews the key architecture of such a model and illustrates the value of 3D presentation and the capability of agent simulations to represent complex environments.

1. Introduction The worlds ports are the cornerstones to the development of the worlds trading economy, and have in recent years been under increasing pressure as more and larger vessels ply the oceans. Port stakeholders (harbour masters, terminal operators and shipping companies), all have a strategic interest in ensuring that navigational safety and port/ maritime security is enhanced to maintain adequate port capacity and safety in the face of rising volumes and ship size. Most frequently port authorities must address a version of the following question: What is the present capacity of the ports fairways and terrestrial infrastructure and how can we predict and manage the impacts from future growth. Focussing initially on the marine side of the port it is possible to identify the following factors that must be addressed within any assessment designed to
E-mail: rdc@bmtasia.com.hk; bey@bmtasia.com.hk

30

Agent Based Risk Management & Operational Modelling of Ports

31

accurately address the impact of future facilities and marine traffic: Geometry (draft, width & airdraft) of navigable channels Traffic Mix (size, speed, type of vessels & manoeuvrability) Metocean Environment (current, wind & wave) Control Regime (TSS, speed limits, VTS advisory) Operational Capability (Mariners familiarity and tolerance to vessel proximity)

These variables are not static; and key drivers for change include: Global & Regional Growth Port Development New Port Facilities Changes in Vessel Size

To accurately address capacity and safety a model must to able to model the physical geography of the water space, the volume, nature and capabilities of the vessels and the response of mariners to different control stimuli. 2. Agent Based Simulation In reviewing the requirements outlined above it is clear that an extremely flexible architecture is required to integrate the ships reaction to the physical environment in which they navigate and the human factors issues of control and reaction to the marine traffic environment. BMT have in recent years been researching just this issue in a bid to develop the next generation of traffic models that meld together marine traffic issues, with the increasingly significant human factors element to map congestion and risk. To achieve this goal BMT have entered into a partnership with Massive Software (www.massivesoftware.com) in order to focus their Academy Award winning autonomous agent animation software towards marine engineering applications. Massive is the premier 3D animation system for generating crowd-related visual effects for film and television. Animators are able to develop characters with a sophisticated set of reactions to their environment, where each agent can be programmed to develop reactive behaviour for the most complex actions. When scaled up into the hundreds or hundreds of thousands the interaction within groups of agents (such as people, orcs or penguins!) that emerges from these individuals is highly realistic. The same processing architecture that has been so successfully applied for film and video is equally applicable for marine applications. The key features of Massive of interest for marine congestion and safety assessment are: Open and scaleable logic structure for the creation of autonomous agent brains. Open script structure for constraint input and output data creation. Fuzzy logic programming to mimic mariner response to navigation.

32

R. D. Colwill and S. L. Yeung

Current and wind fields and lane guidance options. 3D graphic ability and sophisticated element and terrain modelling. 3. Marine Applications Key marine focussed behaviours have been developed within the Massive architecture building on the core logic of BMTs established Dynamic Marine Traffic simulation software, DYMTIRI.1 Principal features include: Near Field Collision Avoidance. Mid Field Grounding Avoidance. Far Field Track Following. An autonomous agent brain is assigned to each vessel navigating within the Study Area to allow it to make decisions according to a predefined rule set. In Massive each brain stores a collection of rule-sets for decision making and controlling the response of the agent. An agent executes a rule-set when criteria associated with the rule are satisfied. The building blocks of Massive brains include input, fuzzy, logical, defuzz and output nodes. When combined, these nodes evaluate the external environment, develop weighted outputs, and ultimately decide on a course of action for the agent. Massive includes a simple and self-explanatory Graphical User Interface for constructing rule-sets in a brain; nodes can be dragged and dropped within a window and rule-sets constructed by connecting nodes in the aforementioned order. A key value of developing decisions making within a fuzzy logic environment is that complex situations can create situations impossible to predict and account for in a deterministic manner. A series of fuzzy rule sets will result in the most true action being identified (as defined by the membership functions) that represents the agents most pressing need. This action will then be taken; in many cases within deterministic software implementations freezing of the programme would occur. Vessels within the model each transmit and receive data on their location; the distance, orientation and relative bearing between a ship agent and the target ship agent can be automatically recognised. A heirachy has been established to identify vessel types so that the correct application of the COLREG2 may be made. Figure 1 shows a simplified brain element of a ship agent for collision avoidance due to the starboard crossing of a target ship. It represents the simple case when a mariner becomes aware of an approaching ship and decides if his own ship is going to collide with this vessel. The decision options include slowing down his own vessel and/or steering to starboard to avoid collision. Once the manoeuvring action has been completed and the rule sets no longer identify that an action is required the vessel regains its course. In real-life, there is a variation in different peoples response to stimuli caused by perception of external factors, degree of response and so on. These can be implemented in agent based simulations by adding variation to membership

Agent Based Risk Management & Operational Modelling of Ports

33

Fig. 1. Brain function for starboard avoiding action.

Fig. 2. Membership functions for direction perception.

functions. It also provides a smooth transition from one action to another one, a characteristic of human decision making. Figure 2 shows the membership function for defining port, ahead and starboard where the trueness of a vessels position relative to the observer is input into the brain. It is feasible to develop extremely complex behaviour by the bundling of rule sets; and the Massive architecture allows the independent testing of key rule sets and their later consolidation. An example of an early development of the mariner brain is illustrated in Fig. 3. Of course the key to any model is its accuracy and in order to develop and calibrate the model it is necessary to identify, in quantitative terms, the concepts for near and far etc, as perceived by the mariners. To achieve this validation

34

R. D. Colwill and S. L. Yeung

Fig. 3.

Basic COLREGS implementation.

Fig. 4. Constricted marine traffic environment.

has been conducted against a number of digital radar data sets of port activity providing quantitative support to the logical reasoning that underpins the mariners response behaviour. Figure 4 illustrates an application of the model where barge movements within a constrained river channel are assessed to review the risk associated with the addition of more berths. The output flexibility available in Massive allows the user to develop specific reports. In this case the frequency of vessel conflicts and collision potential was output against berth utilisation.

Agent Based Risk Management & Operational Modelling of Ports

35

Fig. 5. Example of junction analysis.

4. Terrestrial Applications While BMTs focus has been on marine applications it is straightforward to apply the same techniques to any problem where agent (people or vehicle) interaction is a significant issue; within ports such problems exist at port gates and within the stacking yard. Figure 5 illustrates the models application to a complex junction. While many existing models readily address junction capacity, this particular example featured the assessment of the impact of large-scale roadworks, and the model illustrated the impact of the works on traffic flows and system capacity, and of particular significance any knock-on impacts. 5. Conclusion The continuing development of the worlds ports creates increasing capacity constraints on existing infrastructure. The ability to plan and optimise developments is essential, and recognising that the logistics systems will have significant man-machine interfaces is critical for future success. Autonomous agent models provide the ability to represent complex marine and terrestrial systems and give port stakeholders the tools to enhance the safe and efficient operation of port systems. References
1. Dand I. W., Colwill R. D. (2003), Simulation of Traffic Flows using Dynamic Ship Modelling Proceedings, International Conference on Marine Simulation and Ship Manoeuvrability, Kanazawa, Japan. 2. International Maritime Organisation (1972), Convention on the International Regulations for Preventing Collisions at Sea (COLREGs).