日本財団 図書館


Top > 産業 > 運輸.交通 > 成果物情報

Conference Proceedings Vol. I, II, III

 事業名 海事シミュレーションと船舶操縦に関する国際会議の開催
 団体名 日本船舶海洋工学会 注目度注目度5

By Dr. I. W. Dand and Dr. R. D. Colwill,
British Maritime Technology Limited
 Abstract: This paper describes the dynamic marine traffic simulation model Dymitri. It has been developed for use in marine risk studies in areas where traffic densities are high, principally in port development studies. After discussing the model, data collection and post processing for estimating risk, matters related to the modelling of decision-making for avoidance manoeuvres are discussed. This is at the heart of the dynamic model and the ways in which avoidance decisions are implemented are outlined by means of examples.
 The simulation of marine traffic flows is not new. Indeed the operational research aspects of the topic have their origins in the second World War. However, with marine risk assuming an increasingly important role in the design and operation of ports and waterways, the simulation of traffic flows provides a valuable capability for the engineer, port operator or legislative authority seeking to enhance safety and efficiency, in increasingly busy and congested areas.
 This paper describes a model which goes a little further than some of its contemporaries in simulating the manoeuvring behaviour of ships within a traffic model. This of itself is of some interest, but, as is often the case, the development of such a model has raised a number of intriguing questions in addition to providing some answers. These questions relate, almost inevitably, to questions of control in encounter situations, and highlight the multiplicity of decisions that mariners must make when in close proximity to other ships.
 Some of these matters are explored after an introduction to marine traffic simulation and the "Dymitri" dynamic model, developed by British Maritime Technology (BMT).
 As far as is known, marine traffic simulation has its roots in Operational Research studies carried out in World War II (Reference 1). Motivation for them was a need to determine the effectiveness of various countermeasures and this was achieved by measuring (and then estimating) the surface density of combatant vessels. The next major motivating force, which probably led to the development of present-day marine traffic studies, occurred some 30 years later after a spate of accidents and sinkings in busy coastal waters in Europe, and congestion in the busy port areas in Japan (References 2 and 3). This led to measures being taken to determine the number of ships in a given area at any one time and the routes they followed. In turn this prompted the development of traffic separation schemes (TSS), navigation information systems and a more focussed and widespread use of Vessel Traffic Services (VTS).
 Traffic modelling at this time was largely confined to the use of algebraic, rather than numeric, formulations, but in spite of this, great progress was made. The development of computer processing speeds in the meantime led to the development of more sophisticated traffic models working in the time domain and it was not long before these were providing valuable operational information for busy ports and waterways.
 These models used the well-established concepts of encounters and domains, developed in earlier traffic models, and provided the very valuable ability to explore "what if?" scenarios. By this means it was possible to explore the effect of increases in future traffic levels, ship size and speeds, or changes to VTS and port operating procedures. It became clear that the key measure of the success or otherwise of such operational changes lay in changes to the marine risk environment, and it was to the development of a representative and accurate measure of this entity that traffic modelling was directed.
 Marine risk, of course has two main components - frequency and consequence - the overall measure of risk comprising a combination of the two. Marine traffic modelling provides valuable information on the frequency component which, by the nature of such models, is restricted to the frequency of collision, striking or, in some situations, grounding. Other forms of accident such as fire, explosion or foundering, are not covered. Although traffic models provided among other things, estimates of the frequency of encounter, there was still some ground to be made up before this information could be used directly for the estimation of risk. This involved:
・Determination of the link between frequency of encounter and frequency of collision - a prime requisite to use the models in a quantitative manner, and
・Provision of realistic models of ship behaviour
 The former proved elusive, often due to lack of sufficient historical accident data, but for some busy ports a definitive link has been made and this will be discussed further in the later stages of this paper.
 The latter point related to the assumption, made in some computational models, that ships moved from waypoint to waypoint in a linear fashion, changing track heading instantly in concert with the changes in track. Furthermore each ship in the population was assumed to maintain a steady speed (or a speed from a fixed speed profile) throughout its run with no attempt being made to avoid other ships. An encounter was usually defined when one ship entered the domain of another in such a way that one or other would have had to change course. This took no account of the manoeuvrability of either ship in an encounter, or whether one or more ships would slow down to avoid collision.
 It became clear therefore that time domain marine traffic simulation needed some input from the world of ship manoeuvring simulation. This in conjunction with some simple on-board "intelligence" would, it was reasoned, provide a more realistic, dynamic, traffic simulation model.
 The Dymitri model is data-driven so it is of vital importance to collect accurate present and future traffic data at the outset. Without this, the dynamic simulation model cannot run so, before describing the model, the important aspect of data collection is discussed.
 Collection of the necessary data can be achieved by careful attention to the design of the survey campaign and by mixing various survey, techniques to ensure maximum coverage of temporal and spatial data. Vessel movement histories (timing), paths (courses) and types must all be identified to collect a statistically representative sample of activity. The availability of this data from different sources is illustrated in Table 1:
Table 1 Available Survey Data Sources
Data Source History Path Type
4   4
Operator vessel
4   4
Capture from radars 4 4 4
Visual surveys 4   4
 While each data set provides useful information on vessel activity no one type fully covers all three aspects of history, path and type. In general survey data must be collated, compared and blended to form the necessary datasets for development of the dynamic marine traffic simulation model. These datasets cover the following key areas:
・Route structure;
・Vessel classes and types;
・Speed profiles, and
・Numbers of vessels on each route throughout the day.
 The accurate assignment of vessels within their correct routes is the single most important aspect of data collection as it impacts the traffic density and position of vessels with respect to each other developed in the final simulation. The use of radar data is key in this respect and BMT have deployed a series of video frame grabbing and vessel tracking techniques to capture and isolate the path of vessels as they move across a radar screen. (See Figure 1).
Figure 1 Traffic Routes from Radar
 The advent of digital radars which record the position of all vessel movements within their surveillance area, and provide more accurate information on vessel size, greatly assists the development of traffic input data for the simulation model. Route structures featuring vessels with similar class, speed and destination may be extracted rapidly from the digital output with the use of ship type assignment algorithms. Visual surveys, conducted at a series of gates, may then be used to "ground truth" the remotely sensed traffic regime to ensure its accuracy.
 The digital radar data set may then be processed to identify the key characteristics identified above, and output developed in a format directly importable by the traffic model. Such automated acquisition, attribution and compilation of data is of particular importance in busy port areas where vessel movements may easily exceed 10,000 individual records per day.
 Having outlined the method by which data is collected, we now discuss the dynamic model itself.
 Once every ship in a traffic simulation model is given the ability to manoeuvre, the complexity of the modelling significantly increases. It is necessary not only to model the manoeuvring behaviour of the ship with sufficient realism, but also to provide some representation of the way mariners may be expected to behave in an encounter situation. While the former may be reasonably straightforward, the latter is far from easy and it is fair to say that some approximation is needed. Ideally a complete "mariner" model, of the type discussed in earlier MARSIM Conferences should be used, but this was not thought feasible for Dymitri because:
・It is not at all clear whether a sufficiently comprehensive model of this type exists, or, indeed, can ever exist
・In order for Dymitri to perform its function, the mariner model which chooses when and how much to turn, and whether to slow down or speed up, must work extremely rapidly in computational terms. This required a simple model, a need which drove much of the Dymitri development.
 In essence, therefore, Dymitri is a time-stepping model in which a number of vessels in the play area move along prescribed routes and manoeuvre according to a set of rules. The rules determine when and how to change course and when to slow down or increase speed, thereby providing rudder and engine commands to each active vessel. These are then used by the manoeuvring model to alter course or speed before, during and after an encounter.
 Such actions are triggered by computing the Closest Point of Approach (CPA) which is monitored continuously for all active ships. If the CPA drops below a prescribed minimum value, avoiding action is initiated. To avoid ships in the far distance needlessly taking avoiding action with nearby ships because the (distant) CPA is less than the minimum, CPA checks are not initiated unless both ships are within an area whose size is prescribed in the initial input data.
 Sterring and turning is accomplished by a simple two-term Nomoto model, combined with a linear acceleration/deceleration model. Although the steering model is fairly crude, it is extremely fast in computational terms, a necessary feature when over 1,000 active vessels may be in the play area at any one time. To date it has served its purpose well.
 The Steering Rules from the International Rules for the Prevention of Collisions at Sea (the ColRegs) are involved for many avoidance manoeuvres, although these may be superseded by "local rules" in congested waters. In such cases it will be more appropriate for one vessel to keep out of the way of the other by virtue of the relative sizes of the other and regardless of whether it is the "give-way vessel or not. For example, a small motor launch or ferry will keep out of the way of a large cruise liner in harbour waters, even if it is, strictly speaking, the stand-on vessel.
 Such considerations depend largely on the size and type of each vessel and are represented in Dymitri by a "give-way" or decision matrix, the default option of which is to use the ColRegs.
 Berthing, unberthing and the associated waiting when no berth is available are all modelled as in other traffic simulations. There is no need for any decisions regarding avoidance (other than by passing craft) to be made in this aspect of the simulation so it will not be discussed further.
 Two types of ferry operations are modelled. One is where the route is a long-haul route and ferries may pass in or out of the study area. All ferries in such a model behave according to the decision rules regarding encounters and avoidance manoeuvres. The only difference between them and other vessels is that they operate to a regular schedule which may, in some cases, remain fixed throughout the study period.
 The other type of ferry operation relates to a local, perhaps river, operation in which each ferry visits a number of stops along a route which lies entirely within the study area and starts and finishes at the same terminus. Such a feature causes the ferries, by avoiding other traffic, to incur delays, thereby demonstrating the effect of traffic density on such a route. It also has high-lighted another decision-related issue. If a ferry is approaching a stop and has to make an avoidance manoeuvre, how best should such a manoeuvre be approached? It is a situation seen in short cross-harbour routes across the world where solutions are found by slowing down, altering course or delaying departure/arrival. To incorporate such decisions into a model is complex and in Dymitri a simplified approach is adopted. The ferry carries out an avoidance and/or slowing manoeuvre as required. If this brings it within the defined "capture area of the berth, a berthing routine takes over and the ferry is berthed. No deliberate delays on departure are modelled.
3.7 Manual Operation
 The discussion above has concentrated on the automatic operation of the Dymitri model. This is its normal mode of operation and takes full advantage of the speed of the host computer.
 However, on some occasions it is of value to be able to "steer" a designated ship "manually" through the traffic and this capability has been incorporated in the model. When using this option, the program speed reduces to a chosen multiple of real time and a rudder bar appears on the screen. The next ship generated on the chosen route is de-coupled from the automatic pilot in the simulation and may be steered from the screen using the rudder bar. Other vessels in the traffic stream behave as before and take avoiding action.
 When the steered ship leaves the study area the program reverts to its free-running speed and the run continues.

サイトに関するご意見・ご質問・お問合せ   サイトマップ   個人情報保護




更新日: 2020年9月19日


1.MARSIM'03 開催報告
3.九州における離島住民からみた交通バリアフリー化に関する調査研究?鹿児島県をモデルケースとして? 報告書
  [ 同じカテゴリの成果物 ]








ご質問は こちら から