当前位置:首页 >> 工学 >>

Case-Based Snow Clearance Directive Support System for Novice Directors


Case-Based Snow Clearance Directive Support System for Novice Directors
Yoshinori Ikeda1 and Yoshio Nakatani2
1 Graduate School of Science and Engineering, Ritsumeikan University, Kusatsu, Shiga, Japan rs002025@se.ritsumei.ac.jp 2 College of Information Science and Engineering, Ritsumeikan University, Kusatsu, Shiga, Japan nakatani@is.ritsumei.ac.jp

Abstract. Experts have issued directives for snow clearance in snowy regions.These directives are implemented in accordance with the experience of the director in charge, and snowplow drivers are reliant on directors to carry out their tasks efficiently. However, many experienced directors will be retiring when they reach retirement age in the next few years and their successors will be faced with difficult circumstances because of a lack of training and experience. In order to provide support to such neophyte directors, we propose a case-based directive support system, which reuses past directive cases. Each such case stores data on weather, snow, number of snowplows available, the time taken to complete a snow clearance task and an evaluation of the tasks carried out by the director. When a neophyte director specifies current task conditions, the system searches and presents cases similar to the current conditions. When the director selects the most similar case, the system estimates the time required to complete the snow clearance task, based on the snow and road conditions. The director can also manage the progress of snow clearance tasks with the system. Keywords: case-based reasoning, artificial intelligence, snow clearance.

1 Introduction
Every winter, 60% of Japan’s cities are covered in snow and snow removal becomes central to the maintenance of everyday life. Many studies have attempted to solve these problems by, for example, implementing efficient snow removal techniques or road heating systems, however, human experts have directed these tasks. Until recently, directing these tasks has been unproblematic but now many experienced directors are approaching mandatory retirement age. Their successors will be faced with difficult, unfamiliar situations but will be required to maintain quality standards, despite a lack of training. New directors cannot permit service levels to decline, whatever the weather conditions, because citizens have high expectations of them and their ability to provide quality snow clearance services. In order to provide support to neophyte directors, we propose a case-based directive support system that reuses past cases to derive a new snowplow deployment plan.
M.J. Smith, G. Salvendy (Eds.): Human Interface, Part II, HCII 2007, LNCS 4558, pp. 893–902, 2007. ? Springer-Verlag Berlin Heidelberg 2007

894

Y. Ikeda and Y. Nakatani

The system’s database records past data on the amount of snow; the task area; the number of snowplows required and available; time and distance; and evaluation of the operations. When a novice director uses the system, it automatically searches for similar past cases and adjusts the available data for the current snowfall situation. The system then suggests a solution in the form of a snow clearance plan and provides the director with visual feedback of snow removal progress. We have developed a prototype of the system for the Windows PC platform. At the end of this paper we have enumerated advice received from experts who have evaluated the system and our intended future work on the system.

2 Tasks and Their problems
We interviewed Sapporo city government officials to analyze their current situation regarding snow removal tasks and to determine the system’s specifications. The following is a summary of the analysis performed. (a) Sapporo city has 39 snowplow centers and 10 construction technology centers that manage snow removal centers. Snowplow directors control the snow removal centers and manage the snow clearance centers, snow removal planning, management of snow removal contractors and claim processing. Snowplow directors and snowplow drivers belong to the same private-sector construction companies and all snow removal operations (snow observation, plowing, instructions, among others) are outsourced to them. City government officers merely organize the budget because the city government officers are reassigned every four years, making it difficult for them to accumulate sufficient operational experience. (b) Snow removal tasks commence when more than 10 cm of snow falls and operations take place at night to avert traffic congestion. (c) Snowplows are slow moving vehicles so work must be completed outside of peak traffic times (7:00 to 8:30 am and 5:30 to 7:00 pm.) (d) There are many weather forecasting sensors around Sapporo that provide prediction data about estimated snow depth supplemented by information from the Meteorological Agency. This provides predictions such as there is a 30% chance of a 10 cm snowfall in the next three hours. We have listed four snow removal systems as shown in the organizational chart in Fig 1. Many local governments seem to use similar systems, but without weather forecasting sensors that are unique to Sapporo because it is the largest city in Japan that experiences heavy snowfall. In our research, we could not find other municipalities with a better system. It is impossible for city officers to perform snow removal tasks so they have no choice but to outsource the work to construction companies. When snow falls: (1) the director drafts a snow removal plan; (2) snowplow drivers are alerted; (3) dispatched device; and (4) if necessary, the director repeatedly updates the snow removal plan until the task has been completed. Basically, snow removal is deficit work for construction companies. Companies previously covered their deficits by performing public construction in summer but public construction projects have recently declined

Case-Based Snow Clearance Directive Support System for Novice Directors

895

Fig. 1. Snow removal organizational chart in Sapporo

in number due to a shortage of local government funding. This has mitigated the bargaining power of local government with construction companies because the threat of withholding public construction orders is no longer effective and a number of construction companies have started rejecting snow removal work. Therefore, to maintain service levels, it seems that construction companies will have to work more efficiently, despite the problems of a rapidly aging and shrinking industry. According to reports from the Cabinet Office, more than 80% of construction companies have less than 10 employees. Additionally, their employee turnover ratio is higher than in any other industry. This shows that their business bases are very weak and it is difficult to train new employees to replace old, experienced directors. The current experts are of the baby boomer generation and they are facing mandatory retirement within the next three years. In general, construction companies are facing some difficult maneuvering in the near future, with regard to organizational reformation. During research interviews, we discovered some of the issues that the people involved in snow removal face. a) Construction companies find it onerous to keep large fleets of snowplows so they have to efficiently use a limited number of snowplows. b) Many expert directors will retire at the official retirement age limit within a few years and their expertise must be passed on to their successors. It is likely that snow removal officers will be changed every year because they are selected by means of a bidding process. There is a possibility that each new director’s management tactics will differ from those of his/her predecessor. c) Directors use wireless communications to communicate with snowplow drivers and gathering snowfall data. They also have to manage task progress by identifying the current locations of snowplows via GPS. For these reasons, whenever there is a heavy snowfall, the director is overwhelmed by a flood of calls. For the issues raised in a) above, it appears impractical to persist with the previous tactic of maintaining a large fleet of snowplows in the face of a declining construction industry. Therefore, better performance with fewer snowplows and drivers is required and several studies have been conducted on this approach. Potential solutions

896

Y. Ikeda and Y. Nakatani

suggested in these studies are, for example, one-operator snowplows (current snowplows require two operators) and retaining lane assist systems. For the issues raised in b) above, every industry is subject to the problems of aging and generation shift, especially the construction industry. It has been speculated that snowplow drivers are comparatively young, whereas directors are close to retirement age so there will be a major change in directors in the near future. Every effort should be made to minimize the damage caused by many directors retiring in a short space of time, even if the construction industry’s turnover ratio is high. These problems are exacerbated by a decline in the number of workers in the industry and growth in the areas requiring snow removal services. When the city experiences a heavy snowfall and all snow clearance tasks cannot be completed before rush hour, highways have to be prioritized for clearance work, bearing in mind that school routes have to be attended to as well. In short, the work must be done effectively under time pressure. Chaos may ensue if an inexperienced director is faced with the complex task of switching plowing areas or requesting backup from other areas. Moreover, it is foreseeable that snow removal decisions will change every year because officers, being subject to appointment by bidding, will change annually. In addition, construction workers are employed seasonally, seldom remaining with the same company for a long period of time. This affects efficiency because snowplow drivers do not have time to become acquainted with any particular area. Despite these problems, snow clearance work must be efficiently and quickly carried out. For the issues raised in c) above, directors must deal with a communications overload during periods of heavy snowfall and information gathering is at risk of breaking down. There are no GPS locating systems or cameras fitted to the snowplows in Sapporo, so headquarters is obliged to communicate with snowplow drivers to ascertain the actual conditions on the ground. All communications are voice-based. The city hopes to motivate workers (directors and snowplow drivers) to dedicate themselves to their tasks by fitting GPSes and several other devices to snowplows in the near future, which automatically report snowplow conditions and locations. We have identified three problems to solve, apart from the issues raised in c) for which many studies have suggested potential solutions. However, there have been few studies aimed at helping directors. We have focused especially on the issues raised in b) and propose a fast generation-shift methodology by using case-based reasoning. We believe that it is effective to provide neophyte directors with a suggested plan of action that is based on references to past, similar cases. Consequently, we have developed a prototype suggestion system that has a database of past snow clearing operations. A neophyte director faced with fresh snowfall can consult the system for guidance.

3 Solution: Case-Based Reasoning
Case-based reasoning (CBR) is a process of solving new problems based on the solutions applied to similar, past problems. In the early 1980s a lawyer, Roger Schank, and his students at Yale University developed a CBR system to develop trial strategy from legal precedents and judge-made law.

Case-Based Snow Clearance Directive Support System for Novice Directors

897

Case-based reasoning has been formalized, for purposes of computer reasoning, as a four-step process: 1. retrieve; 2. reuse; 3. revise; and 4. retain (Fig 2.) Therefore, CBR has flexible, analogy-searching features. If no existing case perfectly matches a new problem, CBR is capable of making suggestions to the user. Other artificial intelligence methods, for example, rule-based reasoning and model-based reasoning, are less flexible because these are based on confirmed rules and exact matching with past cases. CBR stores past and current solutions in a database, together with the solutions and processes employed, as resources for solving similar problems in the future. CBR improves in accuracy as each new scenario and solution is stored in the database, making it easier to provide new suggestion patterns based on real cases.

Fig. 2. Model of Case-Based Reasoning

Rule-based reasoning makes it difficult to retrieve similar cases when faced with an unknown situation. To address this problem, we have to use another reasoning method. The problem to be solved is to suggest a snow removal plan to a novice director by using past cases. Given this scenario, it is difficult to implement rulebased reasoning because of its heuristic rules and lack of basic scientific and physical evidence. In these circumstances, we have tried to provide neophyte directors, who must manage snow clearing operations, with a system that applies case-based reasoning.

4 Reasoning Algorithm
In order to complete snow removal tasks before rush hour, a director should prepare a snowplow operating schedule in advance. Therefore, it appears that predicting snowplow working hours would be an effective strategy. There is enough time to prepare for snow removal, such as to call to snowplow drivers, because no work commences before 10 cm of snow have fallen. Sufficient time helps in formulating a flexible response to rapidly changing climatic conditions. Fig. 3 shows an outline of the system. By checking current snowfall against past snowfalls in the database, the system suggests that the director should advance the schedule to finish the task on time. Sapporo weather bulletins contain many data items but to simplify the system, we have used the amount of snow that has fallen; the task area; the numbers of

898

Y. Ikeda and Y. Nakatani

Fig. 3. System outline

snowplows available; the time/distance involved; and evaluation of the operations, as relevant items from past cases. The system rejects irrelevant cases from targeted information by performing evaluations and this will help to manage special circumstances that were hard to quantify at the time. The amount of accumulated snow is the primary factor in snow removal, so the system searches for similar accumulated amounts of snow cases first. Next, it searches for similar weather conditions and the number of snowplows available from the first search obtained and predicts likely solutions by comparing the search results with the current snowfall. The following shows the specific reasoning method. Chart 1 shows the format of the case database.
Table 1. Format of the Case Database Date: Year/Month/Day (3 types) Weather: Sunny/Cloudy/Snowy (3 types) Amount of snowfall: Accumulated snow depth snowplows started their task (Unit: cm) Time/distance: Time/distance for the task (Unit: minute) when

Evaluations of operations: Director evaluates his/her own work on a scale of 1 to 5 (bad) 1 - 5 (good) Comments: Director’s comments. E.g., special circumstances

Case-Based Snow Clearance Directive Support System for Novice Directors

899

Basically, time/distance is inversely related to the number of snowplows available. Therefore, the prediction of time/distance (T) is such that:

r = time/distance in the past case ( t ) × number of snowplows available ( c )
prediction of time/distance ( T ) =

(1) (2)

r
Numbers of snowplows ( w )

First, the system searches its database for similar weather conditions with snowfall within 5 cm of the current snowfall. We have used 5 cm for our own convenience and hope to narrow this number when the system goes into official use. The reason we have limited the search to the same weather conditions is to take into consideration that weather conditions may change during operations, and we have classified only three types of weather. In any event, the system will be capable of a flexible response when the database contains a sufficient number of cases. The matching of past scenarios to the present has to be based upon similarities and not precise matches because it is very rare for cases to match exactly. To accomplish this we have introduced a weighted reasoning element, k into the system as follows:

k=

Cs ? Cs ? Ps E × 5 Cs
(3)

k: Weight Cs: Current snow depth Ps: Snow depth in past case E: 5 step evaluation of the case

Consequently, the maximum value of k is 1. The closer the match, the closer K will approach 1. We obtained time/distance by using equations (1,) (3) and (4.)

∑ (k × t ? c ) 1 T= × w ∑k
i i i n i n

K: t: C: W: N:

Weight Time/distance in the past case Number of snowplows in the past case Available snowplows Numbers of cases used for reasoning

(4)

The assumption is that the accuracy of the reasoning element will be enhanced when the number of cases (n ) rise, ask + k + ....+ k approaches n . In these circumstances the system will cope with a record snowfall. The system will then use the most similar case (the heaviest case recorded) and retain the data. Even if the

900

Y. Ikeda and Y. Nakatani

number of stored cases further increases, the system’s accuracy being reduced by this increase, the system will still be sustainable as a reference system.

5 System Implementation Plans
We have derived plans for a snow removal reference system from the perspectives of time/distance,weather condition, and the number of snowplows available. In virtually all cases the amount of the snow falls over an entire region but icy roads. Therefore, the snow removal procedure remains virtually static and the system suggests a removal plan for highway and parallel roads. However, when snow falls shortly before rush hour and the clearing task has to be completed before traffic congestion builds up, directors have to prioritize highways for snow clearance operations. That means that it is important to make an early decision to exclude less frequented streets from the clearance plan. We decided to focus on highways when the system predicts that currently deployed resources are insufficient to complete clearing all areas in time for rush hour traffic. The snowplow echelon can later time crunch as the area remaining to be cleared begins to shrink.

6 Prototype and Evaluation: System Outline
In order to support neophyte directors, we developed a prototype system of the snow removal directive support method described in this article. The system automatically collects the relevant data for conditions, such as depth of snow accumulated and temperature. Next, the system searches for cases with conditions that are similar to the current conditions and selects the most similar case. Then, the system estimates the time required to complete the task. After the snow removal work starts, the system acquires the current locations of snowplows via GPS and plots task progress on the display to show how long it will take to complete the task. Fig. 4 shows a screen image of the prototype system.

Fig. 4. System screenshot

Case-Based Snow Clearance Directive Support System for Novice Directors

901

We relegate the final task decisions to the director to avoid the frame problem in artificial intelligence as the system cannot understand all comments in the database and if the system provides all the answers, the director loses the opportunity to hone his/her own skills. When a director orders a snow removal operation, the database retains this information and the weather data for the new case. Next, the database retains the time/distance involved, the director’s evaluation and the director’s comments. While the snowplows are operating, the snowplow locations, roads that have already been cleared and those that have not, are shown on the system map, to make assessment of the situation easier. In addition, the system displays roads that have not yet been plowed in a lighter color during highway operations.

7 Evaluation and Future Work
We conducted interviews with employees of the Sapporo city government snow removal section and asked them to evaluate the system. The results reveal the advantages and disadvantages of the system. (a) There has been little research on the issue of supporting neophyte directors. Casebased approaches to support directors are few in number and this system is expected to be helpful to them. (b) The system should take additional conditions such as clouds and the quality of snow into consideration. (c) Cities are constantly changing. New roads are constructed and the temperature changes from year to year. The system, however, does not take these factors into consideration and relies on past case scenarios, which occurred before the city changed. The evaluation in (c) above is one of the principle problems with case-based systems. Of course, past cases can provide at least outline plans, which can be of some help to directors. This is the primary advantage of using a case-based approach: it can provide suggestions even if the historical cases do not match the current situation. The staff interviewed also observed that the reasoning order is quite simple to predict. The system first searches for a similar amount of snowfall and then performs a fivestep evaluation, but employees suggested that the system should use a specific temperature, weather conditions and the quality of snow as priority criteria. The initial reasoning method is limited to comparing current and past cases. Criteria such as the quality of snow and snow conditions will have a beneficial effect on the elements of snow weight and time/distance. These tasks remain as future work. We designed the initial system to support novice snow removal directors. However, there are other problems in Sapporo, for example, a severe shortage of dump trucks and laborers. Our study may be designed to address future problems, but these are currently the primary tasks for Sapporo.

8 Conclusion
We propose a case-based directive support system for neophyte snow removal directors, based on an interview with the Sapporo city government. The basic idea is

902

Y. Ikeda and Y. Nakatani

to support snow removal directors by searching for cases similar to the prevailing snow conditions. The system can be helpful by providing suggestions based on past cases, even if the current condition is not exactly the same. Future tasks include taking subsidiary conditions such as clouds and the quality of snow into account, in addition to the depth of snow in judging similarity among cases and also to incrementally improve similarity metrics, based on the evaluation of tasks by directors.

References
1. Cabinet office, Government of Japan: Report of local economy (2001) 2. T. Yuki, T. Inoue, H. Okada: Control of maintenance for extraordinarily heavy snowfalls (2005) 3. M. Maruyama: Suggestion for a method that of planning for contract snowremoval (2005) 4. Kristian, J.: Hammond: Case-Based Planning. Academic Press, San Diego (1988)


相关文章:
更多相关标签: