Critical factors for emergency vehicle routing expert systems

Pergamon Expert Systems With Applications, Vol. 7, No. 4, p. 589-602, 1994

Copyright © 1994 Elsevier Science Ltd Printed in the USA. All rights reserved

0957-4174/94 $6.00 + .00

Critical Factors for Emergency Vehicle Routing Expert Systems

ROBERT GOLDBERG

Queens College of The City University of New York, Flushing NY

PHILIP LISTOWSKY

The Graduate School and University Center of The City University of New York, New York, NY

Abstract--A survey of professionals in the dispatching field was conducted to determine critical factors for the construction of an expert system knowledge base. The design goal is the Road Utilization Learning Expert System (RULES), an expert system that selects an appropriate path utilizing thought processes that the human expert would follow given various dynamic inputs and their expert knowledge. The survey results indicated that about 70% of police and fire chiefs nationwide would trust a computer with the task of dispatching their units, and that these respondents chose a computer system containing the knowledge of all human experts in the department as a method that elicits the greatest confidence and trust. In addition, many respondents expressed the need for verification of the suggested routes by human expert consultation. High degrees of relevance to the prediction of traffic flow were assigned to time of day, day of week, construction, and weather patterns. The validity of the results was verified by statistical analysis of responses to questions that had been designed for that purpose. We report on survey data critical to the design of the expert system and suggest that future emergency service dispatching systems requires expert computer assistance.

1. INTRODUCTION

1.1. Background

RECENT POLICY INITIATIVES o f President Clinton and Vice President Gore include emphasis on focusing national efforts toward certain critical technologies. The executive office of the President of the United States recently released a strategy paper on "Technology for America's Economic Growth, A New Direction to Build Economic Strength" (Clinton & Gore, 1993). The presidential directive emphasized that "providing a world class transportation sector will require the na- tion to meet the challenges posed by increased congestion." The document specifies the need to increase research into technologies that lead to the development of "smart highways." Specific technologies supported by this initiative include automated traffic monitoring and the distribution of planning information to vehicles.

Similar concerns have previously been expressed by

Requests for reprints should be sent to Robert Goldberg Department of Computer Science, Queens College of The City University of New York, 65-30 Kissena Boulevard, Flushing, NY I 1367.

589

the U.S. Department of Transportation in its strategic plan report to Congress on an Intelligent Vehicle Highway System (Card, 1992). The basis of this IVHS plan includes providing enroute emergency vehicles with traffic information and route guidance. This critical application domain must include human expertise in order to satisfy the "smart" system criteria. During crises, these public service professionals rely on prior experiences to supplement traffic volume and congestion information provided by historical or purely statistical and simulated data approaches. It is these considerations that suggest the development of an expert system for emergency vehicle routing. We report the results of a survey of emergency service professionals conducted nationwide to determine the critical factors affecting their instinctive expert decisions, which will then form the foundation of future dispatching systems.

Most major metropolitan regions throughout the world are currently facing severe traffic flow problems on their highways and access roads. Construction of new roads, and the addition of lanes to existing tho- roughfares have only seemed to exacerbate the problem, as any improvement in traffic flow seems to attract more vehicles to such roads and any flow gains are soon lost. Motorists caught in this inevitable traffic are

590 R. Goldberg and P. Listowsky

often faced with a choice of routes. Those who are very familiar with the region they are traversing gather input from radio traffic reports, weather reports, perhaps CB transmissions, and other sources. This information is then assimilated (in their minds) using recollections of the various factors affecting flow on particular roads. The expert motorist then makes route decisions, which will usually affect their arrival at the destination in the minimum amount of time and with the fewest disrup- tions. For critical applications, such as dispatching emergency service vehicles, effecting good response times demands the choice of the most appropriate route in an efficient manner.

1.2. Motivation for Survey

The Road Utilization Learning Expert System (RULES) project is an effort to design an expert system to effectively replicate the thought processes of a human expert in charge of dispatching vehicles and planning routes. The goal of the RULES project is the production of a knowledge-based system that determines the best route to get from a starting location to a given destination at different times and under varying conditions. A rule-based system is being formulated from knowledge of significant factors that may influence path choices as well as expert considerations of how these factors interact with each other. This paper reports the results of a survey conducted to gather the information that will form the core of RULES. The data obtained by this survey of public service professionals also includes information about these experts' receptiveness to computers in general and expert systems in particular.

A representation of the structure and operation of the RULES system can be seen in the information flow diagram in Figure 1. The survey reported in this paper was designed to gather data that shall be a critical component of the "SELECTION OF EXPERT KNOWL- EDGE RULES FOR PROBLEM SET" block in Figure 1. The dynamic inputs (clock, traffic, weather, and special events) feeding into the "SYSTEM UPDATE" block of RULES were determined from the relative rankings assigned by the human experts responding to the survey. RULES maintains a database of historical

USER CLOCK TRAFFIC WEATHER EVENTS

REQUEST [ SYSTEM P PROCESSING,~, L ~

SELECTION] [ APPLY CURRENT] OF EXPERT ~ CONDITIONS ~ I D~-~'~'-~,- II KNOWLEDGE r "J TO RULES I ~ t . . . . . JJ RULES FOR I ' PROBLEM SET [

FIGURE 1. Overview of Information Flow in RULES.

traffic-flow data, both for validation purposes and to assist in the generation of the best route as well. Road- side traffic flow sensors are in place and operational in many arteries of major metropolitan areas and directly assist in providing real-time traffic flow data (Zove, 1982). Similar mechanisms are being installed in other important roadways. In addition, data describing typical hourly traffic flow for major roadways across the country for all the months of a given year has been collected and maintained by the U.S. Department of Transportation (Curry, 1990). If traffic flow data were not available dynamically for a given region, then RULES will consider the historical data provided by those government studies. With each run of the expert system, the data set will thus be refined by utilizing statistical data feedback by-products as discussed by (Sobol, 1992). The system will thus learn more each time it is run.

Resources such as road-side sensors currently exist to gather dynamic traffic flow data. A knowledge base constructed from the survey results of human experts in this field complement such resources and provide for the selection of the best route to the destination. Thus, the dispatching task can be automated through the production of a computer expert system. We are producing the Road Utilization Learning Expert Sys- tem (RULES) to meet this need and to satisfy the aforementioned requirements. RULES is thus an expert system that recommends the best route from some point location to a specified destination based on rules of inference contained in a knowledge base and various dynamic inputs including time, weather, traffic flow, and special events such as construction or protests.

1.3. System Design Strategies and Objectives

Algorithmic techniques have previously been employed to solve an array of vehicle routing problems. These computerized methodologies have been successfully deployed by operations research professionals in several specific routing applications. Most such systems have been applied to commercial delivery operations, as in the food delivery vehicle system used by Southland Corporation's 7-Eleven stores (Golden & Assad, 1986). Issues such as inventory, payload, and profit naturally play a more vital role in the computations executed in such cases (Anily & Federgruen, 1993; Golden & Wasil, 1987; Baker & Schaffer, 1986). In addition, purely mathematical approaches do not readily accommodate dynamic updates of knowledge and new factors affecting the environment of the application domain. The learning processes of the expert system are better suited for this role. Emergency services operations involve many other issues and concerns that cannot be directly formulated by mathematical equations alone.

The expert system employed in the RULES project addresses issues critical to the emergency services rout-

Emergency Vehicle Routing 591

ing function. There are three phases in the emergency vehicle dispatching process. First, dispatching personnel determine which units RULES will consider in its evaluation by factoring in unit locations and current unit activity/inactivity. Second, the dispatcher then uses RULES to determine which available units would be capable of responding in a timely fashion. Finally, when responding, the assigned unit uses the route recommendations provided by RULES. The first issue has previously been addressed. Goldberg and Szida- rovszky (1991) described the use of the Hypercube Model in the evaluation of busy probabilities for EMS vehicles. Ball and Lin (1993) addressed the need to deploy emergency services resources in particular numbers at specific locations to provide for response to incidents in a minimum amount of time. The second and third phases, however, hold unique positions in the case of emergency services dispatching and hence are the main focus of RULES.

Although it is true that part of the police officer's thought processes involve consideration of where the heavy traffic is, the experienced officer will have expert knowledge that supersedes such information. Stochastic models to "determine vehicle flows on each link at each instant of time resulting from drivers using in- stantaneous minimal time routes" (Ran, Boyce, & LeBlanc, 1993) seem to have a rational and compre- hensive foundation. However, all of the previously described systems ignore real world considerations, which frequently are very relevant to emergency vehicle operations. For example, if an accident occurs at point X on road A, the historical data, the road-side sensors, and the mathematical model, may all indicate to police officer Jones that road A is a poor choice, since traffic flow is 0 mph on it, and that the unit should detour via road B. Officer Jones, the human expert, however, knows that radio traffic reporters have just begun telling motorists to use road B (hence it too shall shortly be congested), and that road A has a wide shoulder by which the officer can speed to the accident site. Many other "smart" factors come into play in the emergency vehicle realm. If there is no shoulder, the expert officer might sometimes determine that it would be safe and effective to travel west in the eastbound lanes (lights and sirens on, of course). Similarly, due to their official right-of-way, emergency service vehicles can force im- mobile traffic to part to the sides and hence travel between lanes (in a narrow two lane tunnel, for example). The nature and variety of the data comprising the officer's expert knowledge defy the capabilities of existing stochastic models. The expert officer is aware of subtle items that have dramatic impact on traffic patterns. These may include the nature of the drivers on roads at different times ("Sunday drivers"), in different sea- sons (effects of the officer's snow tires, etc.), and inti- mate knowledge of direction of congestion at the various rush hours.

The above described limitations of traditional models indicates the need in the emergency services profession for a new approach in the application of computer systems, in much the same way that certain aspects of the stock market crash of October 1987 illuminated flaws in program trading. As Donald B. Matron, Paine Webber chairman and former governor of the New York Stock exchange said, "[from the crash] we learn that you can't let computers do all your thinking for you" (Sanger, 1987). There were several stages of program trading that caused the problem. Nearly everyone in the market had a computer system that sensed price discrepancies between the New York and Chicago markets. All of these systems would then automatically run sell programs for large portfolios of stocks worth millions of dollars causing a snowballing effect that manifested itself in the crash. There were two major lessons, however, that this experience taught. The first lesson is that some people made small fortunes from the events of the market crash. These were the investors who ignored the models' recommendations and used their own "expert knowledge" to determine the best purchase opportunities. An additional lesson was that when many users simultaneously consult a system that in theory is the best "model," the net result under some circumstances, can be disaster for everyone. The strategic goal of our survey was therefore to gather a foundation of expert knowledge from a large sample of experts.

1.4. Data Collection Strategies

A fundamental element of the daily routine of public service professionals such as fire and police personnel is the dispatching of vehicles to incidents. For obvious reasons, these vehicles have a critical need for choosing the best route to an incident. The survey was therefore mailed to the offices of police and fire chiefs in the United States. These individuals were asked to apply their expertise in the field, in answering the survey questions. In some cases the persons responsible for dispatching and communications completed the survey. The survey therefore sampled the best experts on the routing decisions any individual emergency, service driver would face. A list of the chiefs of every department in the country (in excess of 6,000 cities) was obtained, and a survey was mailed to police and fire chiefs of 250 of the largest and most representative cities, which were selected to insure a statistically significant sample (a total of 500 surveys were mailed out). The survey was designed to provide a detailed description of the organization responding (service type: police, fire, etc., and service size: personnel, vehicles, area covered, vehicle dispersal), and data regarding their organization's performance (response times and distances typically traveled). The survey also contained inquiries about the methodologies and equipment currently used


to choose routes, and asked the respondents to rate their current systems. A vital component of RULES, the weights of the critical factors involved in choosing the most appropriate path to a site, is determined by the relative weights chosen by the experts. Participants also reported features desired in future dispatching systems, and indicated their plans for upgrading current equipment.

Because of the importance of choosing the best route when police and fire vehicles are dispatched, the opinions of relevant personnel in these roles would be important for contributing to the knowledge base of our system. The relevance of various factors in predicting traffic flow was therefore one of the sections of the survey most useful for implementation of RULES. In addition, the participants in the survey were asked if they would trust a computer-based system to perform tasks currently implemented by human experts. Hence, the survey serves a dual function: to solicit the input of expert professionals in the design of the RULES system, as well as to determine whether a system like RULES would gain widespread acceptance among the professionals who would be most likely to use it.

2. SURVEY DESIGN

The concerns expressed in the previous section were incorporated into the design of the survey now described. The reader is referred to Appendix 1 for a copy of the cover letter and survey questionnaire. The survey results are reported and analyzed in subsequent sections. As described in the cover letter mailed to survey participants, our goals were to determine the critical factors for designing better dispatching systems for the future. The participants described what types of dispatching systems are currently used and how their users rate them. In addition, the respondents expressed their degree of confidence in expert systems, and indicated which items should be considered by the expert system in deciding the best route. The structure of the survey is as follows:

Questions 1-5: general descriptive data (department type, size, etc.)

Questions 6-10: incident response descriptive data (distance, time, etc.)

Questions l 1-13: queries about trust in computer- based systems

Question 14: survey of factors relevant in predicting traffic flow

Questions 15-19: evaluation of dispatching system currently in use

Questions 20-25: suggestions for future dispatching systems

Questions I-5 accumulate descriptive data from the respondents. Question l is about the department type (Police, Fire, or other). Question 2 asks about the de-

partment size. Although the responses indicated that the choice of cutoff points between each of the four categories of department size was well founded, it was discovered that the fourth category should be expanded in future surveys (since the largest police departments have in excess of 20,000 officers). Question 3 is about the number of vehicles, question 4 about the area covered by the department, and question 5 about the number of vehicles in use per unit of area.

When the data in question 3 and question 4 are considered together and compared with the data in question 5, a validity check of the survey's answers may be executed; since most departments follow similar guidelines for unit dispersal, the number of cars for a given area should be fairly predictable. However, there are some mitigating factors, which result in deviation of the data from such patterns. For example, in more densely populated regions there may be a higher dispersal number (l police car may cover .25 square miles in New York, and 25 square miles in Ne- vada).

The next five questions requested incident response data. Question 6 asks for average response time, and question 7 for the distance traveled in a typical incident response. Question 8 asks how communication is performed (radio, computer, or both); question 9 is about methods currently used to select units for a response, and question l0 asks how the best route is chosen.

The next group of questions gauge human expert trust for a computerized dispatching system in a "real world" capacity: question 11 determines the most trusted dispatching method, (question 12 solicits a reason for that response), and question 13 surveys if computers can be trusted with the dispatching task. Al- though there is an obvious distinction between questions I l and 13, another reason both were included was to provide another of the several cross-checks of the validity of the response data. Question I 1 solicits whether the respondent would trust a computer more than a human, and question 13 corroborates the human confidence in expert computer systems. By correlating the response of these two questions, we have a way of verifying this portion of the survey that measures the degree to which a computer would be trusted in dispatching tasks (as shall be demonstrated in section 4.2).

Question 14 asks the respondent to rate the relevance of 12 critical factors for predicting traffic flow. The elements covered in this question are foundations of the knowledge base in the RULES system. Prior to this survey, extensive research had been performed to determine the factors that effect a person's choice of roads. The factors previously considered included the following items: lanes available, intersections, speed limit, slow moving vehicles, signal delays, turns, pave- ment, road work, unexpected congestion, traffic, weather, accessibility, freeway, time of day, volume,


travel time, route distance, route setting, habit, auto- mobile condition, and flexibility of arrival time (Kwon, 1990). The aforementioned study surveyed commuters' decision making at their trip origin and at the ramp to their primary corridor. Our survey incorporates many of the factors determined by Kwon to be relevant in modeling "on-line traffic demand." However, the population of this survey (by the nature of their occupa- tions) represents expert opinions in these matters, and their responses incorporate the unique nature of emergency vehicle routing. This knowledge thus obtained is applied to our expert system's development.

Part two of the survey, comprising questions 15- 18, asks the respondents to rate the effectiveness of the system they currently use. The knowledge gained from these data was used to facilitate an effective design of the RULES system. Question 19 asks about upgrade plans, if any. Part three of the survey measures the importance the respondents would assign to inclusion of various critical components in future systems (questions 20-23) and offers the respondents space to write in suggestions that were not covered by the survey.

3. RESULTS

Details of results in the six areas delineated in the survey design specification (previous section) are now reported. Of the 500 surveys distributed, there were 178 respondents (35.6%). The first two areas of the survey describe general information about the capabilities and responsibilities of each of the respondents. The subsequent groupings evaluate the dispatching systems currently in use and solicit the expert opinions on issues such as trust in and critical factors of computer-based dispatching systems. The results of each area follow.

3.1. General Descriptive Data

This set of questions (1-5) requested general information about the availability of typical resources to the emergency vehicle dispatcher. The responses were nearly evenly divided between police (41.6%) and fire departments (42.1%) (see appendix 2a for details). It may be noted that although the survey was sent to the police and fire chiefs of 250 cities (i.e., for each selected city, both departments received the survey), and though the responses appear to break down almost evenly, these police and fire respondents were not necessarily from the same city. In fact, responses were received from both police and fire departments of a particular city in 42 cities. The third department type choice, "other" was indicated by 29 respondents (16.3%). These were comprised of general purpose 911 dispatching facilities used either by both fire and police, or by some combination of fire, police, and emergency medical services departments.

Responses to the department size question indicated that most had between 100 or 500 persons (see Ap- pendix 2b for details). In addition to surveying the opinions of such average-sized departments, we also received responses from major cities including: Los Angeles, Chicago, Baltimore, Boston, Detroit, Min- neapolis, Cleveland, Philadelphia, Dallas, Houston, and Milwaukee. The number of vehicles allocated to the departments fell in the 10 to 100 vehicle range for the majority of respondents (see Appendix 2c). Thus, the survey population consisted of departments having a full range of resources available.

The area covered by departments of question 4, yielded the following information: The minimum area covered by a department was .75 square miles and the maximum was 840 square miles. The median value was 32 square miles and the mean was 85.2 square miles. 173 respondents answered this question 4. The minimum vehicle dispersal rate (question 5) was .07 vehicles per square mile, and the maximum was 16 vehicles per square mile. The median value was l vehicle per square mile, and the mean was 2.033 vehicles per square mile. Although only 144 answered question 5, in most cases a reason was offered by the respondent for omitting such data; it was explained that the dispersal rate can vary significantly from day to day due to numerous unpredictable factors inherent in the emergency services field.

3.2. Incident Response Descriptive Data

Having obtained general information about department resources, the next set of questions (6-10) determined objectively the capabilities of each department to respond to an incident site. Question 6, about the average incident response time, yielded 164 responses. The minimum was 1 minute, and the maximum was 24 minutes. The median response time was 4 minutes, and the mean was 4.869 minutes. In question 7, the distance typically traveled for an incident, results were a minimum of .25 miles, a maximum of 11.6 miles, a 2-mile median, and a mean of 2.209 miles, with standard deviation of 1.714 for 174 responses.

All (178) respondents answered question 8, How does your department currently dispatch and communicate with units?: 98.3% of the departments communicate with units by radio, and 24.2% of all respondents use computer-based communication links (appendix 2d). It is important to note that most of the 75.8% who answered "via radio only" do in fact have computers in their vehicles; they were just not used as a means for the personnel in the vehicle to talk to the base (the compu te r s - - "MDT' s " - - a r e used as mobile display terminals for tasks such as license plate queries). This observation can be confirmed based on the data received in the responses to the following question, question 9. When asked what methods are used to decide


which units to dispatch to an incident, of 173 responses, 26.5% (46) checked (A): base broadcasts, units tell base if they can respond, 47.3% (82) checked (B): base personnel choose units to assign incident to, and 51.4% (89) checked (C): computer advises department on best unit to assign. A number of respondents checked more than one box, the breakdown of responses can be found in Appendix 2e.

When asked how dispatched units currently determine the best route to an incident (question 10), 177 (of 178 responses) checked (A): personnel in unit decide route based on their own knowledge, 7 checked (B): personnel at base tell personnel in unit best route, and 3 checked (C): computer recommends best route. A number of respondents checked more than one box; the breakdown of responses can be found in Appen- dix 2f.

3.3. Queries About Trust in Computer-Based Systems

When the chiefs were asked which method they would have the greatest confidence and trust in, the over- whelming majority of respondents (117--67.2%) said a computer system containing the knowledge of all of their human experts would be best (see Appendix 2g). In question 12, they wrote that their choice of ideal system was based on personal experiences with current computer-based dispatching systems. Interestingly, many dispatchers who selected the expert system as ideal, specified comments such as "more inclusive of all factors affecting unit selection." Some of these respondents further suggested that as a method of validation, human expert consultants should be involved in this decision making process: "A dispatcher should have sufficient knowledge to make the final call." This also satisfied their concern about computer systems shutting down in the midst of a crisis. Question 13 asked i fa computer could be trusted with the dispatching of units. The responses (detailed in Appendix 2h) indicated that 68.4% of the emergency vehicle dispatchers would trust a computer with that task.

3.4. Survey of Factors Relevant in Predicting Traffic Flow

The most important factors in predicting traffic flow were: time, day, and construction and weather patterns (see Appendix 2i). Factors found to be of low signifi- cance were road speed limit and accident rate. Since emergency vehicles have the right of way in any traffic situation, road conditions that depend on physical lay- out of the roadways was less significant than traditional traffic flow predictors (as indicated by time on a given day) and uncontrollable patterns (such as weather and construction).

3.5. Evaluation of Dispatching Systems Currently in Use

The results of this section indicate that more than half of the dispatching systems currently in use have response times that were rated only moderately effective at best. This indicates that there exists a real world need for dispatching and routing tools that will provide better response times. Most respondents did report that their current systems are at least moderately effective at providing units with all data needed and effecting good resource utilization. The survey participants had also been asked how effective their current system is at reducing the number of units needed. Though most said that it was "ineffective" in this function--i t was frequently indicated (hand written notes on the survey) that this question was inappropriate since "reducing the number of units" is not necessarily a desired goal-- emergency services want to have more units than they typically need (to cover unforeseen developments). A detailed breakdown of the responses is in Appendix 2j.

3.6. Suggestions for Future Dispatching Systems

In this final section of the survey, the participants were asked to suggest special features that would enhance the fundamental dispatching and routing systems. Most (76.9%) of the respondents indicated that it is important for a computer-based routing function be included in any future dispatching system. A unit monitor feature was also deemed desirable. An analysis of the results indicated a high level of correlation for many respondents between the "CHOOSE BEST UNIT" option and "MONITOR UNIT LOCATION" while simultaneously demonstrating a high level of correlation between "RELAY TRAFFIC CONDITIONS" and "SUGGEST BEST ROUTE." This trend occurs es- pecially in the emergency service field because units in the field dispatched must remain on site to attend to the incident reported. The detailed responses to this section are presented in Appendix 2k. The responses to questions 24 and 25 contained information specific to the concerns of particular departments once at a site, but did not directly relate to dispatching per se.

4. ANALYSIS

The three most significant findings from the survey data are the high degree of trust in computer-based dispatching, the choice of an expert system as the ideal tool, and the relevance of various critical factors affecting the choice of routes. The validation of these findings is demonstrated by correlating the responses to questions designed for that purpose.


4.1. Significant Findings

The survey data indicated the high degree of trust emergency services personnel would place in an expert system (see Appendix 2g). It is significant that the survey population comprised of police and fire chiefs and not computer professionals, whereas prior surveys reporting trust in expert systems have queried popula- tions primarily composed of computer professionals (Diderich, 1992). Thus, the implementor and the end- user provide a favorable rating for computer expert systems. This was determined from the responses to questions l l through 13. In question 13 only 10.5% said they would not trust a computer, and 21. 1% said that they would trust a computer with some skepticism. The remaining 68.4% answered that "yes," "very definitely," or that they would trust the computer "completely." In addition, 67.2% of the respondents said that the ideal system would be a computer that contained the knowledge base of their expert employees (question 11).

In question 12, where respondents were asked to provide any additional comments about their response to question 1 l, many of the professionals said that a human expert monitoring the output of the computer expert would be an indispensable verification and validation component of such a system. The RULES system, for this reason, incorporates several independent means of verifying that the best route chosen is appropriate. For example, the best route determined from the knowledge base of historical traffic flow data is compared with the best route determined by analyzing the data from dynamic inputs such as road-side sensors.

It is apparent from the evaluations of systems currently employed that most of the respondents are only moderately satisfied with the systems they use (see Ap- pendix 2j). Routing systems were found to be in demand because when asked to suggest features desired in future dispatching systems, 76.9% said that suggest- ing the best route to get to an incident is "moderately important" (37%), "very important" (33.5%), or "essential" (6.4%). As indicated by the responses to questions 21 and 23, features most desired for such routing systems are choosing the best unit for an incident (41.3% said "essential") and monitoring the location of units (39.5% said "essential"). Thus, in addition to expressing a need for better systems, most police and fire department personnel in charge of dispatching have indicated that they want a system that suggests a best route, and would trust computers to make that decision. In fact the professionals considered an expert system ideal. They further suggested as a method of verification and validation that human experts monitor the computer-based decision.

The 12-part question 14 surveyed the relevance of various factors in predicting traffic flow on roadways

(see Appendix 2i). The responses have proven to be valuable in the design of RULES. It was found that time of day, day of week, and construction and weather patterns were recognized as the most important predictors. The least significant predictor, according to the respondents, was accident rate. These results are consistent with the impressions logic would dictate: If a car is to get from point A to point B over route Q, the most important things to consider is if it is rush hour on a regular workday, and whether there is any dis- ruptive construction or weather condition on route Q. Similarly, because accidents are not continuous phe- nomena on most roads most of the time (though they may significantly affect traffic flow when they do occur), this relevant factor was ranked low by the survey respondents. Thus, in RULES traffic reports are considered as a dynamic input (part of events). The rankings assigned by the respondents describe the weights assigned each factor in the RULES system.

4.2. Validation of Survey Results

Various methods of verifying the results of the survey were employed. Although question 11 ("Ideally which method of dispatching would you have the greatest confidence and trust in?") is different from question 13 ("Would you trust a computer with the task of dispatching your units?"), the data gleaned from the two questions may be used to check for consistency. Re- sponses 1 and 2 to question 11 indicated trust in a human, and responses 3 and 4 indicated preference for a computer. It would therefore be reasonable to con- clude that the same number of respondents who indicated trust in computers in 13 would choose 3 or 4 in question 1 l, and vice-versa. In fact, 71.2% chose 3 or 4 in question I l, and 68.4% indicated that they would trust the computer, in question 13. These per- centages, coupled with the high correlation observed between the responses confirmed our hypothesis.

A correlation matrix of the responses to all questions of the survey was obtained to determine whether any unusual or unexpected correlations existed, and to ver- ify expected correlations. No significant unexpected correlations were found, demonstrating that the survey questions predicted properly the current trend in dispatching systems. Expected correlations did however appear, and validated the data. Specifically, the correlation between area covered by department (question 4), and department size (question 2) was .473, and the correlation between area (4) and number of vehicles (question 3) was .448. These confirm expectations, since these numbers are statistically significant correlations (Sekaran, 1984) for our response rate (178) and sample size (500), and one would expect that larger areas would be covered by larger departments with a greater number of vehicles. In question 14, the corre-


lation between day of week and holiday was .464 (which is logical, because weekend traffic is expected to be similar to holiday traffic). Similarly, the highest correlation found was also in question 14: day of week and time of day was .642, indicating that the experts consider traffic flows a predictable regular event.

A further check of the data was conducted by di- viding the data into categories based on department size: all statistical analysis performed on the entire data set was then performed for each of four groups defined by the responses to question 2, department size. Though there were few responses in the category of 3001- 20,000, the distribution among the remaining categories ( l - 100: 48, l 01-500: 84, 501-3000: 40) was sufficient to provide significant data. In fact, there are very few departments in the country that belong to this last category. In the category 1-100, 68% indicated a trust in computers, in the category 101-500, 73% indicated such trust, and in the 501-3000 category the number was 72%. Thus, the results met expectations that expert systems will play a significant role in future dispatching systems and were consistent with the over- all results mentioned earlier.

The correlation matrices relating various factors within the different ranges/categories of the department size also confirmed consistency in the data. In the 1- 100 category the correlation between time and day was .730, and in the 101-500 category it was similarly high: .673. In the 1-100 category the correlation between speedand lanes was .507, and in the 501-3000 category it was similarly high: .653. In the 101-500 category the correlation between alternates and exits was .629, and in the 501-3000 category it was similarly high: .680. These and similar correlations throughout the analysis of the survey data confirmed the existence of expected relationships between and relative importance of critical factors affecting emergency vehicle routing decisions.

5. DISCUSSION

This section provides a perspective on prior work deal- ing with issues related to RULES. Section 5. l contains an overview of two related surveys: the first conducted of emergency service personnel in major cities about communication facilities, and the second conducted of computer scientists over Internet about trust in expert systems. Also, a brief description is included of some in-vehicle computer mapping systems currently in operation. Section 5.2 describes future enhance- ments to RULES by incorporating a corresponding knowledge base to determine the particular units most versatile in handling a specific incident. A networked configuration of RULES is then considered to allow for cooperating dispatching sites to interact effectively.

5.1. Related Work

Surveys of emergency services communications facilities have previously been conducted. The Los Angeles Police Department conducted a survey of major cities' communications facilities (Walsh, 1990). Police departments surveyed therein included only those of Bal- timore, Chicago, Dallas, Detroit, Milwaukee, New York, Philadelphia, Phoenix, Tulsa, and Washington, D.C. Nine cities used some form of Computer-Aided Dispatching (CAD) system. Features provided in these CAD systems included unit recommendation, incident histories, and unit status data. Some elements desired for future systems were computer mapping and unit locating. It was also found that less than half of the communications systems had an interface with their radio system. Our survey verifies these concerns. In addition however, the current survey provides specific details that will assure that RULES meets real-world requirements.

Acceptance and trust in the information generated by an expert system has been previously estimated in a survey conducted on Internet (Diderich, 1992). To that survey, 534 persons responded. The population sample was a cross section of Internet users. Asked if they would trust a decision of an expert system, the respondents replied as shown in Table 1. Though these data would seem to indicate that computer professionals have much less faith in expert systems than the police and fire chiefs participating in our survey, it should be noted that the Internet survey solicited responses from the general population, albeit limited to people with Internet access. Our survey was mailed to a population carefully selected to represent a valid cross section of public service dispatching professionals. Di- derich had no way ofcontroUing his survey population.

Computer-aided dispatching systems have been employed by emergency services units to cut response times, disperse vehicle fleets, and automate incident reporting for Police and Fire departments throughout the country. Most of these systems are still primarily dependent on a human expert's presence "in the loop." The Pennsylvania Emergency Management Agency (Gantz, 1990) for example, has a network of systems that contain current maps and database information. This information is used to choose equipment for de- ployment as well as for assigning dispatching routes.

TABLE 1 Results of Diderich Survey

Would You Trust an E.S.? # of Respondents (%)

Yes 83 (15.5) Partially 399 (75) No 40 (7.5) No reply 11 (2)


The system does not make those choices independently, though it provides the human expert with all the information needed to make the choice.

A number of systems provide as output an estimate of the best route. None, however, model the human experts reasoning process; rather, these systems attempt to provide a mathematical model based on static for- mulae and simplistic interpretations of data sampled from dynamic sources. John Tibbets, system status manager of Albuquerque Ambulance (AA) explains this need (Sena, 1990): "It's difficult to find good para- medics. It's even more difficult to find crews who have an expert knowledge of all the streets in a city the size of Albuquerque." The Etak emergency response system that AA uses provides traffic flow data which human experts then consider in executing routing decisions. Similarly, the Integraph DMS system (Sena, 1990) al- lows the user to select a best route based individually on such parameters as minimum time, minimum distance, minimum number of turns, minimum number of intersections, or minimum risk to vehicle. A pref- erable solution would allow dispatchers (or drivers) without expert knowledge to consult an expert Com- puter-Aided Dispatching system for effective rapid de- ployment. The AA system currently used (Etak) paid for itself in less than a year and improved response time dramatically. The addition of expert systems technology would yield an even more effective dispatching system.

5.2. Future Work

The system described here uses a centralized computer that provides route recommendations based on derived expert knowledge, historical experience, and dynamic inputs of data affecting current traffic flow. The knowledge obtained by RULES can be utilized by the dispatcher in two ways: (l) to determine which units would be able to respond most rapidly due to the current conditions, patrol sectors and unit locations, and available equipment; (2) to recommend customized route planning for each unit responding. Similarly, an enroute unit may also initiate a request for efficient vehicle routing (Figure 2).

Traditionally, the focus in the emergency service profession has been exclusively on reducing response times. The investigations of Chelst and Jarvis (1979) have determined however that there is a complex re- lationship between response time and probability of arrest (for police), or prevention of death from accidents and myocardial infarction (for EMS). The implication of that finding is that though time is obviously a critical factor, there are other items affecting the success of responses. Determining the most appropriate set of responding units might be best defined with the help of a future expert system implementation. This comple- mentary system chooses the number of units and the best units for the response, based on the personnel and equipment in a given unit, and the circumstances of the incident. The user of such a system would again be the base/dispatching personnel.

An additional consideration would be any advice that RULES had concurrently offered to other users. For example, the very presence of an emergency service vehicle with lights and sirens activated may directly affect traffic flow on a road (motorists slow down when they see it). RULES would then be able to determine which roads may soon become congested, in addition to its fundamental capability to detect currently congested roads. This can be further complicated by the existence of multiple dispatching centers (Police, Fire, EMS) responsible for overlapping areas, the need of neighboring sectors to utilize some of the same arteries, and the occasional need for resource sharing between sectors. To accomplish this we would implement a network configuration of RULES. The vehicle routing network operations would be analogous to similar data- flow and routing problems in a computer network, but with roads taking the place of network lines, intersections in place of gateways and nodes, and individual vehicles filling the role of individual data packets.

6. CONCLUSIONS

There exists a real-world need for emergency vehicles to arrive at incident locations in the shortest amount of time. Dynamic factors that predict traffic flow affect this decision and include weather and roadway con-

BEST ROUTE TO INCIDENT

/ IEMERCENC

' SERVICE I .

UNIT )(L

RULES BEST ROUTE~

MODULE .) ~ BESTUNITS

UNIT ] DATABASE I

*patrol S¢Ctorl * Uni t Loc l t i on l

Communications Link )~t CENTER )J FIGURE 2. Usage of RULES by Emergency Services.

598 R. GoMberg and P. Listowsky

di t ions and t ime o f day. This survey o f pol ice and fire chiefs requested that these experts rate thei r current d ispa tching me thods and provide in fo rmat ion abou t the ideal future d ispatching system. The op in ions offered by these h u m a n experts also conveyed their un ique perspect ives o f the re la t ionships o f traffic flow factors.

It was found that near ly two thirds o f pol ice and fire chiefs in the na t ion ' s ma jo r cities ra ted their current d ispa tching system as only modera te ly effective, or worse, at p rovid ing units in the field with all da ta needed. In addi t ion , ha l f o f the depa r tmen t s felt s im- i larly that their cur rent systems were only modera te ly effective at p roduc ing good response t imes, and in excess o f two- th i rds ra ted their cur rent system as only modera te ly effective at provid ing good resource utili- zat ion. This demons t ra tes that there is a critical need for bet ter d i spa tch ing systems. In addi t ion , this survey provides the relat ive rankings o f 12 factors cri t ically affecting the d ispatching decisions. The archi tecture o f the Road Uti l izat ion Learning Expert System considers d y n a m i c inputs p rovided f rom external sources and, based on the above weights and the knowledge base conta in ing the history o f the par t icu lar paths involved, provides an exper t decis ion on the best selected path. Fur the rmore , abou t 70% of pol ice and fire chiefs said that they would trust a c o m p u t e r with the task o f dis- pa tching thei r units (and 68% said that a c o m p u t e r expert system would be ideal). The results o f this survey clearly indicate that the p roduc t ion o f the Road Uti- l izat ion Learn ing Exper t System is an impor t an t and necessary endeavor .

Acknowledgment--This research was supported in part by PSC- CUNY Grant #24-66408. We thank the reviewers for their helpful suggestions. In addition, Dr. S. Habib and Dr. T. Wesselkamper (CUNY) and Dr. M. Davis (NYU) provided encouragement for our research.

REFERENCES

Anily, S., & Federgruen, A. (1993). Two echelon distribution systems with vehicle routing costs and central inventories. Operations Re- search, 41(I), 37--47.

Ball, M.O., & Lin, F.L. (1993). A reliability model applied to emergency service vehicle location. Operations Research. 41, 18-36.

Baker, E., & Schaffer, J. (1986). Solution improvement heuristics for the vehicle routing and scheduling problem with time window constraints. American Journal of Mathematics and Management Science. 6(3-4), 261-300.

Card, A.H., Jr. (1992). Intelligent vehicle-highway systems strategic plan. U.S.D.O.T. Report to Congress FHWA-SA-93-009, Wash- ington, DC.

Chelst, K., & Jarvis, J.P. (1979). Estimating the probability distribution of travel times for urban emergency service systems. Op- erations Research, 27( 1 ), 199-204.

Clinton, W.J., & Gore, A., Jr. (1993). Technology for America's economic growth, a new direction to build economic strength. Office of Science and Technology Policy, U.S. Government Printing Office, Doc #1993-347-397/80142, Washington, DC.

Curry, J.R. (1990). A review of information on police-reported traffic crashes in the United States--General estimate system 1990. U.S. D.O.T. technical report. Washington, DC: National Highway Traffic Safety Administration.

Diderich, C.G. (1992). Science technology society program: A survey (in French), Department of Computer Studies technical report, Ecole Polytechnique Federale De Lausanne, Suisse.

Gantz, J. (1990, May). Graphical dispatching. Computer Graphics World, pp. 31-33.

Goldberg, J., & Szidarovszky, F. ( 1991). Methods for solving nonlinear equations used in evaluating emergency vehicle busy probabilities. Operations Research, 39, 903-916.

Golden, B.L., & Assad, A.A. (1986). Perspectives on vehicle routing: Exciting new developments. Operations Research, 34, 803-810.

Golden, B.L., & Wasil, E. (1987). Computerized vehicle routing in the soft drink industry. Operations Research, 35, 319-333.

Kwon, E. (1990). Modeling on-line traffic demand and diversion in freeway corridors. Ph.D. Dissertation, University of Minnesota, Minneapolis, Minnesota.

Ran, B., Boyce, D.E., & LeBlanc, L.J. (1993). A new class of instan- taneous dynamic user optimal traffic assignment models. Oper- ations Research, 41( 1 ), 192-202.

Sanger, D.E. ( 1987, October). Limits set on program trades. The New York Times, pp. 3-4.

Sekeran, U. (1984). Research methods for managers. New York: Wiley.

Sena, M.L. (1990, May). Computer aided dispatching. Computer Graphics World, pp. 35-42.

Sobol, M.G. (1992). Using statistical data feedback by-products of expert systems. IEEE Transactions on Systems, Man, and Cy- bernetics, 22, 568-580.

Walsh, P. (1990). Major cities survey of police communications facilities and systems. Los Angeles Police Department Communi- cations Division Survey, Los Angeles, CA.

Zove, P. (1982). Integrated motorist information system. U.S.D.O.T. (Federal Highway Administration Report FHWA/RD-82/108). Washington, DC: U.S. Department of Transportation.


APPENDIX 1

August 10, 1992

Dear Public Service Professional,

We are conducting a survey of dispatching systems used by key Police Departments, Fire Departments, and Emergency Services. This study is being conducted in the Computer Science Department of the Graduate School and University Center of the City University of New York. Our goals are two-fold: First, to determine what types of dispatching systems axe currcndy being used and how their users rate them. Second, we are interested in surveying the needs of these Public Service departments, to design even better systems in the future.

The results will be made available to all survey participants (If you would like to receive a copy of the results please provide the information requested below). We therefore respectfully request that you take a few minutes to complete the questionnaire and return it within the next week. Confidentiality will be maintained, and all responses will remain anonymous. This survey has been carefully designed to facilitate its completion in a minimal amount of time. We estimate that it should take only 3 to 6 minutes to complete. Please note: Questions I1, 13 & 14 are the most important for our surv~ry.

Sincerely Yours,

Philip Listowsky CUNY Grad. Center

Dr. Robert Goldberg Queens College of CUNY

I-1 Yes, we would like a copy of the results. Mail the survey results to:


Survey of Public Service Dispatching and Routing Systems

PART 1; G E N E R A L INFORMATION Please check box(es) or fill in where applicable:

(1) Service Type: [ ] 1. Police Department [ ] 2 Fire Department [ ] 3 Other (please specify)

(2) Department Size: [ ] 1.1-100 persons [ ] 2. 101-500 persons [ ] 3. 501-3000 persons [ ] 4. 3001-20000 persons

(3) Number of Vehicles: [] 1.1-10 [ ] 2. 11-100 [ ] 3 101-500 [ ] 4. 501+

(4) Area covered by department: sq. miles

(5) Estimate of typical unit dispersal: vehicle(s) in use per square mile

(6) Average unit incident response-time:

(7) Distance typically traveled in response (from question 6):

(8) How does your department currently dispatch and communicate with units?: [ ] 1 via radio only [ ] 2. via computer communications link [ ] 3. both radio and computer link to vehicles

(For questions 9 & 10, please check all answers that apply) (9) What method is used to decide which units to dispatch to an incident?:

[ ] a base personnel broadcasts incident location, units in vicinity of incident will let base know they can respond

[ ] b base personnel manually track units at all times and choose units to assign incident to

[ ] c. computer advises department on best unit to assign

(10) How do dispatched units determine the best route to incident?: [ ] a personnel in unit decide route based on their own knowledge [ ] b. personnel at base tell personnel in unit best route [ ] c. computer recommends best route

(11) Ideally, which method of dispatching would you have the greatest confidence & trust in? [ ] 1 A department employee generally acknowledged to be an expert in dispatching [ ] 2. A department employee routinely assigned to that specific task [ ] 3. A computer system bought "off-the-shelf" - but recommended by others [ ] 4 A computer system containing the knowledge of all of your expert employees

(12) Reason for choice in question 11 9


(13) Would you trust a computer with the task of dispatching your units?

With Yes Very Completely No Sk~tlaim Delnndy

1 2 3 4 5

(14) On a scale of l to5 (with 1 b e i n g ~ and5 being ~ , p l e a s e rate the relevance of each of the following in predicting traffic flow on roadways:

time of day day of week season holiday/not holiday

lanes in roadway road condition ,. weather accident rate

construction speed limit altemstes exlUentmnce ramps

PART 2: EVALUATION OF THE DISPATCHING SYSTEMS YOU USE Please rate effectiveness in the following areas, of the dispatching system your deparanent currently uses"

How (15)

slightly moderately very optimally Ineffective effective effective effective effecUve

well does your current system: Provide units in the field with all data needed? 1 2 3 4 5

(16) Produce good response times? 1 2 3 4 5

(17) Consistently provide good resource utilization? 1 2 3 4 5

(18) Reduce the total number of units dept. needs? 1 2 3 4 5

(19) Does your department have plans to upgrade its dispatching system in the near future?

PART 3: SUGGESTIONS FOR FUTURE DISPATCHING SYSTEMS Please rate the importance of including each item below in the design of a dispatching system:

not slighUy moderately vmy ImporUmt impoi~nt important Important e u e n k l

(20) Advise units of traffic conditions in area 1 2 3 4 5

(21) Choose best unit for a given incident 1 2 3 4 5

(22) Suggest best route for unit to get to incident 1 2 3 4 5

(23) Monitor location of all units at all times 1 2 3 4 5

For queslions (24-25), please recommend any other critical items, and rate their imgoflance :

(24) 1 2 3 4 5

(25) 1 2 3 4 5


A P P E N D I X 2

# of Respondents % A C A & B A & C A, B & C

a) departments responding to survey (178 responses) Police Department 74 41.6 Fire Department 75 42.1 Other (general purpose 911 facility) 29 16.3

b) size of departments (personnel) (177 responses) 1-100 Persons 48 27.1 101-500 Persons 84 47.5 501-3000 Persons 40 22.6 3001-20000 Persons 5 2.8

c) vehicles operated (174 responses) 1-10 Vehicles 22 12.6 11-100 Vehicles 98 56.3 l 01-500 Vehicles 46 26.4 501+ Vehicles 8 4.6

d) current communications methods (178 responses) via radio only 135 75.8 via computer com. link 3 1.7 via both 40 22.5

m~ A B C A & B A & C B & C B & C

e) methods used to choose best unit for an incident ( ! 73 responses.)

# 18 55 66 11 7 6 % 10.3% 31.4% 37.7% 6.3% 4% 3.4%

f) methods by which chosen unit determines best route (178 r.)

# 169 1 6 1 1 % 94.9% .6% 3.4% .6% .6%

Human Regular Standard Computer Expert Employee Computer Expert Sys

g) ideal dispatching method (168 responses.) # 38 12 7 117 % 21.8% 6.9% 4% 67.2%

With Very No Skepticism Yes Definitely Completely

h) trust in computers (171 responses.) # 18 36 67 37 13 % 10.5% 21.1% 39.2% 21.6% 7.6%

i) traffic flow predictors: i) time of day: 2) construction:

4.50 3.98 4) weather: 3.48 5) holiday: 3.30

7) lanes: 3.10 10 10) exit/entrances: 5.7% 2.62

3) day of week: 3.95

6) road condition: 3.27

8) season: 2.81 9) alternates: 2.67 11) accident rate: 12) speed limit:

2.57 2.55

Slightly Moderately Very Optimally Ineffective Effective Effective Effective Effective

j) evaluation of dispatching systems currently in use: Provide all data needed I 1 1 l

(6.4%) (6.4%) Has good response times l0 14

(5.9%) (8.2%) Resource utilization is good 9 24

(5.3%) (14%) Reduce number of units 52 39

(31.9%) (23.9%)

84 62 4 (48.8%) (36%) (2.3%)

63 72 11 (37.1%) (42.4%) (6.5%)

78 54 6 (45.6%) (31.6%) (3.5%)

47 23 2 (28.8%) (14.1%) (1.2%)

Not Slightly Moderately Very Important Important Important Important Essential

k) suggestions for future systems: Relay traffic conditions 16

(9.4%) Choose best unit 4

(2.3%) Suggest best route 19

( l l%) Monitor unit locations 8

(4.7%)

35 60 50 10 (20.5%) (35.1%) (29.2%) (5.8%)

5 17 75 71 (2.9%) (9.9%) (43.6%) (41.3%)

21 64 58 11 (12.1%) (37%) (33.5%) (6.4%)

11 38 47 68 (6.4%) (22.1%) (27.3%) (39.5%)

Critical factors for emergency vehicle routing expert systems

Documents

Transcript of Critical factors for emergency vehicle routing expert systems