Jeffrey Scott Forrest, (1997)
Èñòî÷íèê: http://frontpage.hypermall.com/jforrest/DSS/airline_decision.htm
 áèáëèîòåêó
ABSTRACT
Airline operational managers have traditionally relied upon management decision making tools such as "rules of thumb" developed from years of operational and individual experience. Heuristic, experience-based tools are still relied upon by even the largest airline operational centers. Factors such as deregulation, hub and spoke operational scheduling and increased regulatory factors have placed a demand on operational managers to develop decision support systems (DSS) that provide real-time decision support. This paper provides evidence for the migration from heuristic decision making tools to Artificial Intelligence (AI) based DSS platforms by major airline operational departments. Selected case studies are presented for individual airlines. The examples demonstrate the evolution and migration towards AI-DSS by major commercial airlines.
TABLE OF CONTENTS
INTRODUCTION
Background
The Purpose of the Paper
AIRLINE OPERATIONS AND THE DSS
Environmental Analysis
Free Flight
CASE EXAMPLES
Singapore Airlines
Southwest Airlines Co.
Delta Airlines
United Airlines
Cathay Pacific
DISCUSSION
The AI Challenge
Concluding Observations
RECOMMENDATIONS
Future Research
REFERENCES
Footnotes
SECTION 1 Introduction
Background
One of the most challenging examples of operational planning, scheduling and controlling (OPSC) may be found within the domain of the airline industry. This industry, like many other transportation services, is constantly striving to maximize profits within a competitive operational environment. However, the airline industry is acutely affected by temporal, spatial, and conditional factors that greatly inhibit real-time, economic decision making (Gray & Kabbani, 1994).Airline operational managers have traditionally relied upon management decision making tools such as "rules of thumb" developed from years of operational, and individual experience. Heuristic, experience-based tools are still relied upon within even the largest airline OPSC centers, such as the United Airlines System Operations Control (SOC) (Wong, Al, personal communication, July 10, 1997). However, factors such as deregulation, hub and spoke operational scheduling (nodes and arcs), and increased regulatory factors have placed a demand on operational managers to develop quantitative problem solving tools, or decision support systems (DSS), that provide real-time decision support.Efraim Turban (1993) quotes in his book Decision Support and Expert Systems a definition of DSS provided by Keen and Scott-Morton (1978):
Decision support systems couple the intellectual resources of individuals with the capabilities of the computer to improve the quality of decisions. It is a computer-based support system for management decision-makers who deal with semi-structured problems. (p. 12)
Under the guise of this definition, DSS is particularly important to the airline operational manager. The power of a DSS is most useful within the arena of daily flight operational control. Flight operational control decision-making operates within the structured flight schedule planning (long and short run) that has been developed by airline competitive strategy. It then attempts to manage the pre-described flight schedule on a daily bases, contending with a highly dynamic environment. Affects such as weather, unscheduled equipment maintenance, crew shortages, regulatory factors, and aircraft loading can make the profitable deployment and management of a pre-determined flight schedule very difficult.As mentioned, DSS within the airline industry has its’ underpinnings grounded in a heuristic based decision making environment. With the advent of computer processing units (CPUs), airline operation managers combined their heuristic techniques with "what if analysis" by using relatively simple spreadsheet computations (Gray & Kabbani, 1994). By review of the related literature, this paper will show that DSS as applied to airline operations management has evolved past strictly "what if analysis." While still demanding the usage of heuristics, and scenario analysis, the industry has developed very powerful parametric DSS tools that are bordering on the integration of artificial intelligence (AI).
The Purpose of the Paper
This paper serves as a review of the functionality of various DSS systems employed by the airline industry. The examples will show the evolution and migration towards AI by airline flight operations management.
Back to Top
SECTION 2 AIRLINE OPERATIONS AND THE DSS
Environmental Analysis
Long term flight schedule planning is primarily market driven. Routes are selected during processes of long term strategic decision making within the boundaries of profit maximization. DSS incorporating economic modeling algorithms are typically used by airlines during this phase of business planning (Gray & Kabbani, 1994). However, the time allocated for decision making during this stage is usually not critical. Decisions during this phase are not made in real-time, such as they are during daily flight operations. In this environment, decisions must rapidly be made that will solve unexpected problems while minimizing the negative economic impact of those decisions.For these types of decisions, tools must be provided that increase the quality of decision making within a very short time period. Relying strictly upon heuristics may allow an airline to "muddle" through the problem, and perhaps even solve it. However, to stay economically competitive airlines need decision-making tools that can provide qualified, if not quantified information rapidly. The amount of variables that can affect the daily operations of schedule management for any single airline are staggering. The decision to delay, cancel or re-position a flight will depend upon input from multiple sources of information, all providing feedback in real, or nearly real-time. Information needs to be solicited, retrieved and interpreted from sources such as meteorologists, flight dispatchers, crew schedulers and financial analysts (McCartney, 1996).The problem with heuristics in this environment is that they do not always allow you to make a decision that minimizes the negative economic impact while offering the best service possible to the greatest amount of passengers. Spreadsheet analysis combined with heuristics can greatly improve the quality of decision making, although this process can be very time consuming. Traditional management information systems (MIS), in most cases, provide a tremendous amount of filtered information that the operations manager can select for interpretation. The downside to MIS, by itself, is that it can complicate the situation further by providing unnecessary information, thereby degrading the probability of improved real-time decision making.What is needed is a DSS that can solicit the proper types of information and simulate the heuristics, and decision-making processes of the operations manager – all while minimizing costs and maximizing the benefit of service to the passenger.Gray & Kabbani (1994, p. 40) qualify this type of DSS as "Smart simulations supported by databases…" These systems are often referred to as Expert Systems (ES). Expert Systems are comprised of a refined set of algorithms that attempt to replicate the decision-making processes of an expert within a specific domain. ES is a type of AI and in this case attempts to replicate the decision-making processes of the airline operations manager.An example of DSS that almost approaches AI integration is the SITA Flight Operations Services Portfolio (SITA, 1997, June 19). SITA’s FleetWatch is an integrated DSS package that provides the user with real-time decision validation for scheduling, maintenance, management, and crew planning. AI is introduced by the option that allows the user to adjust parameters that complement the decision-making processes of the respective airline.
Free Flight
The concept of free flight is perhaps one of the most compelling environmental considerations that will demand the development and integration of AI-DSS within all areas of airline operations. Free flight, if implemented, will allow the airline to schedule and fly aircraft with minimal interaction with the Federal Aviation Administration’s Air Traffic Control (ATC) service. Freedom from ATC intervention will cost in terms of the ability of the operations manager to "manage by exception" (Nordwall, 1997, p.16). Management by exception will require AI systems that can not only offer solutions, but also make very accurate decisions in real-time (Nordwall, 1997, p.16).Nordwall (1997) describes how Delta Airlines faces the decision-making processes of flight operations for about 570 flights a day. He qualifies that any delay greater than 30 minutes can have significant financial consequences for an airline. Delta has determined that these types of delays cost approximately $450 million a year in losses.The benefit of free flight will be to reduce delays over the entire operational infrastructure of the airline industry. Extraordinary AI-DSS platforms will have to be created in order to handle new levels of information sharing and real-time decision making associated with this new paradigm of commercial flight operations.
SECTION 3 CASE EXAMPLES
Singapore Airlines
AI based DSS platforms are being created as commercial off-the-shelf (COTS) products by commercial vendors (see SITA above) and are being developed internally by major airlines. The major development problem for AI as applied to airline operations is developing algorithms that will accurately integrate all phases of the management process (scheduling, crew, maintenance, etc.). This process is extremely expensive, requires considerable development time, and demands very sophisticated computer resources.As a means of migrating towards AI, Singapore Airlines is integrating a Crew Management System (ICMS) (Ying, 1995, July 5). The ICMS is a stand alone AI based DSS, with limited integration of other management considerations. The ICMS is Unix based, written in C, and supported by an Oracle database. The system is designed to draw upon code that simulated " Natural Intelligence." It can make decisions automatically regarding the rescheduling of crewmembers that might have been affected by such factors as illness or maintenance cancellations. The system was projected to cost between $1.1 million and $4.3 million and was developed by ICL Airlines Group (Ying, 1995, July 5).
Southwest Airlines Co.
Southwest opted not to purchase COTS software as Singapore Airlines did. Keeping within the traditional reputation of Southwest’s efficiency, an internally developed Integrated Flight Tracking System (Swift) was developed that allows 37 dispatchers the ability to track 2,200 daily flights (Hoffman, 1996, March 25). As Of February 1995, this system had only cost Southwest approximately $300 thousand to build.The outstanding benefit of Swift is that it replaced, "…a 17 foot long flow sheet…" that took 15 minutes, or more, to annualize (Hoffman, 1996, March 25). Swift allows the same decisions to be made in approximately 45 seconds. According to Dave Jordan, Director of Flight Dispatch at Southwest, "We were missing opportunities to protect the customer because there was so much data to look for…" (Hoffman, 1996, March 25). Swift provided AI decision-making, removing the need to manually filter through irrelevant information. As of 1997, Southwest planned to add and integrate maintenance, planning, and other functions to Swift.
Delta Airlines
Delta Airlines commissioned Transquest Company to help them develop an AI system that would eliminate the excessive amount of time spent looking through large quantities of information (Nordwall, 1997, p.16). This AI-DSS automatically determines the solutions to problems such as: Which aircraft in a large holding pattern should land first? Or; Which flight(s) should be canceled or re-routed as a result of ATC flow control (Nordwall, 1997, p.16)?
United Airlines
In 1992 United Airlines initiated an AI-DSS called System Operations Advisor (SOA). From a period between October 1993 to March 1994, SOA reduced potential delays by 27,000 minutes. This was translated by United as a savings of approximately $540,000 in related delay cost (Rakshit, Krishnamurthy, Yu, 1996, March - April).At the time of implementation SOA aided operational managers on the deployment of over 2,000 flights a day on over five continents (Rakshit, et al, 1996, March – April, p. 50). SOA allows the operational manager to use a "solve button" on module programs that integrate and provide real-time decision support for delaying a flight, swapping and canceling (Rakshit, et al, 1996, March – April, p. 53). Each module acts as an ES and provides a graphical user interface (GUI) that may be used to set up models for solving specific real-time problems.SOA utilizes a mixed nodal hierarchy model, consisting of nodes and arcs. Each node represents is designated to represent arrival, ground and departure times. Each arc indicates the direction of specific aircraft flight flow, a direct route, maintenance, cancellation or the swap of alternate aircraft. Allocated costs are assigned to each arch created within each model set calculated by SOA. The slope and flow direction of each arc determines the overall affect of the incident on the final solution. After pushing the solve button, SOA provides the optimal cost solution along with other [ranked] alternatives (Rakshit, et al, 1996, March – April, pp. 54 - 57).In 1996 United decided that further efforts were warranted in the development of additional integrated AI-DSS modules that would be added to SOA. Specifically, United wanted decision assistance with the management of gates, passenger flow, and overall staffing (Rakshit, et al, 1996, March – April, p. 55). Conversations with Al Wong of United’s Operational Control Center indicated that this process in AI development was continuing in 1997 (Wong, Al, personal communication, July 10, 1997).
Cathay Pacific
As most other commercial airlines, Cathay has developed a hierarchy for strategic planning. First, Cathay’s long term strategy includes analysis of new aircraft type, evaluation of fleet plan, effects of aircraft on deployed routes, and new route studies (Yau, 1993). Medium and short run planning consist of the same considerations conducted in long range planning with the addition of seasonal and weekly ad hoc modifications. For short term planning Cathay felt that an AI-DSS platform that could offer quantitative solutions was needed in order to remain competitive. Under internal guidance, Cathay developed their Interactive Flight Scheduling System (IFSS) (Yau, 1993, p.1618). IFSS is characterized as a short term planning tool that incorporates a framework of rules and symbolic processing. For example, rules concerning maintenance, ad hoc reports, training flights, charter flights, and many others have been translated into computational values. Symbolic values are generated by the IFSS upon the user’s request. Examples are Shift Flight, Best Fit, Exchange Flights, Cancel Flight, Upgrade Flight (larger aircraft) or Downgrade Flight (smaller aircraft). The IFSS also allows the user to manually Add, Delete, Move, Exchange, and Search various flights (Yau, 1993, p.1619).An interesting feature of the Cathay IFSS is that a utility function is added to the IA processor. This function allows the airline to pre-define the probabilities for utility of affected operational attributes for each calculated solution. This allows the user to determine the highest probability of economic utility when deciding between solutions that will all solve the defined problem. In essence, the operational manager is better able to refine the IFSS selected solution within real-time decision making.
Back to Top
SECTION 4 Discussion
The AI Challenge
The challenge of using AI in airline operations management is the proper development of the mathematical modeling needed for simultaneous and integrated decision making. Vance, Barnhart, Johnson, and Nemhauser (1997, March-April) qualify this problem by describing the magnitude of difference for preparing AI algorithms for scheduling pilot crews, as compared to flight attendants:
The flight attendant problem tends to be much larger since a flight attendant may be qualified to serve on more than one type of aircraft. Thus, we may not be able to consider the various aircraft types individually. Also, the restrictions on the assignment of flight attendants to flight segments may differ significantly from those for the pilots. (p. 266)
Vance, et al (1997, March-April) describe how AI applied to flight attendant scheduling is feasible, however significantly more difficult do to the greater demand for simultaneous solution problem solving.An interesting example of developing AI algorithms that begin to account for selected environmental factors while determiningsimultaneous integrated solutions is the Flight Operations Decision Model presented by Ming and Kanafani (1996). This model integrates delay and cancellation analysis together as one discrete solution. By using this AI model, flight operations managers do not have to mentally integrate the results of a separate delay model with the results of a separate cancellation model. This model is an example of the increased development of AI as applied to airline operations management and the reduction in day to day reliance upon heuristic decision making processes.
Concluding Observations
This paper investigated and qualified the increasing demand by airline operations managers for artificial intelligence based decision support systems. It found that over the past 5 years, significant efforts have been made by the industry to develop AI platforms applied to airline operations. Examples of major airline companies incorporating AI-DSS into their real-time operational concerns have been well documented in the literature.
SECTION 5 Recommendations
Future Research
This paper has documented the partial change from traditional heuristic based decision making tools to AI-DSS platforms. Of concern to the author of this paper was the apparent lack of evidence for training efforts that will help facilitate the manager’s transition, and evolution to these new decision making tools.1 Emphasis needs to be placed on not only the formulation of new AI techniques, but the proper integration of training strategies for AI adoption.It is recommended that additional work be conducted that will offer the airline operations manager a framework of properly designed instructional tools that will help insure the successful migration towards AI-DSS platforms.
Back to Top
REFERENCES
Gray, Douglas & Kabbani, Nadar (1994, April 1). Right Tool, Right Place, Right Time. OR/MS Today, 21, 34.Hoffman, Thomas (1996, March 25). Airline turbocharges schedule efficiency. [Electric Library] Computer World, pp. 1. Infonautics Corporation.McCartney, Scott (1996, September 11). The Inexact Science of Keeping an Airline on Schedule. The Wall Street Journal, pp. B1W, B1E.Nordwall, Bruce (1997, July 16). Operations Centers Are Airline Money-Makers. Aviation Week and Space Technology, 146.Rakshit, A., Krishnamurthy, N., & Yu, G. (1996, March – April). System Operations Advisor: A Real-Time Decision Support System for Managing Airline Operations at United Airlines. Interfaces, pp. 50-58, 26.SITA (1997, June 19). SITA Flight Operations Services Portfolio. [Electronic Library]. M2 PressWIRE. Infonautics Corporation.Turban, Efraim (1993). Decision Support and Expert Systems. NY, NY: Macmillan Publishing Company.Vance, P., Barnhart, C., Johnson, E., and Nemhauser, G. (1997, March-April). Airline Crew Scheduling: A New Formulation And Decomposition Algorithm. Operations Research, 45 (2).Yau, C. (1993). Interactive Decision Support System for Airline Planning. IEEE Transactions on Systems, Man, and Cybernetics 23, (2).Ying, Tong. (1995, July 5). Singapore Airlines Moves Into Intelligent Systems. [Electric Library]. Computer World, pp. 81. Infonautics Corporation.
Footnotes
1 TransQuest, Delta's primary information-technology provider, does provide educational training for the technology associated with their DSS system.