Claudia Linnhoff-Popien, Ralf Schneider and Michael Zaddach (eds.)Digital Marketplaces Unleashedhttps://doi.org/10.1007/978-3-662-49275-8_40
40. Hub Airport 4.0 – How Frankfurt Airport Uses Predictive Analytics to Enhance Customer Experience and Drive Operational Excellence
Rolf Felkel1 , Dieter Steinmann1 and Frank Follert1
(1)Fraport AG, Frankfurt on the Main, Germany
Rolf Felkel (Corresponding author)
Email: [email protected]
Dieter Steinmann
Email: [email protected]
Frank Follert
Email: [email protected]
40.1 Introduction
With more than 61 mi. passengers in 2015, Frankfurt Airport is Germany’s largest airport and a leading international traffic hub in the heart of Europe. At Frankfurt Airport, passenger demand has been growing incessantly over the past decades, pushing the capacity utilization of the airport infrastructure, such as security checkpoints and passport control, towards design limits. Currently, a major expansion program is in progress, including the construction of a new terminal. In parallel, latest advancements in technology are being implemented to support airport operations and to provide the passengers with a great travel experience.
In this context, Big Data technologies, and most notably predictive analytics, are a key success factor to enable Fraport, owner and operator of Frankfurt Airport, to continuously improve operational processes,
to obtain a sound basis for business decisions, and
to increase customer satisfaction.
Based on its integrated business model, Fraport is in the unique position to correlate and integrate data from various business domains such as flight data, ground handling data, and retail data and passenger data. This exclusive source is subsequently used to obtain new insights that serve as a basis for deriving consequential and adequate action strategies.
40.2 Passenger Flow Analysis
Smallest possible waiting times, e. g. at security checkpoints, are a critical factor for passenger satisfaction. In order to minimize waiting times, Fraport developed an integral strategy for advanced passenger flow management. It is based on the fact that the waiting time at a specific process point (such as security checkpoints or border control) is predominantly determined by the capacity, in terms of throughput, of the process point as well as the number of passengers appearing at the process point in a specific time frame. Core strategy of the selected approach is to continuously balance the effective throughput capacity and current passenger demand. Implementation of this procedural method is realized by either adapting the capacity to the expected demand or by redirecting the demand to process points with capacity available. This presupposes to permanently obtain an overview of the current passenger situation in the terminals as well as to predict its future development over the next hours – especially in terms of the demand at the various process points. From a conceptual point of view, Fraport’s approach to passenger flow management comprises three consecutive steps:
The first step is to measure the passenger flow through the terminal at multiple process points. For this purpose, data is collected by means of various sensor technologies including video based solutions and boarding pass bar code scanners.
In a second step, forecasts of the future demand at different process points are calculated. This process uses sophisticated models to predict the number of passengers on each individual flight as well as the transfer relations between all flights. These are used to compute quantitatively corroborated passenger flow prognostications between different gates, gate areas, concourses and terminals. Finally, advanced technologies such as agent‐based behavior simulation are applied to calculate the expected movement of the passengers through the terminal, allowing conclusions on the future demand at individual process points.
In the third step, the resulting data from the measurement and forecasting functions are taken as a basis for decision making processes, either affecting the staffing of the process points or passenger guidance via dynamic signage for wayfinding in the terminal. Each of the individual steps is described in further detail in the following sections.
For an adequate calculation of the future demand in step two an interaction between the three disciplines measuring, forecasting, and controlling the passenger flow is compulsory, as shown in Fig. 40.1. A more detailed insight in the whole passenger flow management is given in [1].
Fig. 40.1Interaction of Disciplines
40.2.1 Measurement of the Passenger Flow
Fraport has been exploring various technologies to measure the actual passenger flow through the terminal building over the last decade, thereby gathering significant expertise in this challenging discipline. While some of the technical solutions that were implemented, tested and evaluated proved to be promising or adequate, others were removed after a few years, since technical developments or changing conditions inside the terminal negatively influenced the results with the technology. Fraport decided to make the experience gained in implementing and utilizing various passenger flow measurement technologies available for airports worldwide by coauthoring a paper on Best Practice on Automated Passenger Flow Measurement Solutions (version 1.1) [2] issued by ACI World1. Please see [3] for further information. Finally, two solutions have proven to deliver sufficient quality and reliability for permanent operational use at Frankfurt Airport: the boarding pass scan and a camera based passenger counting solution.
Passengers proceeding through a boarding pass checkpoint scan their 2D barcodes on the boarding passes they are carrying in digital form on their smartphones, or as print‐outs. After the validity of the boarding pass was successfully checked, an anonymized excerpt of the data is transferred for further processing. Nevertheless, the majority of passenger flow related information is gathered with a video detection technology. It can provide data on the amount of passengers in a queuing area or at the checkpoint in close to real time. The specific solution used at Frankfurt Airport does not deliver video streams over the local IP network, but provides the results of a counting process conducted internally in the cameras.
The raw data delivered by the boarding pass scanners and the passenger flow measurement cameras are used to calculate several performance indicators, such as waiting times, throughput, arrival profiles2, or area occupancy. Most of the measured and all of the calculated data is stored in the central data warehouse BIAF (Business Intelligence Architecture Framework) for further processing and analysis. This powerful BI platform supports all operational processes managed by Fraport at Frankfurt Airport and contains detailed historical data from the last decades as well as up to date process information from all relevant processes including ground handling. Consequently, the second step of the integral passenger flow management, calculation of forecasts on the future demand, is also implemented on the BIAF platform.
40.2.2 Forecast of the Passenger Demand at Process Points
The function calculating the passenger demand forecast is the core of the entire passenger flow management solution. It consists of four modules: the data preparation, the forecast of the passenger numbers on board of a flight and the number of transfer passengers between flights, the simulation of the passenger flow inside the terminal building, and finally the visualization of the results on a sophisticated HMI3.
The data preparation module selects and prepares the historical data for calculating the subsequent forecast. Significant attention to this process step is necessary to achieve high data quality, which is automatically and manually double checked. This process requires diligent quality assurance measures, as inaccuracies at this stage would potentially render the final result inaccurate or incorrect.
The forecast module predicts two important variables: the number of passengers on board a future flight, represented by its flight n
umber, as well as the transfer relations between inbound and outbound flights, in terms of how many passengers from an inbound flight A will transfer to an outbound flight B. This is a complex and comprehensive task, given the between 1200 and 1500 passenger flights operated in Frankfurt per day. The forecast takes into account various external influencing factors such as school holidays, public holidays, seasonal effects, and fairs. The transfer relations are of major importance since more than 50% of all passengers in Frankfurt are changing flights in Frankfurt and continue to their final destination. An unprecedented level of forecast precision was achieved by combining multiple statistical methods, including decision‐trees, linear regression and multiple imputation.
In order to calculate future passenger demand at specific process points, a powerful simulation platform was erected. It operates on a detailed model of the terminal buildings, including security checkpoints, boarding pass control points, boarding gates as well as the corridors, stairs, elevators and terminal areas connecting them. By combining state‐of‐the‐art simulation technologies, i. e. discrete event processing and agent‐based behavior simulation, Fraport was able to achieve an adequate quality of the simulation results required for the subsequent process steps. The event discrete simulation is used to model process points (e. g. security checks). The future capacity of the process points is required as input data for the modelling process and is obtained from the relevant staff planning and scheduling systems. The agent‐based simulation models the movement of passengers through the terminals between the successional process points. For this purpose, probabilities for different connecting paths are assessed in order to represent the complex topology of the terminals, especially in Terminal 1. The processing time needed to simulate the entire airport for one day of operation was gradually reduced to less than two minutes by using an optimized behavior model for the agent‐based part of the simulation platform. An integrated feedback loop analyzing discrepancies between the forecast and the subsequently measured passenger flow, is used to continuously improve the parameters used in the highly sophisticated simulation approach. The simulation phase results in detailed predictions on the number of passengers expected to reach a specific process point at a certain point in time in the future. This corresponds to the expected demand at the process points.
Fraport’s IT department and experts from the Terminal Control Center (TCC) have jointly specified and developed a powerful HMI on the BIAF platform (see Fig. 40.2 Passenger Flow Analysis (PFA)), depicting the predicted demand and the measured passenger flow information at a process point. The HMI provides decision makers in the TCC with all information they need to efficiently manage the complex passenger flow at Frankfurt Airport, including the number of connecting passengers between flights and the corresponding connection times.
Fig. 40.2Passenger Flow Analysis (PFA) HMI
40.2.3 Managing the Passenger Flow
As depicted in Fig. 40.2, the decision makers in the TCC have access to all relevant information, the measured passenger flows, process point specific demand predictions, and transfer information, via a single highly integrated HMI. It enables the responsible staff to take corrective action in case demand and capacity are predicted to get out of balance at specific process points.
In principle, TCC staff has two options to tackle imminent imbalances. First, personnel can be instructed to move from one process point to another or to either open or close specific lanes at checkpoints. Second, passengers can be actively guided to alternative but functionally identical process points with sufficient remaining capacity immediately available. This process is implemented via 102 pre‐defined wayfinding scenarios through the terminal buildings available to the TCC staff. Whenever a specific scenario is selected, in excess of 100 dynamic signage displays in more than 40 locations provide unambiguous guidance information to passengers.
40.2.4 Results and Next Steps
The innovative passenger flow management solution implemented at Frankfurt Airport has been leading to significant service improvements. Moreover, the data generated in the flow management process can be used on various communication channels to inform passengers as well as so called ‘meeters and greeters’ on the current situation in the terminals. Users of the Fraport Mobile Passenger App (FRA App) [4, 5] have for instance access to up to date measured queue times at security checkpoints. The same information is presented to passengers inside the terminal via displays. In the future, passenger flow management information and individual guidance instructions may also be integrated with way‐finding and indoor navigation solutions in the FRA App.
40.3 Smart Data Lab
Fraport has recently started a new and innovative initiative for investigating new ideas for applying Big Data and predictive analytics to business problems (commercial as well as operational): the Smart Data Lab. In this environment, interdisciplinary teams of data analysts, mathematicians, business process experts, and IT specialists successfully work on clearly delimited business challenges with high relevance in a predefined time frame of several weeks. The most promising results achieved during a Smart Data Lab period will be turned into regular implementation projects in order to thoroughly prepare a sustainable utilization of the findings in daily operations or business.
The following elements were identified as key success factors for the Smart Data Lab: it is a mandatory prerequisite to provide the interdisciplinary working groups with unlimited access to all data the company possesses,
the teams need to be composed adequately,
the lab has to constitute a protected environment that accepts initial failure as progress towards the best possible solution, and
a powerful analytic toolset is needed as well as agile working methods.
With regard to the first success factor “unlimited access to all data”, Fraport has established two comprehensive centralized BI data repositories (the Business Intelligence Architecture Framework BIAF for operational data and SAP Business Warehouse for administrative/commercial data), which have been collecting relevant business data for many years and keep them available for detailed analysis through various channels and tools. Based on its integrated business model, Fraport owns data from various areas of the airport business’ value added chain, including flight data, ground handling data, passenger flow data, retail data etc. It is essential that the Smart Data Lab team has access to all data across business units, as this allows for the identification and investigation of interrelations between different data sets and the related business processes.
The second key success factor is related to the best possible composition of the interdisciplinary teams. Besides experts in data analytics, the team needs to comprise representatives with detailed knowledge of the business processes related to the operational problem under investigation, such as business analysts and process experts. Moreover, the ICT department provides technical and methodical assistance – especially concerning the data repositories and the available tools.
Thirdly, it is important that the team is able to operate in a protected and non‐constraining environment. This allows for the formulation of new hypotheses unpersuaded by typical limiting factors such as the previously established approach or political influences. Nothing is stigmatized as off‐limits and the teams can use the available data to evaluate any correlation that appears constructive. Moreover, it is very important that failure is an accepted outcome of the investigation phase. The earlier an approach fails, the better, since it is then more likely that the team is able to find a better approach towards the right answer in the remaining available time.
In addition, the industry‐leading BIAF platform provides a powerful toolset for customized analysis and reporting. One of the modules is a visual analytics tool that enables the user to disp
lay and explore georeferenced data as shown in Fig. 40.3. A typical exemplary application based on these capabilities is a visual analysis of the distribution of retail sales in different terminal areas. With the help of this tool set, data analysts can drill down into data sets in close to real time in order to identify superordinate patterns and trends.
Fig. 40.3Visual Analytics
Finally, the working methods in the Smart Data Lab follow agile principles, e. g. by using a Kanban board to illustrate tasks and progress and by holding short daily (stand‐up) meetings to facilitate team communication and to scrutinize progress. Fraport has been making repeated use of the Smart Data Lab and achieved remarkable results. Some examples are presented in the following subsections.
40.3.1 Recognizing Trends in Retail
Today, non‐aviation revenues contribute significantly to the overall financial success of an airport. An important source of non‐aviation revenue is retail, where airport operators typically earn a share of the retail revenues of the stores in the terminal buildings as a concession fee.
Against this background, the Smart Data Lab was tasked to derive a method to forecast retail revenues and to identify deviations from earlier predictions. Based on a statistical analysis of historical data, it was found that the overall retail revenue can be predicted with high confidence based on only two parameters: number of passengers and time.
Digital Marketplaces Unleashed Page 62