Service level description and assessment

This page is based on ICoMaaS-conference article[1]. It focuses on the assessment of micro transport services that operate a fleet of vehicles. Such services are used in settings where the goal is to bridge the gaps in public transport in rural and urban areas. The main goal is to transport people between rural places and public transport hubs such as train stations. In comparison to taxis, the fees of such services are quite low. However, multiple customers are driven in the same car at once, and demands must be pre-announced. Municipalities might be willing to subside such services since they profit from affordable public transport, decreased local emissions, and avoiding car-induced traffic in neighbouring urban centres. We investigated measures for improving operating costs and user convenience of such services based on methods stemming from operations research. Trade-offs between customer travel times and operating costs are analysed and optimised for a real-world setting.

Service levels and performance indicators. Service levels of demand responsive transport services are distinguished into those that are directly observable by the users of the service, and those only relevant for the operators of the demand responsive transport systems. Regarding the service levels observable to passengers, the following main key performance indicators can be investigated, e.g.:

  • The difference between the desired time of departure (or arrival) and the departure time confirmed by the operator.
  • The punctuality in the sense of the deviation of the confirmed time of departure and the actual time of departure.
  • The detour duration is determined by the difference between the actual duration of the passenger’s trip and the shortest possible duration of that trip. Detours can appear in demand responsive transport services to combine trips of multiple passengers.
  • The number of other passengers and additional stops that take place during the trip.
  • The price of the trip paid by the customer.

In addition, there is a range of performance indicators relevant to the operator of the system, e.g.:

  • The average cost of performing a trip for the operator.
  • The composition of the fleet to be used in terms of vehicles sizes and numbers.
  • The operating hours during which each car of the fleet is used.
  • The usage of the fleet in terms of a percentage of time the vehicle is transporting passengers.
  • The average number of passengers transported at the same time.

Optimization based analysis of real-world data. For this project, a range of data was provided by ISTmobil, the operator of a demand responsive transport service in Korneuburg, Austria. This data comprises transport requests recorded over a period ranging from May 2015 to June 2016. It consists of 257 days for which operational data was available. On average, each day comprised 50 requests. A request states the start and end locations requested, as a well as the time of that request and the number of passengers to be transported. In addition, GPS data of driven routes was available. This data was used to determine instances of a variant of a pickup and delivery vehicle routing problem. Solutions for this problem then specify the routes driven by the fleet of vehicles that are operated. An integer linear programming approach and a Variable Neighborhood Search based approach was used to compute routes for a fleet of vehicles fulfilling the given requests. The considered fleet is composed of taxi vehicles. That fleet is further divided into fixed and flexible vehicles. Fixed vehicles are paid by the hour and have fixed service times. Flexible vehicles are only paid per kilometre in case they are in service. Fixed vehicles are also paid a distance-based fee, which is, however, much lower than that of the flexible vehicles. These properties are taken into account in the model. The solution method and modelling are then used to analyse a range of properties from the given data, which is presented in the following.

We varied time window durations, the vehicle capacity, and the number of flexible vehicles. In the following experimental results, if no other indication is given, time window durations of 15 minutes, a vehicle capacity of 8 passengers and zero flexible vehicles are used. Per default, pooling is allowed. This basic characteristic of micro transport services allows combining rides of multiple users. Splitting, i.e., using more than one vehicle for a request of a group of people, is not considered except when stated otherwise.

Cost reduction depending on different time window flexibility values.

The figure above shows results analyzing the impact of different time window durations. A time window models possible waiting times as well as detour durations. For both, maxima can be specified using a time-window-based modelling. Depending on the allowed duration, we observe an impact on the costs. For example, allowing a maximum combined waiting time and a detour duration of 15 minutes can reduce operating costs by around 10% on average when compared to the solution allowing no waiting time or detour durations at all. The extreme case of 1440 minutes on the right-hand side of the graph is not a realistic time one would allow for waiting. Still, it is interesting since it shows that the fleet has a capacity of subsuming much more requests, which supports the assumptions that such fleets can be operated much more efficiently if more vehicles are used.

Time 2
The frequency of time deviations in the model.

The figure above shows frequencies of time deviations observed in the experiments. It relates to instances where the maximum length of time windows was fixed to one hour. This shows that allowing one hour as a guarantee has a lesser disadvantage for most of the requests. We see that most requests are still served within well below an extra duration of 30 minutes.

Average utilisation (orange line) and the number of vehicles (blue line) depending on the number of requests.

The figure above shows the average utilisation of the cars in the vehicles fleet. It is defined in terms of the number of passengers transported. For example, a value of 0.5 means that a car is 50% empty and 50% of the driven distance filled with at least one passenger. We observe that the utilization improves in cases where there are more requests per day. This is explained by the increased possibility to combine requests. However, starting from around 40 requests per day, this number decreases again. This is explained by the increased number of vehicles that are used.

Percentage of pooled requests.

The figure above shows the percentage of pooled requests. Complementing Figure 21, we observe an increase in the percentage of pooled requests with a growing number of requests per day. This shows that the demand responsive transport service profits from a larger user base resulting in lower costs per request.

Transferability of demand responsive transport services. The aim of demand responsive transport services is to enable the region-wide mobility for all citizens and fill existing gaps in public transport. Therefore, an assessment of how the service would perform in a novel region where it is not in operation yet would be an important factor for facilitating such an extension. This assessment has several steps:

  • Assessment and analysis of individual regional mobility demand and gaps in mobility The number of potential users of the system influences the utilization of the vehicles and consequently the profitability of the whole system. Therefore, a thorough analysis of potential users of the system is the key requirement. This analysis includes factors such as the number of users, and the regularity and predictability of the usage (frequent usage with a priori ordering, or occasional usage with spontaneous ordering), but also willingness to wait before being picked up or after being delivered, e.g., at the train station, as well as willingness to share the vehicle space with others and to accept a detour.
  • Definition of accumulation/access points in the region. Depending on the number of potential users living in the focal area, access points, where users are picked up and delivered, should be determined. The more access points are defined, the more convenient the system is for users in terms of accessibility and flexibility, but the more detours might be necessary / accepted to exploit vehicle capacity usage.
  • Development of a tariff system. For the system to be profitable in a region, the tariff system needs to be adapted to user requirements. The marginal cost of accumulating users on one trip done by one vehicle needs to be traded off against the inconvenience imposed on the users in terms of waiting time or additional ride time because of a detour. Eventually, a different tariff could be offered to frequent users ordering a priori compared to occasional users ordering
  • Involvement of vehicle fleet operators/taxi To guarantee an optimal usage of vehicles, a shared fleet performing multiple trips during the day for different operators could be beneficial. The pool of vehicles would then be integrated into the scheduling system and optimally assigned to (new) requests, allowing for a balanced and cost-minimal usage of vehicles, while respecting user requirements.

Methodology and data for assessing transferability. In the following, an outline of a methodology for assessing the transferability of a service to a novel region is given:

  1. Derive test instances from input data. This includes the geographical coordinates (of the access points) of the requests as well as the distance and travel time matrix to be calculated for an underlying digital roadmap.
  2. Investigate scenarios for relevant parameter settings. This comprises the size of the operating area, the number and capacity of vehicles per time interval, additional temporal considerations (duration between desired and confirmed dates, maximal user ride time), constraints for pooling and splitting requests.
  3. Optimise test instances. Determine plans for the given instance on the scenarios to be investigated using a suitable optimisation
  4. Evaluate solutions based on key performance indicators, e.g., considering fixed and variable cost, utilisation of vehicles in capacity and time, total driven distance.
  5. Present a report containing analysis & figures as decision support.

Scalability. A scalability analysis was conducted to further assess the impact of a potentially larger user base. It addresses the question what would happen to the indicators if the service was used more broadly. For this, we used again the data from 257 days between May 2015 and June 2016 with about 50 requests per day. We generated a larger instance from these recorded requests. For this, we first accumulated all requests into one big set of 13.000 requests. From this, we generated artificial requests by random sampling without replacement. By this method, we obtain instances with 200, 300, 400, and 500 requests per day, which are then used in further computational experiments.

Impact of pooling depending on the number of requests.

The figure above shows the percentage of requests that are pooled for the instance generated as described before. We observe that the percentage of pooled requests increases with the number of requests per day. This confirms the assumption that scaling is an important issue for demand responsive transport services. The wider the user base of such a service means that the more efficiently it is operated and vehicles driven around without any passengers is avoided.

Cost comparison depending on the number of base vehicles.

The figure above compares the costs that arise for the operator of a demand responsive transport operator, based on a cost model of the real-world case in Korneuburg. The fleet is split into base vehicles and flexible vehicles. The base vehicles cause a fixed cost per hour but are cheaper per distance driven. The graph shows that the optimisation model can help as a decision aid to determine the right number of vehicles. Costs depend on the number of base vehicles that are used. If too few base vehicles are used, then too many flexible vehicles are needed to satisfy all requests. This yields higher costs due to the higher cost per kilometre of the flexible vehicles. If too many base vehicles are used, then they have a low utilisation but still, they require a fixed cost per hour to be paid. To summarise, an optimisation based analysis can help in decision making for demand responsive transport services. Various scenarios can be analysed and decisions can be taken based on the results of computational experiments.

Next: Configuration of the optimisation framework


[1] Asamer, J., Biesinger, B., Knopp, S., Nolz, P. & Reinthaler, M (2017). An optimization based analysis of a micro transportation service. 1st International Conference on Mobility as a Service, ICoMaaS, Tampere 28.-29.11.2017 Conference Proceedings 1st International Conference of Mobility as a Service, 48 – 51.

Leave a Reply