Electricity (& Gas) Market Clearing Publications
Designing Day-Ahead Multi-carrier Markets for Flexibility: Models and Clearing Algorithms
Gas and electricity can sometimes be alternative energy sources for the same purpose, for example for heating. Electricity transports more easily/cheaply than gas. On the other hand, gas stores more easily than electricity. In addition gas can be converted to electricity. These three factors: availability, transportability, convertibility determine price per location. In the Magnitude H2020 project, we wanted to capture these carrier coupling flexibilities and modelled such a unified market of gas and electricity. Our simulations for the current size of the European Day Ahead electricity market predict savings of millions of EUROs per year. Note that these savings could be captured if only the market would be organised as such in pratice.
Published Online on September 25th 2020,
by eprint arXiv:2009.12093
Testing TSO-DSO Interaction Schemes for the Participation of Distribution Energy Resources in the Balancing Market: the SmartNet Simulator
In the SmartNet H2020 project, we developed software with 7 companies, modelling large parts of the TSO (Transport System Operator) and DSO (Distribution System Operator) electricity networks in Denmark, Spain and Italy. This encompassed aggregation into market bids, market clearing and bid disaggregation. We did so under various assumed schemes of TSO DSO coordination, for estimated future levels of integration of renewables and for the four seasons, in order to conclude which scheme is best in which circumstances. To me, the main conclusion was that when line congestion occurs at TSO or/and DSO level, the related network part needs to be modelled in detail to be accurate. Network operators will know if congestion already occurs in practice. But also If either network topology or/and components are changed or producers/consumers are added/removed, it is advisable to model at this level detail. The modelled parts of Denmark had less congestion than the modelled parts in Italy and Spain.
ABSTRACT and FULL PAPER
both published by the 25th International Conference on Electricity Distribution, 3-6 June, 2019.
Smart TSO-DSO interaction schemes, market architectures, and ICT solutions for the integration of ancillary services from demand-side management and distributed generation. SmartNet Deliverable 2.2: Network and Models
This project report describes the basis for electricity network modelling towards ‘the SmartNet Simulator’, used in the next paper mentioned here. This encompasses both the bid product definitions, TSO and DSO network modelling, and TSO-DSO coordination scheme modelling.
published on the SmartNet project website on February 5th 2019.
PhD Peter Sels:
Large-Scale, Passenger Oriented, Cyclic Railway Timetabling and Station Platforming and Routing
Journal Publications
The Train Platforming Problem: The Infrastructure Management Company Perspective
Published Online on 17/1/2014
by Elsevier Journal
Transportation Research, Part B
Volume 61, March 2014, Pages 55–72
http://www.sciencedirect.com/science/article/pii/S0191261514000058
http://authors.elsevier.com/offprints/TRB1196/41cb9677601828708356947222e5732f
We deduce an macroscopic integer programming model to assign a platform to each train occupation and a route to/from that platform for each train movement, avoiding any simultaneous use of platforms by more than one occupation and avoiding any simultaneous use of any dependent routes by more than one movement. This method generates usable platforming plans without any conflict in less than a second per station. Original platforming solutions sometimes have planned 2 trains using the same platform track at the same time but even more often have two train movements using dependent routes simultaneously. Our solution have n osuch conflicts and are also more robust against delays than the original ones.
Reducing the Passenger Travel Time in Practice by the Automated Construction of a Robust Cyclic Railway Timetable
Published by the Journal: Transportation Research Part B DOI: 10.1016/j.trb.2015.12.007
http://www.sciencedirect.com/science/article/pii/S0191261515002684
This is the key paper of my PhD. Prior to this journal paper, it was thought that it was impossible to automatically compute a timetable for a whole country automatically. We calculate a time supplement (on top of the minimum needed time) for all ride, dwell and even transfer actions present in the timetable for all (196) hourly trains in Belgium. We deduce that the expected passenger time in practice, including statistically expected primary and secondary delays is 3.8% lower than the then current timetable in operation. We show that the improved timetable has less supplements in planned time on average, so this shows NMBS/Infrabel plan too many of them. They are only incentivised to have good a good robustness KPI so that the execution of the timetable is close to the planning, but the planned trajects are slower than needed. We cross-verify with a commercial simulator that delay propagation for our timetable is much less globally, both passenger weighted and train weighted, than with the then current timetable in operation. So we have both more robustness against delay and more speedy connections than the manually constructed timetable. In addition it takes only 2 hours to calculate while human planners are working 6 months on it with about 20 people.
Automated Platforming & Routing of Trains in all Belgian Railway Stations
Published by the Journal : Expert Systems with Applications
https://www.sciencedirect.com/science/article/abs/pii/S0957417416302676
This journal paper shows that we implemented the method of our first platforming paper at Infrabel as ‘Leopard’ (Lean Platforming including Routing Dependencies) tool and that it can tackle every passenger station in Belgium in one second with the called Gurobi solver, except for 3 stations that still take less than 10 seconds.
With these almost real time solving times, one can imagine this method could be used in real time, even if it is a macroscopic one. Note however, that the model assumes fixed train arrival and departure times, so in a real time context they would have to be accurately measured or/and predicted for the method to be applicable.
Conference Publications (b)
Practical Macroscopic Evaluation and Comparison of Railway Timetables
http://www.sciencedirect.com/science/article/pii/S2352146515002033
It all begins with an idea. Maybe you want to launch a business. Maybe you want to turn a hobby into something more. Or maybe you have a creative project to share with the world. Whatever it is, the way you tell your story online can make all the difference.
Light Rail, Tram and Bus Timetabling Minimising Passenger Time in Practice
Abstract, Accepted by the Conference TRISTAN. Oranjestad, Aruba, 13-17 June, 2016
We show that our cyclic timetable procedure devised for railway timetabling can also be applied on bus timetabling. In fact, bus timetabling is less challenging computationally speaking, since overtaking is possible everywhere at the hunch of a bus driver and so does not have to be planned nor modelled. This could allow to add the spreading in time of alternative buses constraints researched before, but which we found out, was computationally too hard for full incorporation at full scale into the railway cyclic timetabling model.
Optimal Temporal Spreading of Alternative Trains in Order to Minimise Passenger Travel Time in Practice
IAROR CONFERENCE, TOKYO, 23-26 MARCH, 2015
Presentation held at Conference
It is well known that spreading equidistant in time of alternative trains from A to B generates the lowest average waiting time for passengers from A to B. We analytically derive a model that shows how the stochastic expected passenger time depends on the planned time in between these alternative trains. We show this model obtains lower passenger wait times when applied on the set of 26 most important passenger trains in Belgium and 12 sets of alternative routes. However, when applying the same technique for more trains and sets of alternative routes, even with the MILP solver Gurobi solving time gets excessive and we could not produce solutions yet in a practical time.
Towards a better Timetable for Denmark Reducing Total Expected Passenger Time
Presentation held at the CASPT Conference, Rotterdam, 19-23 July, 2015
During his presentation at IAROR Copenhagen, on May 13-15 2013, Daniel Sparing claimed that it was still impossible to produce a cyclic timetable for a full reasonably size country. I challenged this because we had proven it to be possible for Belgium. From this came a cooperation with BaneDanmark and we were able to -very efficiently- produce a timetable for all hourly trains in Denmark in 1 hour of computation time. The paper reports on this work. Interestingly, contrary to the Belgian case, our passenger optimal solution, on average, uses more time supplements than the manual solution in operation rather than less.
Automatically and Quickly Planning Platform and Route of Trains in Railway Stations
Presentation held at IFORS, Barcelona Conference, 14 July, 2014
We integrated our platforming and routing solution at Infrabel, which enables application on it for all stations in Belgium. We show that it returns solutions for 530 out of 533 cases, when we use the Gurobi MILP solver, we get results in less than a second. The 3 other cases solve in less than 10 seconds. With the solver CPLEX we get 1 case taking up to 50s and one up to 130s. With XPRESS, we get 1 case taking up to 20s and 1 up to 130s and one station that is only suboptimally solved after 2 hours.
Manually planned existing ‘solutions’ have errors (two simultaneous movements on the same/dependent routes or two simultaneous occupations on the same platform track), our automatically reached solutions never have errors. Robustness in terms of time in between reuse of the same resources is also better with our solutions than with existing ones.
Conference Publications (a)
Automatically and Quickly Planning Platform and Route of Trains in Railway Stations with a Case Study of Mechelen Station.
International Journal Conference on Civil and Transport Engineering
ISAET'14, Bangkok, Thailand, 1-2 January 2014
We describe our new Integer Programming based method to quickly assign a platform track and a route t/from it for every planned train movement, respecting also its already assumed planned fixed arrival and departure time. We also compare the current manual solutions to our generated solutions. Our graphical representations of both solutions visually clearly indicate errors and robustness warnings for planners. We apply our method on the Mechelen station case with actual traffic. Current solutions for Mechelen have errors (two simultaneous movements on the same/dependent routes or two simultaneous occupations on the same platform track), while our solution does not have a single have errors. This is guaranteed by the constraints coded into the optimisation model. Robustness in terms of time in between reuse of the same resources is also better with our solutions than with existing solutions.
Automated Passenger Time Optimal, Robust Timetabling using Mixed Integer Programming
IWHIR 2011, Shenzhen & Hong Kong, China, 19-22 July 2011
Presentation held at Conference
Weighing with the passenger flow numbers obtained from ticket sales by the method described in our earlier publication, we now attack the problem of cyclic timetable generation for a subset of all hourly train in Belgium.
A Passenger Knock-On Delay Model for Timetable Optimisation
MT_ITS 2013, Dresden Conference, 2-4 December 2013
Presentation held at Conference
For the application of the generation of a timetable that is robust against frequently occurring primary delays, we need to avoid that trains that run subsequently over a common resource (track/switch) too close together. To be able to ensure this, we construct a model that contains the dependency in terms of probabilistic passenger delays on the planned time between the first and second train. We analytically derive this dependency called “knock-on model” in this paper. For primary delays, we assume our earlier model with a single settable parameter.
Calculation of Realistic Station Capacity by Platforming Feasibility Checks
MT-ITS 2011, Leuven, Belgium, 22-24 June 2011
Presentation held at Conference
Some research tries to define a capacity for a station without considering the specific traffic that has to be absorbed by it. We think this cannot be done and neither would it be a practically useful research question. In practice, the real question is: “Can this particular traffic with these origin and destination lines and specific timing be mapped or not?” This is what we study and solve here.
Expected Passenger Travel Time for Schedule Evaluation & Optimisation
IAROR 2013 CONFERENCE,COPENHAGEN, DENMARK, 13-15 MAY 2013
Presentation held at Conference and at PhD Seminar March 26 2013
In optimisation problems, sometimes the goal function used is/has to be a simplification of the one one really wants to strive for in evaluation. This simplification is done so speed up the computationally harder process of optimisation versus evaluation. In our application of time table construction, we simplify by linearisation and by ignoring passenger flows smaller than a threshold number. Of course, post optimisation we evaluate the optimisation solution by using the evaluation function in full detail.
Deriving all Passenger Flows in a Railway Network from Ticket Sales Data
Extended abstract Orbel 2011, Ghent, Belgium
10-11 Feb 2011
Full Paper, IAROR 2011, Rome, Italy1, 16-18 Feb 2011
Presentation held at both conferences
To construct a timetable that minimises expected passenger travel time for all passengers, one must first know how many people are on each train for which ride and dwell actions and also for which transfer actions between each couple of trains. We propose a new method and apply this on all railway passengers in Belgium with subscriptions, mapping them on the available existing train timetable.
Short Course on Railway Planning Solutions, Presented at the IAROR Conference, Tokio, 23 March 2015
Timetabling for Railway Passengers
Presentation held at the Conference
This summarizes our (cyclic) timetabling work, where we stress that the expected passenger time is the objective to be minimised and so, is to be taken as the goal function in MILP optimisation.
Platforming and Routing: Quickly and Automatically
Presentation held at the Conference
We show that with our MILP model we were able, per station, to assign to a maximum subset of the train movements to be planned, both a platform track and a route to/from without any conflicts, and this for every passenger railway station in Belgium.