Older projects

ATLAS: modelling short-term European electricity markets

The ATLAS model simulates the various stages of the electricity market chain in Europe, including the formulation of offers by different market actors, the coupling of European markets, strategic optimization of production portfolios and, finally, real-time system balancing processes. ATLAS was designed to simulate the various electricity markets and processes that occur from the day ahead timeframe to real-time with a high level of detail. Its main aim is to capture impacts from imperfect actor coordination, evolving forecast errors and a high-level of technical constraints--both regarding different production units and the different market constraints.

During my work-study contract at RTE, I was involved with the implementation of the "day-ahead-orders" module of the simulation model. This module consists in simulating the formulation of market orders. This formulation aims at maximizing the market participant's profit, but should reflect the technical constraints of their portfolio of means of productions. A particular emphasis was laid on implementing an optimization program for the formulation of thermal orders, aiming at reflecting the technical constraints of these technologies. This optimization program for thermal units relies heavily on mixed-integer linear programming (MILP).

The deliverable resulting from the project is accessible here, and the supporting preprint is accessible here.

Reference:

• Little, E., Cogen, F., Bustarret, Q., Dussartre, V., Lâasri, M., Kasmi, G., ... & Bourmaud, J. Y. (2024). ATLAS: A Model of Short-term European Electricity Market Processes under Uncertainty. arXiv preprint arXiv:2402.12848.

Counting wolfpacks in Finland

In January 2019 the University of Jyvaskyla and the National Resources Institute Finland (Luke) started a collaboration to implement a new approach for estimating the wolves population in southern Finland. The end goal is to its aim is to propose an online assessment of the Finnish wolf population using a multiple target tracking algorithm based on a dynamic statistical model.

Currently, estimations are carried out using yearly surveys and population simulation models are used to fill the gap between two rounds of estimation. The model is run using different specifications and when new data is available, prediction outside the observed realizations are discarded. Fresh data then serves as the new initial conditions of the model. This method allows for online and frequent estimation of the number of wolves.

On the other hand, as part of the wolves monitoring policy in Finland, Luke maintains a database of wild animals observations. This database, called TASSU, is updated on a daily basis and the idea would be to develop a statistical model that could provide a real-time estimation of the number of wolves packs. This estimation would then be available to the public through a dedicated website. Besides, compared to other approaches (e.g. surveys or GPS tracking) this method, if proven reliable, could be much more cost-efficient.

The Department of Mathematics and Statistics focuses on developing the estimation algorithm based on a statistical model. In order to do so, it is necessary to first construct a dynamic and an observation model. The former describes how new packs are formed or disappear while the observation model links the latent dynamics to the actual observations that feed the TASSU database. Among others, this observation model needs to takes into account the fact that all packs are not necessarily immediately observed.

From a mathematical perspective, this setting can be seen as a filtering problem, i.e., a dynamic situation where incomplete signals (the observations) are received from a latent unobserved system (the wolf packs population). These observations can be noisy and yet one wants to use them in order to provide a ”best estimation” of the current status of the system.

More precisely, the filtering method that was chosen is a particle filter (or sequential Monte-Carlo) that allows computing Monte-Carlo approximations when dealing with filtering problems. The construction of this filter is gradual. The starting point is to simulate the dynamic and observation models in a very simple manner and to see if inference in this context is feasible. The main question of this internship was to know whether inference could be made in the context of wolf packs tracking.

The internship report can be accessed here and the resulting publication is accessible here.

Reference:

• Karppinen, S., Rajala, T., Mäntyniemi, S., Kojola, I., & Vihola, M. (2022). Identifying territories using presence-only citizen science data: An application to the Finnish wolf population. Ecological Modelling, 472, 110101.

APE's master thesis

Abstract: What incentives do politicians respond to? Based on aggregated data of the French National Assembly activities, we constructed a dataset designed for estimating the impact of the majority. Our fixed-effect model allows us to document a 4.67% increase in the probability to attend committee meetings and a decrease in 6.48% in the probability to intervene in floor meetings associated with the majority. The qualitative pattern argues in favor or a majority focused on moving the Government’s agenda forward while leaving to the floor to the opposition during public meetings. Thus, it suggests that the subordination of the National Assembly to the Executive branch originates from the subordination of the majority group. These results provide renewed evidence of what has been highlighted by earlier literature as a distinctive pattern of the National Assembly and show how differences in political statuses can influence otherwise similar deputies.

You can access the master thesis here and the Beamer presentation here. For this project, I used the data gathered from the non-profit association Regards Citoyens and accessible on the website Nos Deputés. The supporting dataset in .dta (Stata) format can be accessed here: