Challenge 1 - Analyzing and predicting football players and team performance with movement tracking technology
Performance management of football players is a complex process that involves optimizing their physical performance, tactical training, influence in the game, and minimizing the risk of injuries while maximizing the overall team performance. The analysis of players’ tracking data has gained a lot of popularity in team sports. However, the use of movement tracking data to inform the design of training programs and understand activity level differences between training and match sessions is still an open research question. The use of GPS (Global Positioning System) incorporating inertial sensors has increased the amount of data available for coaching and backroom staff in order to monitor the activity and performance of players in training sessions and matches.
The software manufacturers of this tracking technology collect multiple physical variables of the training sessions and the performed tasks, such as the distance travelled above 19.8 km/h, the accelerations and decelerations or the number of sprints. However, considering the importance of technical and tactical aspects in football, the analysis of physical data in solitary have limited usefulness.
A GPS provides data on the position, orientation and acceleration of each player every tenth of a second. Therefore, a vast amount of raw data must be processed and analyzed in order to measure important aspects of the game, report any significant changes, obtain more advanced physical information, tactical analysis and combine all that information to generate insights that may help coaches predict how their teams perform in a game.
Considering the tracking technology and the data available, the challenge would contribute to pave the way for data scientist and team managing staff to extract knowledge from in-game data in the form of KPIs (Key Performance Indicators) that may contribute to provide important tactical aspects of football (e.g.: effective playing space, space control gain, pressure passing efficiency, maximum performance peak or reached threshold of exertion of a player, etc.).
Challenge 2 - Digital Twins: Merging physical and experimental knowledge
The development of reliable physical models that allow predicting the behaviour of a system under different circumstances is very important for activities such as condition-based maintenance, improved design, optimisation, etc.
The development of these systems is based usually in physical knowledge (fundamental laws) or in field data (Machine Learning techniques). Fitting both types of model is difficult, on the one hand, because the mathematical laws that represent the real process do not consider all the phenomena and, on the other hand, because some of the parameters of the model are not known accurately (only a rough estimation is available). In general, the merging of both types of knowledge is not done or is done at a very basic level, e.g. by adjusting some parameters of the physical model in order to improve the correlation of data and model results simply by trial and error.
The proposed challenge consists of defining a suitable procedure to produce simulation models that combine the theoretical knowledge about the physical phenomena associated and a big amount of field information on the behaviour of such a system. No constraints are imposed on the selection of the mathematical techniques to be applied for the task. The procedure will be illustrated with an application to the air consumption and generation system in railway vehicles. A physical model of the system will be provided, as well as operational data of its real behaviour will be provided by CAF. Starting from this information a technique to combine these models and the operational data to produce the best digital model will have to be derived.
Challenge 3 - Improving the prediction of customer abandonment
One of the main challenges in a telecommunications company is to reduce the rate of customers who leave the company, so the main objective of this project is to improve the prediction of current abandonment rate at Euskaltel group.
Thus, we propose three complementary work areas that pursue the goal of reducing the number of customers who leave the company, which may be covered by different profiles.
1. Improvement of the current propensity model of Euskaltel. Nowadays, based on internal customer data, every month we study who will be the most likely customers to leave the company to call them the following month. Each of these telephone calls implies an important cost for Euskaltel group and that is the reason why we have to be very efficient.
We would provide such model and the data to: ➢ Rethink the model. Is the model used the most appropriate one? Could we improve it? Should we use another one? Which one? Why and how? We must bear in mind that in addition to predicting, we want to be able to explain which are the variables that most affect the churn rate. ➢ Find solutions that raise an optimal cut-off point. The number of calls that we can assume monthly ranges between 1,000 and 2,000. How can we determine the optimal cut-off point considering this requirement? If we could expand this limitation, how could we do it?
2. Imaginative search for external variables to help us improve the model. The current model is generated from internal variables, the question is what external information could we get to improve such model. How could we extract it? From where? How much improvement would we get with this information?
3. Creative visualization of results. What we are looking for is the most creative and simple way to understand the results of the model and the correlations between the main variables.
Challenge 4 - Optimal design of the electric vehicle’s public charging network
The issue of locating and securing the availability of charging infrastructure becomes a complex question which needs to be answered in order to ensure the correct integration of the EV. In this matter, Iberdrola plans to commission more than 200 charging stations (one per 100km) all along the main motorways, traffic corridors and public areas in the major cities of Spain's geography in 2019.
The charging stations will have 50kW (fast), 150kW (super fast) and 350kW (ultra fast) of power to charge most of the battery in approximately 20 to 30 minutes, or just 5 to 10 minutes at ultra fast recharging stations, depending on the vehicle. Furthermore, each station will be able to charge between 2 and 7 vehicles at a time, bringing the number of fast charging points to 400 throughout the network.
However, the deployment of these charging infrastructures is a challenging issue. It requires strategic planning to locate the right number and sizing of charging stations in the right locations. The correct design of the whole infrastructure will allow to charge the vehicles in a more dynamical way reducing the battery capacity needs and therefore optimizing the net social benefit.
Since studies give different weights to different constraints, the optimal deployment results are different. It is largely subject to geographic constraints, temporal vehicles flow on the mains roads, socioeconomics characteristics on the nodes, EV adoption curve and coordinates matrixes. However, Iberdrola has the target of reaching nationwide coverage of every province capital and main cities of Spain and ensuring the availability to travel with the EV around the Spanish geography.
The expected outcomes are the optimal locations and sizing (total power and level of service considering 150kW and 50kW DC chargers) for a fixed number of public charging stations considering the traffic flows of the Spanish six main roads (A1-A8 and A66) using the mathematical tools that are deemed most appropriate by the researcher (e.g. flow-capturing models, genetic algorithm model or clustering algorithms).
We believe that companies benefit a lot from the interactions between mathematicians and industrialists.
If you think that your company may have an idea for a collaborative project, please do not hesitate to send an email to firstname.lastname@example.org