Safe operation of key infrastructures based on chance-constrained optimization and learning

• Principal Investigator
  • Teodoro Álamo Cantarero / Daniel Rodríguez Ramírez 
• Funded by
  • Plan Estatal 2017-2020 Retos – Proyectos I+D+i

This project deals with the development of methodologies for the safe operation of key Infrastructures (KI) based on chance constrained optimization and learning, focusing on the management of HVAC systems.

To this end, firstly suitable learning methodologies for control of dynamical uncertain systems will be developed considering new estimation methodologies such as Kinky inference, direct weight optimization and kernel methods. In parallel to this task, the research on novel optimization algorithms for the safe management of key infrastructures will be carried out. Novel optimization methods for problems in which a chance-constrained setting is required will be developed. Split methods and adaptive techniques and parallelizable formulations will be researched. Another focus will be on the proposal of algorithms that can be implemented in low-end embedded hardware. These tools will be used in the rest of the subproject tasks which are briefly described next.

One task will be developing strategies to check the probabilistic satisfaction of constraints in a dynamic setting. The characterization of safe regions in which the constraints are satisfied in a chance constrained manner will be addressed by means of randomized techniques. The computation of probabilistic invariant sets for complex, uncertain systems will be addressed.

In addition, novel methodologies for the safe operation of complex systems under stochastic conditions using predictive control techniques based on data will be developed as a specific objective of the project. Safe operation, stability and constraint satisfaction will be secured. Predictive control strategies relying on model free strategies or with an explicit model will be researched, along with scenario and
other probabilistic methods. These strategies will clearly align with the fourth project objective, safe operation of KI.

Finally, techniques to assess and validate in a probabilistic way the performance of safe management and energy efficiency in KI will also be tackled in this subproject. Challenging problems, such as non-convex optimization, scarcity of data with respect to the high dimensionality of the problem and repeated validation with same data will have to dealt with. These techniques will contribute to fulfill the general objective on performance and safety validation.

All the project results will be validated on a relevant case study, a heat ventilation and air conditioning (HVAC) system of the Engineering School buildings of the Universidad de Sevilla (inaugurated in 1998) . The Engineering School of the Universidad de Sevilla will allow the validation of results on an existing infrastructure. This will require the improvement of the existing real-time information system with the addition of new sensors. . A HVAC experimental facility in the Universidad de Sevilla will be also used for a preliminary validation of the results.