The computational simulation of physical phenomena is an extremely complex and expensive process. In particular, simulating the energy performance of a building (how it responds to changes in internal and external conditions regarding HVAC and lighting systems) requires the development of costly models based on energy transfer equations. Traditional simulation models do not provide results in sufficient quantity and speed to enable automatic decision-making. Conversely, a data-driven simulation (DDS) model requires less effort to build and less time to execute. However, existing approaches in the literature, based on classical statistical techniques for time series prediction, have numerous limitations: they require encoding knowledge of the underlying physical system (the building, in this case), handle a limited number of variables, and are difficult to extend to different scenarios.

D3S will explore new Deep Learning techniques to build faster, more accurate, and realistic DDS models using neural networks to simulate the effects of HVAC systems in buildings. The project has three main objectives: (1) the development of new Deep Learning architectures for data-driven simulation, (2) the characterization of the relationship between a Deep DDS model and the corresponding physical system, and (3) the integration of capabilities to handle imperfect information. This approach represents a paradigm shift compared to the traditional modeling process, ensuring that the project outcomes, published as high-quality scientific contributions, will have a significant impact at both regional and international levels, opening new opportunities for improving efficient building control.

Computational simulation of physical phenomena is an extremely complex and expensive process. In particular, the simulation of the energy behavior of a building (how it reacts to changes in the internal and external conditions of the air conditioning and lighting elements), requires the creation of expensive models based on energy transfer equations. Building controllers based on Artificial Intelligence often require the running of long series of simulations representing what-if scenarios.

With traditional simulation models not allowing to generate data in sufficient quantity and speed, and being executed sequentially, a paradigm change is required. A first approach comprises of reduction in the time taken by individual simulations. In another project, we leverage Deep Learning techniques to automatically create accurate simulations models from historical data. One of the key properties of such simulation models is that they can be run almost instantaneously.

A second approach consists of executing all simulations at the same time employing a massively parallel approach. Commodity cloud computing offerings enable this. Consequently, SPEEDYs departing hypothesis is that it is possible to highly reduce the overall execution time of a large set of energy simulation runs by means of leveraging commodity cloud resources and employing massive distribution of work.

The project is specifically focused on the development of new algorithms that allow a fast, accurate, and realistic simulation model of a building to learn automatically from the data.

The project develops new tools and methods to manage and analyze multisource data to predict and optimize energy production and demand.

DeepSim researches and develops new Deep Learning techniques to automatically build faster, more accurate, and realistic data driven simulation models(DDS) using neural networks(Deep DDS, D3S). DeepSim will generate various methods, computational models, algorithms and software for learning D3S simulators. These contributions will be applied and validated in three use cases: the action of a robotic arm, the energetic behavior of a building and the dispersion of polluting particles in the air.

The PERCENTAGE project pursued the development of advanced Deep Learning techniques to model and predict energy consumption in protected buildings, with the aim of optimizing energy efficiency in those locations.

This project has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement no 754446.

Analysis of the propagation of multimedia disinformation during the pandemic. Social Data Science: advanced data analysis to characterize a social phenomenon.

Collaboration between universities and research centers in Spain and Portugal. Our group leads an activity on technologies and participates occasionally in others (strategic analysis, coordination, digital literacy, dissemination of results).

This project has received funding from the European Commission under the CEF-TC-2020-2 (European Digital Media Observatory) agreement with reference 2020-EU-IA-0252.

The project focuses on minimizing the possibilities of spreading false information to give credibility to the digital transformation and achieve the development of sustainable and resilient societies, based on sound economic and social structures. In line with this main objective, this project will provide relevant knowledge to public and private actors to address misinformation processes in general and in particular in the field of health.

Grant PLEC2021-007681 funded by MCIN/AEI / 10.13039/501100011033 and by European Union NextGenerationEU/PRTR.