SIAM CSE23 conference took place between February 26 – March 3, 2023, in Amsterdam, the Netherlands, which included a joint mini-symposium of HiDALGO2 and EXCELLERAT P2 CoE of the EuroHPC Joint Undertaking. This event granted us a great opportunity to showcase the results and methods of two HiDALGO2 use cases for the scientific community and to build an international network of HiDALGO2 stakeholders.
The Society for Industrial and Applied Mathematics (SIAM) is an international community with more than 14.000 individual members. SIAM has been established in the US in 1954 and aims to foster the development, industrial and practical application of applied mathematics and computational sciences. The biggest Activity Group of SIAM is Computational Science and Engineering, CSE with 2000 members. Its conference has been organized first time outside US, in Europe this year.
The organizing co-chairs of SIAM CSE23 were Karen Devine (Sandia National Laboratories, US), Dirk Hartmann (Siemens Industry Software, Germany), and Wil Schilders ( Eindhoven University of Technology, The Netherlands). Prof. Zoltán Horváth (Széchenyi István University, Hungary), leader of WP5 and use case Urban Air Project in the HiDALGO2 project also participated in the organizing committee.
More than 420 mini-symposia took place at the conference, and among them was the HiDALGO2 – EXCELLERAT P2 joint mini-symposium, MS192 HPC Applications for Tackling Global and Engineering Challenges Towards Exascale. This event was co-organized by Zoltán Horváth (SZE) and Wojciech Szeliga (PSNC) and included 5 presentations.
Mátyás Yves Constans presented “Real-Time Digital Twin for Urban Air Pollution: Method, Software, and a Use-Case” as the representative of the HiDALGO2 SZE team.
SZE has been developing and optimizing a workflow with automated deployment of containerized solutions to HPC platforms with preprocessing (OpenStreetMap), simulation (OpenFOAM – scales up to 100k CPU cores), and visualization, which will help decision-makers and urban planners to detect and evaluate local hot-spots and decrease negative health effects in the Urban Air Project (UAP) use case and previous projects (HiDALGO, FIEK, TKP).
The own REDSIM GPU code has been presented for urban airflow simulation, full and reduced order, including mathematical background, visualization, and benchmarks.
Real-Time Digital Twin for Urban Air Pollution: Method, Software, and a Use-Case
Zoltán Horváth, Mátyás Constans, Ákos Kovács, László Környei, and József Bakosi Széchenyi István University, Győr
Abstract. Urban air pollution attributes to millions of annual deaths. In city planning and policy-making procedures, computational models are of significant importance. High-resolution models in space and time need large-scale computations. In this talk, we are presenting the details of a simulation digital twin for the airflow and NO2 dispersion over a 3D domain of 25 km^2 ground area in the city Gyor, Hungary including building geometries. This solution uses the in-house code Fluid-Solver that runs efficiently on CPU and GPU nodes. The algorithm is based on a flux-vector splitting finite volume scheme over unstructured tetrahedral meshes and solves the compressible Euler and Navier-Stokes equations in the second order. For model order reduction we applied the Proper Orthogonal Decomposition with the snapshot matrix generated by runs for velocity inlet boundary conditions of which wind directions samples the wind rose. Boundary conditions are obtained by wind measurement sensor data assimilation. Results are visualized in real-time by viewing images of dedicated slices at sampled time points and also in an interactive way. Some remarkable numerical properties of the reduced model simulations will be discussed as well, namely the greatly increased
Wojicech Szeliga represented the team of PSNC in HiDALGO2 with the presentation, “Digital Twin in Modeling of Renewable Energy Sources”.
The use case “Renewable Energy Resources (RES)” will develop more advanced models for wind and solar energy and study the impact of uncertainties in the model, which was not ever done before in the HiDALGO2 project. The goal is scalability is to develop and run ensemble uncertainty quantification on 98,000 CPU cores. RES answers the question of how much energy from wind and sun can be produced considering specific weather conditions. It is worth mentioning that RES has dedicated modules that consider urban and extra-urban conditions.
Digital Twin in Modeling of Renewable Energy Sources
Wojciech Szeliga, Michał Kulczewski, and Ariel Oleksiak
Poznan Supercomputing and Networking Center, Poland
Abstract. Renewable energy sources are undeniably a key component of a global transformation required for limiting anthropogenic climate changes. While harvesting energy from wind or solar radiation has its own limitations in terms of accessibility of the energy source, there arises a necessity of optimization of the harvesting process. The amount of energy that can be produced depends on the weather conditions which are a result of complex multiscale phenomena in the atmosphere. We present a digital twin for renewable energy sources, a comprehensive tool named RES for modeling fine-grained weather conditions for wind and photovoltaic farms and providing estimates of energy production. The framework introduces a multiscale approach by using two models for different scales: WRF and EULAG. With both models coupled, it is possible to perform complex simulations in HPC environment, tailored towards renewable energy sources. Great level of details significantly increases the quality of the digital twin and its energy production estimations done at the post-processing stage. Described framework has been introduced to one of the largest Polish energy operators. As the operator owns multiple wind farms, their energy production estimation is detailed with created digital twin of each. RES is also used by the operator to create a digital twin of energy infrastructure in order to predict its failures due to upcoming wind gusts, excessive temperatures or icing.
In the mini-symposium participants could also listen to interesting talks by the representatives of EXCELLERAT P2 CoE: 1. “Study of Spatial-/Time-Varying Parameter Estimation Using Genetic Algorithm and Overlapping Sliding Window Approach” by Huan Zhou and Ralf Schneider, Universität Stuttgart, Germany; Sebastian Klüsener and Andreas Backhaus, Federal Institute of Population Research, Germany, 2. “Efficient Improvement for Monte Carlo Methods Based Uncertainty Quantification” by Qifeng Pan, Universität Stuttgart, Germany.
About EXCELLERAT P2
EXCELLERAT P2 CoE is the European Centre of Excellence for Engineering Applications. As a single point of access for expertise it aims to support the solution of complex engineering problems with the help of HPC solutions and Exascale computing.
The last presentation at the mini-symposium was a guest talk, “A Study of Prediction Performances of Econometric Arima and Soft Computing Ann Models in Forecasting Black Carbon Concentration Data” by Kulwinder Singh and Jatinder Kaur, Punjab Technical University, India; Sarbjit Singh, Guru Nanak Dev University, India; Kirti Soni, CSIRO Minerals, Australia.