Current research areas

1. Machine learning based rapid response tools for real-time operation prediction and uncertain analysis. The objectives are:

  • Development of a new simulator based on data centric modelling for multiscale geophysical dynamical modelling (in combination with advanced machine learning, data assimilation and reduced order modelling techniques), which is capable of a) representing and analysing uncertainty in models; b) real-time updating of uncertainties, parameter estimation/optimization, and supervised calibration; and c) providing uncertainty analysis for risk assessment, decision making and policy-making processes risk.
  • Development ofa highly efficient but detailed, AI-enabling reduced order model based on both large datasets and multi-scale physics models to provide a rapid response in emergency situations.
  • AI tools for optimising observational networkto guide where/when to collect data.
  • Applications focus on air pollution and atmospheric modelling, virtual city modelling for air flows and flooding - providing the computational infrastructure required to assist with operational decision making, flooding, coastal and ocean modelling, as well as renewable energy (off- and on- shore wind turbines, tidal and wave energy).

2. Digital Twin environment forecast and optimal operational system- “how to effectively can we manage our growing cities in an operational sense (hourly/daily) in response to increased pollution, climate change, health and economic impacts”. The research aim is to develop an advanced digital twin decision support framework using hybrid-AI and physical modelling tools which will enable the urban population as well as policy makers to make both strategic and everyday decisions that are essential for generating a sustainable environment by 2050. An effective and efficient city operational system of smart cities requires the integration of complex nonlinear physical and dynamic processes (meteorological, chemicals, particle matters, transportation emissions etc) and the interactions between them, as well as improving health and supporting economic impact. The digital twin system using integrated AI and physical modelling continuously learns and updates itself from real-time datasets/sources as their physical conditions change. Such an operational system can help create virtual city models to better operate, analyse and optimise how they respond to challenges in the city environmental management. Our EPSRC funded project: AI-physical modelling tools for urban decarbonisation (Details see https://ai4urban.github.io/). 

3. Hybrid-AI and multiscale physical modelling in renewable energy- develop a hybrid-AI and physical modelling tool for renewable energy. The main objectives are (1) improved accuracy of weather forecast – one of challenges in renewable energy, thus helping to predict renewable energy capabilities for up to weeks in advance. For this purpose, advanced spatial-temporal machine learning and dynamically adaptive mesh techniques are used; (2) AI techniques for efficient energy consumption (smoothening the flow of energy from generation to consumption); (3) AI tool for optimal control/management of national energy grids by adjusting customer energy consumption and coordinating with the photovoltaic generation in the neighbourhood) -- helping balance demand and supply of the energy sector.

4. Hybrid-AI and physical informed warning system for nature hazard: The research will provide a rapid and accurate tool for emergency planning and managing the coincidence of 2 or more natural disasters by improving the modelling and predictive capabilities of integrated components of a natural disaster. Improvements in predictive capabilities in such cases are critical in populated areas, and especially in cities.