The Synthesis of Mathematical and Data Driven Modelling of Complex Physical Systems
Lead Research Organisation:
Imperial College London
Department Name: Mathematics
Abstract
Forecasting and prediction are operations of critical importance in the energy sector. Whether it is predicting the output of a solar or wind farm, forecasting the energy demand at city and regional levels, predicting the operational condition of assets such as gas turbine plants or the long term environmental impact of engineering operations, they all have to optimize under great uncertainty in a coherent and consistent manner. This proposed PhD posits that an overarching theoretical, methodological and practical framework to integrate both data driven and physics based models will provide greater modelling capability and representation as well as superior predictive capability in the presence of uncertainty.
Organisations
People |
ORCID iD |
| Andrew Duncan (Primary Supervisor) | |
| George Wynne (Student) |
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| EP/S513635/1 | 30/09/2018 | 29/09/2023 | |||
| 2130619 | Studentship | EP/S513635/1 | 30/09/2018 | 31/12/2022 | George Wynne |
| Description | Functional data, where one data point is a discretised function, is important and becoming common place due to advances in data collection technology. For example, there are multiple applications to engineering, finance and medicine such as scan image data, time series and evolving surfaces. In my work, I have developed novel theoretical foundations and methodology to analyse data specifically of a functional nature, as opposed to existing methods for functional data which are largely ad hoc and use classical techniques that do not leverage the inherent structure of such data. The algorithms I have derived provide superior testing power beyond state of the art and lays the foundations for other machine learning based methods such as inference for functional data. |
| Exploitation Route | This work break the ground in the use of statistical machine learning methods for functional data. The developed methodology opens the door to new avenues in research to predictive health monitoring of engineering assets where pervasive data sources can be interpreted as snapshots of functional data. In particular, we anticipate this would push forward the state of the art performance in this area leading to superior estimates of remaining useful life (RUL) as well as fault and anomaly detection. |
| Sectors | Aerospace Defence and Marine Energy Manufacturing including Industrial Biotechology |