Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Time-series Analysis to monitor Methane Emissions and identify Leakages
In the project, we will develop a new approach based on machine learning to monitor emissions and detect gas leakages. Two possible focus points are i) developing new gate structures for deep sequential models, ii) combining deep sequential models with the classic bayesian modelling approaches.
This Master thesis project is a collaboration between the EcoVision Lab and Kayrros.
With the launch of Sentinel-5P satellite (Precursor of Sentinel-5) in 2017, globally monitoring gases concentrations has been brought to a new level. The selected wavelength range for TROPOMI, the main instrument of Sentinel-5P, allows observation of key atmospheric constituents, including ozone (O3), nitrogen dioxide (NO2), carbon monoxide (CO), sulfur dioxide (SO2), methane (CH4), formaldehyde (CH2O), aerosols and clouds.
Combining this new critical data with the latest machine learning research on time-series analysis and Kayrros’ unique expertise in the energy market creates an exciting opportunity to explore novel research directions whilst tackling important global issues.
We decided to focus this master thesis on the topic of monitoring emissions at industrial asset level (coal power plant, gas, refinery…) and gas leak detection.
Recent literature suggests that deep sequential models such as LSTMs have unnecessarily complicated gate structures that limit the effectiveness and stability of the training process, the student could work on developing new models that are able to learn long-term dependencies while limiting these issues. Another area that could be explored is the combination of deep sequential models with the classic bayesian modelling approaches (deep Kalman filter, bayesian RNNs).
This Master thesis project is a collaboration between the EcoVision Lab and Kayrros. With the launch of Sentinel-5P satellite (Precursor of Sentinel-5) in 2017, globally monitoring gases concentrations has been brought to a new level. The selected wavelength range for TROPOMI, the main instrument of Sentinel-5P, allows observation of key atmospheric constituents, including ozone (O3), nitrogen dioxide (NO2), carbon monoxide (CO), sulfur dioxide (SO2), methane (CH4), formaldehyde (CH2O), aerosols and clouds. Combining this new critical data with the latest machine learning research on time-series analysis and Kayrros’ unique expertise in the energy market creates an exciting opportunity to explore novel research directions whilst tackling important global issues. We decided to focus this master thesis on the topic of monitoring emissions at industrial asset level (coal power plant, gas, refinery…) and gas leak detection. Recent literature suggests that deep sequential models such as LSTMs have unnecessarily complicated gate structures that limit the effectiveness and stability of the training process, the student could work on developing new models that are able to learn long-term dependencies while limiting these issues. Another area that could be explored is the combination of deep sequential models with the classic bayesian modelling approaches (deep Kalman filter, bayesian RNNs).
The goals of this project are to attribute and quantify observed emissions to a specific source, to predict the emissions of a specific power plant and to build an embedding of different classes of industries based on spatio-temporal features in order to output the location of different types of industries.
The goals of this project are to attribute and quantify observed emissions to a specific source, to predict the emissions of a specific power plant and to build an embedding of different classes of industries based on spatio-temporal features in order to output the location of different types of industries.
Riccardo De Lutio (riccardo.delutio@geod.baug.ethz.ch)
Dr. Jan D. Wegner (jan.wegner@geod.baug.ethz.ch)
Riccardo De Lutio (riccardo.delutio@geod.baug.ethz.ch) Dr. Jan D. Wegner (jan.wegner@geod.baug.ethz.ch)