Full Program »
Initial Observations From Machine Learning Approaches Using Uk Temperature Data For Mine Water Thermal (mwt)
Until now, machine learning has been underutilised in shallow geothermal and more specifically, mine water thermal applications. Mine water thermal is a low-carbon, self-replenishing solution to providing space heating and cooling yet does not have significant uptake in the UK relative to other European country counterparts. This slower development is partly due to a lack of understanding of heat movement and behaviour in existing and abandoned mine systems. Where machine learning could offer a different perspective to more traditionally used numerical modelling techniques is in its ability for algorithms to quickly synthesise and process large volumes of available data. In this paper, we present machine learning methods applied to a UK subsurface temperature dataset to begin to minimise some of the knowledge gaps currently hindering development. We collated publicly available temperature data from The Coal Authority, British Geological Survey, Glasgow UK GeoEnergy Observatory, Scottish Environment Protection Agency, and the North Sea Transition Authority alongside additional licenced data from The Coal Authority, to form a dataset with over 2.4 million datapoints from 800 distinct locations. Both k-means clustering and multiple linear regression algorithms are presented alongside preliminary results of three models where geothermal gradients are predicted. The average geothermal gradient predicted by these models was 22°C/km (model A) and 23°C/km (model B and C), which is lower than the average UK-wide geothermal gradient of ~26°C/km modelled by previous authors. Furthermore, mine water temperatures were predicted to be 2°C warmer than other groundwaters in unmined aquifers at the same depth. All models were statistically significant with R-squared values of >0.85. We suggest that the data is likely presenting skewed results due to the inclusion of temperatures taken from pumping boreholes and mine shafts that mixes the warmer water at depth with colder, shallower water, creating a more suppressed temperature with depth. The simple relationships between temperature and depth illustrated in this paper form the beginnings of a machine learning platform from which more features can be added to further understand heat flow in mines. This study has also highlighted the importance of data sharing: a national or even global temperature database of the mined subsurface would allow for the uncertainties surrounding mine water thermal resources to be more easily addressed.