Designing Convolutional Neural Networks to Make Successful Predictions of Windfarm Geotechnical Parameters Using Field Geophysical Data
conference paper
We use convolutional neural networks to predict geotechnical parameters (synthetic CPTs) from 2D ultra high resolution seismic data for an offshore windfarm site. The site is characterized by laterally varying channel deposits in the upper 20m with a thick but non-uniform unit underneath. We investigate methods to improve predictions in the poorly imaged lower unit and to improve the capture of geotechnical property variation within units. We use interpreted geotechnical soil units as an input parameter to capture the geological character of the site. The lateral variation of these units introduces data challenges and uncertainties, which we capture in a new method to calculate prediction confidence intervals. The confidence intervals also take into account uncertainties in data acquisition, processing and CNN training, to provide robust and realistic confidence intervals for the predictions at this site.
TNO Identifier
992396
Publisher
European Association of Geoscientists & Engineers (EAGE)
Source title
4th EAGE Global Energy Transition Conference and Exhibition, Nov. 2023
Pages
1-5
Files
To receive the publication files, please send an e-mail request to TNO Repository.