A machine-learning approach for noise reduction in parameter estimation inverse problems, applied to characterization of oil reservoirs

Scientometric review of artificial intelligence for operations & maintenance of wind turbines: The past, present and future

“…Lack in diversity of alarm messages Natural language generation techniques for generating contextual information on faults Few-shot learning techniques [99] to learn from low-diversity and limited training data; Generalized language models pretrained on large corpuses of information such as Bidirectional Encoder Representations from Transformers (BERT) [104], Generative Pre-trained Transformer (GPT-2/GPT-3) [164,165] Change-point grouping and quartile algorithms [153], Least Median of Squares (LMedS) method [151], LMedS with random search [152], statistical and control filtering techniques like Kalman filters [154], specialised loss functions in deep learning, data re-weighting and training procedures [166], class noise and attribute noise identification techniques (especially ensemblebased noise elimination) [167], specialised ML-based noise reduction techniques such as Multi-step finite differences, Splines, Mixture of sub-optimal curves etc. [168] Missing values in datasets Supervised and unsupervised classification (for fault prediction) and regression (for forecasting vital operational parameters) techniques; Reinforcement learning techniques for O&M planning; Natural language processing techniques for classifying alarm messages;…”

A New Inverse Modeling Approach for Hydraulic Conductivity Estimation Based on Gaussian Mixtures

“…A downside for the use of the derivatives is that the noise level in the observed head values could be increased, but this problem is addressed by choosing a small value of N k , or by using a noise reduction method such as cubic splines (e.g., Illman & Neuman, 2001) or the one proposed by Minutti et al (2018), which is designed to address noisy data in inverse problems when the derivative is required, by approximating the data in a similar way to regression splines, but using different solutions of the forward model as a basis (or space of functions) of the regression model.…”