Exploring the potential of neural networks to predict statistics of solar wind turbulence

Abstract

Small artificial neural networks (ANNs) are good at predicting large scale values of structure functions.• An ANN with only 20 hidden neurons statistically outperforms simple imputation techniques for large fractions of missing data. • More work is needed to improve the ANN's performance in predicting both large and small scale values.

Interpolating Strange Attractors via Fractional Brownian Bridges

“…We expect the presented research to be useful for predicting and analyzing sparselysampled time series data, e.g., in agriculture or other fields where fine-grained measurements are expensive. Furthermore, our methodology could be applicable to a broad range of other real-world problems such as the filling of gaps in solar wind measurements [2] or spatiotemporal wind fields [3] for the assessment of wind turbine loads. As our method can be considered as some "hybrid" between a stochastic (by virtue of the fractional Brownian bridge interpolation) and a deterministic algorithm (by the embedding and genetic algorithm), it should be highly relevant for the filling of such time series or spatial fields, which often exhibit deterministic and stochastic elements at the same time.…”

“…Typical examples where such data augmentation techniques are deployed include gaps in time series from solar wind measurements [1,2], spatio-temporal wind fields from meteorological mast arrays [3], as well as the study of particle transport in intergalactic magnetic fields [4]. An enhancement of data in these examples is commonly achieved by relying on one of the various interpolation techniques, such as linear, polynomial, fractal, or stochastic interpolation methods.…”