Improving Wind and Solar Power Prediction with Efficient Wrapper-based Feature Selection: An Empirical Study
Merged summary
TL;DR — An empirical ML-methods paper proposing Cluster-based Sequential Feature Selection (CSFS), a model-agnostic wrapper method for picking input features in renewable energy (wind/solar) power prediction pipelines. It matters because it offers a systematic, cheaper alternative to ad-hoc feature selection in a domain with many monitoring variables.
- Motivated by two structured literature reviews (wind turbine power-curve modeling; PV power prediction) finding that feature selection is currently limited or unsystematic despite many available variables.
- CSFS is a clustering-based wrapper method aimed at automatic, efficient, reliable feature selection; open-source implementation provided on GitHub.
- Benchmarked against sequential feature selection (SFS), filter-based methods, and Random Forest embedded importance; wrapper-based methods gave overall better feature selections.
- CSFS matches SFS predictive performance while cutting computational cost by ~21% on average.
Sources (1)
Improving Wind and Solar Power Prediction with Efficient Wrapper-based Feature Selection: An Empirical Study
TL;DR — An empirical ML-methods paper proposing Cluster-based Sequential Feature Selection (CSFS), a model-agnostic wrapper method for picking input features in renewable energy (wind/solar) power prediction pipelines. It matters because it offers a systematic, cheaper alternative to ad-hoc feature selection in a domain with many monitoring variables.
- Motivated by two structured literature reviews (wind turbine power-curve modeling; PV power prediction) finding that feature selection is currently limited or unsystematic despite many available variables.
- CSFS is a clustering-based wrapper method aimed at automatic, efficient, reliable feature selection; open-source implementation provided on GitHub.
- Benchmarked against sequential feature selection (SFS), filter-based methods, and Random Forest embedded importance; wrapper-based methods gave overall better feature selections.
- CSFS matches SFS predictive performance while cutting computational cost by ~21% on average.