Reproduce “Wind speed prediction using multidimensional convolutional neural networks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kevin Trebing, Siamak Mehrkanoon
A novel deep learning architecture leveraging depthwise-separable convolutions across multiple spatio-temporal input dimensions enhances short-term wind speed prediction. This multidimensional convolutional neural network achieved superior or competitive accuracy, marked by lower MAE and MSE, compared to traditional CNNs and a persistence model across real-world datasets from Denmark and the Netherlands for forecast horizons up to 24 hours.
Key insights
- Applying depthwise-separable convolutions across multiple permutations of spatio-temporal, multivariate input data (e.g., cities-time steps, cities-features) effectively captures complex non-linear dynamics.
- The multidimensional CNN architecture demonstrates enhanced generalization capabilities for multi-step-ahead wind speed forecasts, exhibiting less pronounced error increases with longer prediction horizons.
- Data-driven deep learning models, particularly the proposed multidimensional CNN, can significantly outperform traditional persistence models and provide competitive or superior results compared to other advanced CNN architectures for short-term meteorological forecasting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.