Analysis of spatio-temporal contexts for short-term wind forecasting
Description
Description This dataset contains the spatiotemporal meteorological time series and analysis outputs generated during a study aimed at optimizing short-term wind forecasting and meteorological network design. The core objective of the research is to empirically identify the maximally informative Spatiotemporal (ST) context—defined as the optimal subset of spatial locations and historical time lags—that maximizes the accuracy of local wind predictions for short-term horizons (1, 3, and 5 hours ahead) while minimizing data redundancy. The dataset reflects a multi-horizon forecasting framework applied to a regional case study in Argentina.
Files
Steps to reproduce
Repository File Structure and Analysis Components This section details the notebooks and scripts that make up the repository. Each code provides detailed instructions on the parameters to set to ensure reproducible results. 1. Seleccion y analisis de datos de parques (Park Data Analysis) Description: Handles initial data ingestion, and exploratory data analysis of wind speed and direction time series from the Argentine study region. 2. Pruebas capacidad de red LSTM (LSTM Capacity Tests) Description: Preliminary calibration and evaluation of the Long Short-Term Memory (LSTM) network. 3. Búsqueda de fronteras espaciotemporales (ST Boundary Search) Description: Executes the systematic parametric sweep across geographic influence radii and historical time-lag windows to feed the forecasting models. 4. Resultados de la búsqueda de fronteras espaciotemporales (Boundary Search Results) Description: Compiles and visualizes prediction errors across the short-term horizons to identify initial trends and physical boundaries of the ST context. 5. Incertidumbre en ventanas temporales (Time Window Uncertainty) Description: Analysis of variability in the model’s performance as a function of different historical time lags. 6. Análisis SHAP (SHAP Analysis) Description: Implements Permutation Shapley Importance to quantify the contribution of spatial coordinates and the time window. 7. Metodología cluster + Greedy parques (Park Clustering + Greedy Methodology) Description: Implements the two-stage optimization algorithm. It uses time-series clustering to group similar meteorological behaviors and filters spatial redundancies, then initializes the Greedy Selection (GS) process. This step concentrates on determining the spatial locations that contribute significantly to accurate prediction for each forecast horizon. 8. Resultados metodología Cluster + Greedy (Clustering + Greedy Results) Description: Evaluates the metrics and reduction achieved by the optimized selection. 9. Metodología Greedy multihorizonte (Multi-Horizon Greedy Methodology) Description: Final optimization phase that consolidates and compares results across all forecasting horizons. It definitively identifies the optimal locations for physical sensor placement.