Wind Power Forecasting and Fault Analysis System

Industrial software for wind turbine operators: it cleans SCADA data, forecasts generation 1–48 hours ahead using numerical weather prediction (NWP), detects anomalies and generates maintenance decisions.

The system trains six different machine learning models (XGBoost, LightGBM, CatBoost, Random Forest, Extra Trees, HistGB) with time-ordered validation and combines them with a conformal 90% prediction interval. Through the Open-Meteo integration, weather forecasts extrapolated to hub height are used as features at the target hour. Isolation Forest-based anomaly detection and physics-based fault classification in six categories produce an alarm list and estimated generation loss (MWh) without requiring labeled data. Results are delivered through a SCADA-style web dashboard, a REST API (OpenAPI) and CSV exports; drift monitoring (PSI), a model registry and a scheduled retraining loop keep the operation running on its own.