Advanced harmonic forecasting in offshore wind farms with permanent magnet synchronous generators using a hybrid deep and machine learning architecture

Abstract

Wind energy is crucial for reducing fossil fuel dependence and promoting sustainability. Offshore wind farms (OWFs) benefit fromhigher, stable wind speeds but pose challenges such as harmonic distortion and voltage fluctuations when integrated into powergrids. This study develops an advanced model for accurate harmonic forecasting in OWFs using permanent magnet synchronousgenerators (PMSG). Real meteorological data from Zonguldak and Sinop in the Black Sea region of Turkey were used to simulatepower output, voltage, and current waveforms. Harmonic components, including total harmonic distortion for voltage (THDV)and current (THDI), were extracted and predicted. Various machine learning (ML) and deep learning (DL) algorithms wereapplied, including Linear Regression, Decision Tree, Random Forest, Gradient Boosting, XGBoost, KNeighbors, LSTM, GRU, andCNN. Additionally, hybrid ML-DL models were explored to enhance forecasting accuracy. A comparative analysis of these modelsdemonstrated their effectiveness in improving harmonic prediction. Results indicate that hybrid models, particularly LSTM+GBand GRU+GB, improve harmonic forecasting accuracy by reducing RMSE by approximately 15% compared to traditional MLmethods. This enhancement contributes to better power quality management and grid stability, making offshore wind farmsmore viable for large-scale renewable energy integration. The findings of this research provide a fundamental basis for futureinvestigations into offshore wind harmonic forecasting.