Harmonic forecasting in offshore wind systems utilizing DFIG on bozcaada island: a hybrid machine learning and deep learning approach

Abstract

In today's energy market, wind power is pivotal for reducing reliance on fossil fuels in electricity generation while promoting environmental sustainability. Also, Offshore wind farms (OWFs) play a key role by capitalizing on higher and steadier wind speeds than onshore options. Yet, integrating OWFs into power grids poses challenges like harmonic distortion and voltage fluctuations. Furthermore, harmonics prediction is among techniques used to mitigate these issues, particularly in cases of extensive OWF integration. The main goal of this project is to develop an innovative forecasting model for accurate and reliable prediction of harmonics in offshore wind farms (OWFs). To achieve this, Double-Feed Induction Generator (DFIG) configurations are utilized. The meteorological data from Bozcaada, Aegean Sea, Turkey, including wind speed, serves as input parameters for the model. The model simulates output power, along with current and voltage waveforms, which are then used to extract and forecast harmonics. Also, Total Harmonics Distortion (THD) parameters are specifically forecasted for voltage and current waveforms using various machine learning and deep learning algorithms. Common machine learning algorithms like Linear Regression (LR), Decision Tree (DT), Random Forest (RF) and Gradient Boosting are applied, along with deep learning algorithms such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). To improve prediction accuracy, hybrid models are proposed, combining machine learning techniques with outputs from deep learning models. The study aims to compare the effectiveness of different machine learning, deep learning, and hybrid techniques on THDV and THDI forecasting using meteorological data from Bozcaada, Aegean Sea, Turkey. Results indicate the influence of specific algorithms on prediction performance, offering valuable insights for future studies in renewable system harmonic prediction. Particularly, certain machine learning algorithms show effectiveness in THD prediction, while proposed ML-DL hybrid models demonstrate superior performance other than unique ML and DL models. These findings contribute to advancing harmonic forecasting capabilities and related applications.