Hybrid Statistical, Machine Learning, and Deep Learning Framework for Accurate Wind Power Forecasting using Big Data

Abstract— To accurately predict wind power, the suggested study employs a comprehensive hybrid system that combines statistics, ML, and deep learning models. Improved prediction accuracy and model robustness are primary goals of effectively modeling large-scale, time-series data for both linear and non-linear connections. The most applicable meteorological variables in the 568,139 records contained in ten CSV files that constitute the Kaggle Wind Power Repository database are wind speed, direction, temperature, humidity, and air pressure.Data preprocessing (cleaning, standardization, and extraction of temporal features) and dynamical model creation based on statistical models (ARIMA, SARIMA, SARImax), learning regressors ( Random forest, XGBoost, LightGBM, SVR, El), and other methods make up the mentioned technique.The model's performance was validated and thoroughly tested using a number of metrics,including RMSE, MAE, MAPE, R 2, and NSE, which were evaluated using the experimental data. In comparison to more conventional methods of machine learning and traditional approaches, experiments have shown that deep learning models, and CNN-LSTM hybrid, in particular, do much better. CNN-LSTM was the best model to use when it comes to utilizing time trends and generate accurate and reliable predictions due to the best results in relation to the RMSE, MAPE, R2, and NSE. The results highlight the opportunities of hybrid deep learning models in enhancing predictive renewable energy and grid stability.