• A novel SVR-NARX hybrid model outperforms traditional machine learning methods like LSTM and GRU in long-term wind power forecasting. • Shannon entropy quantifies input and output data uncertainty, directly influencing the accuracy of wind power predictions • Conditional Value at Risk (CVaR) is employed to benchmark financial risks, helping investors anticipate potential revenue fluctuations. • Revenue predictions indicate that July, August, November, and December yield the highest returns, while March, June, and September pose the greatest financial risks. • he proposed approach provides investors with data-driven financial foresight, enabling informed decision-making for wind energy projects. Accurate wind power forecasting is essential for grid stability and energy planning. This study proposes a hybrid SVR–NARX model, applied to data from Akkanayakanpatti, Tamil Nadu, India, to benchmark wind energy generation potential and quantify forecast uncertainty. Using meteorological inputs, the model assesses power generation over a 100-hour horizon, evaluated through MSE, RMSE, and cross-entropy metrics. Probabilistic performance is further validated with Value at Risk (VaR) and Conditional Value at Risk (CVaR). Comparative results show that the hybrid model outperforms baseline LSTM and GRU architectures under seasonal variability. This study aims to develop a hybrid SVR–NARX model to improve wind power forecasting accuracy and quantify uncertainty through entropy-based and risk-aware (VaR/CVaR) measures. Using data from Akkanayakanpatti, India, the model outperformed benchmark LSTM and GRU models, reducing RMSE by 10–15% and improving reliability in high-risk months. The findings support grid operations and investment decision-making under variable conditions.
Sanyal et al. (Sun,) studied this question.