Forecasting of Solar Irradiance and Power in Uncertain Photovoltaic Systems Using BiLSTM and Bayesian Optimization

The dynamic and intermittent nature of solar energy presents significant challenges for its stable integration into current energy systems. Moreover, photovoltaic (PV) solar power generation is consistently accompanied by uncertainties due to measurement errors, parametric variation, and noise, which complicate energy management (EM). To address these issues, accurate and robust solar irradiance and PV power forecasting are essential for the seamless integration of solar energy into the power grid and for managing uncertainties. Short-term forecasting, in particular, plays a crucial role in enabling real-time power dispatching. The proposed approach includes two main contributions. Firstly, it introduces a Bidirectional Long Short-Term Memory framework enhanced by Bayesian Optimization (BiLSTM-BO), where BiLSTM is applied for time series prediction, and the BO is used to fine-tune the hyperparameters of the BiLSTM model. This approach addresses common challenges such as time-intensive manual tuning, the risk of suboptimal parameter selection, and the complexity of navigating a high-dimensional search space. Additionally, the BiLSTM-BO-based model uses interval-valued data to effectively handle uncertainties (e.g., measurement errors, noise, variable variability,...). The primary objective of the proposed methodology is to ensure accurate predictions of solar irradiance and PV power, even under uncertain conditions. Various performance indicators, including mean square error (MSE), root mean square error (RMSE), normalized RMSE (nRMSE), and the coefficient of determination ( \(R^2\) $R^2$ ), were employed to assess the efficacy of this approach. The results consistently demonstrate high performance across these evaluation metrics, affirming the effectiveness of the proposed method.