Predictive control technique for solar photovoltaic power forecasting

An accurate estimation of photovoltaic (PV) power production is crucial for organizing and regulating solar PV power plants. The suitable prediction is often affected by the variable nature of solar resources, system location and some internal/external disturbances, such as system effectiveness, climatic factors, etc. This paper develops a novel strategy for applying a predictive control technique to PV power forecasting applications in a smart grid environment. The strategy develops the model predictive control (MPC) under demand response (DR) and some data-driven methods. It has been found that it is challenging to model an MPC for solar power forecasting regardless of its robustness and ability to handle constraints and disturbance. Thus, an optimal quadratic performance index-based MPC scheme is formulated to model a forecasting method for a PV power prediction. This strategy is then compared with some machine learning approaches. The developed strategies solve the problem of accurately estimating the direct current (DC) power yielded from the PV plant in a real-world implementation. The study also considers external disturbances to evaluate the significance of the developed methods for a suitable forecast. Therefore, this study optimally demonstrates that an accurate solar PV DC power prediction can relatively be estimated with an appropriate strategy, such as MPC and MLs, considering the system disturbances. This study also offers promising results for intelligent and real-time energy resource estimation that assist in developing the solar power sector. • Develop optimal forecasting strategies to estimate power from solar PV plants. • Implement model-based and data-driven strategies to predict power from solar PV systems. • Apply state-space modelling to formulate the forecasting strategy-based MPC design. • Apply various machine-learning approaches to the solar PV power forecasting scheme. • Drop-out layers are added to provide better accuracy and prevent data overfitting. • Evaluate the accuracy of the forecasting methods during and without disturbance.