Machine learning–driven forecasting and robustness analysis of hybrid wind–solar energy systems for urban street lighting applications

Engineering notes

Engineering notes will be added by the Power for Solar editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为光伏效率、钙钛矿太阳能电池、储能技术、太阳能热利用、BIPV、并网技术等高价值论文补充中文说明。

Original abstract

This study proposes a data-driven optimization and analytical framework based on artificial-intelligence techniques for a hybrid wind and solar energy system designed specifically for urban street-lighting applications in İzmir, Turkey. The system integrates a vertical axis wind turbine with a helical blade configuration and photovoltaic (PV) panels to exploit complementary renewable resources under urban coastal conditions. Long-term meteorological assessment indicates favorable hybrid potential in the study area, characterized by strong wind availability and sufficient solar irradiance. To support reliable short-term power estimation, ensemble-based machine learning (ML) techniques are employed to forecast short-term power generation, including Extreme Gradient Boosting (XGBoost), Gradient Boosting, and Random Forest algorithms. Model performance is evaluated using standard regression metrics, with XGBoost demonstrating the most consistent generalization performance across both wind- and solar-driven outputs. To enhance interpretability and address the limited transparency often associated with data-driven models, SHapley Additive exPlanations (SHAP) analysis is applied to quantify the relative influence of environmental variables. The results reveal that wind speed, solar irradiance, and ambient temperature are the dominant drivers of hybrid energy production. Beyond predictive accuracy, the analysis incorporates urban operating constraints, demonstrating that hybridization substantially mitigates output variability under realistic urban wind conditions. Statistical and distributional analyses show that the hybrid configuration reduces overall power-output variability by approximately thirty-eight percent compared with standalone photovoltaic operation, thereby improving supply continuity for public lighting services. The novelty of this work lies in combining physics-informed energy modeling with explainable ensemble-based forecasting, explicitly tailored to urban street lighting applications rather than generalized power systems. By emphasizing predictive reliability, interpretability, and urban microclimatic relevance, the proposed framework provides a practically oriented basis for decentralized renewable energy planning in similar urban environments.

Links and sources

Comments