Multi-label machine learning for power forecasting of a grid-connected photovoltaic solar plant over multiple time horizons

Because of solar power's inherent intermittency and stochastic nature, accurate photovoltaic (PV) generation forecasting is critical for the planning and operation of PV-integrated power systems. Thus, accurate power forecasting becomes vital for maintaining good power dispatch efficiency and power grid operational security. Several PV forecasting methods based on machine learning algorithms (MLAs) have recently emerged. This paper presents machine learning methods for multi-label forecasting of PV and AC power delivered to the grid of a building-applied PV plant. Various algorithms representing multiple groups are evaluated, including linear regression (LR), polynomial regression (PR), neural networks (NN), deep learning (DL), gradient-boosted trees (GBT), random forests (RF), decision trees (DT), k-nearest neighbor (k-NN), and support vector machines (SVM). The models use real-time collected data from sensors over one year for solar irradiance, ambient temperature, wind speed, and cell temperature to predict PV and AC power outputs. Forecast performance over multiple time horizons is validated using four datasets: 24 h, one week, one month, and sudden variations. Models are evaluated based on performance metrics such as absolute error (AE), root mean square error (RMSE), normalized absolute error (NAE), relative error (RE), relative root square error (RRSE), and correlation coefficient (R). Results show that RF, DT, and DL consistently achieved the highest accuracy (R ≈ 99.8-100%) with minimal errors (RMSE within 0.014-0.022, AE within 0.008-0.015) across various forecasting scenarios. These models demonstrated strong adaptability and predictive reliability across short-term, medium-term, and long-term forecasts, making them the most effective choices for PV and AC power prediction. The accurate forecasts generated in this study have the potential to aid grid operators in forecasting PV power output variability and planning for integrating intermittent PV power into the grid. Understanding how PV generation will fluctuate given different meteorological conditions allows operators to ensure the consistent integration of this weather-dependent power source. Moreover, multi-label prediction of DC and AC power enables inverter efficiency optimization and grid integration analysis. The average actual and predicted efficiencies of the inverter are 0.96688 and 0.9638, providing valuable insights.