The increasing penetration of large-scale photovoltaic power plants has intensified the challenge of managing power variability caused by rapid changes in solar irradiance. Short term and ultra short-term power fluctuations, mainly driven by cloud movement, can lead to severe ramp rate violations and grid instability, particularly in weak and island power systems. This review critically examines existing solar power forecasting approaches with a focus on short-term forecasting methods relevant to ramp rate control. Conventional physical and statistical models are discussed alongside machine learning and deep learning techniques, highlighting their strengths and limitations across different time horizons. Special emphasis is placed on vision-based forecasting methods using satellite imagery and ground-based sky imagers, which have demonstrated superior capability in capturing fast irradiance transients. The review further analyses commonly used input parameters, forecasting horizons, and cloud motion prediction techniques. Based on comparative assessment, the paper identifies key research gaps related to real-time deployment, data integration, and model generalization. The findings provide a structured foundation for developing accurate and practical forecasting frameworks to support grid stability in high solar penetration environments.