Abstract
This paper presents a comprehensive approach to forecast solar power generation in the Kazakhstani region, leveraging historical and weather data. The main challenge with solar power lies in its dependency on weather conditions, which significantly impacts day-ahead output planning and grid operations. To address this, we have employed advanced timeseries forecasting models like ARIMA, Facebook Prophet, and XGBoost to predict the power output for the next day. By utilizing the real data from two solar plants located in Kazakhstan and analyzing region's historical weather patterns, our proposed model were carefully validated with walk-forward model validation approach. Obtained results highlight superior performance of XGBoost model, outperforming plant dispatcher's predictions and demonstrate the potential of utilizing machine learning in enhancing the reliability of solar power forecasts. This research provides practical insights for both academics and practitioners, including grid managers and plant dispatchers, aiming to optimize solar power integration and reliable operation of the power grid.