Solar Power Forecasting and Uncertainty Quantification Using SARIMA: Case Study of Tacna Solar Plant, Peru

Solar power forecasting is essential for grid stability and energy management in renewable energy systems. This study develops a comprehensive SARIMA-based forecasting framework for the Tacna Solar Plant, a 20 MW photovoltaic installation in Peru's Atacama Desert. Using daily generation data spanning one complete annual cycle (April 2024-March 2025, 365 observations), we implement a SARIMA(2,1,1)(1,0,1)₇ model optimized for weekly seasonal patterns. The methodology encompasses rigorous stationarity analysis through Augmented Dickey-Fuller and KPSS tests, systematic model specification, and multi-horizon evaluation across 1-day, 3-day, and 7-day forecasts. Performance benchmarking against modern deep learning alternatives (LSTM, Transformer, Prophet) demonstrates SARIMA's superior uncertainty quantification capabilities with 94.2% prediction interval coverage compared to <90% for neural networks. While achieving competitive point accuracy (12.8 % MAPE), SARIMA provides optimal cost-benefit ratio with high interpretability and low computational requirements. Operational regime analysis reveals systematic performance variations, with typical production days achieving 95.1 % coverage versus 77.8 % during consecutive low-production periods. The framework addresses critical grid integration challenges by delivering reliable probabilistic forecasts essential for risk-informed energy dispatch and operational planning in variable renewable energy environments.