<![CDATA[Solar power plants are highly dependent on solar radiation intensity, which fluctuates due to changes in atmospheric conditions. To maintain system stability and efficiency, an accurate short-term solar radiation prediction model is essential. This study developed a model for forecasting global solar radiation one hour ahead using the Backpropagation Neural Network (BPNN) method. The dataset was obtained from a photovoltaic (PV) system at Building A8 of Surabaya State University, recorded over four days (June 14-17, 2025) at two-minute intervals. Five input variables were used: clearness index, solar radiation, air temperature, air humidity, and PV output power, resulting in a total of 3,020 data samples. The model was trained through a trial-and-error process by varying the number of neurons, hidden layers, and epochs to determine the optimal configuration. The forecast capability of the model was assessed through four statistical indicators: Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and the coefficient of determination (R²). The best performance was achieved with a network architecture of 15 input neurons representing input variables resulting from data transformation using the sliding window method, one hidden with 25 neurons, and a single unit in the output layer trained for 2000 epochs, resulting in R2 = 0.98, MAPE = 5.89%, and MSE = 0.00027. The novelty of this research lies in the integration of meteorological data with actual PV power output as model input, enabling the network to capture more realistic nonlinear temporal relationships. The proposed short-term forecasting model provides a practical approach to predicting solar radiation based on historical data and can support efficient energy management and photovoltaic system performance analysis.]]>
