Enhanced global solar irradiance and temperature forecasting for rooftop PV power generation prediction using multi-step LSTM and GRU based hybrid models

• Developed and evaluated LSTM-based, GRU-based, and MLP models to forecast solar power. • Conducted two scenarios; a one-year univariate forecast and a ten-year multivariate forecast for accurate comparison. • Evaluated the feasibility of a 319-kW rooftop solar photovoltaic system for a building with 2,300 m 2 roof area in Shiraz which has 1,260 MWh/year electrical load demand. • Predicted that the rooftop photovoltaic will generate 7.797 GWh of electricity from 2024 to 2033, that reducing LPSP and CO 2 emissions by 62% and 504.55 tons, respectively. The fluctuating and intermittent characteristics of solar radiation impose significant constraints on several applications. The precise predicting of solar irradiance is an important element in estimating the output power of a solar panel. The study assesses the viability of a rooftop solar photovoltaic system for a 2,300 square meter roof area of a hotel building and utilized monthly data from 1984 to 2023 and hybrid models comprising LSTM-based and GRU-based architectures, along with the MLP model and the standard forecasting methods, to forecast solar radiation and temperatures. Two different scenarios are examined, differing in forecasting horizon and input data type; scenario A features a one-year forecast horizon with univariate input data, while scenario B encompasses a ten-year forecast horizon with multivariate input data. The CNN-GRU model in scenario B demonstrates high accuracy relative to other models across two scenarios, utilizing six input parameters: solar radiation, temperature, wet bulb temperature, surface pressure, relative humidity, and surface temperature, along with two outputs: solar radiation and temperature. The RMSE and R 2 values were 0.0368 W/m 2 and 98.54 % for solar radiation, and 0.0445 °C and 97.69 % for temperature, respectively. As a result, the mean annually output power for the ten years from 2024 to 2033 is forecasted to be 779.7 MWh for the 319-kW rated power rooftop PV system. Also, the building load demand was calculated to be 1,260 MWh/year using Openstudio and energyPlus. Finally, with these characteristics, this production would cover about 62 % of the hotel annual load demand, and lower the LPSP from 1 to 0.38 and CO 2 emissions from 822.78 tons to 318.23 tons per year.