Comparison of CLOT-Adjusted AHI-8/9 and FY-4A Solar Irradiance Products for Solar PV Power Output Forecasting Using LSTM

Accurate solar photovoltaic power output (PPV) forecasting in grid planning is crucial to ensure sufficient power supply during peak hours and periods with high electricity demand. Indeed, it is a challenge to balance supply and demand in the grid, especially when using variable and unreliable power output from solar PV systems. The said challenge may be overcome using various forecasting techniques, including statistical, machine learning, and hybrid. This work uses a long short-term memory (LSTM) architecture in solar PV output forecasting to capture long-term dependencies and temporal patterns in time series data. This study examines the performance of the LSTM model in forecasting solar PV output using two satellite-derived solar irradiance data, namely, Short Wave Radiation (SWR) from Advanced Himawari Imager 8/9 (AHI-8/9) and Surface Solar Irradiance (SSI) from Fengyun 4A (FY-4A), adjusted for cloud effects using cloud optical thickness (CLOT) from AHI-8/9. The study evaluates the accuracy of the models using root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). Results show that using the adjusted SSI (R′SSI), a more accurate forecasted solar PV output was produced than the adjusted SWR (R′SWR). Moreover, the R′SWR outperforms the original RSWR across all metrics. Meanwhile, the original RSSI performs better than the R′SSI. Results in this work recommend using the R′SWR in forecasting PPV for improved predictability and optimal use of solar resources. For future studies, CLOT correction can also be derived from FY-4A to adjust SSI products and improve forecast accuracy.