Two-Stage Data-Driven Modeling for Day-Ahead Solar Power Forecasting: A Case Study of the SUT Campus Floating Solar System

This paper presents a comparative study of forecasting methods for solar PV power generation, using a two-step forecasting approach. The data includes solar irradiance, global tilted irradiance, temperature, humidity, and wind speed, analyzed with LSTM, Bi-LSTM, GRU, and RF models, 7,665 samples of 30-minute recordings from 2023. In the first part of the study, the important factors affecting power generation, irradiance and temperature, are forecasted using the LSTM method to consider the relationship between external factors and power generation. In the second part of the study, the results from the aforementioned forecasting are used to predict power generation. The forecast results obtained by LSTM, Bi-LSTM, GRU, and RF, were investigated and compared. This study not only helps to better understand the relationship between external factors and power generation but also significantly improves the overall forecasting efficiency.