Long-term Solar Irradiance Forecasting Based on TimesNet with Multiple Weather Parameters

In renewable energy grids, solar irradiance prediction is significant for electric operation planning. The emphasis of our work is forecasting irradiance changes over multiple days ahead using hourly irradiance data in association with weather parameters such as temperature, humidity, barometric pressure, wind speed, and cloud cover. Specifically, a framework based on TimesNet was proposed for 4-day global horizontal irradiance (GHI) forecasting and a comparative experiment was conduct with three well-performing models in long-term forecasting tasks. The experimental results indicate that the suggested method outperformed the accuracy when compared to other models, with MSE 0.198, MAE 0.259, RMSE 0.445 and R² 0.787.

I. Introduction

Energy is the basis and driving force of sustainable economic and social development. In recent years, the energy crisis and climate warming have become global problems. As an inexhaustible, clean and environmentally friendly new energy source, solar energy is one of the most reliable alternatives to widely used fossil fuels. The solar photovoltaic sector is expected to have the largest installed capacity (8,519 GW) by 2050, generating about 25% of total electricity demand [1]. However, the power output of a solar PV system depends on the variation of solar irradiance, which is affected by a variety of conditions such as seasonality and geography [2]. Global horizontal irradiance (GHI) has a linear relationship with the power output of PV cells, therefore, accurate GHI forecasts are important for grid planning, operation and management.