Photovoltaic System Power Output Forecasting Using Vector Auto-Regression (VAR)

EasyChair Preprint 15480

Abstract

As the penetration of solar energy sources into a power system increases, the significance of precise short-term forecasts for solar power plants becomes paramount. However, the erratic and non-periodic nature of solar poses challenges in accurately predicting the output power. The main objective of this paper is to study power output forecasting in photovoltaic systems based on vector autoregression (VAR) and SCADA data. The study of forecasting aims to predict the condition of specified parameters e.g., power output for the next time ahead to set up the planning for anticipation action if the predicted parameters fall into bad conditions. In this study, the data utilized to train the VAR algorithm originates from seven months of historical solar power plant operations collected from 1 December 2022 to 30 September 2023. The trained VAR is then applied to forecast one month ahead of the power output. The results are evaluated with performance forecasting measures namely nRMSE and nMAE which are relative measures with a basis on maximum power output generated in photovoltaic systems. Referring to the normalized RMSE, the maximum active power in this study reached 1350 kW for July 2023, and then the nRMSE and nMAE were 0.06 and 0.04, respectively. The nRMSE and nMAE give small values that mean the proposed method is admissible and could be applied to real solar power plant systems in forecasting the output power.

Keyphrases: Forecasting, Performance, Photovoltaic, SCADA, VAR

@booklet{EasyChair:15480,
  author    = {Agussalim Syamsuddin and Andrew Cahyo Adhi and Achmad Widodo and Toni Prahasto and Amie Kusumawarhani},
  title     = {Photovoltaic System Power Output Forecasting Using Vector Auto-Regression (VAR)},
  howpublished = {EasyChair Preprint 15480},
  year      = {EasyChair, 2024}}