Applying Johansen VECM cointegration approach to propose a forecast model of photovoltaic power output plant in Reunion Island

• Main Authors: Yannick Fanchette, Harry Ramenah, Camel Tanougast, Michel Benne
• Format: Article
• Language: English
• Published: AIMS Press 2020-05-01
• Series: AIMS Energy
• Subjects: solar photovoltaic; time series; statistical model; johansen cointegration; long run equilibrium; residuals
• Online Access: https://www.aimspress.com/article/10.3934/energy.2020.2.179/fulltext.html
• ISSN: 2333-8326; 2333-8334

Since 2007 Reunion Island, a French overseas region located in the Indian Ocean, aims to achieve energy self-sufficiency by 2030. The French government has made this insular zone an experimental territory for renewable energy resources (RES) by implementing great powers photovoltaic (PV) plants. However, the performance of PV conversion is highly climate dependent, and there have been many research contributions to show that the two main factors that influence PV cell efficiency are solar radiation and cell temperature. Moreover, considering the high variability of environmental factors on PV plants, the high penetration of PV in electric systems may threaten the stability and reliability of the electrical power grid. In this study, a linear relation analysis of time series data collected over one year is performed in order to investigate the dependent variable of PV power output from explanatory variables such as solar irradiance, cell temperature, wind speed and humidity. The originality of this paper is to apply cointegration methods, usual tools of econometrics, to PV systems. More precisely, this research work lies in the use a robust statistical method to model a vector cointegrating relationship linking the PV power output and the four environmental parameters mentioned above, to make accurate forecasts in a tropical area. The Johansen vector error correction model (VECM) cointegration approach is used to determine the most appropriate PV power output forecasting when the desired model is concerned with N explanatory variables and for N > 2. This long run equilibrium relationship has been tested over many years of data and the outcome is more than reliable when comparing the model to measured data.