Application of a hybrid model based on multiple linear regression -principle component analysis (MLR-PCA) for electricity export forecasting

Abstract

International electricity trade as a strategic commodity plays a prominent role in the foreign trade market of countries.Electricity export forecasting leads to better production planning, supply security, blackouts reduction, and obligations fulfillment. This paper aimed to provide a model for electricity export forecasting. In this regard, electricity consumption in different consumer sectors, gas consumption, population, GDP, and electricity prices have been entered into the multiple regression model as predictor variables. Although R 2 =.976, F=66.110, and SIG<.05 indicate the model appropriateness, the high correlation between the predictor variables created collinearity. In other words, Tolerance, VIF (variance inflation factor), Eigenvalue and the Condition Index are less than .2, more than 10, close to zero, and more than 15 respectively. To solve this problem, two hybrid methods of Multiple Regression-First Difference Function and Multiple Regression-PCA have been used. In the first hybrid method (R 2 =.553) the Tolerance and VIF index still show the presence of collinearity. In the second hybrid method (R 2 =.936, F=169.9, SIG<.05) due to all the mentioned indicators, the collinearity has been completely resolved. So, the MLR-PCA method is the most appropriate model for electricity export forecasting. The data collected from Iran have been used to illustrate the model.