Electricity Consumption Forecasting Based on Load Peak Data: A Comparative Analysis

Electricity is the backbone of modern industry and technological progress, with global demand continuing to increase. The complexity and high costs of electricity generation, transmission and distribution demand effective grid management strategies to ensure efficient and cost-effective supply. Over the years, various statistical and computational techniques have been applied to improve forecasting accuracy, one of which is ARIMA. In addition, hybrid approaches such as ARIMA-ANN have been explored to improve short-term forecasting accuracy. Despite their widespread use, both traditional ARIMA models and hybrid approaches have limitations in capturing the dynamic and complex nature of electricity consumption patterns, especially in a multivariate context. To address these shortcomings, this study compares the performance of single-input and multi-input transfer function methods for electricity consumption forecasting. It aims to investigate whether there is a significant difference in forecasting results using univariate and multivariate methods with statistical model approaches. The methods are applied to two subsystems, Krian and Krian-Gresik, to evaluate their effectiveness in improving forecasting accuracy. The results show that the single-input transfer function performs better, with the highest R² being 0.90 on both training and testing data. On the other hand, the multi-input transfer function method only performs well on training data, with the highest R² being 0.76. In contrast, it decreases on testing data with the highest R² only reaching 0.69. These findings indicate that the single-input transfer function method is more effective for electricity consumption modeling than the multi-input transfer function method in the analyzed scenarios. However, the R² results are less than satisfactory, this occurs because the electricity consumption data pattern is non-linear.