Comparative Analysis of Time Series and Artificial Intelligence Algorithms for Short Term Load Forecasting

• Published in: Conference proceedings - Canadian Conference on Electrical and Computer Engineering pp. 1 - 7
• Main Authors: Beydoun, Hana, Khan, Ahmad, Ali Arefifar, S.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 12.09.2021
• Subjects: ARIMA; Artificial intelligence; Data models; distribution systems; Exponential Smoothing; Facebook Prophet; Holt-Winters; Load forecast; Load forecasting; neural network; Predictive models; random forest; smart grid; Smoothing methods; STLF; Time series analysis; Training
• ISSN: 2576-7046
• DOI: 10.1109/CCECE53047.2021.9569166

Short-term load forecasting (STLF) has been an interesting subject for power utility companies in the last few decades. In this paper, five different time series forecasting models are employed and integrated with Artificial Intelligence (AI) models to study the applicability of the methodologies for short term load forecasting. The AI models are employed and integrated with the seasonal ARIMA methodology to study their effects on STLF, whereas the emerging Facebook Prophet model is utilized along with the conventional Holt-Winter's to evaluate against the original and hybrid ARIMA models. The hybrid ARIMA approach involves modeling the ARIMA residuals and increasing the input features space to include weather parameters in addition to previous days load data. For our case study pertaining to modeling and predicting the load on a short-term basis using the data from a Michigan based utility, the Holt-Winters' model performance exceeded the performance of the remaining methodologies, where it achieved an accuracy of 99% and 98.6% in both the training and testing phases, respectively.