Electricity Sales Forecasting Using Hybrid Autoregressive Integrated Moving Average and Soft Computing Approaches in the Absence of Explanatory Variables

Abstract

Electricity is important because it is the most common energy source that we consume and depend on in our everyday lives. Consequently, the forecasting of electricity sales is essential. Typical forecasting approaches often generate electricity sales forecasts based on certain explanatory variables. However, these forecasting approaches are limited by the fact that future explanatory variables are unknown. To improve forecasting accuracy, recent hybrid forecasting approaches have developed different feature selection techniques (FSTs) to obtain fewer but more significant explanatory variables. However, these significant explanatory variables will still not be available in the future, despite being screened by effective FSTs. This study proposes the autoregressive integrated moving average (ARIMA) technique to serve as the FST for hybrid forecasting models. Aside from the ARIMA element, the proposed hybrid models also include artificial neural networks (ANN) and multivariate adaptive regression splines (MARS) because of their efficient and fast algorithms and effective forecasting performance. ARIMA can identify significant self-predictor variables that will be available in the future. The significant self-predictor variables obtained can then serve as the inputs for ANN and MARS models. These hybrid approaches have been seldom investigated on the electricity sales forecasting. This study proposes several forecasting models that do not require explanatory variables to forecast the industrial electricity, residential electricity, and commercial electricity sales in Taiwan. The experimental results reveal that the significant self-predictor variables obtained from ARIMA can improve the forecasting accuracy of ANN and MARS models.

IntelliHome: An internet of things‐based system for electrical energy saving in smart home environment

“…During the past decade, hybrid statistical and ML‐based methods have been studied for time series forecasting . These are receiving increased attention as researchers at transportation network companies such as Uber Technologies, Inc. are betting on them to solve various forecasting problems across different use cases.…”

“…During the past decade, hybrid statistical and ML-based methods have been studied for time series forecasting. [17][18][19] These are receiving increased attention as researchers at transportation network companies such as Uber Technologies, Inc. are betting on them to solve various forecasting problems across different use cases. In this context, we studied the applicability of the hybrid time series forecasting method by Smyl,20 a Data Scientist at Uber Technologies, Inc., to the case of our energy usage forecasting problem; this method was submitted to the 2018 edition of the popular open forecasting competition known as M (Makridakis) Competition, 21 demonstrating an impressive accuracy that gave it victory.…”

Input Modeling and Uncertainty Quantification for Improving Volatile Residential Load Forecasting

“…For example, in [1] and [2], standard multiple linear regression models are enhanced by different techniques to tackle large datasets for load forecasting. In [3] and [4], modified ARIMA models are used to capture customers' load patterns. In [5] and [6], support vector regression algorithms are modified to achieve a high predictive accuracy especially when handling high-resolution distribution systems.…”