Forecasting Hierarchical Time Series in Power Generation

Tiago Silveira Gontijo,Marcelo Azevedo Costa

doi:10.3390/en13143722

Tiago Silveira Gontijo, Marcelo Azevedo Costa

Open Access

https://doi.org/10.3390/en13143722

Copy DOI

Journal: Energies	Publication Date: Jul 20, 2020
Citations: 7	License type: CC BY 4.0

Affiliation: Universidade Federal de Minas Gerais

Abstract

Academic attention is being paid to the study of hierarchical time series. Especially in the electrical sector, there are several applications in which information can be organized into a hierarchical structure. The present study analyzed hourly power generation in Brazil (2018–2020), grouped according to each of the electrical subsystems and their respective sources of generating energy. The objective was to calculate the accuracy of the main measures of aggregating and disaggregating the forecasts of the Autoregressive Integrated Moving Average (ARIMA) and Error, Trend, Seasonal (ETS) models. Specifically, the following hierarchical approaches were analyzed: (i) bottom-up (BU), (ii) top-down (TD), and (iii) optimal reconciliation. The optimal reconciliation models showed the best mean performance, considering the primary predictive windows. It was also found that energy forecasts in the South subsystem presented greater inaccuracy compared to the others, which signals the need for individualized models for this subsystem.

Highlights

The advent of Industry 4.0 revolutionized factories worldwide, since it allowed the connectivity between measuring machines and the automation of companies, distributing the capacity to collect massive volumes of data [1]
The present paper presents a case study using a power generation data set from Brazil (2018–2020)
Note that the model was estimated, taking the main hierarchical adjustment approaches into account, for the following levels: (i) total power generation in Brazil (Level 0), (ii) total energy generation by electrical subsystem (Level 1), and (iii) total energy generation by the energy generating source (Level 2)

Summary

Introduction

The advent of Industry 4.0 revolutionized factories worldwide, since it allowed the connectivity between measuring machines and the automation of companies, distributing the capacity to collect massive volumes of data [1]. In high-level data analysis, forecasting models allow the extraction of behavior patterns, as well as the prediction of future values for the collected data set [2]. In the above-mentioned scenario, the construction of predictive models is gaining prominence in the literature [3,4,5], since economic agents deal with uncertainty in multiple spheres and aim to achieve the best results using available resources [6]. Developing acceptably accurate models presents a meaningful challenge, as prediction is a technique that deals with risk and there will always be a fundamental error associated with it. The best model is the one that most adequately represents the phenomenon of interest. In relation to the object of our study, power generation, there are several forecasting applications:

Methods

Results

Conclusion