Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An adaptive standardisation methodology for Day-Ahead electricity price forecasting

Carlos Sebastián, Carlos E. González-Guillén, Jesús Juan

TL;DR

The paper addresses dataset shift and non-stationarity in Day-Ahead electricity price forecasting by introducing adaptive standardisation with rolling-window mu_d and sigma_d estimation, transforming prices and features to enable robust learning. It evaluates two learning architectures, LEAR and DNN, across five markets (including two new 2019–2023 datasets), and shows consistent improvements, especially when combining adaptive and traditional methods. Ensemble strategies that blend adaptive and non-adaptive models yield the strongest forecasts, confirmed by multivariate Diebold-Mariano tests. The work provides public datasets and a practical forecasting procedure that retrains daily, offering a robust framework for forecasting under regime shifts in modern electricity markets.

Abstract

The study of Day-Ahead prices in the electricity market is one of the most popular problems in time series forecasting. Previous research has focused on employing increasingly complex learning algorithms to capture the sophisticated dynamics of the market. However, there is a threshold where increased complexity fails to yield substantial improvements. In this work, we propose an alternative approach by introducing an adaptive standardisation to mitigate the effects of dataset shifts that commonly occur in the market. By doing so, learning algorithms can prioritize uncovering the true relationship between the target variable and the explanatory variables. We investigate five distinct markets, including two novel datasets, previously unexplored in the literature. These datasets provide a more realistic representation of the current market context, that conventional datasets do not show. The results demonstrate a significant improvement across all five markets using the widely accepted learning algorithms in the literature (LEAR and DNN). In particular, the combination of the proposed methodology with the methodology previously presented in the literature obtains the best results. This significant advancement unveils new lines of research in this field, highlighting the potential of adaptive transformations in enhancing the performance of forecasting models.

An adaptive standardisation methodology for Day-Ahead electricity price forecasting

TL;DR

The paper addresses dataset shift and non-stationarity in Day-Ahead electricity price forecasting by introducing adaptive standardisation with rolling-window mu_d and sigma_d estimation, transforming prices and features to enable robust learning. It evaluates two learning architectures, LEAR and DNN, across five markets (including two new 2019–2023 datasets), and shows consistent improvements, especially when combining adaptive and traditional methods. Ensemble strategies that blend adaptive and non-adaptive models yield the strongest forecasts, confirmed by multivariate Diebold-Mariano tests. The work provides public datasets and a practical forecasting procedure that retrains daily, offering a robust framework for forecasting under regime shifts in modern electricity markets.

Abstract

The study of Day-Ahead prices in the electricity market is one of the most popular problems in time series forecasting. Previous research has focused on employing increasingly complex learning algorithms to capture the sophisticated dynamics of the market. However, there is a threshold where increased complexity fails to yield substantial improvements. In this work, we propose an alternative approach by introducing an adaptive standardisation to mitigate the effects of dataset shifts that commonly occur in the market. By doing so, learning algorithms can prioritize uncovering the true relationship between the target variable and the explanatory variables. We investigate five distinct markets, including two novel datasets, previously unexplored in the literature. These datasets provide a more realistic representation of the current market context, that conventional datasets do not show. The results demonstrate a significant improvement across all five markets using the widely accepted learning algorithms in the literature (LEAR and DNN). In particular, the combination of the proposed methodology with the methodology previously presented in the literature obtains the best results. This significant advancement unveils new lines of research in this field, highlighting the potential of adaptive transformations in enhancing the performance of forecasting models.
Paper Structure (19 sections, 6 equations, 5 figures, 10 tables)

This paper contains 19 sections, 6 equations, 5 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Previous work
  3. Electricity price forecasting (EPF)
  4. Adaptive transformation schemes
  5. Proposed methodology
  6. Experiments
  7. Learning algorithms
  8. LEAR
  9. DNN
  10. Datasets
  11. Evaluation metrics
  12. Forecasting procedure
  13. Results
  14. Statistical testing
  15. Results
  16. ...and 4 more sections

Figures (5)

  • Figure 1: Spanish Day-Ahead market prices through the years
  • Figure 2: Comparison of the current price in the Spanish Day-Ahead market during hour 20 (top), its median-arsinh transformation (middle) and the resulting process derived of applying our methodology with $v=7$ days, 168 hours (bottom)
  • Figure 3: Markets considered. Observe the differences between the variability of current markets and previous markets.
  • Figure 4: MAE per month and per dataset for each model set. Each individual model is shown in translucent form.
  • Figure 5: Multivariate DM test between each analyzed model, including ensembles, for the every dataset considered