Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

OrderFusion: Encoding Orderbook for End-to-End Probabilistic Intraday Electricity Price Forecasting

Runyao Yu, Yuchen Tao, Fabian Leimgruber, Tara Esterl, Jochen Stiasny, Derek W. Bunn, Qingsong Wen, Hongye Guo, Jochen L. Cremer

TL;DR

This work tackles probabilistic intraday electricity price forecasting in continuous intraday markets by modeling the full buy–sell interaction structure of the orderbook. It introduces OrderFusion, an end-to-end, parameter-efficient framework that encodes orderbook data, applies dual masking and cross-side fusion via iterative attention, and outputs non-crossing, multi-quantile forecasts through a hierarchical head. Empirical results in German and Austrian markets show that OrderFusion outperforms Naïve, 1D- and 2D-encoded baselines, with ablations confirming the contribution of each component and the reduction of quantile crossing. The approach demonstrates the value of injecting domain priors into forecasting models and provides an open-source implementation for practical deployment in CID electricity markets.

Abstract

Probabilistic intraday electricity price forecasting is becoming increasingly important with the growth of renewable generation and the rise in demand-side engagement. Their uncertainties have increased the trading risks closer to delivery and the subsequent imbalance settlement costs. As a consequence, intraday trading has emerged to mitigate these risks. Unlike auction markets, intraday trading in many jurisdictions is characterized by the continuous posting of buy and sell orders on power exchange platforms. This dynamic orderbook microstructure of price formation presents special challenges for price forecasting. Conventional methods represent the orderbook via domain features aggregated from buy and sell trades, or by treating it as a multivariate time series, but such representations neglect the full buy-sell interaction structure of the orderbook. This research therefore develops a new order fusion methodology, which is an end-to-end and parameter-efficient probabilistic forecasting model that learns a full interaction-aware representation of the buy-sell dynamics. Furthermore, as quantile crossing is often a problem in probabilistic forecasting, this approach hierarchically estimates the quantiles with non-crossing constraints. Extensive experiments on the market price indices across high-liquidity (German) and low-liquidity (Austrian) markets demonstrate consistent improvements over conventional baselines, and ablation studies highlight the contributions of the main modeling components. The methodology is available at: https://runyao-yu.github.io/OrderFusion/.

OrderFusion: Encoding Orderbook for End-to-End Probabilistic Intraday Electricity Price Forecasting

TL;DR

This work tackles probabilistic intraday electricity price forecasting in continuous intraday markets by modeling the full buy–sell interaction structure of the orderbook. It introduces OrderFusion, an end-to-end, parameter-efficient framework that encodes orderbook data, applies dual masking and cross-side fusion via iterative attention, and outputs non-crossing, multi-quantile forecasts through a hierarchical head. Empirical results in German and Austrian markets show that OrderFusion outperforms Naïve, 1D- and 2D-encoded baselines, with ablations confirming the contribution of each component and the reduction of quantile crossing. The approach demonstrates the value of injecting domain priors into forecasting models and provides an open-source implementation for practical deployment in CID electricity markets.

Abstract

Probabilistic intraday electricity price forecasting is becoming increasingly important with the growth of renewable generation and the rise in demand-side engagement. Their uncertainties have increased the trading risks closer to delivery and the subsequent imbalance settlement costs. As a consequence, intraday trading has emerged to mitigate these risks. Unlike auction markets, intraday trading in many jurisdictions is characterized by the continuous posting of buy and sell orders on power exchange platforms. This dynamic orderbook microstructure of price formation presents special challenges for price forecasting. Conventional methods represent the orderbook via domain features aggregated from buy and sell trades, or by treating it as a multivariate time series, but such representations neglect the full buy-sell interaction structure of the orderbook. This research therefore develops a new order fusion methodology, which is an end-to-end and parameter-efficient probabilistic forecasting model that learns a full interaction-aware representation of the buy-sell dynamics. Furthermore, as quantile crossing is often a problem in probabilistic forecasting, this approach hierarchically estimates the quantiles with non-crossing constraints. Extensive experiments on the market price indices across high-liquidity (German) and low-liquidity (Austrian) markets demonstrate consistent improvements over conventional baselines, and ablation studies highlight the contributions of the main modeling components. The methodology is available at: https://runyao-yu.github.io/OrderFusion/.

Paper Structure

This paper contains 28 sections, 26 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Preliminary
  3. OrderFusion
  4. Encoding
  5. Backbone
  6. Dual Masking Layer
  7. Iterative Fusion Layer
  8. Aggregation Layer
  9. Head
  10. Loss
  11. Baselines
  12. Naïve
  13. 1D Encoding
  14. 2D Encoding
  15. Evaluation Metrics
  16. ...and 13 more sections

Figures (3)

  • Figure 1: Analysis of orderbook and price indices. A, Buy-sell interactions for delivery at 18:00 on 2024-07-23. Buyers and sellers adjust bids and offers based on the opposite side, reflecting strategic interactions. As delivery time approaches, prices exhibit downward jumps. B, Histogram of ID$_3$ from 2022 to 2024, illustrating a shift toward greater price stability in recent years. C, Seasonal boxplot of ID$_3$, highlighting seasonal fluctuations across years. D, Count of high prices ($>$500 €/MWh) for ID$_1$, ID$_2$, and ID$_3$ from 2022 to 2024. A sharp decline in high prices is shown after energy crisis in 2022. E, Count of negative prices ($<$0 €/MWh) for ID$_1$, ID$_2$, and ID$_3$ from 2022 to 2024. Negative-price events increase substantially over time, indicating growing market imbalances.
  • Figure 2: Structure of OrderFusion. The model takes the 2$\times$2D encoding of orderbook as input. The buy-side input and sell-side input are masked through dual masking layers and iteratively fused to form representations of buy–sell interactions in the latent space. The representations are aggregated across different degrees of interactions and then passed through a hierarchical head to generate multiple quantile estimates, enabling end-to-end probabilistic forecasting.
  • Figure 3: Result analysis on the testing data for three price indices. A, True price versus median prediction for the German market. B, Testing loss (AQL) versus cutoff length $L$ for the German market, with the optimal cutoff indicated. C, True price versus median prediction for the Austrian market. D, Testing loss (AQL) versus cutoff length $L$ for the Austrian market, with the optimal cutoff indicated.