Day-Ahead Electricity Price Forecasting Using Merit-Order Curves Time Series

2025-12-19

AI Summary

Frequent negative day-ahead electricity prices—particularly during low-load, high-generation periods under high renewable penetration—degrade forecasting accuracy. Method: This paper proposes a novel price forecasting paradigm that employs supply-demand marginal cost curves as an intermediate representation. We first construct time-series marginal cost curves, then apply functional principal component analysis (FPCA) to capture their morphological evolution in reduced dimensionality; subsequently, a regularized vector autoregressive (VAR) model captures curve dynamics, enabling both point and quantile probabilistic forecasts. Contribution/Results: Unlike conventional direct market-clearing price forecasting, our framework achieves both mechanistic interpretability and improved accuracy. Experiments on Italian day-ahead market data from 2023–2024 show an average reduction of ~5% in prediction error overall, with gains up to 10% during midday photovoltaic generation peaks.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Planning, Routing, and Scheduling: Model-Based Reasoning
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

We introduce a general, simple, and computationally efficient framework for predicting day-ahead supply and demand merit-order curves, from which both point and probabilistic electricity price forecasts can be derived. Specifically, we leverage functional principal component analysis to efficiently represent a pair of supply and demand curves in a low-dimensional vector space and employ regularized vector autoregressive models for their prediction. We conduct a rigorous empirical comparison of price forecasting performance between the proposed curve-based model, i.e., derived from predicted merit-order curves, and state-of-the-art price-based models that directly forecast the clearing price, using data from the Italian day-ahead market over the 2023-2024 period. Our results show that the proposed curve-based approach significantly improves both point and probabilistic price forecasting accuracy relative to price-based approaches, with average gains of approximately 5%, and improvements of up to 10% during mid-day hours, when prices occasionally drop due to high renewable generation and low demand.

Problem

• Forecasting day-ahead electricity prices using merit-order curves
• Comparing curve-based and price-based forecasting models
• Improving accuracy of point and probabilistic price predictions

Innovation

• Functional PCA for low-dimensional curve representation
• Regularized vector autoregressive models for curve prediction
• Deriving price forecasts from predicted merit-order curves