TLDR: This research paper reviews the evolution of probabilistic electricity price forecasting, highlighting the shift from single-value predictions to uncertainty-quantifying methods. It covers techniques from Bayesian models and Quantile Regression to advanced Conformal Prediction, examining their application across Day-Ahead, Intra-Day, and Balancing Markets. The paper emphasizes the growing need for adaptive, real-time forecasting, especially with renewable energy integration, and identifies key challenges and future research directions for improving forecast validity and economic value.
Electricity price forecasting is a crucial element in today’s energy markets, especially with the increasing presence of renewable energy sources like wind and solar. These renewables, while vital for a sustainable future, introduce significant volatility and uncertainty into electricity prices. Historically, energy market participants relied on ‘point forecasting,’ which provides a single predicted price. However, as markets become more complex and dynamic, the need for a more comprehensive approach has grown: ‘probabilistic forecasting.’
Probabilistic forecasting moves beyond a single price prediction to offer a range of possible future prices, known as Prediction Intervals (PIs). This approach allows for a much better understanding of the inherent uncertainty and helps market players assess risks more effectively. This shift is essential for optimizing trading strategies, managing financial risks, and maintaining the stability of the electricity grid.
Understanding the Markets
The paper reviews forecasting techniques across three main electricity markets, each with its own unique characteristics and challenges:
- Day-Ahead Market (DAM): This is a forward market where electricity trades are scheduled one day in advance, typically in hourly blocks. It generally has lower price volatility than real-time markets, but the integration of weather-dependent renewables still makes forecasting complex.
- Intra-Day Market (IDM): Operating in the hours leading up to physical delivery, the IDM allows participants to adjust their positions based on updated forecasts. It features multiple trading sessions and is highly sensitive to short-term deviations, leading to higher volatility and shorter decision horizons.
- Balancing Market (BM): This is a real-time market managed by the Transmission System Operator (TSO) to correct supply-demand imbalances after the IDM closes. It doesn’t involve voluntary trading; instead, the TSO dispatches energy as needed, often with very short settlement periods (e.g., 5 minutes). This market is characterized by extreme volatility and frequent price spikes.
The Evolution of Forecasting Methods
The research traces the evolution of probabilistic forecasting methods, from traditional statistical approaches to cutting-edge machine learning techniques:
- Early Approaches: Methods like Bayesian models, Distribution-Based forecasts, Monte Carlo simulations, Bootstrap, and Historical Simulation were among the first to quantify uncertainty. Bayesian methods incorporate prior knowledge and update beliefs with new data, while Monte Carlo simulates multiple future price paths. Bootstrap methods resample historical data to approximate forecast distributions, and Historical Simulation uses past forecast errors to create prediction intervals. While valuable, these methods often face limitations in rapidly changing, non-stationary environments or can be computationally intensive.
- Quantile-Based Techniques: Quantile Regression (QR) and Quantile Regression Averaging (QRA) emerged as significant advancements. QR predicts conditional quantiles of electricity prices, offering insights into uncertainty across different price points, especially during extreme events. QRA combines outputs from multiple point forecasting models, applying QR to estimate quantiles and improve robustness. These methods have become dominant, particularly in the Day-Ahead Market, for their ability to capture non-linearities and tail risks.
- Conformal Prediction (CP): A more recent and increasingly popular approach, Conformal Prediction offers a distribution-free framework for uncertainty quantification. It provides prediction intervals with guaranteed coverage, meaning the true price will fall within the predicted range with a predefined confidence level, regardless of the underlying data distribution. This is a major advantage in volatile markets where data distributions can be complex and constantly evolving. Variants like Split Conformal Prediction (SCP), Ensemble Batch Predictive Interval (EnbPI), and Sequential Predictive Conformal Inference (SPCI) have been developed to better handle time-series data and dynamic market conditions.
Also Read:
- Forecasting Electricity Prices with AI: A New Model for Extreme Market Conditions
- Enhancing Power Grid Optimization Proxies with Constraint-Informed Active Learning
Challenges and Future Directions
While significant progress has been made, particularly in the Day-Ahead Market, the Intra-Day and Balancing Markets remain less explored. These real-time markets present intensified challenges due to higher temporal granularity, greater price volatility, and real-time operational constraints. The paper highlights a crucial trade-off between ‘validity’ (achieving the desired coverage level for prediction intervals) and ‘efficiency’ (producing sharp, narrow intervals).
For traders and grid operators, reliable and actionable forecasts are paramount. The research suggests that while quantile-based models (QR, QRA) are effective in the DAM, adaptive Conformal Prediction methods (especially SPCI and EnbPI) show superior performance in the highly volatile Balancing Market, offering better coverage and financial gains. The Intra-Day Market, however, remains an open frontier for these advanced adaptive methods.
Future research needs to focus on developing CP techniques that can adapt to non-exchangeability and temporal drift in real-time streaming data. There’s also a strong call for systematic evaluation of the economic value of these forecasts, directly linking forecast quality to trading profits and system balancing costs. Finally, the field needs standardized benchmark datasets for the Intra-Day and Balancing Markets to foster reproducibility and accelerate methodological progress. For more in-depth information, you can refer to the full research paper: The Evolution of Probabilistic Price Forecasting Techniques: A Review of the Day-Ahead, Intra-Day, and Balancing Markets.
In conclusion, the journey of electricity price forecasting is moving towards more flexible, data-driven, and adaptive probabilistic methods. Conformal Prediction, with its robust validity guarantees and adaptability to non-stationary data, is poised to play a critical role in navigating the complexities of modern, renewable-dominated electricity markets, especially in the challenging real-time trading environments.
