TL;DR
This paper tackles the challenge of energy storage arbitrage in electricity markets, where price uncertainty significantly impacts profitability and risk. The authors propose a computationally efficient framework that incorporates various uncertainty models and risk preferences, offering a robust decision-making tool for market participants.
What Problem Does This Paper Address?
The increasing integration of renewable energy sources into electricity grids creates significant operational challenges. Energy storage systems are crucial for balancing supply and demand, and a key strategy for their economic viability is arbitrage: buying electricity when prices are low and selling when they are high. However, accurately forecasting electricity prices is difficult, and this uncertainty introduces substantial risk for storage operators. If prices don't move as predicted, operators can face significant losses, deterring participation, especially from non-professional players. This paper aims to provide a framework for making informed arbitrage decisions that balance profit potential with risk management, considering different levels of conservativeness and market conditions.
Methods
The paper proposes a general uncertainty-incorporated storage arbitrage formulation that can accommodate various price uncertainty models and risk preferences. They introduce two main approaches:
- Robust Decision Making: This is a deterministic, worst-case approach. The optimization problem is formulated to maximize performance under the most unfavorable price realization within a defined uncertainty set. They explore polyhedral and ellipsoidal uncertainty sets for this.
- Chance-Constrained Decision Making: This approach works with probabilistic uncertainty models. It ensures that the stochastic objective reaches optimality with a certain probability (confidence level). They specifically consider normal and log-normal distributions for price forecasts.
The authors also develop a performance evaluation framework using "efficient frontiers." These frontiers graphically represent the trade-off between risk (e.g., number of negative profit days) and profit (e.g., expected profit) for different strategies and uncertainty levels.
Datasets
The study uses hourly price data collected from the New York ISO (NYISO) for the New York City (NYC) zone from 2014 to 2023. Uncertainty quantification was performed using data from 2014 to 2021, with the last two years (2022-2023) reserved for out-of-sample testing.
Key Insights and Ideas
- Robustness vs. Profitability Trade-off: As strategies become more conservative (higher uncertainty budget Γ), worst-case and expected profits generally decrease. This highlights the "price of robustness."
- Polyhedral Strategies Excel in Volatility: Polyhedral uncertainty representations (Poly-Quantile, Poly-Mean-Std) and Ellip-Cov tend to cover a broader range on the efficient frontier, especially in volatile markets. This suggests they are better at managing risk without sacrificing too much profit potential.
- Chance-Constrained Limitations: While chance-constrained strategies can guarantee positive profits even in conservative scenarios, they might not bound the magnitude of risk as effectively as robust strategies.
- Market Conditions Matter: The efficient frontiers shift significantly based on market conditions. High volatility (like in 2022) leads to a high-profit, high-risk frontier, while more stable periods (like in 2023) result in a low-profit, low-risk frontier.
- Data Sensitivity: Ellipsoidal strategies show higher sensitivity to the historical datasets used for characterization, while polyhedral strategies appear more resilient.
Key Results
- The paper demonstrates that robust strategies generally outperform chance-constrained ones in risk management, particularly in volatile markets, while offering comparable profit expectations.
- The proposed framework allows for the selection of strategies that align with individual risk preferences, providing a structured decision-making tool.
- The study quantifies the impact of uncertainty on arbitrage strategies, showing how different uncertainty representations and conservativeness levels affect worst-case profit, expected profit, and the ratio of non-negative profit days. For instance, at Γ=0.8, the non-negative profit ratio for Poly-Quantile and Poly-Mean-Std remains high (0.94), while for Chance-Normal it drops to 0.57.
Limitations and Future Work
- Computational Cost: While the paper aims for computational efficiency, robust and chance-constrained optimization can still be computationally intensive, especially for longer time horizons. The paper mentions that some methods still require substantial computational power.
- Model Simplifications: The arbitrage model omits operational costs for simplicity. The paper also notes that approximating sums of lognormals by a single lognormal variable is a simplification.
- Future Work: The authors suggest that the framework can be adapted to evolving market conditions and offers practical implications for deploying risk-averse solutions in distributed storage. Further research could explore more sophisticated uncertainty models and integrate operational costs for a more comprehensive analysis.
Summary updated