Abstract
This paper applies computational techniques of convex stochastic optimization to optimal operation and valuation of electricity storages in the face of uncertain electricity prices. Our approach is applicable to various specifications of storages, and it allows for hard constraints on storage capacity and charging speed. Our valuations are based on the indifference pricing principle, which builds on optimal trading strategies and calibrates to the user's initial position, market views and risk preferences. We illustrate the effects of storage capacity and charging speed by numerically computing the valuations using stochastic dual dynamic programming.
Introduction
As renewable energy becomes a larger part of the power grid, electricity prices are becoming increasingly volatile and seasonal. This creates a massive incentive for energy storage, but valuing these assets is notoriously difficult. Standard financial models often fall short because energy markets are "incomplete"—you can't perfectly hedge away all risks. Furthermore, physical storages have "hard constraints": a battery has a maximum capacity and a maximum charging speed. Traditional mathematical tools, such as the Hamilton-Jacobi-Bellman (HJB) equations, often struggle to incorporate these state constraints effectively. This research provides a computationally efficient way to determine the "indifference price"—the maximum amount an agent should be willing to pay for storage without being worse off—while respecting the laws of physics and the agent's specific risk tolerance.
Methodology
The storage management problem is formulated as a convex stochastic optimal control problem over a discrete time horizon. The key components of this approach include:
- Indifference Pricing Principle: Instead of looking for a single "market price," the price is calculated based on the agent's utility, comparing optimal "well-being" with the storage versus without it.
- Stochastic Dual Dynamic Programming (SDDP): To solve the multi-stage optimization problem efficiently, SDDP approximates the "cost-to-go" functions using a collection of linear inequalities (cuts), handling hard constraints on storage levels and charging speeds.
- Markovian Discretization: Electricity prices are modeled as a mean-reverting Ornstein-Uhlenbeck process, discretized into a finite-state Markov chain using Gauss-Hermite integration quadratures.
- Decision-Observation-Decision Format: The model allows for a sophisticated control structure where decisions can be made both before and after new information is revealed, mimicking real-world day-ahead and intraday trading.
Data and Model Calibration
The model is calibrated using historical electricity price data from the European Power Exchange (EPEX). Specifically, the authors utilize hourly data for both Day-ahead and Intraday (ID1) prices to capture the market dynamics that storage systems exploit.
These data series are used to estimate parameters for a mean-reverting Ornstein-Uhlenbeck process, accounting for annual, weekly, and daily seasonalities. This model then generates 1,000 out-of-sample scenarios for evaluation.
Key Results and Insights
- Risk Aversion Saturation: For highly risk-averse agents, the value of storage saturates. Beyond a certain point, additional capacity does not increase the asset's value because the agent is unwilling to take the price risk associated with holding massive amounts of energy.
- Sensitivity to Charging Speed: Valuation is highly sensitive to charging and discharging speeds. A battery that can react quickly to price spikes is significantly more valuable than a slow-charging one of the same capacity.
- SDDP Efficiency: The SDDP algorithm is remarkably efficient, providing accurate valuations that converge quickly (often in fewer than 10 iterations for deterministic cases and stable within 500 for stochastic cases).
- Volatility and Value: Higher price volatility directly increases the indifference price. The more chaotic the market, the more valuable the ability to store energy becomes.
Conclusion and Future Work
While the current framework provides a robust tool for storage valuation, there are avenues for expansion. Future work could incorporate more complex physical models (e.g., battery degradation), stochastic interest rates, or participation in multiple markets simultaneously (like frequency regulation). Additionally, exploring different utility functions, such as Conditional Value at Risk (CVaR), could provide further insights into diverse risk preferences.
Summary updated