TL;DR
This paper introduces a novel approach to the Security-Constrained Economic Dispatch (SCED) problem by formulating it as a cardinality minimization problem. This new model aims to minimize operational costs while simultaneously reducing the number of transmission lines operating in emergency zones during contingency events, thereby enhancing system reliability. The authors develop both a difference-of-convex approximation and a mixed-integer programming formulation to solve this complex problem, demonstrating significant reductions in operating costs and avoiding prolonged periods of high electricity prices.
What Problem Does This Paper Address?
The core challenge addressed is the inherent conflict between economic efficiency and system reliability in power grids. Independent System Operators (ISOs) and Regional Transmission Organizations (RTOs) are tasked with ensuring the cheapest electricity possible while maintaining grid stability, especially during unexpected events like equipment failures (contingencies). Traditional SCED models often focus on minimizing cost, but may allow transmission lines to operate in "emergency zones" (exceeding normal thermal ratings for limited periods) to achieve this. This can be risky, and current methods for managing these emergency operations are either too conservative (leading to higher costs) or lack precise control over the duration and extent of emergency line usage. The paper tackles the need for a SCED model that explicitly balances cost reduction with strict, yet flexible, reliability requirements, particularly concerning how long lines can operate outside their normal thermal limits.
Methods
The paper's main contribution is the formulation of SCED as a cardinality minimization problem (CMP). This means the objective function not only minimizes operational costs but also penalizes the number of transmission lines operating in emergency zones (both long-term emergency (LTE) and short-term emergency (STE)).
To tackle the non-convexity and non-smoothness of this CMP, the authors propose two key approaches:
- Difference-of-Convex (DC) Approximation: They reformulate the cardinality terms into a difference-of-convex function. This allows them to use a Difference-of-Convex Algorithm (DCA), a well-established method for solving such problems, which iteratively linearizes concave parts of the objective function to find a solution.
- Mixed-Integer Programming (MIP) Formulation: They also develop a MIP formulation using binary variables to represent whether a line is in an emergency zone. This allows them to leverage off-the-shelf MIP solvers.
The paper also introduces a rolling-horizon framework for the multiperiod SCED problem, enabling the system to plan for future periods while implementing decisions in the current period.
Datasets
The paper evaluates its proposed models using three standard test power systems:
- RTS-96 test system: An updated version with 73 buses, 108 lines, and 158 generators.
- IEEE 118 test system: With 118 buses, 179 lines, and 54 generators.
- IEEE 300 test system: The largest, with 300 buses, 409 lines, and 57 generators.
These are publicly available benchmarks commonly used in power systems research. The authors also mention using load data transformed to a 15-minute resolution and artificially increasing it to emulate contingency scenarios.
Key Insights and Ideas
- Quantifying "Emergency" Usage: The core innovation is moving beyond simply allowing or disallowing emergency line operation to actively minimizing the number of lines in these states. This provides a more nuanced control over reliability.
- Balancing Cost and Risk: The CMP formulation explicitly models the trade-off between reducing operational costs (by using lines more flexibly) and increasing system risk (by operating lines in emergency zones). The weighting parameters ($\gamma^l, \gamma^s$) allow operators to tune this balance.
- Duration Matters: The model incorporates constraints on how long lines can remain in emergency zones, reflecting real-world reliability standards (e.g., California ISO's 4-hour LTE and 15-minute STE limits). This is crucial for preventing equipment damage and ensuring grid stability.
- DC Approximation for Tractability: The development of a DC approximation and the use of DCA provide a computationally efficient way to solve the otherwise intractable CMP, offering a practical solution for real-time operations.
- Location Marginal Prices: The DC approximation approach allows for the computation of location marginal prices, which are vital for market operations and understanding the economic impact of grid constraints. The MIP formulation, while efficient, does not directly provide this information.
Key Results
- Reduced Operating Costs: The proposed CMP models (both DC approximation and MIP) consistently achieved lower total operating costs compared to a "strict" model that enforces normal thermal limits at all times. The savings were significant, up to 6.14% on some days.
- Reduced Demand Shedding: The increased flexibility offered by the CMP models also led to a reduction in the amount of demand shed (between 2-7%), indicating improved reliability in meeting demand.
- Efficient Emergency Zone Usage: The models effectively utilized emergency zones but did so in a controlled manner, minimizing the number of lines operating in these states and adhering to duration constraints.
- Scalability: The CMP models demonstrated good performance across different system sizes, with computational times becoming more comparable between the DC and MIP approaches for larger systems.
- Price Volatility Reduction: The CMP-DC model was shown to reduce the duration and severity of scarcity pricing (high electricity prices due to grid constraints), a key benefit for system operators.
Limitations and Future Work
- Static Thermal Ratings: The current formulation relies on static thermal ratings for transmission lines. The authors acknowledge that incorporating dynamic thermal ratings (DTR), which vary with ambient conditions, could further improve system operations. However, this would introduce non-convex, non-smooth, and stochastic optimization problems, which is an area for future research.
- Rolling-Horizon Length: The paper uses a rolling-horizon setup where the lookahead period ($T'$) is set to zero for current ISO practices. A thorough investigation into the optimal choice of $T'$ for the CMP model is suggested as a fruitful research direction.
- Stochastic Optimization: The inclusion of DTR would necessitate moving towards stochastic optimization, which is currently in its infancy and is a subject of the authors' ongoing work.
- MIP vs. DC Computation Time: While the MIP formulation is generally faster for solving individual instances, the DC approximation offers valuable insights like location marginal prices, which the MIP does not directly provide. Further research could explore hybrid approaches or optimizations for the DC solver.
Summary updated