Abstract
Modern power systems face significant challenges due to stochastic fluctuations from renewable energy sources and discrete switching events like generator failures. This paper presents a framework for synthesizing controllers that provide formal guarantees for satisfying Metric Temporal Logic (MTL) specifications in such switched stochastic systems. By employing a stochastic control bisimulation function (SCBF), the complex probabilistic control problem is transformed into a robust deterministic optimization, solvable via mixed-integer programming. Case studies on 4-bus and 9-bus power systems demonstrate the effectiveness of the approach in maintaining grid stability under disturbances.
Introduction
Ensuring the reliability of critical infrastructure like power grids requires managing both continuous noise (e.g., wind power variability) and sudden structural changes (e.g., line trips). These systems are best modeled as switched stochastic systems. Traditional control methods often struggle with time-sensitive safety and performance requirements, such as returning grid frequency to nominal values within a specific window after a fault. These requirements are formally expressed using Metric Temporal Logic (MTL).
The primary challenge lies in the computational complexity of synthesizing controllers that guarantee probabilistic satisfaction of MTL specifications. This research addresses this by creating a mathematical bridge between noisy, complex dynamics and simplified deterministic models.
Methodology: The Bisimulation Bridge
The proposed framework relies on three core components:
- Stochastic Control Bisimulation Function (SCBF): The authors define an SCBF to quantify the "distance" between the actual stochastic system and a nominal deterministic proxy. This function provides a probabilistic bound: if the nominal system's trajectory remains within a calculated safety margin, the real system will satisfy the original specification with a high, pre-defined probability.
- Robustification of Specifications: Instead of applying the MTL constraints directly to the noisy system, the researchers "robustify" the formula. They shrink the safety margins based on the SCBF's error bounds. If the deterministic proxy satisfies this stricter, robustified MTL formula, the stochastic system is mathematically guaranteed to meet the original requirements.
- Optimization-Based Synthesis: The controller synthesis is formulated as a Mixed-Integer Linear Programming (MILP) or Quadratic Programming (MIQP) problem. This allows for efficient calculation of optimal control inputs, such as power injections from Energy Storage Systems (ESS), to keep the system within the robustified bounds.
Case Studies
The framework was validated using two standard power system benchmarks:
- Four-bus system: Includes a thermal plant, a wind farm with 200 turbines, and an ESS.
- Nine-bus system: A more complex network used to test scalability and the inclusion of power flow constraints on transmission lines.
In the 9-bus scenario, the authors implemented an iterative constraint addition method. Rather than checking all line constraints simultaneously, they added them only when violations occurred in the nominal trajectory, significantly reducing computational overhead.
Key Results and Discussion
In simulations of both systems, the synthesized controllers successfully managed disturbances such as sudden generation loss.
Key findings include:
- Guaranteed Performance: While the theoretical lower bound for success was set at 95%, empirical tests across 100 random realizations showed a 100% satisfaction rate. This indicates that the SCBF-based bounds are safe, though potentially conservative.
- Multi-Objective Control: The controllers effectively balanced multiple competing objectives, including grid frequency regulation, wind turbine rotor speed stabilization, and transmission line power limits.
- Handling Instability: By incorporating an exponential term in the bisimulation function, the method remains applicable even when the underlying linear dynamics of certain modes are unstable, broadening the range of controllable scenarios.
Conclusion and Future Work
This work provides a robust path toward provably safe control of complex, noisy infrastructure. While the current results are promising, the inherent conservativeness of the mathematical bounds suggests room for optimization to allow for more efficient control strategies. Future research will explore extending this framework to fully non-linear stochastic dynamics and improving scalability for extremely large-scale power grids.
Summary updated