Abstract
This study presents an optimal control strategy designed to maximize the profit of a multi-fuel power plant while strictly adhering to a predefined production schedule. By integrating Pontryagin’s Maximum Principle with a novel iterative feedback/feedforward algorithm, the research demonstrates a significant profit increase—up to 45% in noisy environments—compared to standard discrete optimization methods.
Introduction
In liberalized electricity markets, such as the Nordic "Nord Pool," power plant operators must balance the fulfillment of a strict production plan with the economic realities of varying fuel costs and physical system constraints. The primary challenge lies in optimizing the use of multiple fuel sources:
- Coal: Low cost but slow response and measurement uncertainty.
- Gas: High controllability but expensive and lower efficiency in specific boiler types.
- Oil: High cost, typically limited to start-up phases.
Traditional control systems often prioritize stability (disturbance rejection) over economic efficiency. This paper addresses the need for a controller that maximizes profit while maintaining the required electricity output.
Methodology
The researchers modeled a 400MW coal-fired plant using a third-order linear dynamic model. The optimization problem was framed using Pontryagin’s Maximum Principle (PMP) to identify the optimal "switching function" ($\sigma(t)$) for fuel selection.
To overcome the computational difficulties of continuous-time optimization and the presence of "singular arcs," a multi-stage approach was implemented:
- Initial Guess: A discrete-time version of the problem was solved using quadratic programming.
- Iterative Refinement: A feedback/feedforward algorithm refined the initial trajectory. A heuristic was introduced to stabilize the solver by selecting the fuel with the largest switching function value at any given time.
- Penalty Functions: The production plan was incorporated as a quadratic penalty term in the objective function to ensure tracking without excessive computational overhead.
Key Results
The proposed strategy was tested using real-world data from Nord Pool and a production plan from Energinet.dk. The results indicate substantial improvements over baseline methods:
- Economic Performance: The continuous feedback/feedforward strategy achieved a 30% higher profit than discrete optimization. When 5% Gaussian noise was introduced to simulate real-world measurement errors, the profit gain increased to 45%.
- Tracking Precision: The mean tracking error relative to the production plan was reduced from 14.73 MW to 3.40 MW.
- Fuel Dynamics: The optimization successfully identified strategic uses for gas to handle rapid ramps where coal was insufficient, while avoiding the use of expensive oil for steady-state production.
Conclusion
The integration of PMP-based feedback control allows power plants to navigate market volatility more profitably. By identifying singular arcs and utilizing a robust iterative algorithm, operators can significantly enhance economic output while maintaining high tracking accuracy. Future work will focus on incorporating mechanical switching costs to reduce equipment wear and exploring real-time implementation via pseudo-spectral techniques.
Summary updated