Abstract
Accurate dynamic modeling of power plant components is critical for ensuring grid stability and reliable transient performance analysis. This paper proposes a hybrid identification procedure that integrates the Least Squares (LS) method with the nature-inspired Cuckoo Search (CS) algorithm to optimize the parameters of turbine-governor and excitation systems. By utilizing real-world field data from the 320 MW Chabahar power plant, the study demonstrates that this hybrid approach significantly improves parameter estimation accuracy while reducing the need for manual tuning. The resulting models exhibit high fidelity, with validation errors maintained below 0.5%.
Introduction
Power systems are increasingly complex, necessitating precise mathematical models for stability studies. Gas power plants, in particular, consist of intricate subsystems where standard models like the IEEE GGOV1 (governor-turbine) and ST6B (exciter) are commonly used. However, the specific parameters—such as gains, time constants, and limits—vary significantly between individual units. Traditional manual tuning via trial-and-error is often insufficient for capturing non-linear and multi-modal dynamics. This research addresses the need for a robust, automated identification process using metaheuristic optimization.
Proposed Hybrid Methodology
The authors developed a four-stage identification workflow designed to bridge the gap between "white-box" physical models and "black-box" data-driven approaches.
- Data Acquisition: To avoid triggering plant protection systems with noisy signals, the researchers employed square pulse waves (5 MW steps) to excite the system, recording data at a high-resolution 1ms sample rate.
- Preprocessing: Raw data underwent low-pass Butterworth filtering and normalization (per-unitization) to ensure consistency across different measurement scales.
- Linear Pre-identification: The Least Squares (LS) method provided an initial "warm start" estimate of the system parameters.
- Hybrid Cuckoo Search (CS) Optimization: The CS algorithm, inspired by the brood parasitism of cuckoos and utilizing Lévy flights for efficient global search, was used to fine-tune the parameters. The objective was to minimize the Mean Square Error (MSE) between simulated outputs and field recordings.
Experimental Results and Validation
The methodology was tested using field data from the Chabahar power plant in Iran. The study compared the performance of the Cuckoo Search algorithm against other popular metaheuristics, including Genetic Algorithms (GA) and Particle Swarm Optimization (PSO).
Model Accuracy
The hybrid CS approach achieved exceptional precision. During the training phase, error indices for all subsystems were consistently under 0.2%. More importantly, the models demonstrated strong generalization capabilities; when tested against a separate validation dataset from different operating points, the error remained below 0.5%.
Comparative Analysis
While both CS and GA provided high-quality results, CS was identified as the superior choice for industrial applications due to its simplicity. Unlike GA or PSO, CS requires fewer user-defined parameters, making it easier to implement without extensive algorithmic expertise. Furthermore, a "whiteness test" on the residuals confirmed that the models captured all meaningful system dynamics, leaving only random environmental noise.
Conclusion
The hybrid Cuckoo Search identification procedure offers a reliable and efficient method for tuning complex power plant models. By combining the speed of classical Least Squares with the global search power of the CS algorithm, the researchers achieved high-fidelity models suitable for grid stability analysis. Future work will focus on incorporating Inlet Guide Vane (IGV) dynamics to improve model accuracy at lower loads (below 70%), where airflow control becomes a significant factor.
Summary updated