Abstract
Rapid and accurate detection of power quality disturbances (PQDs) is essential for maintaining grid stability, especially with the increasing integration of renewable energy sources. This paper proposes a classification method based on a Kernel-based Extreme Learning Machine (KELM) optimized by a Chaotic Particle Swarm Optimization (CPSO) algorithm. By leveraging chaos theory to improve global optimization performance, the proposed method achieves higher classification accuracy and reliability compared to traditional models.
Introduction
Modern electrical grids face growing challenges from Power Quality Disturbances (PQDs) such as voltage sags, swells, and harmonic distortions. These disturbances, often caused by non-linear loads and renewable energy inverters, can lead to equipment failure and significant economic losses. Effective grid management requires rapid identification and classification of these events. While traditional Artificial Neural Networks (ANNs) and Support Vector Machines (SVMs) have been used, they often suffer from slow training speeds or sensitivity to parameter selection.
Methodology
The proposed approach combines the speed of Extreme Learning Machines (ELM) with the robustness of kernel methods and chaotic optimization.
Kernel Extreme Learning Machine (KELM)
The KELM is a single-hidden-layer feedforward network that replaces the random hidden layer mapping of standard ELMs with a kernel function, typically a Radial Basis Function (RBF). This modification improves generalization and eliminates the instability caused by random weight initialization.
Chaotic Particle Swarm Optimization (CPSO)
The performance of KELM is highly dependent on two parameters: the regularization coefficient ($\lambda$) and the kernel width ($s$). To find the optimal values, the researchers employed Particle Swarm Optimization (PSO). To prevent the algorithm from converging on local optima, a chaos search mechanism was introduced. Chaos theory's properties of ergodicity and non-repetition allow the "particles" to explore the search space more thoroughly, ensuring a global optimum is reached.
Experimental Results
The model was tested using a dataset of 350 power quality disturbance samples across seven categories (e.g., voltage swell, harmonics, flicker). The dataset was split into 280 training samples and 70 test samples.
The results demonstrated that the CPSO-KELM model significantly outperformed baseline methods. In the 70-sample test set:
- CPSO-KELM: 6 classification errors.
- PSO-KELM: 8 classification errors.
- Standard KELM: 13 classification errors.
The CPSO-KELM model showed superior stability and accuracy across all disturbance types, proving its effectiveness for real-time grid monitoring.
Conclusion
By integrating Chaotic Particle Swarm Optimization with a Kernel Extreme Learning Machine, this research provides a robust framework for PQD classification. The method overcomes the limitations of traditional neural networks by offering faster training and higher global optimization capabilities. Future work will focus on testing the algorithm's performance in environments with high background noise and overlapping disturbances.
Reference: Xu, L., et al. (2024). "Power quality disturbance detection method based on optimized kernel extreme learning machine." MATEC Web of Conferences 399, 00022.
Summary updated