基于极限学习机的谐波电流检测方法

谐波电流检测的实时性和准确度直接影响有源电力滤波器的谐波补偿效果.针对基于传统神经网络谐波检测方法的不足,提出了一种基于极限学习机的谐波电流检测新方法.首先详细给出了极限学习机的训练样本的组成和训练方法,然后构造检测模型实现对谐波电流幅值和相位的检测.仿真结果表明,该谐波电流检测方法的检测精度普遍达到10-6,在有白噪声影响的情况下检测精度达到10-4,与基于传统神经网络的谐波检测方法相比具有更高的检测精度和更强的泛化能力,更加适用于谐波源固定的场合.

Harmonic Current Detection Algorithm Based on Extreme Learning Machine

The real-time and accuracy of harmonic current detection influenced the harmonic compensation performance of active power filter directly. A novel harmonic current detection algorithm based on extreme learning machine （ELM） was proposed to overcome the shortage of traditional neural network-based approach for harmonic current detection in this task. Firstly, the training sample composition and training method were presented in detail. Secondly, detection model was constructed to detect harmonic amplitude and phase. Finally, simulation results demonstrated that ELM-based approach could demonstrate better performances in some aspects than traditional neural network-based approach for harmonic current detection, such as computational complexity, calculation speed, the abilities of function approximation and generalization. And the proposed approach would be especially applied to fixed harmonic source.