Tang Minan, Wang Hongjie, Qiu Jiandong, Tao Zhanglong, Yang Tong
College of New Energy and Power Engineering, Lanzhou Jiaotong University, Lanzhou, China.
College of Electrical and Mechanical Engineering, Lanzhou Jiaotong University, Lanzhou, China.
PLoS One. 2024 Dec 5;19(12):e0314720. doi: 10.1371/journal.pone.0314720. eCollection 2024.
The large-scale integration of offshore wind power into the power grid has brought serious challenges to the power system power quality. Aiming at the problem of power quality disturbance detection and classification, this paper proposes a novel algorithm based on fast S-transform and crested porcupine optimizer (CPO) optimized CNN. Firstly, the intrinsic mechanism and waveform characteristics of offshore wind power grid-connected disturbances are analyzed, and the simulated disturbance signals are feature extracted and time-frequency diagrams are obtained by fast S-transform. Secondly, the CPO algorithm is used to optimize the convolutional neural network and determine the best hyperparameters so that the classifier achieves the optimal classification performance. Then, the CPO-CNN classification model is used for feature extraction and feature selection of the time-frequency diagrams and classification of multiple power quality disturbances. Finally, a simulation experimental platform is established based on MATLAB to perform simulation verification and comparative analysis of power quality disturbance classification. The experimental results show that the model established in this paper is effective, and the classification accuracy is improved by 3.47% compared with the CNN method, which can accurately identify the power quality disturbance signals, and then help to assess and control the power quality problems.
海上风电大规模接入电网给电力系统电能质量带来了严峻挑战。针对电能质量扰动检测与分类问题，本文提出一种基于快速S变换和冠豪猪优化器（CPO）优化卷积神经网络（CNN）的新型算法。首先，分析海上风电场并网扰动的内在机理和波形特征，通过快速S变换对模拟扰动信号进行特征提取并得到时频图。其次，利用CPO算法优化卷积神经网络并确定最佳超参数，使分类器达到最优分类性能。然后，将CPO-CNN分类模型用于时频图的特征提取与特征选择以及多种电能质量扰动的分类。最后，基于MATLAB搭建仿真实验平台，对电能质量扰动分类进行仿真验证和对比分析。实验结果表明，本文建立的模型有效，与CNN方法相比分类准确率提高了3.47%，能够准确识别电能质量扰动信号，进而有助于评估和控制电能质量问题。
