基于CNN-BiLSTM-Attention 神经网络的电能质量扰动分类

打开文本图片集

中图分类号：N031 文献标志码：A doi： 10.3969/j .issn.1673-5862.2025.05.013

Power quality perturbation classification based on CNNBiLSTM-Attention neural network

ZHANG Dong1，FU Jiayu²，WANG Xiaofeng 3 ，GUO Quanli1，GU Yutong²， LIUQuan4，LI Haoxuan4，NING Zhaoqiu 4 ，FANG Wenmo³，SUNMing³，SUN Zhiqiang？

（1.Institute of Carbon Peak and Carbon Neutrality，Shenyang Institute of Engineering，Shenyang 10136， China；2. Colege of Electric Power，Shenyang Institute of Engineering，Shenyang 11Ol36，China；3. Shenyang Aircraft Industry Company，Shenyang 11oo34，China；4. State Grid Tieling Power Supply Company，Tieling 112000，China）

Abstract： The classification of power quality events is usually divided into two stages： feature quantity extraction and reference classification. In modern signal processing，the accurate extraction of various disturbance signals is the core link of power quality disturbance detection. To address the challenges posed by complex power quality problems and excessive test requirements，an innovative solution is proposed in this study： the key information is captured through the use of mixed wavelet decomposition technology，and the attention and CNN-BiLSTM model are used to judge the dataset to realize the electrical energy fluctuations in the distributed energy system. The experimental results show that the feature sequence of information entropy and energy entropy is used as the input of the model，the average accuracy reaches up to 97.25% . The comparative experiment of the dataset shows that the model is more generalized after adopting the present feature sequence.

Key words： power quality; variational mode decomposition; convolution network; bi-directional long-short term memory network

新能源产业的飞速发展和电力市场的逐渐开放使得电网的电能质量面临着严峻挑战。（剩余9235字）