Deep Learning based Collection and Analysis of Power Quality in Power Supply Stations

With the development of new energy sources such as distributed photovoltaics and wind power, it has also brought various problems such as frequency deviation and voltage flicker to the power grid. Therefore, accurate collection and analysis of power quality is very important. This article proposes a deep learning based method for analyzing power quality disturbances. Firstly, a power waveform dataset is generated using mathematical models of eight types of power quality disturbances; Then, the waveform data is converted into a two-dimensional image using the Gram Angular field (GAF); Afterwards, use the SE attention module to focus on the key wave points of the waveform data; Finally, SE-ResNeXt50 is used to extract features and classify power quality disturbance signals. The experimental results show that the deep learning based power quality disturbance analysis method achieves a recognition accuracy of 99.25% for single disturbance signals and 95.3% for dual disturbance signals at a signal-to-noise ratio of 15dB.