Power Quality Disturbance Analysis Based On Improved HHT And CNN-LSTM

With the rapid development of modern society,the frequency of electricity use is also increasing,and the requirements for electricity are becoming increasingly stringent.In fact,considering the influx of a large number of loads and the non-standard use of user groups,various disturbances often occur in daily life,including single and composite disturbances,in the types of disturbances that occur in the power grid,However,there are still issues such as inaccurate disturbance analysis and poor computational efficiency in the basic methods of power quality composite disturbance analysis.We need to use some new algorithms to identify and classify power quality disturbances.Conducting in-depth research on power quality analysis is beneficial for people to take timely measures to control disturbances and reduce unnecessary economic losses,which is of great significance for people’s normal lives.This article analyzes power quality disturbances from three aspects: signal denoising,feature extraction,and disturbance classification.The main research content is as follows:Firstly,based on literature research and findings,the history of power quality analysis has developed for a long time.The analysis of power quality disturbances is mainly divided into three parts: noise reduction,feature extraction,and disturbance recognition.The existing research on these three parts is summarized and summarized,and improvement methods are proposed to address the shortcomings of existing methods.In terms of signal denoising,a self-adaptive improved kernel regression method is proposed to address the problems of poor denoising performance and large computational workload in classical kernel regression methods.It can automatically select the optimal bandwidth within a certain range.The experimental results show that this method can not only improve the denoising effect of disturbance signals,but also accurately extract the features of disturbances,with good improvement effects.In terms of feature extraction,due to the endpoint effect and modal aliasing problem of the traditional Hilbert Huang transform(HHT),we adopt CEEMDAN decomposition and parallel extension methods to improve these two aspects.The experimental results show that the improved method can effectively suppress endpoint effects and modal aliasing problems,and can more accurately extract disturbance signal feature information,providing data for subsequent disturbance recognition.In the aspect of disturbance identification,because the accuracy of single neural network for disturbance classification and identification is not high,an improved network model combining convolutional neural networks and long-term memory networks(CNN-LSTM)is proposed.According to the fact that the correlation between various disturbances is not taken into account during classification,a multi task multi learning method is added to the combined neural network,divide the classification results into four subtasks,and ultimately represent the disturbance type by a one-dimensional vector.The experimental results show that this method has higher accuracy compared to traditional methods.