Research On Power Quality Disturbance Recognition Based On Compressed Sensing And Deep Learning

With the development of social economy,high-precision power electronic equipments are widely used,and users' demands for the stability of power quality are becoming more and more important.However,the uthe use of non-linear and shock loads and the integration of various new energy sources,e.g.,wind energy and solar energy,have seriously polluted the power quality and furhter caused power quality disturbances.Various power quality disturbances will seriously affect the use of electrical equipments and even directly threaten the safe operation of the power grid.Therefore,in order to improve and enhance the power quality,it is necessary to eliminate various disturbance signals in the power grid.Fast and accurate detection and classification of various disturbances in the power grid is a prerequisite for the improvement of power quality problems.Based on those mentioned above,this paper recognizes power quality disturbances through two deep learning models:Convolutional Neural Network(CNN)and Convolutional-Bidirectional Long-Short Memory(CNN-BiLSTM).Furthermore,compressed sensing theory is used to replace traditional Nyquist sampling to obtain power quality disturbance signals,achieve the disturbance classification combined with the deep learning method.The specific works carried out in this paper are as follows.(1)Recognition of power quality disturbance based on CNN.The CNN of deep learning is used to power quality disturbance recognition.The CNN network model was built on the Tensorflow platform to generate a batch of disturbed signals with different signal-to-noise ratios.After being converted into grayscale images,the classification results were obtained through the CNN model.Experimental results show that the CNN has good accuracy and certain noise resistance for disturbance classification.(2)Recognition of power quality disturbance based on CNN-BiLSTM.The CNN-BiLSTM hybrid model was built on the Tensorflow platform to classify the power quality disturbance signals.The model uses CNN to automatically extract disturbance feature vectors,convert the feature vectors,which are converted into time series.Then,the time series are input into BiLSTM for further processing to realize the classification of disturbance signals.The experimental results show that the CNN-BiLSTM model has a higher recognition rate and better noise resistance than the CNN model alone.(3)Power quality disturbance recognition based on compressed sensing and deep learning.Compressed sensing is used to compress the perturbation signal by means of the theory of compressed sensing,and the perturbation signal is recovered through the reconstruction algorithm.The experimental results show that when the compression rate is 25%,the characteristical information of the original signal very well.After being converted into a grayscale image,the reconstructed signal will be as input of the CNN and CNN-BiLSTM,respectively.The average recognition rate of CNN is 97.2%,which is 1.9%lower than the original signal input.Furthermore,the average recognition rate of the CNN-BiLSTM is 97.6%,which is 1.6%lower than that of the original signal input.The experimental results show that after the signal is compressed and reconstructed,the recognition rate is slightly reduced compared with the original signal input,but the accuracy rate is more than 97%,which can satisify the accuracy requirements of the classification results.