Research On De-noising And Recognition Of Power Quality Disturbance Signal Based On Wavelet Transform And Deep Learning

With the improvement of the degree of electrification in power system,the use of a large number of impact and non-linear power electronic devices and the access of renewable energy grid-connected technologies have caused the operating environment of the power grid to become increasingly complex.The resulted power grid will lead to various power quality disturbances,which threaten the safe operation of the power systems.Therefore,it is an important prerequisite to improve power quality that realize accurate classification and identification of power quality disturbance signal.However,it is well known that the electromagnetic interference existing in the operation of the equipment,and the disturbance signal is easy to be mixed with noise in the acquisition process.All these will make the disturbance characteristics be covered by noise,which will reduce identification accuracy.Based on the above reasons,this paper starts from reducing the interference of noise to the disturbance features,improving the effectiveness of features,and optimizing the network structure of the deep learning model,and then explore the problem of power quality disturbance identification under environment of noise interference and noiseless.The specific works carried out in this paper are as follows:(1)Power quality disturbance signal de-noising based on improved wavelet threshold de-noising algorithm.Wavelet threshold is improved according to the characteristics of noise distribution after wavelet decomposition.Then an adjustable factor is introduced to make constructed threshold function has the advantages of the function both traditional soft and hard threshold.Finally,the appropriate wavelet basis and decomposition layer are selected to realizing de-noise of power quality disturbance signal.The simulation results show that the improved algorithm can get higher signalto-noise ratio for all kinds of disturbed signals under different degrees of noise interference.Meanwhile,the disturbance characteristics can be better preserved,it is making the recognition accuracy of the subsequent disturbance classification experiment is greatly improved compared with that before de-noising,which verifies the effectiveness of the algorithm.(2)Power quality disturbance signal identification based on wavelet time-frequency graph and AlexNet network.Wavelet continuous transform is used to extract the time-frequency features of the disturbance signal to generate the disturbance time-frequency map.Then,AlexNet network is construct by using the advantage that deep learning model can automatically extract features of deeper layers,and the disturbance time-frequency map is used as the input of the AlexNet model.Finally,the network parameters are fine-tuned through pre-training,and the disturbance identification is realized after continuous optimization of the classification performance of the model.The simulation results show that the recognition accuracy of the model has been effectively improved under noise-free environment,and it has a certain degree of noise resistance.(3)Power quality disturbance signal identification based on feature fusion CNN-SENet-LSTM.1D-CNN and 2D-CNN are respectively used to extract the local features of the disturbance signal and the time-frequency map that in order to enhance the spatial correlation of the feature extraction of the convolutional network,and the SENet channel attention mechanism is introduced to enhancing the ability of network model to identify the importance of features.The fused features are input into LSTM in the form of sequence to give it time dependence,which make the network achieve better classification performance.The simulation results show that the model has an advantage in classification performance compared with its partial structure sub-models.Meanwhile,the recognition accuracy of power quality disturbance signals is further improved under environment of noise interference and noise-free compared with the methods in the previous two chapters,which verifies the accuracy and superiority of the model.