A New Convolutional Network Structure For Power Quality Disturbance Identification And Classification In Micro-Grids

Micro grid is an effective technical way for large-scale application of distributed generation,the first step for smart grid to gradually replace the traditional grid,and an effective way to realize active distribution network.However,due to the use of a large number of power electronic equipment,distortion,fluctuation,flicker and three-phase voltage imbalance in the power grid seriously affect the safety,stability and economic operation of the power system.In this paper,a recognition and classification method based on deep convolution neural network is proposed for power quality disturbance recognition and classification of microgrid,and a neural network is designed for power quality disturbance recognition and classification.Aiming at the feature extraction and classification of one-dimensional continuous signals,this dissertation proposes a 1D-MIR model based on the concept Res Net,which is used to extract the features of power disturbance signals,cooperate with the neural network to form a 1D-MIR deep convolution neural network to identify and classify power quality disturbances.The network extracts the features of one-dimensional continuous signals through five-layer perception Res Net,and then classifies the features by a three-layer neural network,which improves the accuracy of power disturbance classification.The recognition and classification can be carried out simultaneously,simplifies the steps of power disturbance recognition and classification,and reduces the classification time.Aiming at the training and optimization of 1D-MIR deep convolution network,this paper proposes a method of sample calibration based on the idea of word vector technology,which integrates sample sampling,sample data enhancement and calibration,reduces the labor cost of training,and solves the problem that the training of convolution neural network with one-dimensional continuous signal as sample can not be calibrated.Next,adaptive moment estimation(Adam)and Dropout technology are used to train and optimize the network.The adaptive moment estimation method is used to help the network find the global(local)optimal solution,which makes the network converge quickly.At the same time,Dropout technology is used to filter the nodes to prevent the network from over fitting.Finally,on the basis of the above research,build power quality database,build 1D-MIR depth convolution neural network,and train the network.The data and charts generated by simulation experiments on the network experiment platform prove that the network can realize the generation,sampling,calibration,recognition and classification of power quality.This study not only lays a theoretical foundation for the application of artificial intelligence in power quality disturbance classification,but also provides a novel and effective method and scientific basis for power quality disturbance identification,which can achieve the purpose of rapid and accurate identification of power quality disturbance.