Research On Classification Of Multiple Power Quality Disturbances Based On Kalman Filter And Deep Learning

Power quality problems will not only affect the normal and safe power supply of the power supply system,but also bring many potential hidden dangers to the power supply system.In a more serious situation,even our life and property safety will be threatened,and at the same time,the country’s economic loss and socio-political status will also be affected.Therefore,it is particularly important to control and deal with power quality problems.In addition to a single disturbance,the actual power system also includes a composite disturbance formed by the combination of multiple single disturbances.Therefore,accurate identification of them is the basis for solving power quality problems.To sum up,this paper has carried out exploration and study on the classification of composite power quality disturbances:(1)An adaptive maximum likelihood Kalman filter algorithm is initially proposed to remove the noise in the original distorted power quality disturbance signal.Firstly,the state space model of the power quality disturbance signal is established,and then the Kalman filter algorithm is used to process the noise covariance matrix and initial condition parameters of the maximum likelihood adaptive optimization to obtain the optimal estimation of the signal.(2)In this paper,a deep belief network is built to feature extraction、train and classify the noise-removed disturbance signals.The deep belief network is composed of three layers of restricted Boltzmann machines stacked,and the deep belief network is used to train and classify the power quality disturbance data denoised by the adaptive maximum likelihood Kalman filter algorithm.Since the quality of the network structure directly affects the accuracy of classification,the network structure includes the number of hidden layers and the number of nodes in each layer.Therefore,the network structure is continuously adjusted during the training process,so that the classification accuracy can be optimized.(3)In this paper,a lightweight convolutional neural network is built to feature extraction、train and classify the noise-removed disturbance signals.The lightweight convolutional neural network extracts features from images to realize the classification of power quality disturbance signals.Therefore,the denoised power quality disturbance data is firstly mapped into a time-frequency image and a grayscale image through continuous wavelet transform,and then the network is used to train and classify the generated time-frequency images,and then convert data to grayscale image by Matlab simulation software.The simulation results of the network are compared with other classification algorithms such as convolutional neural network and deep belief network to verify the superiority of the algorithm.This paper proposes an algorithm based on adaptive maximum likelihood Kalman filtering,which overcomes the shortcomings of traditional Kalman filtering algorithms that are difficult to determine the statistical characteristics of noise.The deep belief network and the lightweight convolutional neural network are proposed to classify composite power quality disturbances.The experimental results show that the two types of neural networks have good recognition effect and good robustness,and contribute a feasible method to solve the power quality problem.