Power Quality Disturbance Identification Based On Convolutional Neural Network And Support Vector Machine

With the increasing industrialisation of the world and the widespread use of modern intelligent information technology,many electrical equipment and protection devices are being used more and more in life,thus causing an increasing number of power sources and electrical loads in the power grid,which leads to a series of power quality disturbance problems and safety hazards.Therefore,in order to ensure the efficient operation of the power system and the safe use of electricity by customers,it is of great importance to extract and classify the power quality disturbance signals accurately.This study focuses on the feature extraction of power quality disturbance signals using the Gram’s corner field principle and sparse self-encoder theory,the introduction of two-dimensional convolutional neural networks and support vector machine classifiers with particle swarm optimisation to identify the disturbance signal models,and the testing of the proposed method with simulation experiments.This paper firstly introduces the research background and and development significance of the power quality disturbance problem,and summarises its research status at home and abroad on the premise of reviewing a large amount of literature,while establishing a simulation signal model in Matlab according to the power quality definition and criteria,and generating sample data randomly.Secondly,the basic principle of Gram’s corner field is analysed,and the one-dimensional power quality disturbance signal data is transformed into Gram’s corner field image by encoding the value of the time series as cosine angle and the time as radius to generate a new time series in polar coordinates.The structure of neurons and the specific operation method of each layer in the convolutional network and BP learning algorithm are introduced,and the structure and training of the two-dimensional convolutional neural network are analysed.The structure and training process of 2D convolutional neural networks are analysed.Then,the two-dimensional convolutional neural network model is built by the Tensorflow/Keras learning framework,and the perturbation recognition simulation experiments are carried out with the input of Gram’s corner field images,and the perturbation signal classification recognition effects under different noise conditions are obtained by adding random noise,and the network training and simultaneous testing are carried out.Finally,the principle and algorithmic steps of the sparse self-encoder are introduced,which is used to obtain a new feature quantity by dimensionality reduction feature extraction of the perturbed original signal to prevent local optimisation.A support vector machine model is established analytically and its parameters are optimised using the particle swarm algorithm.The extracted feature quantities are fed into the particle swarm algorithm optimised support vector machine based model for simulation experiments and compared with the standard support vector machine model,the Gram’s Corner field and the two-dimensional convolutional neural network model for classification and recognition effect plots,which show that the power quality disturbance signal classification model of this study recognises The accuracy is higher.Figure[27]table[10]reference[92]...