Research On New Methods Of Dynamic Power Quality Disturbances Detection And Classification

In recent years,the grid-connected power generation of new energy has increased dramatically,and a large number of nonlinear loads,impact loads and power electronic equipment have been injected into the power grid,which makes various kinds of complex power quality disturbances frequent.Complex power quality disturbances are composed of single disturbances,which cause different degrees of pollution to the modern power grid.To establish a safe and reliable power supply system and promote power grid pollution control,it is imperative to detect and classify power quality.Researching novel methods for dynamic power quality disturbance detection and classification and building a dynamic power quality disturbance detection system have great theoretical and practical significance for improving power quality detection and strengthening stability of the power system.For the analysis requirements of dynamic power quality disturbances detection,the basic line of this dissertation is “research background-time frequency analysisdisturbances classification-hardware design”.With the basic line,this dissertation is devoted to the research of dynamic time-frequency characteristics for power quality disturbance analysis algorithm,dynamic power quality disturbance identification method,and dynamic power quality disturbances detection and classification and so on.In addition,a power quality disturbance detection system based on ADC+DSP+PC is developed.The main research work of this dissertation is as follows:(1)To achieve accurate and fast time-frequency analysis for dynamic power quality disturbance signal,a time-frequency analysis method of disturbance signal based on improved double-resolution S transform is proposed.The improved double-resolution S transform divides the signal into high and low frequency domains with 1.5 times of the fundamental frequency as the boundary.It has the characteristics of high time resolution in low frequency band and high frequency resolution in high frequency band.Then,combined with the principle of maximum energy concentration,an adaptive genetic algorithm is proposed to realize the adaptive selection for window parameters of the time-frequency analysis algorithm,thereby improving the time-frequency performance,which can achieve the characteristic analysis for dynamic power quality disturbance signal.Moreover,to improve the execution efficiency of the algorithm and facilitate embedded implementation,the time-frequency transformation is performed on the key characteristic frequency points.The analysis results of simulation experiments under different noise levels show that the improved double-resolution S transform has the advantages of strong anti-noise,high accuracy of time-frequency analysis,and fast execution speed.(2)Aiming at the time-consuming and redundant design of artificial features,and solving the problem that the accuracy of traditional classifiers is greatly disturbed by noise,a fast identification method for dynamic power quality disturbance classification based on improved convolutional auto-encoder is proposed.The basic principle of convolutional auto-encoder and its existing limitations are introduced,and then use the residual structure to improve the convolutional auto-encoder.Then,propose a residual network structure with weight distribution which can improve the deep feature expression ability of the disturbance signal and make the network converge quickly.To simplify the classifier,a single-layer convolutional neural network model with batch normalization function is designed to replace the decoding structure of the model and complete the classification and recognition of single and composite disturbance signals.The experimental results of simulation comparison with other advanced dynamic power quality disturbances classification methods show that the improved convolutional auto-encoder has higher accuracy and stronger robustness under more serious noise interference,and can achieve efficient detection and classification for dynamic power quality disturbances.(3)To solve the problem of complex and lengthy training process caused by the separation of decoding and classification operations in improved convolutional auto-encoder,and realize the accurate end-to-end detection between the input and the output for nonlinear power quality disturbances,a novel method named FFNet is proposed for the classification and identification of dynamic power quality disturbances.A feature fusion module based on CNN is designed to extract the deep features of the disturbance signal and realize the full fusion of the deep features extracted from different sub-models at the two-dimensional level.To increase the local receptive field,a composite convolution kernel is proposed to replace the traditional convolution kernel in the feature fusion module.Additionally,to avoid the loss of feature information,the traditional max pooling is replaced by the improved pooling layer,so that improving the classification performance of the FFNet.Simulation and comparison experiments under different noise interference conditions show that the FFNet has superior detection performance for various power quality disturbance signals,and the intermediate parameters are directly involved in the training,so the training process is more concise which is suitable for the efficient classification of dynamic power quality disturbance signals.(4)On the basis of the dynamic power quality disturbances analysis and classification algorithm simulation verification,a hardware detection system of dynamic power quality disturbances based on ADC+DSP+PC architecture is built,and the detection and classification algorithm of dynamic power quality disturbances proposed in this dissertation is realized on the detection system.The overall framework and specific completion steps of the hardware test platform are given,and the main hardware module circuit design ideas and software implementation process are introduced.According to the test conditions stipulated by the relevant national standards,the hardware detection system of dynamic power quality disturbances designed in this dissertation is tested,and the actual results are analyzed and discussed.The experimental results show that the dynamic power quality disturbances hardware detection system designed in this dissertation can measure the actual disturbances and realize the high-precision detection.