Research On Power Quality Prediction Of Active Distribution Network Based On Hybrid Deep Learning Model

In response to the national environmental protection and sustainable development strategy,the penetration rate of various distributed generation in distribution network has continued to increase with rapid development of smart metering and information and communication technologies.It has become a trend that passive control is transformed to active control in distribution network.Active distribution network has become one of the most important development models for intelligent distribution networks in the future.In addition,with the rapid increase of sensitive loads in active distribution network,as well as the widespread use of power electronics and impact loads,the power quality in the power grid is suffering more and more intense interference.Therefore,in order to cope with power quality problems in power grid,it is necessary to comprehensively consider various influencing factors in power grid,analyze and predict the power quality accurately,so as to achieve the purpose of anticipating the trend and avoiding greater losses by active control.In order to mine effective information from massive high-dimensional data,and improve the accuracy of power quality prediction in active distribution network,a hybrid-deep-learning-based power quality index prediction method is proposed in this thesis according to the sequential and nonlinear characteristics of power quality data in a long time span.Firstly,the power quality data containing various influencing factors is processed,which is transformed into square graphs with time scale by using sliding windows.Secondly,each square graph sample is extracted by using the feature extraction advantage of convolution neural network,and the extracted feature information is transformed into the input of the long short-term memory network in a time series sequence.Finally,the long short-term memory network is used to complete the power quality prediction of the active distribution network based on the output of the convolutional neural network.This method on the one hand uses a convolutional neural network to solve the problem of extracting and analyzing the effective feature information of the massive high-dimensional data in active distribution network.On the other hand,it uses a long short-term memory network,which is good at memorizing long-term and short-term information of time series.The advantage solves the problem that the traditional prediction algorithm lacks the correlation analysis of time series.The analysis shows that this method combining the convolutional neural network and the long short-term memory network makes the feature extraction of power quality and the prediction task decoupled,and the prediction work simplified.Most importantly,compared with the selected models in contrast,this method remarkably improves the prediction accuracy.It helps to discover the problems in the active distribution network in time and provides effective help to ensure the stable operation of the active distribution network.