Research On Power Fingerprint Recognition Algorithms Based On Deep Learning

Now,the school logistics management is more and more important,and the students’ dormitories implement the open power management system.With the improvement of people’s living standards,students often use electric kettle,induction cooker and other electrical appliances in their dormitories.Because of the high power of these electrical appliances,students’ dormitories are on fire.Therefore,it is necessary to identify the prohibited electrical appliances in students’ dormitories.This thesis takes this as the research topic,combining deep learning methods to complete the identification of illegal appliances.The work done is as follows:(1)The data acquisition platform is built.When the electric kettle and induction cooker are started,their current and voltage values are collected.The current and voltage data are transformed into V-I fingerprint trajectory by MATLAB.(2)Power fingerprint recognition based on deep learning.The noise information in the V-I fingerprint image affects the training results of the model.The noise information of V-I fingerprint is removed by threshold segmentation.and uses Gan counter neural network to enhance the data.Then,combined with convolutional neural network,The data is trained iteratively on the classical network models VGG19、Res Net50 and Inception V3,And setting three different sets of important parameters such as learning rate and Dropout value to train the model,so as to complete the parameter setting of the model.(3)Network model with transfer learning.Due to the low recognition accuracy of the original CNN model and the small sample of V-I fingerprint data,the problem of inaccurate recognition may occur.Therefore,transfer learning technology is introduced,The weights trained on the Image Net complex network model of large natural data set are transferred to V-I fingerprint recognition,which are used as the initialization weight of V-I fingerprint training,constructing a new full connection layer and updating the network with a small learning rate.Therefore,the features of V-I fingerprint are extracted and a better performance model is established,and the model can solve the problem that new electrical appliances cannot be identified.(4)On this basis,a segmented network recognition model is proposed.Considering the incompleteness of collecting V-I fingerprint.according to the local fingerprint waveform,the rule of segmentation is proposed to divide the fingerprint,and combine with three models to train,a segmented network recognition model is established,the experimental results are analyzed.