Research On The Identification Mechanism Of Electric Power Business Based On Deep Learning

With the rapid growth of the power grid from automation to intelligence,the electric power communication network has place a greater demand on the flexibility,security and the resource scheduling control capability of the network.Accurate identification and classification of power business is the key to achieve the above requirements.Therefore,a power business identification and classification method that can accurately identify businesses and improve network efficiency is proposed,which is helpful to the construction of smart grids.A method of power businesses identification and classification based on deep learning suitable for electric power communication network is proposed in this thesis.This thesis mainly includes the following contents:1.Analyzed the structural features of the electric power communication network,the division of power businesses types,the features and the communication index requirements of power businesses.Based on this,an overall scheme of identification and classification of power communication network businesses is proposed.2.A power communication network business feature extraction and outlier detection method based on convolutional autoencoder is proposed.According to the requirements of identification and classification of power communication network businesses,a convolutional autoencoder is selected as the basic model,and a 15-layer convolutional autoencoder is designed.The model takes the flow statistical characteristic data of the power business as input,automatically analyzes and extracts the hidden characteristics of the business,and compares the reconstruction loss obtained by the model with the set loss threshold to achieve the outlier detection.In order to verify the effectiveness of the method proposed in this thesis,the experiments are designed on the Tensorflow simulation platform.The simulation results demonstrate that the features of the power business can be availably extracted by the proposed method,the outlier data in the input can be accurately detected.Moreover,the accuracy of anomaly detection can reach 97.42%.3.The normal business classification method based upon light GBM and the abnormal data classification method based upon the LOF algorithm is proposed.Because of the highspeed and high-accuracy characteristics of power business classification,the light GBM classifier is selected to classify and output the business features extracted from the autoencoder.Aiming at the abnormal data in the network,the LOF algorithm is selected to distinguish the real abnormal data from the unknown type.The simulation results demonstrate that the strategy in this thesis can quickly and efficiently classify the business in the power communication network,and achieves an overall classification accuracy of95.27%.At the same time,it can effectively analyze abnormal and unknown data.The overall accuracy of abnormal classification is 92%,which reduces the labor consumption in abnormal data processing.According to the experimental results,it can be concluded that the strategy proposed in this thesis can finely identify and classify the power communication network businesses.The results of the identification and classification can not only provide a reference for network managers to perform flow control and network resource optimization configuration,but also improve the abnormal prediction ability and security of the power communication network.