Research On Data Augmentation For Communication Network Based On GANs Model

The high-speed iterative upgrading of communication network brings multi-dimensional information confusion problems,such as inconsistent format,lack of information and so on.It is easy to cause"poor quality" and "small quantity" of effective data(such failure)in communication network,which is difficult to support the normal implementation of data-driven algorithms.Data enhancement technology can be used to expand and optimize small-scale data sets to remove the limitations of the algorithms at the data level..This paper focuses on the data enhancement technology of self optimizing communication network,and focuses on the three core challenges of data enhancement:self optimizing,conditional and high fitting.It mainly completes two research works:single condition fixed length data enhancement and multi condition variable length data enhancement.It alleviates the industry problem of critical data shortage in communication network,and significantly improves the performance of artificial intelligence model in communication network scenario.The main work and innovation achievements are as follows:(1)Aiming at the technical challenge of effective data self optimization enhancement in fixed length communication network under single condition,this paper studies and proposes a data enhancement method based on multi-layer perceptron under the framework of countermeasure generation network.Under the semi supervision mechanism of countermeasure generation network,one multi-layer perceptron adjusts and learns the distribution law of original measurement data by self optimization,and the other multi-layer perceptron uses it In order to reduce the risk of schema crash,it is necessary to generate data normalization parameters.(2)Aiming at the technical challenge of effective data self optimization enhancement in multi condition variable length communication network,a data enhancement method based on convolutional neural network and recurrent neural network in the framework of countermeasure generation network is studied and proposed.On the one hand,based on the fixed length data generation model,the recurrent neural network is introduced as the enhancement model generator,so that the model can be used to enhance the data of variable length communication network On the other hand,convolutional neural network is introduced to identify different label data including topological data types,and the corresponding label communication network data is generated by cooperating with cyclic neural network.In this algorithm,a small batch generation strategy training cycle generation network model is introduced,so that the model can better capture the time characteristics of communication network data.(3)In this paper,we use the data collected from the backbone transmission network of an operator and the traffic data of the UK backbone network for academic purposes as data sets to simulate and verify the two algorithms.The results show that in algorithm 1,the overlapping degree of single type fixed length data distribution before and after enhancement is more than 90%,and the accuracy of the trained single target prediction model is as high as 100%.In algorithm 2,the overlapping degree of multi type variable length data distribution histogram before and after enhancement is more than 95%,and the accuracy of multi-target prediction model is improved to more than 98%.Compared with algorithm 1,algorithm 2 improves the utilization ratio of original dirty data by more than 50 times,and the enhanced data autocorrelation is highly fitted with the original data.