Reasearch On Lightpathquality Of Transmission Prediction Method Based On The Ann With Loss Function Optimization

With the rapid development of communication technology,the access of a large number of users and network terminal equipment has led to a rapid increase in the amount of network data,which puts forward higher requirements for the network capacity.Low margin optical network deployment has gradually become an effective way to improve network capacity,intelligent lightpath quality of transmission(QoT)prediction is the basis of low margin optical networks.Inaccurate QoT prediction will lead to more margin,and thus reduce available optical network capacity.Therefore,how to improve the accuracy of intelligent prediction model of QoT to reduce the network margin is the key to improve the capacity of optical network.The design of the intelligent prediction model has an important influence on the prediction results,and the model parameters are determined by the continuous optimization of the loss function in the training process.Therefore,this thesis studies the prediction of QoT method based on artificial neural network(ANN)with the loss function optimization,and improves the accuracy of the prediction model by optimizing the design of the loss function.The main contents and innovations of this thesis are as follows.(1)In order to solve the problems of maximum overestimation of the current intelligent prediction model based on ANN is too large,resulting in inaccurate prediction,a prediction of QoT method based on user-defined asymmetric loss function in ANN is proposed.The loss function proposed in this paper introduces a penalty term based on Mean Absolute Error(MAE)and Mean Square Error(MSE).When the predicted value is greater than the real value,the model will be given additional punishment.The simulation results show that compared with the ANN prediction model based on MSE and MAE,the proposed model effectively reduce the design margin,and improve the allocation efficiency of lightpath modulation format and optical network capacity.(2)In order to solve the problems of low average prediction accuracy caused by the introduction of penalty term into the user-defined loss function in the above model,a prediction of QoT method based on multitask learning is proposed.In the multi-task learning mechanism,the maximum overestimation and average prediction accuracy of the model are considered simultaneously.A joint loss function is designed based on the proposed asymmetric loss function and the existing symmetric loss function,and the model is trained by joint training.The simulation results show that the prediction of QoT model based on multi-task learning can further reduce the design margin and average error,and improve the network capacity.