Parameter Optimization Of Quantum Key Distribution Network System Based On Machine Learning

As a key technology for practical quantum communication networks,quantum key distribution(QKD)can provide a pair of theoretically unconditionally secure keys for both sides of communication,thereby ensuring communication security.Among them,measurement-deviceindependent QKD(MDI-QKD)thoroughly closes all measurement-side vulnerabilities in quantum key distribution,twin-field QKD(TF-QKD)breaks the key rate upper limit of no relay,and the decoy-state scheme solves the problem of Photon-Number-Splitting(PNS)attack caused by weak coherent light sources.Currently,transmission distance has become a key factor in the practical process of QKD.Before quantum repeaters mature,combining satellite communication can significantly extend the transmission distance in specific cases and achieve long-distance QKD.With the development of machine learning,machine learning can efficiently solve the regression and classification problems in the field of QKD.Combining machine learning with QKD has also become a hot spot in current research on the practical application of QKD.This paper mainly analyzed the performance of MDI-QKD and TFQKD and used a variety of machine learning algorithms to optimize the system parameters.Combining the relevant technical theories of QKD and machine learning,this paper deeply studied the parameter optimization performance of various machine learning systems of different QKD network systems,and obtained relevant research results.Considering the signal selection probability and finite key length under actual conditions,it is found that with the increase of the total number of transmitted pulses,the key rate will increase and the transmission distance will be significantly longer by simulating the satellite-ground MDI-QKD system.Using the local search algorithm(LSA)to optimize all parameters of the system will also significantly improve the key rate of the system.Full-parameter optimization of a quantum key distribution system using random forest(RF)and back propagation neural network(BPNN)can also reduce the optimization time to microseconds.Through simulating the TF-QKD system,this paper analyzed the impact of different environmental parameters on the TF-QKD key rate and performed full parameter optimization using exhaustive traversal.Using the exhaustive traversal data to train and test the extreme gradient boosting(XGBoost),BPNN and RF model,it is found that XGBoost has the shortest optimization time and the highest accuracy among the three machine learning algorithms.By using XGBoost,the environmental parameters of TF-QKD are ranked in importance.The results obtained can help the practical process of quantum key distribution and provide a theoretical reference for the construction of quantum communication networks in the future.