Chaotic dynamics generated by vertical-cavity surface-emitting lasers(VCSELs)has stimulated a variety of applications in secure communication,random key distribution,and chaotic radar for its desirable characteristics...Chaotic dynamics generated by vertical-cavity surface-emitting lasers(VCSELs)has stimulated a variety of applications in secure communication,random key distribution,and chaotic radar for its desirable characteristics.The application of machine learning has made great progress in the prediction of chaotic dynamics.However,the performance is constrained by the training datasets,tedious hyper-parameter optimization,and processing speed.Herein,we propose a heterogeneous forecasting scheme for chaotic dynamics in VCSELs with knowledge-based photonic reservoir computing.An additional imperfect physical model of a VCSEL is introduced into photonic reservoir computing to mitigate the deficiency of the purely data-based approach,which yields improved processing speed,increased accuracy,simplified parameter optimization,and reduced training data size.It is demonstrated that the performance of our proposed scheme is robust to the deficiency of the physical model.Moreover,we elucidate that the performance of knowledge-based photonic reservoir computing will fluctuate with the complexity of chaotic dynamics.Finally,the generality of our results is validated experimentally in parameter spaces of feedback strength and injection strength of reservoir computing.The proposed approach suggests new insights into the prediction of chaotic dynamics of semiconductor lasers.展开更多
基金National Key Research and Development Program of China(2021YFB2801900)National Natural Science Foundation of China(U22A2089,62104203,62375228,62431024)+1 种基金Sichuan Science Fund for Distinguished Young Scholars(2023NSFSC1969)Fundamental Research Funds for the Central Universities(2682022CX024,2682023CG003)。
文摘Chaotic dynamics generated by vertical-cavity surface-emitting lasers(VCSELs)has stimulated a variety of applications in secure communication,random key distribution,and chaotic radar for its desirable characteristics.The application of machine learning has made great progress in the prediction of chaotic dynamics.However,the performance is constrained by the training datasets,tedious hyper-parameter optimization,and processing speed.Herein,we propose a heterogeneous forecasting scheme for chaotic dynamics in VCSELs with knowledge-based photonic reservoir computing.An additional imperfect physical model of a VCSEL is introduced into photonic reservoir computing to mitigate the deficiency of the purely data-based approach,which yields improved processing speed,increased accuracy,simplified parameter optimization,and reduced training data size.It is demonstrated that the performance of our proposed scheme is robust to the deficiency of the physical model.Moreover,we elucidate that the performance of knowledge-based photonic reservoir computing will fluctuate with the complexity of chaotic dynamics.Finally,the generality of our results is validated experimentally in parameter spaces of feedback strength and injection strength of reservoir computing.The proposed approach suggests new insights into the prediction of chaotic dynamics of semiconductor lasers.
