Predicting the dynamical behaviors for chaotic semiconductor lasers by reservoir computing

Abstract

We demonstrate the successful prediction of the continuous intensity time series and reproduction of the underlying dynamical behaviors for a chaotic semiconductor laser by reservoir computing. The laser subject to continuous-wave optical injection is considered using the rate-equation model. A reservoir network is constructed and trained using over 2 × 104 data points sampled every 1.19 ps from the simulated chaotic intensity time series. Upon careful optimization of the reservoir parameters, the future evolution of the continuous intensity time series can be accurately predicted for a time duration of longer than 0.6 ns, which is six times the reciprocal of the relaxation resonance frequency of the laser. Moreover, we demonstrate for the first time, to the best of our knowledge, that the predicted intensity time series allows for accurate reproduction of the chaotic dynamical behaviors, including the microwave power spectrum, probability density function, and the chaotic attractor. In general, the demonstrated approach offers a relatively high flexibility in the choice of reservoir parameters according to the simulation results, and it provides new insights into the learning and prediction of semiconductor laser dynamics based on measured intensity time series.