Prediction of Optical Chaos Using a Comparative Adaptive Extreme Learning Machine

In this letter, a comparative adaptive extreme learning machine (CAELM) is proposed for continuous prediction of optical chaos with a simple updating rule and low computational complexity. A recursive least square (RLS) with a adaptive forgetting factor (AFF) updating method is devised to track the dynamics of the optical chaos. The results demonstrate that the proposed CAELM can effectively execute the time-varying optical chaos predictions, and delivers much better performance in terms of normalized mean squared error (NMSE), with a value of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$2.4\times 10 ^{-4}$ </tex-math></inline-formula>. It also demands fewer training samples than state-of-the-art adaptive methods. Finally, we validate CAELM’s generalization capability under the condition of changing laser parameters, and the proposed CAELM remains accurate and adaptive to predict the time-varying optical chaos with very short training length for the model update.