Optical chaos prediction based on adaptive extreme learning machine

Prediction of optical chaos has been a key enabler in fields including random number generation and private secure communication. Hereby, we propose an Extreme Learning Machine (ELM) based approach to forecast the chaotic phenomenon of semiconductor lasers effectively. Then, by taking advantage of the features of the ELM, we propose to use a circulant structure for the input weight matrix and the Fast Fourier Transform (FFT) for implementation, leading to significant computational complexity reduction. To meet the the need of dynamic forecasting with less samples, an adaptive ELM is introduced for continuous prediction of optical chaos. To achieve this, recursive least square is employed to update the ELM with chaotic data arriving one-by-one or batch-by-batch for dynamic prediction. Simulation results demonstrate the proposed methods can forecast the optical chaos effectively in dynamic forecasting scenario.