Investigation Of Chaotic Reservoir Computing Subject To Optoelectronic Feedback And Recognition Application

Reservoir computing is the improvement of recurrent neural network (RNN). Because of its unique dynamic characteristics and simple training methods, it has been successfully applied in a lot of fields. The reservoir computing system is simplified by chaos, and the chaotic reservoir computing has more powerful information processing capabilities due to the initial conditions sensitivity and high bandwidth characteristics of chaotic signal. The reservoir computing based on optical feedback chaotic system is one of implementation methods of chaotic reservoir computing. It has advantages of high-speed processing, low interference, low power consumption and its hardware structure is simple and easy to achieve. Accordingly, in this thesis, handwritten numeral recognition and optical packet header recognition based on optoelectronic feedback chaotic reservoir computing were theoretically simulated and deeply investigated. The works mainly completed in this thesis are as follows:(1) Handwritten numeral recognition utilizing optoelectronic feedback chaotic reservoir computing is proposed and investigated. From zero to nine, ten kinds of handwritten numerals have been successfully recognized by numerical simulations. The influence of system parameters on the recognition results has been investigated by adjusting the feedback coefficient, input gain, system delay time and virtual nodes of reservoir. The optimum parameters for the simulation are achieved.(2) Optical header recognition using optoelectronic feedback chaotic reservoir computing is proposed and investigated. Optical headers have been successfully recognized by numerical simulations from 3 bit to 32 bit. The parameters of the system are optimized and the noise tolerance is also evaluated.(3) The experiment system of optoelectronic feedback chaotic reservoir computing is constructed.3 bit optical packet headers were successfully recognized by experiment, which verifies the feasibility of this scheme. The feedback intensity and the bias voltage of MZM had been analyzed to obtain the influence on the recognition results.