Aiming to reduce peak-to-average power ratio (PAPR) in coherent optical orthogonal frequency division multiplexing (OFDM) systems, several swarm intelligence algorithms are investigated, including particle swarm optimization algorithm (PSO), bat algorithm (BA) and birds swarm algorithm (BSA), and these algorithms can be used for optimizing the sub-carrier phase of OFDM symbols. Moreover, BA and BSA are modified by changing cognitive coefficient and learning factor dynamically. The simulation is carried out in a 100 Gb/s, binary quadrature amplitude modulated (4QAM) coherent optical OFDM system and the results show that PSO, BA and BSA can effectively reduce PAPR of the system, and the PAPRs for the modified BSA and the modified BA can be reduced by about 5.11 dB and 5.48 dB, respectively, compared with that of the original OFDM. The modified intelligence algorithms show better performance. The intelligence algorithms can also improve the bit error ratio performance, and the performance improvement is more obvious when the optical signal to noise ratio increases.