To address the complex nonlinear relationship between the bias of fiber optic gyroscopes and temperature, a temperature bias compensation method based on an extreme learning machine enhanced by particle swarm optimization algorithm is proposed. Firstly, the physical mechanism of temperature-induced errors in fiber optic gyroscopes was analyzed. Subsequently, the particle swarm optimization algorithm was employed to optimize the hidden layer parameters of the extreme learning machine, establishing a high-precision nonlinear mapping relationship. Through compensation experiments, the PSO-ELM model was compared with traditional ELM and multiple linear regression methods. The results demonstrate that the PSO-ELM model achieves higher compensation accuracy in mitigating temperature-induced errors in fiber optic gyroscopes.