An online prediction model for the energetic electron flux at the geostationary orbit was built based on the PSO (particle swarm optimization) algorithm and the LSSVM (least squares support vector machines) method. To overcome the premature convergence problem in PSO, a new diversity measure was put forward. The improved PSO was utilized to optimize the LSSVM′s parameters. Through a sliding time window strategy, a variable selection invoking threshold and a model re-training mechanism, the online characteristic of the model was realized. 1～3 day ahead prediction experiments were done on the basis of the electron flux data, solar wind parameters and geomagnetic parameters in 2000, and the analysis results show that the proposed online PSO-LSSVM model works well and has practicable value for prediction.