According to the fact that the existing icing prediction methods has a slow convergence speed and poor prediction accuracy, a method based on particle swarm optimization with extended memory (PSOEM) is proposed under the consideration of the icing thickness influence to optimize parameters. It is applied to the least squares support vector machine (LSSVM) to predict icing thickness. The proposed method introduces an extended memory factor into the traditional particle swarm algorithm to make the particles have stronger search capabilities, thereby speeding up convergence and improving prediction accuracy. Finally, the actual line icing data is utilized to test the accuracy of the prediction model. It is shown that the average relative error of the prediction model based on PSOEMLSSVM is less than 3%. Compared with other models, the prediction effect is the best.