Abstract: Logistics demand forecasting is an important step in the process of logistics system planning and development. The existing models for logistics demand forecasting often encounter problems of low forecasting accuracy and slow convergence speed. Consider these problems, a hybrid model called KPCA-LSSVR-TOOPSO was proposed to improve the forecasting accuracy and accelerate the convergence speed. The hybrid model integrated the Kernel Principal Component Analysis (KPCA), the Two-order Oscillating Particle Swarm Optimization (TOOPSO) and least squares support vector regression. First, the nonlinear features of the influential factors of logistics demand were extracted by KPCA. Then, the kernel principal components were put into LSSVR and a LSSVR model was established for logistics demand forecasting. Finally, TOOPSO algorithm was used to optimize the parameters in LSSVR. Empirical results from the China logistics demand indicate that the proposed model decreases the dimension of the modeling data. The minimum and maximum relative prediction errors of the proposed model are only -0.33548 and 5.3270%, respectively. The NRMSE, NMAE and MPE of the model are 0.1890, 0.1367 and 0.0182, which are smaller than ones of the other three models. The convergence speed of the proposed model is the fastest among the four models.