To solve the problem that particle swarm optimization and most of its improved algorithms are easy to fall into local convergence, standing on bionics and psychological point of view, a human simulated particle swarm optimization(HSPSO) algorithm based on deep extended memory is presented. The knowledge of each particle is shared by introducing the knowledge sharing part, which is accumulated by the deep extended memory, and the human simulated forgetting function is used to set up the weight of knowledge of different period. The simulation analysis shows that HSPSO algorithm is highly sensitive to the forgetting function and the forgetting factor, and has a better performance in convergence precision, success ratio and reducing the cost of algorithm compared to the standard PSO and its improved algorithm when applied to the optimization of multi-dimensional and multi-extreme value functions.