Abstract: Based on the theory of dynamic low-rank approximation, we use the singular value decomposition to find the approximate expression of particles on low-rank manifolds, and derive the time integral of the factor matrix by tangent space orthogonal projection. The proposed dynamic low-rank approximation particle simulation method achieves the reduction of particle computation by truncating the rank. A typical physics problem of plasma two-stream instability is simulated to verify the feasibility of the method and evaluate the computational efficiency. The results show that in the one-dimensional electrostatic two-stream instability case, the method reduces the particle computation by about 48% while guaranteeing the maximum variation ratio of the Hamiltonian by 0.065%.