In order to improve particle size inversion accuracy of the dynamic light scattering (DLS), considering non-negative characteristic of the particle size distribution (PSD) and strong anti-noise characteristic of truncated singular value decomposition (TSVD) method, TSVD inversion methods of the trust-region (Trust) and interior point Newton (IPN) non-negative constraints are compared in this paper. Combining characteristics of two methods, a particle size inversion method based on Trust-IPN and TSVD (Trust-IPN-TSVD) is proposed. This method inherits the advantages of the Trust-TSVD and IPN-TSVD methods. The inversion of Trust-TSVD, IPN-TSVD and Trust-IPN-TSVD method for 200~500 nm unimodal distribution particles and 100~700 nm bimodal distribution particles is simulated. Results show that compared with the Trust-TSVD, for unimodal and bimodal distribution particles, Trust-IPN-TSVD can improve the peak value error and the relative error of inversion PSD by 1.68%, 0.2461 and 1.41%, 0.0587, respectively, and its PSD is smoother. Compared IPN-TSVD, for unimodal and bimodal distribution particles, Trust-IPN-TSVD can improve the peak value error and the relative error of inversion PSD by 4.52%, 3.710 and 9.47%, 0.4229, respectively, and its PSD is significantly narrower. Therefore, the inversion PSD of Trust-IPN-TSVD method has higher accuracy and better smoothness, and is more consistent with the theoretical distribution. Finally, the inversion of experimental particles confirms this conclusion.