The effects of Brownian motion and turbulent motion on particle agglomeration in turbulent agglomerator were evaluated. Particle agglomeration was numerically simulated by establishing a KTGF (Kinetic Theory of Granular Flow)-RSM (Reynolds Stress mathematical Model)-PBM (Population Balance Model). Capture efficiency a and fractal dimension were introduced to modify the turbulent agglomeration kernel model. The RMS (Root Mean Square) method used to obtain a kernel model suitable for turbulent agglomeration of coal-fired PM2.5. The model was compared with the traditional turbulent agglomeration model. The results show that the dissipation rate in the vortex area clearly increased, and the particle motion became aggregative fluidization to enhance the agglomeration effect. When the fractal dimension is 2.5, it is suitable for the turbulent agglomeration of nonuniform structure, and the Brownian motion in the calculation of kernel function cannot be ignored. The RMS model is closer to reality. Compared with the experimental results, the maximum error of ideal turbulent agglomeration model is similar to 8%, whereas the maximum error of new turbulent agglomeration model is only 3%.