The invention discloses a global new coronavirus propagation prediction method based on an optimized SEIRD model, and relates to the technical field of infectious disease data processing, and the method comprises the steps: A, collecting data, and building an improved SEIRD cabin model on the basis of an epidemiological SEIR model: 1, increasing the index of latent crowd infection rate, subdividing the susceptible population conversion rate in the SEIR model into a latent population infection rate and a definite population infection rate; 2, subdividing the removal rate in the SEIR model intoa death rate and a cure rate; 3, expressing the cure rate and the death rate as a time sequence function of time t; dividing crowds in the bin into five types, namely susceptible people, symptomatic latent infected people, symptomatic infected people, cured people and dead people; B, respectively carrying out parameter estimation and model fitting on the model according to a least square method, estimating an inflection point and an end date of an epidemic situation, and carrying out model inspection by using actual data. The method has the advantages of accurate prediction and good reliability.