The invention discloses a method for optimizing the main steam pressure of a large steam turbine set, which belongs to the field of learning, modeling and optimizing algorithms of machines. The method comprises the following steps of: firstly, learning and modeling a heat loss characteristic sample set obtained by a machine set performance test by a support vector machine method; obtaining a parameter relation of machine set heat loss, main steam pressure, main steam temperature, ambient temperature and a machine set load; then learning a subset in the sample set, establishing a heat loss model and carrying out increment analysis; establishing a steam turbine set pressure optimizing model and optimizing and solving the steam turbine set pressure optimizing model by a genetic algorithm to obtain main steam pressure optimizing value and curve and an optimizing data set; and finally, carrying out secondary modeling on the optimizing data set to obtain a pressure optimizing characteristic relation. The method gives an optimized main steam pressure set value according to the current running condition of the machine set, greatly amplifies the optimizing range and is particularly beneficial to the energy saving and the consumption reduction of the large power station steam turbine set.