Abstract: The accuracy of numerical result of dynamic simulation is very important to power system planning and operation. This paper proposed a new numerical integration algorithm named spectral deferred correction（SDC） with symplectic property. Based on the low-order integration algorithm, the new method constructed a high-order spectral correction by Gauss quadrature integration and refines the numerical result of the low-order algorithm iteratively to obtain high-order convergence and large step capability. The method has good programming compatibility and easy implementation to reform the traditional low-order simulation program using alternating solution method. The high-order convergence and symplectic property have been verified in a small system. Based on the platform of power system full dynamic simulation program PSD-FDS, an IEEE standard 140 buses system was also performed to compare the simulation property of the trapezoidal rule and the SDC method for different simulation steps. Numerical results show that the SDC method are more accurate and reliable, and that SDC method can not only have high precision, but also can avoid the accumulating error and phase delay of trapezoidal rule during long time simulation.