In order to meet the demand of integrated power flow calculation speed

an improved Newton method is proposed. In view of the seriously ill-conditioned Jacobian matrix and poor convergence and so on

an adaptive Levenberg-Marquardt method is used to choose the initial values

and incomplete LU decomposition method is used to preprocess Jacobian matrices

which effectively guarantees the numerical stability and improves the convergence of Newton method. To solve the problem of large-scale and low computing efficiency after integration of transmission and distribution networks

the GPU parallel acceleration technology is used to accelerate some steps with intensive computation including the formation of Jacobian matrix

matrix-vector operation and so on. The numerical experiments show that

the proposed algorithm can significantly improve the speed and accuracy of the large-scale integrated transmission and distribution networks

and it has a strong universality for different situations such as multiple distribution network areas

distribution network with loops or distributed generators

ill-conditioned systems.