The low computational efficiency and poor optimality of solutions are all along two key problems plaguing the research and application of large-scale hydrothermal optimal power flow(HTOPF). To solve these problems

an accurate and effective decomposition algorithm for HTOPF is proposed. Based on approximate Newton direction method that decouples the first order Karush-Kuhn-Tucker(KKT)conditions of the original problem

a HTOPF problem with cascaded hydro plants is decomposed into a thermal plant sub-problem and a hydro plant sub-problem. The thermal sub-problem is decomposed into T period optimal power flow(OPF)problems over a certain time horizon

and each OPF problem is further divided into multi-area sub-problems. According to different types of hydro plants

the hydro plant sub-problem is combined with fixed head hydro plant sub-problems

variable head plant sub-problems and one cascaded plant sub-problem. Each sub-problem needs only a single iteration instead of optimal solution

which results in a very high efficiency. In order to verify the effectiveness of the proposed algorithm

numerical tests are performed on three large-scale test systems of. Test results show the proposed algorithm has excellent performance in convergence and stability. Not only is memory usage significantly reduced

but also the central processing unit(CPU)time is decreased by about 65%~75%. Under the condition of parallel computing

it is capable of achieving 10-20 times or even 1 000 times faster without loss of optimality and ensuring the error between optimal target value and exact value is below 10-8

which ensures the optimal result of decomposition and coordination.