Probabilistic dynamic dispatch（PDD） can coordinate power system operating economy and reliability， and is thus superior to the traditional deterministic dynamic dispatch. However， the large scales and difficulties in calculation associated are the main obstacles for PDD's engineering applications. A novel solution method based on Benders decomposition is proposed to solve the large scale linear problem of PDD. According to the coupling between operating states， the proposed method decomposes the original PDD problem into a dynamic economic dispatch master problem（MP） at the normal operating states and a number of post-contingency adjustment sub-problems（SPs） at contingency states based on the Benders decomposition techniques. The solution of the original problem can be realized by the iteration of MP and SPs， and each optimization solution scale is effectively reduced. Moreover， a SP pre-selection technique by checking the adaptability of MP results is also deduced in the context to reduce the number of SPs needed in each iteration. The proposed method can increase the calculation speed and solve relatively large scale power system PDD problems. Case studies on a power grid show the validity and efficiency of the proposed method.