Abstract: To overcome the difficulties existing in the conventional numerical methods, a novel neural network (LDPNN) for solving large-scale dynamic programming problems is developed in the paper. The dimension of the neural network is reduced greatly by imbedding the dynamic equations of the subsystems into the corresponding local optimization networks. Moreover, the coordination and local optimization are carried out simultaneously. Therefore, the LDPNN not only has concise and compact structure ,but also has high efficiency for problem-solving. An demonstrative example shows that the LDPNN is suitable for real-time optimization of large-scale systems.