Power system reactive power optimization based on direct neural dynamic programming

Reactive power optimization in power system is a complex nonlinear combinatorial optimization problem with multiple constrained conditions. However, direct neural dynamic programming (direct NDP) approach based on on-line measurements can be employed in this situation, which is independent of models. In this paper, on the basis of applicable analysis to reactive power optimization, this algorithm is improved in the expansion of dimension, and then a direct NDP model is established for reactive power optimization. It is mainly composed of two neural networks: action network (AN) and critic network (CN), AN is used to control, and CN is used to evaluate current system states and update AN. At last, the improved algorithm is tested in the IEEE 6-bus system and compared with the GA optimization algorithm, the results demonstrate that the new algorithm is a feasible and effective way to solve the reactive power optimization problem.