ON-LINE POWER SYSTEM LOAD-FLOW USING NEURAL NETWORKS

This paper presents a new neural network based method for power system load-flow analysis. The outputs of a load-flow program are obtained by solving a set of nonlinear algebraicequations. Assuming that the parameters of the system are known, these outputs are only dependent on the initial conditions (values). Therefore, we may view the outputs of the load-flow program as functions of initial conditions. Indeed, we are faced with a function approximation problem. This can be done by neural networks. In fact, in order to implement an on-line power system load-flow analysis, we may employ a multilayered feedforward neural network with the initial conditions as the inputs and the outputs of the load-flow program as the outputs of the network. To train the neural net, we let the initial values vary over specified ranges.For fast training purpose, we employed the Marquardt based backpropagation algorithm. Finally, the proposed method has been applied into a three machines test system. The performance of the method has been fully discussed.