| The objectives of transient stability assessment (TSA) are to assess the transient stability of a power system subject to a set of pre-assigned contingencies, and to provide the operators with efficient control measures to assure the system transient stability while maintaining economic operation. TSA has always been a challenging problem since the methodology has to be able to cope simultaneously with modeling complexities, the large dimensions of power systems, and a fast assessment of system behavior under a variety of contingencies.; In this work, a new method based on artificial neural networks (ANNs) has been developed to estimate the security margin on-line. The security margin, i.e., the distance between the current operating point and the security limit, for a given power system is obtained by applying Western Electricity Coordinating Council (WECC) Disturbance-Performance Criteria for transient response to off-line time domain simulations. These simulations then form a database that can be used to train the ANNs. Feature selection using statistical approaches is applied to overcome the dimensional problem of applying the ANN to larger systems. If the estimated security margin is less than requirements, then preventive control actions that assure dynamic security of the power system are needed. This is achieved by rescheduling the generation with the given constraints on the network power flows and the transient security margins as estimated by the ANN. This requires a modified optimal power flow (OPF) solution that allows the trained ANN to act as a security objective function.; Various techniques of ANNs based on supervised learning, i.e., backpropagation with L2 error function, backpropagation with L1 error function, and radial basis function network, are investigated for transient stability assessment application. Further, to improve the convergence of the steepest-descent algorithm, the learning-rate adaptation method is incorporated into standard backpropagation to find appropriate step lengths. The numerical results of the New England 39-bus and the WECC systems clearly establish the viability of applying artificial neural networks to on-line dynamic security assessment for today's power systems. |
