Real-time closed-loop system controlled by an artificial neural network for estimation of the optimal load shedding

Electrical power systems (EPS) are constantly exposed to various disturbances that can significantly affect their operation. Hence, the load-shedding philosophy was proposed in order to relieve the overloaded infrastructure in cases of imbalances between generation and demand. However, conventional methods of shedding are slow and inaccurate. In this context, where shedding processes must be optimized, there is a need to pursue new methods and technologies to provide fast and optimized management. Therefore, an artificial neural network (ANN) is proposed to estimate the minimum amount of load to be shed in order to recover load-generation balance. The ANN training and testing data were extracted from an EPS simulated using a real-time digital simulator (RTDS). The best ANN topology, selected through a cross-validation technique, was able to estimate the load shedding quantity with high precision. The system was then configured to work in a real-time closed loop so that the efficiency of the ANN was tested. The results demonstrate a good generalization over the presented overload situations. As a contribution, this work presents the dynamics of a load-shedding scheme controlled by an ANN in a real-time closed-loop system performed only in one step, offering a fast and effective alternative for restoring the frequency close to its nominal value.