Intelligent load shedding schemes for industrial customers with cogeneration facilities

In this paper a neural network-based methodology is proposed to develop load shedding schemes for industrial power systems. For a given post-disturbance scenario, an appropriate architecture of several neural networks suggests the nets necessary to provide the requested load-shedding action. It is assumed that, depending on the type of contingency and pre-disturbance network configuration, the post-disturbance scenario can involve the islanding of the entire power system or the decomposition into subsystems. Depending on the current generation-load mismatches of each subsystem, the exact amount of load to be shed is provided as output to a defined set of input signals. Since each net is trained using input-output patterns obtained from extended transient stability studies, the load shedding action is based on a 'dynamic' criterium, overcoming the limit of a load shedding action computed on the basis of the 'static' criterium of the anticipated overload. The effectiveness of the proposed procedure is tested on the power system of a petrochemical plant.