Load Margin Assessment of Power Systems Using Recurrent Neural Networks

Several factors have propitiated the expansion of electrical systems and this expansion causes a greater complexity of operation. The operating complexity of current power systems increasingly requires the development of indices capable of identifying instability mechanisms in the long term. The load margin of power systems is an index to evaluate the distance that the power system is from a stability problem. Usually, this index employs static models that govern the power flow of the system and aims to determine the voltage collapse condition. However, low-frequency oscillation modes present in dynamic models can also cause instability in power systems when they are not well damped. Thus, the load margin traditionally calculated by static models can provide information on the existence of operating points in the system, but these operating points can present oscillation modes with low damping rates that can be observed in the dynamic analysis of systems often subject to contingencies and these modes can affect the stability of operation of power systems. Thus, it is important to inform the operation center of a load margin index value that satisfies the requirements of the existence of equilibrium points in voltage stability studies and well-damped oscillation modes in small-signal rotor angular stability studies. This paper proposes a new method based on Recurrent Neural Networks (RNNs) to evaluate the load margin with evaluation of the requirements of small-signal and voltage stability. The proposed method of this paper was applied on IEEE 39-bus for different scenarios. The obtained solutions demonstrate a good ability to generalize the RNN and classify the load margin ranges.