Representation of power system load dynamics with ANN for real-time applications

Among severe power system disturbances degrading power quality are voltage sags and transient power supply interruptions. Dynamic behaviour of loads under these types of disturbances must be taken into account in the development of mitigating devices such as dynamic voltage restorer (DVR), active filters etc. This paper presents a representation of load dynamics based on non-linear black box approach with artificial neural networks (ANN). Two types of load models, neural network autoregressive moving average with exogenous inputs (NNARMAX) have been developed. These models have been trained and tested to predict dynamical behaviour of the loads especially at bulk supply point under voltage sag conditions. Off-line trained as the power system in which these models are included nevertheless exhibits a random behaviour. A moving window based approach has been adopted in real-time parameter updating in the proposed load models.