New multivariate linear regression real and reactive branch flow models for volatile scenarios

Nonlinearity of power flow equations is one of the major underlying factors in a power systems operation complexity. The need for a robust and less complex models rises in a volatile, dynamic and real time scenario. This paper introduces new empirical models using multivariate linear regression (MLR) methods with least squares for both real and reactive branch flows. The models do not make prior assumptions and do not depend on a particular base case. Instead they are trained on either simulated or historical data. Tests using the IEEE 14 bus system show that given similar input variables to DC models, the MLR models performs significantly better. They also show that the MLR models have good prediction accuracy in scenarios with high volatility.