This paper presents an artificial intelligence (AI) approach to the op
timal reactive power (var) control problem. The method incorporates th
e reactive load uncertainty in optimizing the overall system performan
ce. The artificial neural network (ANN) enhanced by fuzzy sets is used
to determine the memberships of control variables corresponding to th
e given load values. A power flow solution will determine the correspo
nding state of the system. Since the resulting system state may not be
feasible in real-time, a heuristic method based on the application of
sensitivities in expert system is employed to refine the solution wit
h minimum adjustments of control variables. Test cases and numerical r
esults demonstrate the applicability of the proposed approach. Simplic
ity, processing speed and ability to model load uncertainties make thi
s approach a viable option for on-line var control.