TO POWER SYSTEM MODELLING AND CONTROL
R.R. Zakrzewski, R.R. Mohler · 1992
Artificial feedforward networks arestudied asnonlinear function approximators usedtoidentify forward andinverse mappings ofdiscrete timedynamic systems. Theyarefound toprovide significant advantages overother modelling techniques suchaspolynomial approximations, especially ifthe extrapolation beyond theregion covered bythelearning data isinvolved. Weapply theneural network methodology toa simple second orderapproximation ofa single-machine infinite-bus powersystem controlled bymeansofmodifying thereactance oftheline. Accurate off-line identification of forward andinverse dynamics ofthesystemn isperformed by meansofsingle hidden layer neutral networks, andbothmodels arethenusedinadirect inverse control configuration. The controller simulations showverygoodquality oftransients for severe short-circuit fault.