Enhanced Power System Security Assessment Through Intelligent Decision Trees - Complex System Modelling and Control Through Intelligent Soft Computations - page 356

Fig. 15 Critical monitoring locations from decision tree: MVD versus MVN

6 Conclusion

The proposed ef cient sampling method based on importance sampling idea is one

of the first to be used in power systems for making decision tree based learning

methods effective. The thrust of the proposed sampling procedure was to re-orient
the sampling process to focus more heavily on points for which post-contingency
performance is close to the threshold, i.e., boundary region that contains operating
conditions in uential for rule formation. The primary goal was to increase the

information content in the learning database while reducing the computing
requirements, and consequently obtain operational rules that are more accurate for
usage in real-time situations.
The developed ef

fl

cient training database approach was applied for deriving
operational rules in a decision tree based voltage stability assessment study on
RTE-France

s power grid. The results showed that the generated training database
enhances rules

'

accuracy at lesser computation compared to other traditional
sampling approaches, when validated on an independent

'

test set. The chapter

also emphasized the signi cance of sampling from non-parametric correlated-

multivariate load distribution obtained from historical data, as it is more realistic.
Doing so also ensures generating operating rules that provide higher classi cation

accuracy and economics, and selecting interesting monitoring locations that are
closer to the contingency event, as corroborated by the results. In order to reduce