Enhanced Power System Security Assessment Through Intelligent Decision Trees - Complex System Modelling and Control Through Intelligent Soft Computations - page 354

Table 5 Performance based
on sampling bias

p Size Entropy Accuracy False

alarm

Risk

Base 17,748 0.7423 92.51 0.063 0.091

0.25 13,840 0.7716 93.4211 0.064 0.068

0.50 9,932 0.8181 94.9899 0.049 0.051

0.75 6,025 0.9038 96.0526 0.038 0.041

1.0 2,852 0.9993 97.5202 0.021 0.03

Figure 14 shows operational rule formed using two attributes, namely reactive
reserves at Chevire unit and Chinon unit respectively. The operating conditions
shown in the Fig. 14 are from the entire database. It can be noticed that the rules
formed using the database exclusively from the boundary region is providing more
operating conditions to be exploited in real time situations, than the rule derived
using the database from entire region; because of the increased knowledge and
clarity of the boundary limits.

Sampling Strategies Comparison

Table 7 shows the comparison results of two different sampling approaches,
namely,

1. Importance sampling (IS) of boundary region, with load distribution modeled
with multivariate normal (MVN) distribution (pruned tree).
2. Importance sampling of boundary region, with load distribution modeled with
correlated non-parametric multivariate distribution (MVD) (pruned tree).
3. Same as case 2, with un-pruned tree.

Table 6 Economic benefit from efficient sampling
Top Node

p = 0 p = 1

Cordemais voltage 401.64 kV 399.88 kV

Domloup voltage 397.56 kV 394.51 kV

Louisfert voltage 399.1 kV 396.46 kV

Plaine-Haute voltage 392.26 kV 387.21 kV

Chevire unit reactive reserve 131.38 Mvar 90.76 Mvar

Chinon unit reactive reserve 1,127.54 Mvar 694.62 Mvar

Cordemais unit reactive reserve 70.97 Mvar 16.23 Mvar

Total SEO region reactive reserve 7,395.88 Mvar 6,510.36 Mvar

Plaine-Haute SVC output 11.82 Mvar 13.64 Mvar

Poteau-Rouge SVC output 16.3 Mvar 22.03 Mvar