Artificial Intelligence: Summary and Hybrid Schemes - Digital Signal Processing in Power System Protection and Control - page 304

• Generating rules and inferences in linguistic form,
• Extracting ill-defined image regions, primitives and properties and describing
relations among them as fuzzy subsets.
The ANNs provide natural classifiers having:
• Resistance to noise,
• Tolerance to distorted patterns/images (ability to generalize),
• Superior ability to recognize overlapping pattern classes or classes with highly
nonlinear boundaries or partially occluded or degraded images,
• Potential of parallel processing.
The genetic/evolutionary algorithms are characterized as efficient, adaptive and
robust search processes, producing near-optimal solutions and have a large amount
of implicit parallelism.
The expert systems are capable of making decisions in the face of many
arguments, coming to the solution on the base of knowledge and reasoning rules
defined by a human expert as well as explaining a line of reasoning and providing
the details when necessary.
As usual, each technique is not free from drawbacks that may prevent them
from being applied for some technical problems. The most important difficulties
arise from the fact that:
• The FL and ES schemes are basically not trainable,
• Interpreting of ANN internal signals is difficult, if not impossible,
• Robustness of the ANN scheme is difficult to ensure,
• Improving or extending operational range of an ANN scheme is possible usually
with renewed training of the neural network with new patterns,
• Troubleshooting in case of an ES requires careful examination of all rules (and
adding new rules, if necessary),
• Computational burden by implementation of vast expert systems or big ANN
structures may exceed technical capabilities of available hardware.
15.2 Hybrid Solutions
Application of a single AI technique (with its strengths, capabilities and assump-
tions) to solve a real-world power system problem may not bring satisfactory results.
Integration of two or more techniques may be required in some cases. As mentioned
in previous section, a combination of various AI techniques in one scheme may
contribute to mutual reinforcement of advantages and elimination or at least sig-
nificant reduction of their weaknesses. Numerous topics (e.g. [ 2 , 8 , 13 ]) and papers
(cited later) are available on that topic.
A combination of various AI techniques are aimed at providing flexible
information processing capability for representation and evaluation of various real-
life ambiguous and uncertain situations. The hybrid systems: