Multiple bad data processing by genetic algorithms

Abstract

The identification of multiple bad data, especially when mutually interacting, may be difficult to handle, since the well known procedures based on the normalized or weighted residuals may become faulty. The identification problem is formulated here as that of picking bad data from a set of suspect measurements in order to fulfill the requirements of maintaining observability and eliminating the minimum number of measurements. Three non-deterministic solution procedures based on the use of genetic algorithms are proposed. Aiming at reducing the computation burden, the possible advantage deriving from working with small populations has been investigated by implementing a micro-genetic approach and an evolution strategy in which a single individual population is employed. Numerical efficiency is improved by reducing the number of state re-estimations; a database of already computed cases is used and a filtering mechanism has been designed to skip non promising solutions. Tests are carried out with reference to the IEEE standard test systems.