Heuristic method in searching for global minima relative to high voltage ac error

Searching for global minima is difficult and is still considered an unsolved problem in the study of optimization algorithms. In this paper, a heuristic method for searching for global minima relative to high voltage (HV) AC error is discussed. This heuristic method is important in the design of a HV power fault classification system such as the one described by L. Han et al. (1999). In this system signals, AC voltage error and 6 pulse signals etc., are preprocessed in feature information needed to calibrate a neural network for fault type classification. The feature information is ranged into low, medium or high distributions. Each partition should be optimized according to a criterion, namely, forcing the data points from the same type of fault to fall into the same range. Based on the AC error signals and associated fault-type, a 2-dimensional optimization problem with inequality constraints can be set up. To optimize the partition, we apply the new global minimum/maximum search algorithm. The comparison with the traditional bracketed interval method is discussed and this new method's simulation result is provided.