Development of an Artificial Immune System for Power Plant Abnormal Condition Detection, Identification, and Evaluation

In this paper, the artificial immune system paradigm is used to develop a computational scheme for the detection, identification, and evaluation of abnormal operation of advanced power plants. The self/non-self generation relies on a novel approach consisting of partitioning the Universe and representing clusters as integer strings that can be produced and used with reduced computational effort. The design of the proposed scheme utilizes a positive-selection-type approach combined with a dendritic cell mechanism. The methodology is demonstrated using a high performance model of the acid gas removal unit implemented in Dynsim® that is part of the power plant simulation environment available at West Virginia University AVESTAR Center. Fourteen different abnormal conditions have been considered including solid deposits and leakages occurring at typical locations throughout the system. The proposed monitoring scheme provides excellent performance in terms of false alarm and detection, identification, and evaluation rates.