Artificial Dendritic Cell Algorithm for Advanced Power System Monitoring

Abstract

The functionality of dendritic cells, as part of the biological immune system, exhibits significant capabilities for class discrimination, pattern recognition, data processing, and information structuring and storage. These properties provide the basis for developing biomimetic techniques for monitoring and control of complex dynamic systems and processes. In this paper, a novel artificial dendritic cell algorithm in conjunction with the hierarchical multi-self strategy is developed within the artificial immune system paradigm to support advanced power plant health management. The proposed algorithm is addressing the specific characteristics of modern power plants, which include complexity, multidimensionality, and high coupling. The self/non-self representing the power plant is defined as a set of lower dimensional projections corresponding to the constituent sub-systems and capturing their inter-connections. The artificial dendritic cell is defined as a computational unit that collects the outcomes of the self/non-self discrimination process produced by all projections, structures and stores information, and presents it to the system health monitoring logic for abnormal condition detection and identification. The functionality of the proposed algorithm is illustrated using the self for an acid gas removal unit by detecting the occurrence of several abnormal conditions and identifying the affected sub-systems. Excellent detection and identification performance is obtained in terms of detection time, false alarms, detection rate, and identification rate for all cases considered. It is concluded that the proposed approach can potentially represent an effective and efficient tool for complex system health monitoring.