Locating unknown number of multi-point hazardous gas leaks using Principal Component Analysis and a Modified Genetic Algorithm

Abstract Identifying multi-point hazardous or contaminating gas leak sources is important for emergence treatment and pollution control, whose difficulty, however, may increase if the number of sources is unknown a priori. This study proposed a novel method to estimate the number and locations of several leak sources using Principal Component Analysis (PCA) and a Modified Genetic Algorithm (MGA). PCA works by counting the number of leak sources and providing zones possibly containing each source. MGA is then implemented sequentially to accurately locate each source in those zones. This method was tested in a leak field generated by a steady-state two-dimensional Gaussian plume model with one, two and three leak sources. The effects of concentration sensor array size, leak source location and measuring noise on PCA and MGA performance were analyzed. Using more sensors increases the identification accuracy of PCA but reduces the MGA calculation speed. PCA cannot identify leak sources locating too downstream or having spreading fields with a large overlapping part. The measuring noise generated by Gaussian Noise has little effect on PCA performance, but increases MGA estimation error when identifying source locations.