Abstract

For signal processing in sensor networks there is an on-going challenge for filling missing information when it is either incomplete, uncertain or biased, in ways that are both efficient and with confidence. This paper reviews three established and additional newly developed techniques addressing the problem. Considering sensor signals that are highly correlated in a sensor network, one sensor measurement can be reconstructed based on measurements from other sensors. In such cases, three signal reconstruction methods are considered: 1) principal component analysis (PCA) based missing value approach; 2) self-organizing map neural network (SOMNN) based algorithm; and 3) an analytical optimization (AO) technique. To demonstrate the efficacy of the methods, temperature data are studied on an industrial gas turbine system, where, especially, a faulty sensor signal is utilized to be reconstructed from the other sensor measurements.