AHU sensor minor fault detection based on piecewise ensemble empirical mode decomposition and an improved combined neural network
Xiuying Yan, Boyan Zhang, Guangyu Liu, Kaixing Fan
Abstract
Faulty or inaccurate sensors may lead to increased energy consumption, thermal discomfort, and even failure of control systems. An improved piecewise ensemble empirical mode decomposition (PEEMD)-improved combined neural network (CoNN) is presented for air handling unit sensor fault detection. Piecewise ensemble empirical mode decomposition is used to enhance data quality by denoising raw data sets. The coupling relationship between variables is determined through a data-driven method, and the variables with the lowest correlation between the basic neural network and the auxiliary neural network are removed to reduce the dimensionality of the system, which can shorten detection times. Based on the CoNN, a modified relative error is proposed to improve the fault detection rate, which is called MOCoNN. The results show that, compared to CoNN, the fault detection rate of PEEMD-MOCoNN improved by 49.21%, 34.85%, 90.70%, and 18.60% in the 5%–10% minor bias fault and 0.01/unit–0.02/unit hidden drift fault. Meanwhile, in the same fault condition, the fault detection rate of PEEMD-MOCoNN improved by 43.41%–85.71% and 16.16%–54.65% compared with those of empirical mode decomposition threshold denoising principal component analysis and kernel principal component analysis and double layer bidirectional long short-term memory, respectively.