Contaminant monitoring for industrial emission-diffusion process based on non-negative tensor factorization

This paper considers the contaminant emission diffusion monitoring problem by applying the tensor approach for the industrial-environmental protection. The emission contaminant concentration is used from an embedded sensor for computation of features and monitoring of contaminant diffusion. A reduced feature subset, which is optimal in both estimation and clustering least squares errors, is then selected using a new dominant feature monitoring algorithm to reduce the signal processing and number of sensors required. The matrix based subspace method can't capture the spatiotemporal characteristics effectively. The tensor space method is proposed to be used to monitor the contaminant concentration in environmental protection area. In order to fit the multiple invariance of the measurement output tensor data processing for contaminant emission diffusion monitoring, the non-negative tensor factorization model is proposed to analyze the tensor data, which stems from uniqueness of low-rank decomposition of higher-order tensor. By using the non-negative tensor factorization, the estimated latent contaminant concentration data structure combing with the covariance-based algorithm are given to derive the metric of contaminant concentration in the environmental protection area. Contaminant concentration is then measured using non-negative tensor factorization with observable data based on the reduced features. A simulation example is provided to test the effectiveness and advantages of proposed method using tensor method with only the dominant features measurement.