Outliers Detection in Environmental Monitoring Data

Environmental monitoring in today an important task in any industrial plant operation, because high concentrations of residuals products produced by the process can cause negative impact on human health and over the environment. Usually industrial complexes have environmental monitoring systems based on a network of air quality and meteorological stations providing measurements. These measurements can be contaminated by outliers, which must be diScircded in order to have a consistent set of data. This work presents a nonlinear procedure for outliers detection based on residual analysis of regression with Partial Least Squares and Artificial Neural Networks. This procedure was tested on a simulated nonlinear process and on real data from environmental monitoring contaminated with synthetic outliers. In addition, it is also applied to a real environmental monitoring data having natural outliers, which has been obtained from a monitoring station. The results are encouraging and further developments are foreseen for including information from neighboring stations and emissions source operation. In addition, the method can also be applied in other tasks such as soft sensor design data-driven modeling or validation of operational variables in industrial databases.