Prediction of PCDDs/PCDFs emissions from municipal incinerators by genetic programming and neural network modeling

Abstract

The potential emissions of PCDDs/PCDFs from municipal incinerators have received wide attention in the last decade. Concerns were frequently addressed in the scientific community with regard to the aspects of health risk assessment, combustion criteria, and the public regulations. Without accurate prediction of PCDD/PCDF emissions, however, reasonable assessment of the health risk and essential appraisal of the combustion criteria or public regulations cannot be achieved. Previous prediction techniques for PCDD/PCDF emissions were limited by the linear models based on a least-square-based analytical framework, such that the inherent non-linear features cannot be explored via advanced system identification techniques. Recent development of genetic algorithms and neural network models has resulted in a dramatic growth of the use of non-linear structure for optimization and prediction analyses. Such approaches with the inherent thinking of artificial intelligence were found useful in this study for the identification of non-linear structure in relation to the PCDD/PCDF emissions from municipal incinerators. Examples were drawn from the emission test of PCDDs/ PCDFs through the flue gas discharge from several municipal incinerators in both Europe and North America. Although the neural network model may exhibit better predictive results based on the performance indexes of percentage error and mean square error, model structure cannot be directly identified and expressed for illustrating the possible chemical mechanism with respect to the PCDD/PCDF emissions. However, the tree-structured genetic algorithms, or so-called genetic programming, can rapidly screen out those applicable non-linear models as well as identify the optimal system parameters simultaneously in a highly complex system based on a small set of samples.