Among all the macro-pollutants released by waste combustion, acid contaminants such as sulphur dioxide, hydrogen chloride and hydrogen fluoride have the lowest emission standards in environmental regulations in EU, USA and China. Their removal is thus a key step of flue gas treatment in waste-to-energy (WtE) plants. A widespread approach for acid gas removal is by in-duct injection of dry powdered sorbents, which neutralize the acid pollutants by gas-solid reaction. However, systems based on dry injection, albeit cost-effective and easy to operate, suffer from a limited knowledge of the gas-solid reaction process at industrial operating conditions. High excess of sorbent feed rate is generally required to obtain high acid gas removal efficiencies.
The present study proposes a multivariate statistical approach to the modelling of acid gas treatment units, with the aim of extracting information from real process data in order to derive a predictive model of dynamic acid gas removal efficiency. Specifically, process data regarding the composition of the flue gas, the sorbent feed and other operating conditions were elaborated to characterise the different phenomena that influence acid gas abatement. Eventually, a partial least squares (PLS) regression was set up to predict the outlet concentration of hydrogen chloride as a function of the measured process variables. The resulting model is a step forward with respect to previously available stationary models. Its simplicity and low computational cost could make PLS a promising candidate for model-based process control. Nonetheless, a linear approach such as PLS still comes short of predicting large instantaneous deviations from the typical range of operation (e.g. abrupt peaks in inlet acid gas load), for which a modification of the PLS model to incorporate non-linear behaviour is envisaged.