What is the basis for the PEMS predictive engine? Is it a neural network or a first principle (physical correlation and gas theory)?

The engine is a unique statistical hybrid model. It is neither a neural network nor a first principle type, but it is an empirical method. All CMC PEMS run the same core module – the statistical hybrid engine. The model does not use a theoretical methodology such as a first principle formula nor does it require an iterative model development and testing regimen with experts onsite. There is no specialized staff required to build or maintain the emission model. Continuous Emissions Monitoring (CEM) or Reference Method (RM) test data is used to build the initial model. The empirical SmartCEMS®model utilizes historical data, paired emissions data and process data, to generate predictions in real-time. The predictions are derived directly from the historical training dataset using input parameters from the process that are available from the existing control system and are configured in the PEMS model.


Unlike more complicated empirical systems such as neural network and first principle formulations, the statistical hybrid model can be developed for any given process without any knowledge of the underlying combustion chemistry or of the pollutional controls. The system uses the historical data collected during normal operations and during startups, shutdowns, or transitional states to accurately predict emissions for compliance purposes over the full load range of the unit. This feature allows the system to predict the value of any failed input (or pollutant emission data in lieu of CEMS) with the accuracy of the system entirely dependent on the range and quality of the original data in the training dataset. The system is deterministic in that a given set of inputs (process variables or lack of them) along with a given fixed training dataset will yield a single result for each value to be predicted. The model is unique in its ability to be developed by non-specialized staff that have no familiarity with the process, pollution control devices, or the methodology used by the model. Customers and third party consultants can update the model without support of the manufacturer’s engineering staff.