Nonlinear Partial Least Squares Modeling for Instrument Surveillance and Calibration Verification

An Instrument Surveillance and Calibration Verification (ISCV) system primarily consists of a process model, which is used to verify the output of the measurement instruments in that process. Artificial Neural Networks (ANNs) and Partial Least Squares (PLS) are two methods, which can be used for model development. The linear transformation of the PLS method provides a supervised reduction of the process system's data while preserving the information most relevant to the specific parameter(s) to be modeled. Following this transformation, the PLS method prescribes a linear regression to relate the reduced data to the parameter(s) being modeled. By replacing the linear regression with a single hidden layer ANN, nonlinear relationships can also be incorporated into the model. The combination of PLS and ANNs, referred to as Non-Linear Partial Least Squares (NLPLS), results in a more accurate model of the process than the standard PLS method with linear regression. Using data from Tennessee Valley Authority's Kingston Unit 9, an ISCV system has been developed for 140 measurement instruments with an average estimation error of less than 1% of the measured value.