A convex optimization framework for the identification of homogeneous reaction systems

Abstract

A new convex optimization framework for the simultaneous identification of reaction stoichiometries and kinetics for homogeneous reaction systems is developed. The identification problem is posed as a non-negative Lasso program that incorporates both a set of hypothesized reaction stoichiometries and a set of hypothesized reaction rate laws. Estimation error bounds are established and an iterative technique for finding a proper value of the Lasso tuning parameter for model selection and parameter identification is also developed. The technique was tested via simulation on a continuous stirred tank reactor process demonstrating recovery of stoichiometry, reaction rate laws and near recovery of model parameters using noisy reactor species concentration measurements. This was demonstrated even for a case that was previously described to be structurally unidentifiable. The same was also demonstrated for the case when a subset of reactor species concentration measurements is only available.