A Simple, Powerful, Binning Approach to MPC Model Gain Conditioning

There is power in the plant testing/statistical regression approach to building dynamic models used in model-based controllers. Using regression to derive a raw APC model will maximize its degrees of freedom, and the result is a model that contains small or non-existent degrees of freedom. Control issues can arise when optimization techniques use these small degrees of freedom to calculate targets for plant operation. Techniques like RGA, SVD, and pivoting are helpful for analyzing control interactions, and conditioning issues can be fixed relatively easily in small models. However, these techniques become difficult and tedious for larger, more complex models. This paper describes a novel, non-iterative binning technique for quickly solving 2x2 conditioning issues for any size model, while guaranteeing gain percentage changes below a certain threshold. Higher order interactions are also discussed.