Accurate inference of variables using artificial neural networks

Accurate measurements are essential for feature extraction and accurate control of systems. Often, however, accurate measurements cannot be performed directly due to the inaccessibility in systems to place sensors or non-ideal sensor characteristics. Inference measurements (also called soft sensors) enables estimates of variables in a system from directly measurable variables. This paper shows which directly accessible variables should be used for accurate inference by exploring various systems. Neural networks have universal mapping capabilities, and are used to perform inference measurements on nonlinear dynamic systems. The results show which variables should be used in order to perform accurate inference of unknown variables.