Complexity-penalized Model Selection for Feedwater Inferential Measurements in Nuclear Power Plants

Inferential sensing is a method which can be used to evaluate the parameters of a physical system based on a set of measurements related to those parameters. The most common method of inferential sensing uses mathematical models to infer the parameter value from correlated sensor values. However, inferential sensing is an ill-posed problem because of non-uniqueness and instability of the solution. This research shows that complexity-penalized model selection in the form of the minimum description length principle can be used to produce a unique solution and regularization – to stabilize the solution. The important example of monitoring the nuclear power plant feedwater flow rate is given using data from Florida Power Corporation's Crystal River Nuclear Power Plant.