Optimal sensor location for distributed-sensor systems using multivariate regression

The performance of monitoring and control techniques for distributed-sensor systems is affected by the choice of the measurement sensor location. In this paper a methodology is suggested to solve the optimal sensor location problem. The suggested algorithm does not require any explicit knowledge of the plant model, and is based upon a sequential procedure selecting at each iteration the most informative measurement input, and updating the input and output spaces by subtracting the information explained by the computed regressor. The effectiveness of the proposed algorithm is assessed by means of two simulated case studies concerning the location of sensors in tubular reactors where product composition need to be estimated from temperature measurements.