Simultaneous Model Selection, Robust Data Reconciliation and Outlier Detection with Swarm Intelligence in a Thermal Reactor Power calculation

Data Reconciliation (DR) and Gross Errors Detection (GED) are techniques of increasing interest in Nuclear Power Plants and used in order to keep Mass and Energy balance into account. These Techniques have been extensively studied in Chemical and Petrochemical Industry due to its benefits, which include closing the mass and energy balance and the yield of promising financial results. Many techniques were developed to solve Data Reconciliation and Outlier Detection, some of them use, for example, Quadratic Programming, Lagrange Multipliers, Mixed-Integer NonLinear Programming and others use Evolutionary Algorithms like Genetic Algorithms (GA) and Particle Swarm Optimization (PSO). Nowadays, Robust Statistics is also increasing in interest and it is being used in order to surpass some methods limitation, e.g., assuming that the errors are Normally Distributed, which does not always reflects real problems situation. In this paper we present a novel method to perform simultaneously: (a) the tuning of the Hampel’s Three Part Redescending Estimator (HTPRE) constants; (b) the Robust Data Reconciliation and (c) the Gross Error Detection. The automatic tuning procedure is based on the minimization of the Robust Akaike Criteria and the Particle Swarm Algorithm is used as a global optimization method. Simulations were made considering a nonlinear process commonly used as a benchmark by several authors and also in calculating the Thermal Reactor Power based on a simplified example. The results show the potential use of the technique even in an on-line Process to solve Data Reconciliation and Gross Error Detection problem and do not need a separate procedure to tune first redescending estimator and later perform the DR and GED method.

► We present a Model Selection, Robust Data Reconciliation and Outlier Detection method.
► The novel method directly minimizes the Robust Akaike Information Criteria.
► We use Hampel’s redescending estimator and a slight modified objective function.
► We obtain a good performance using the Particle Swarm Optimization Algorithm.
► Simulations are made including a simplified Thermal Reactor Power calculation.