Identification of MIMO-ARMAX models for glycemia and sodium ions tests, through Particle Swarm Optimization

In this article, the problem of multivariable data autoregressive identification is approached, in view of prediction. Such data are especially provided by natural phenomena with geographical distribution, but could also be acquired from concentrated systems with several outputs, like medical records of one or more patients. Presumably, data are generated by the same source and exhibit more or less correlations to each other. Instead of processing each data set as an independent time series, the whole data block can be seen as the reaction of some open MIMO system to unknown stimuli. Therefore, modeling and forecasting of system behavior can rely on multi-dimensional ARMAX models. Optimal structural indices of MIMO-ARMAX model are found by means of an evolutionary strategy: particle swarm optimization.