Cloud-based identification of an evolving system with supervisory mechanisms

The paper deals with identification of a cloud based evolving system. The antecedent part of a fuzzy rule-based system is defined by clouds and density distribution as proposed by Angelov and Yager [1], [2]. But in the current paper Mahalanobis distance is used rather than the Euclidean one when calculating the density. The idea behind is that the shape of the clouds should be reflected in the density calculation. Covariance matrix of the elements in the clouds is needed for this purpose and it is obtained by a recursive algorithm. An important part of the paper is devoted to supervisory mechanisms that enable higher robustness of the identification. The use of buffers of data are promoted that enable balanced use of batch and recursive identification. This part is still work in progress. The proposed algorithms are illustrated on a simulated pH-neutralisation process.