Abstract:
An arrangement for controlling a system (S) according to the deviation (ERR) between the value measured on the system (VR) and the value (VS) estimated by means of a model of the controlled system (S) of at least one control parameter. The arrangement comprises a neural network (12), which generates the estimation (VS) of the control parameter according to a set of characteristic parameters (I, Q, T) of the controlled system (S) and of respective configuration parameters (W, B). The neural network (12) has associated thereto a training module (11), which can train said neural network (12) by modifying said configuration parameters (W, B) according to a set of updating data (I t , T t , Q t , V t ). An acquisition module (21, 221) acquires the actual value, as measured on the controlled system (S), of a set of sensing parameters comprising at least one from among said control parameter (VR) and said characteristic parameters (I, Q, T) of the controlled system. A variation module (22) is sensitive to the variation of said control parameter (V k , a ; V k , m ) and generates an update-enable signal (EU) when the control parameter falls outside a pre-set tolerance range (228). The acquisition module (21, 221) being sensitive to said update-enable signal (EU) for transferring to the training module (11), as updating-data set, said set of sensing parameters (V t , Q t , I t , T t ). A preferential application is for the control of fuel-cell stacks.