Multi-agent methodology for distributed and cooperative supervisory estimation subject to unreliable information

In this work, a novel multi-agent framework for cooperative supervisory estimation of linear time-invariant systems is proposed. This framework is developed based on the notion of subobservers and a discrete-event system (DES) supervisory control and is applicable to large-scale systems. We introduce a group of subobservers where each subobserver is estimating certain states that are conditioned on a given input, output and state information. The cooperation among the subobservers is managed by a DES supervisor. The supervisor makes decisions regarding the selection and configuration of a set of subobservers to successfully estimate all the system states, while the feasibility of the overall integrated cooperative subobservers is verified. When certain anomalies (faults) are present in the system, or the sensors and subobservers become unreliable, the supervisor reconfigures the set of selected subobservers so that the impacts of anomalies on the estimation performance are minimised to the extent that is possible. The application and capabilities of our proposed methodology in a practical industrial process is demonstrated through numerical simulations.