USING NEURAL-NET COMPUTING TO FORMULATE REAL-TIME CONTROL STRATEGIES

In this paper, we propose a new neural-net approach to learning appropriate control actions. It is not based on the more commonly accepted approach of learning a system emulator and a control-action generator with supervised learning implemented through minimization of errors. Instead, the net observes and records, and adjusts local activation and attention to reflect the frequency of occurrences. Such a net can track the time variational characteristics of physical plants and is compatible with the learning of real-time fuzzy controls.