A dynamically evolving learning network for intelligent control

Abstract

A variation on the adaptive learning network (ALN), which is used for dynamic system identification is discussed. The dynamically evolving ALN (DEALN) is self-organizing and operates online to generate a model of a dynamic plant. The network evolves the necessary structure and parameter values to mimic and predict the plant to within a specified tolerance. An intelligent controller can use the DEALN to simulate the plant, perform diagnoses, and plan coarse and fine control strategies. A high-level intelligent planner can also generate and program lower-level control laws to be implemented by the network, much as a human automates a skill. Results of an initial implementation which indicate that an online self-structuring learning network can be developed are presented. >