An Ontogenic Polynomial Neural Network/Expert System Modeling Evnironment for Chaotic Avionics Systems Behavior Prediction

Ontogenic neural networks, those which automatically discover their topology during synthesis, offer substantial advantages over traditional machine learning techniques. They also have the potential to offer many benefits in several areas of avionics research. Among potential application areas is the ability to predict or forecast the behavior of a seemingly random or chaotic system. This proposal presents an existing ontogenic neural network paradigm - polynomial networks - and discusses both substantial enhancements to its learning algorithm and the development of an evolutionary modeling environment. This will be achieved by implementing system modeling strategies in a production-rule environment using the CLIPS expert system tool. This proposal also discusses a feasibility demonstration for a specific avionics application to clearly demonstrate the utility of the ontogenic paradigm. During Phase I, AbTech will design major enhancements to the existing ontogenic neural network paradigm, implement and deliver a substantial subset of these enhancements, and demonstrate the results of applying the resulting Phase I learning system to a well-defined avionics application. During Phase II, AbTech will implement the full Phase I design and apply the resulting modeling environment to an operational solution of the chaotic avionics system behavior prediction application.