EXPERT SYSTEM FOR IMPROVING LARGE FUSION CODE EFFICIENCY

THE EXPANDING UTILIZATION OF LARGE SCALE COMPUTATIONAL MODELS WITHIN THE MAGNETIC FUSION PROGRAM IS PLACING INCREASING STRESS ON THE PROGRAM'S COMPUTATIONAL RESOURCES. WE PROPOSE TO GREATLY IMPROVE THE EFFICIENCY WITH WHICH FUSION REACTOR DESIGN AND OTHER LARGE COMPUTER CODES MAY PRESENTLY BE OPERATED BY MEANS OF AN ARTIFICIAL INTELLIGENCE-BASED EXPERT SYSTEM. SPECIFICALLY, THIS EXPERT SYSTEM WILL IMPROVE THE EFFICIENCY WITH WHICH THE PARAMETER SPACE CHARACTERIZING COMPUTATIONAL MODELS CAN BE EXAMINED. IT WILL INTEGRATE EMPIRICAL KNOWLEDGE GAINED FROM PREVIOUS RUNS OF THE MODEL, ANALYTICAL KNOWLEDGE ABOUT SPECIAL PARAMETRIC REGIMES OF THE MODEL, AND HEURISTIC KNOWLEDGE SUPPLIED BY THE MODEL DESIGNER OR OPERATOR TO SUGGEST EFFICIENT WAYS OF EXAMINING THE PARAMETRIC DEPENDENCE OF THE MODEL. A SIDE EFFECT OF THE EXPERT SYSTEM WILL BE THE MAINTENANCE OF A DATA BASE OF INFORMATION PREVIOUSLY GAINED FROM THE MODEL WHICH WILL BE READILY ACCESSIBLE BY NONEXPERT RESEARCHERS. THIS IS EXPECTED TO NOT ONLY SAVE COMPUTER TIME, BUT ALSO THE VALUABLE TIME OF THE OPERATING SCIENTIST/ENGINEER. THE INITIAL APPLICATION OF THE SYSTEM WILL BE TO FUSION REACTOR DESIGN CODES WHICH REQUIRE EXTENSIVE, COMPUTATIONALLY EXPENSIVE PARAMETER SURVEYS. THE SUCCESSFUL APPLICATION OF THIS METHODOLOGY TO COMPLEX FUSION CODES CAN BE READILY EXTENDED TO OTHER GOVERNMENT PROGRAMS OR TO LARGE SCALE COMMERCIAL COMPUTING FACILITIES.