The Expert Governed Numerical OPtimisation concept (EGNOP) has been developed to address problems with non-linear numerical optimisation software. Using a balanced combination of procedural and expert system programming languages, a totally new approach for numerical optimisation has been created. Expert system technology is used to govern numerical optimisation with as tasks to changing to other optimisation methods and tuning the operational parameters to ensuring a proper convergence. Much knowledge on using numerical optimisation methods is captured in the system. Convergence of an optimisation run is closely monitored by the system and immediate actions can be taken to improve the convergence. The system can choose the best method for the specific problem at any stage in the optimisation process. As a result, the performance of the optimisation process is increased, because it is closely monitored and changes are effectuated immediately by the system. An application dependent layer can be developed to create a communication between the user, the analysis routines, and the numerical optimisation programs. This layer can perform an initial evaluation of the obtained results for the user and may indicate a new optimisation run to be performed.
The total layout of Expert Governed Numerical Optimisation (EGNOP) can be seen as a sphere, consisting of several shells (see figure below).
Core | The core is a library of elementary numerical routines that can be used by the optimisation methods, by the analysis routines, etc. These routines are not the original optimisation routines, but elementary numerical operations, such as determination of a search direction and convergence analysis. Usually these are rather short routines, which pass on the results of their numerical operations to other routines or to the inner shell. Within the routines, almost no decision making takes place. |
Inner Shell | The inner shell (or inner shells, as they can be subdivided into several layers) contains the application independent control structure for the optimisation process and co-ordinates the optimisation process. The inner shell controls the convergence of the optimisation and can intervene in the optimisation by changing the values of the operational parameters, switching from one optimisation method to another, choosing another line minimisation method, etc. All these changes can be made either temporarily or permanent, and are imposed on the optimisation process immediately whenever necessary. |
Outer Shell | The outer shell performs an application dependent interpretation and evaluation of the obtained results. Actively using available domain knowledge, decisions can be made on the number of design variables and constraints, the starting values of the design variables, etc. |
The principle is that the user interacts with the optimisation program via the outer shell. This shell prepares the optimisation and passes on the control to the inner shell, that in turn invokes numerical routines from the core whenever suitable. The inner shell returns control to the outer shell when a solution has been found, or the optimisation got lost and maybe application dependent action from the outer shell is needed. It is in particular the inner shell - core combination that offers the new view on optimisation techniques. The optimisation methods now tune "themselves" through the inner shell, without interference from the user. Furthermore, EGNOP has been designed modularly, and as such can be extended easily for addition of new optimisation methods and/or line minimisation methods.
EGNOP can solve a good part of the existing problems. Using a balanced combination of procedural and expert system programming languages, a totally new approach for numerical optimisation has been created.
The inner shell governs the numerical optimisation in an application-independent manner, making the optimisation process more robust and saving computational time. This entails: