Optimization of size of informions for designing nonlinear vibration Neuro-modellers of dams

In this paper a Multi-Layer Feed Forward Neural Network (MLFFNN) has been designed to be used in nonlinear vibration analysis of Koyna dam which is a concrete gravity dam, located inKoyna, India. The dam has extensively been studied by many authors in the past and there is quite a large amount of information and data available on its modeling, behavior and response. Thedesigned neural network which is called Neuro-modeller, is capable of dynamic analysis of thedam similar to an analysis software which is capable of receiving a seismic record through out thetime, as input, to perform dynamic analysis as output. The success of the Neuro-modeller ishighly dependent upon the data it receives at each analysis time step to start the next time step of analysis as well as its architecture. The size and content of the vector of input information, whichis called informion, should be optimized so that the Neuro-modeller can both simulate the response with desirable accuracy and also can be capable of generalization.

Keywords:

informion , neuro-modeller , dam , dynamic analysis , nonlinear response