| 초록 |
When the system under consideration is very difficult to identify or complex tomodel mathematically, the neural network modeling is very attractive if the input/outputdata can be easily obtained. Previous research into the design of artificial neuralnetworks for process modeling has largely ignored the existing knowledge about thetask at hand. To take the existing knowledge and the simplicity of artificial neuralnetworks, it is proposed to make use of a neural network model of the process withinputs derived from operation data and existing knowledges. A complex reactingsystem was studied and applied the proposed hybrid neural network model. Thesystem dynamics represented by catalyst activity, can be estimated from the firstprinciples model and can be a input to the neural model. The hybrid model showsbetter prediction capability than the first principles model.
|
