Abstract: |
When developing an artificial neural net model of a system, the most efficient way to obtain training and test data is often to generate a large set of random inputs and run them through the model. But that is not the only way to do it. We demonstrate the use of genetic algorithm-generated data as a source of input-output pairs for training an artificial neural network. If the genetic algorithm and neural network are being developed together – for example, to provide system identification in support of a control system – this data is readily available and performs as well as a random search of the state space. |
