A Perfect Integration of Neural Networks and Evolutionary Algorithms
Evolutionary computation techniques need to interact with a fitness function or an objective function for the selection to be made properly. In cases objective functions are not well defined, evolutionary algorithms may not be able to perform properly. In this paper, we propose to use a well-trained neural network model to provide estimates of objective values at points that we have not experienced to support the selection process of the evolutionary algorithm. An example of gasoline blending task is used to demonstrate that such an integrated system functions as a powerful tool for product design.
KeywordsEvolutionary Algorithm Input Ingredient Evolutionary Computation Technique Octane Rate Neural Network Module
Unable to display preview. Download preview PDF.
- D. B. Fogel, D.B.: System Identification Through Simulated Evolution: A Machine Learning Approach to Modeling. Ginn Press, 1991.Google Scholar
- Yip, P. P. C. and Pao, Y.H.: IEEE Trans. SMC 9, 1383 (1994).Google Scholar
- Aarts, E. and Korst, J.: Simulated Annealing and Boltzmann Machine, John Wiley & Sons, 1989.Google Scholar
- Hochhauser, A.M., Benson, J.D., Burns, V., Gorse, R.A., Koehl, W.J., Painter, L.J., Rppon, B.H., Reuter, R.M., and Rutherford, J.A.: SAE Transactions JJQQ, 748 (1991).Google Scholar