Integrated Structure and Parameter Selection for Eng-genes Neural Models

摘要

A new approach to the construction and optimisation of 'eng-genes' grey-box neural networks is investigated. A forward selection algorithm is used to optimise both the network weights and biases and the parameters of the system-derived activation functions. The algorithm is used for both conventional neural network and eng-genes modelling of a simulated Continuously Stirred Tank Reactor. The resulting eng-genes networks demonstrate superior simulation performance and transparency over a range of network sizes.