Abstract:

In order to improve the generalization ability of ensemble algorithms and the objectivity of the generating process of individual networks, an algorithm of heterogeneous neural network ensemble is proposed based on a new evolutionary programming. In this algorithm, several heterogeneous optimal neural networks are generated based on an improved evolutionary programming and are further integrated to obtain a solution. Simulated results indicate that, as compared with the traditional ensemble algorithms with fixed network structure and low individual precision, the proposed algorithm is of stronger generalization ability and fewer random elements.

Key words: evolutionary programming, neural network ensemble, heterogeneous neural network, Bootstrap sampling, generalization ability