An adaptive evolutionary modular neural network with intermodule connections

To approach the brain-like neural network and further improve the performance of the modular neural network (MNN), an adaptive evolutionary modular neural network with intermodule connections (EA-ICMNN) is proposed in this study. The EA-ICMNN is composed of a group of multilayer neural networks. Unlike traditional MNNs, in addition to the intramodule connections of subnetworks, intermodule connections are built for EA-ICMNN. All the parameters of the EA-ICMNN are learned by the improved Levenberg-Marquardt algorithm, and the optimal structure is adaptively determined by the improved mutation operator in the multiobjective optimization algorithm NSGAII. To verify the effectiveness of the proposed model, the EA-ICMNN is tested on several benchmark datasets and a practical prediction problem for biochemical oxygen demand in wastewater treatment process. The experimental results show that the proposed model has better generalization ability than other MNNs and that its structure is simplified by its sparse intermodule connections.