基于遗传算法的进化神经网络

The Evolving Artificial Neural Network Based on Genetic Algorithm

提出一种基于遗传算法的多层前向神经网络的自动化设计方法(genetic multiplayer feedforward neural network, GMFNN),用以同时完成对网络结构空间和权值空间的搜索。该算法利用双种群权值优化、结构进化自适应变异率等方法来加快算法的收敛速度,改善解的性能。仿真结果显示本文提出的算法能够有效抑制遗传算法初期收敛的发生,有效地提高多层前向神经网络的收敛精度,并可获得更为简洁的网络结构。

This paper presents a genetic multiplayer feedforward neural network in order to evolve neural network architectures and connection weights simultaneously. In the method, the learning efficiency and the convergence are improved greatly by bi-population weights learning and self-adaptive mutate ratio of structure learning. The simulation results show that the premature convergence in genetic algorithm is restrained effectively and the learning efficiency and the convergent precision for the weights of the multi-layer forward neural networks are improved greatly. These results also show that the proposed method in this paper can produce very compact artificial neural networks in comparison with other algorithms.