基于分层遗传算法的BP神经网络学习算法

针对BP算法局部搜索能力强，而分层遗传算法全局搜索优势突出的特点，结合二者优势构造了一种分层遗传算法与BP算法相结合的前馈神经网络学习算法．将分层遗传算法引入到前馈神经网络权值和阈值的早期训练中，再用BP算法对前期训练所得性能较优的网络权值、阈值进行二次训练得到最终结果．仿真结果表明，该混合学习算法能够较快地收敛到全局最优解，优于BP算法、分层遗传算法，具有一定的实用价值．

Learning Algorithm of BP Neural Network Based on Hierarchic Genetic Algorithm

Combining the good local searching ability of BP algorithm and the strong global searching capacity of hierar- chic genetic algorithm, the learning algorithm of feed-forward neural network is constructed. The feed-forward neural network＇s weights and thresholds are firstly trained by the hierarchic genetic algorithm. Then we get the final result by using I3P algorithm to train the weights and thresholds of the best network, which is obtained in early training. Simulating tests show that the mixed learning algorithm can quickly converge to the global optimal solution. It is superior to BP algorithm and the hierarchic genetic algorithm, and, of practical value.