中文摘要：

针对BP算法的缺陷以及标准遗传算法优化BP神经网络权值和阈值的不足，为了提高算法的全局搜索能力，提出了采用自适应动态调整权值和阈值区间的多子代遗传算法的BP神经网络算法。上述算法由权值和阈值构成的父代种群交叉产生多于父代个体数量的多子代种群，根据当前多子代种群个体适应度值的变化而自适应调整搜索区间，并通过种群内部竞争操作保留具有更高适应度值的优秀个体，从而使整个神经网络具有较小的误差和较快的收敛速度。仿真发现，与采用标准遗传算法的BP神经网络算法相比，采用多子代遗传算法的BP神经网络算法具有更高的学习精度和更快的收敛速度。仿真结果证明，自适应动态调整权值和阈值区间的多子代遗传算法对BP神经网络的优化优于基本遗传算法。

英文摘要：

Considering that the shortcomings of BP algorithm and the deficiency of using simple genetic algorithm （GA） to optimize the weights and thresholds of BP neural network, in order to improve the global search ability of al- gorithm, an optimized BP algorithm based on multi-child genetic algorithm （MCGA） with adaptive dynamic adjust- ment of the interval of weights and thresholds is proposed. The search interval is adjusted according to the changes of fitness of current multi- child population which are generated by the parent population consisting of weights and thresholds. Excellent individuals with higher fitness value are retained in the process of population internal competi- tion. Thus, entire neural network has smaller error and faster convergence speed. It is found from simulation that the BP algorithm based on MCGA has higher learning precision and faster convergence rate than the BP algorithm based on GA. The results show that using the MCGA with adaptive dynamic adjustment of the interval of weights and thresholds to optimize BP network is better than using GA to optimize BP network.