动量化GA-BP算法的全局极小值搜索

针对传统BP神经网络算法存在收敛速度慢、易于陷入局部极小的缺点,提出了动量化GA-BP算法,即对传统BP算法引入动量因子,以提高网络的收敛速度,并将其与BP神经网络算法结合起来,充分利用遗传算法的全局搜索性能进行"粗"搜索.当搜索到全局最优点的附近时,再采用BP算法进行局部搜索,这样避免陷入局部极小,从而得到了全局最优解.煤矿安全等级预测的实验表明:动量化GA-BP算法能快速、精确地收敛,最终得到问题的全局极小值,取得了满意的预测结果.

Global extreme little value search of momentum GA-BP algorithm

Directed at overcoming the disadvantage of slow convergence rate and local extreme little value of BP neural network, this paper proposes the momentum GA-BP algorithm which involves addition of momentum faetor to traditional BP nearal network in order to ilnprove convergence rate, the combination of genetic algorithm with BP neural network algorithm, and the use of widely capability of global searching of genetic algorithm to search generally. When getting closer to the optimized position, BP algorithm is adopted to local seareh to avoid local extreme little walue and obtain global optional solution. The method proposed in this paper is used to precise coal mine satety rank. The resuhs indicate that the algorithm offers fast and accurate convergence and leads to global extreme little solution, and yields satisfactory foreeast,