摘要
目前利用最优化算法中的Marquardt法改进BP神经网络正受到越来越多的人们的注意 ,但该方法的网络初始权值是随机选取。由于初始权值选取不当将对整个网络的性能产生严重影响 ,因此提出将遗传算法与Marquardt法结合 ,先利用遗传算法全局随机搜索寻优的特性来寻找网络最佳初始权值 ,再用Marquardt法使网络权系数稳定收敛 ,同时应用该方法对油气水多相流流型进行智能识别 ,结果表明该方法能有效学习模式样本 ,学习稳定 ,推广能力强 ,适合于在流型识别等神经网络为中小规模的场合下应用。
Nowadays it has focused more and more people's attention on using Marquardt algorithm to modify BP neural network, but the initial weights of the modified neural network is selected randomly. As inappropriate selection of the initial weights will have serious influence on the performance of the whole network, genetic algorithm and Marquardt algorithm are combined, firstly the characteristics of global and random search for optimum is used to search for the optimal initial weights of the network, then Marquardt algorithm is used to make weights of network converge to optimum. This method is applied to intelligent identification of flow regime of oil-gas-water multiphase flow. Results show that this method can study model effectively and have merits such as stable training and good extension ability, therefore this method is especially suitable for the fields such as flow regime identification whose neural network is relatively moderate.
出处
《化学工程》
EI
CAS
CSCD
北大核心
2001年第1期30-32,36,共4页
Chemical Engineering(China)
基金
国家自然科学基金重点项目资助! (批准号 :5 9995 46 0 )
