摘要
针对BP神经网络自身收敛速度慢、容易陷入局部极小点的缺点,引入粒子群优化算法,建立地表下沉系数的PSO-BP选取模型。利用粒子群算法反复优化BP网络的权值和阈值,将其作为BP网络的初始值,并将上覆岩层岩性、开采深厚比、松散层厚度、覆岩中坚硬岩层所占比例、是否为重复采动和顶板管理方法等主要影响因素作为网络输入,进行BP算法,直至网络达到训练指标。利用实测资料数据,建立PSO-BP预计模型,并同普通BP神经网络预计结果对比。结果表明:PSO-BP神经网络不仅训练速度快,而且预测精度明显提高,该模型对地表下沉系数选取具有一定的应用价值。
In view of the shortcomings of rate slow,easy to fall into the partial minimum point of the BPneural network,Particle Swarm Optimization is introduced,and the PSO-BP select model of the surface submersion coefficient is established.The article repeatedly uses Particle Swarm Optimization algorithm to optimize the BP network weights and thresholds,then they are taken as the BP network the initial value.Then the main factors including the overburden rock character,the ratio of mining depth and thickness,loose layer thickness,the proportion of hard rock in the overburden rock,whether to repeatedly to pick moves and roof management methods,etc.are taken as network input,and the BP algorithm is carried on until the network training to achieve targets.Useing the actual material data,the PSO-BP estimate models established,and contrasted with ordinary BP neural network estimate result.The result indicated： The PSO-BP neural network not only trains in a fast speed,but also forecasts precision distinct enhancement,and this model to select the surface submersion coefficient has certain application value.
出处
《测绘工程》
CSCD
2010年第6期57-60,共4页
Engineering of Surveying and Mapping
关键词
粒子群
BP神经网络
地表下沉系数
Particle Swarm Optimization
BP neural network
surface submersion coefficient
