地表下沉系数选取的LWPSO-BP方法研究

LWPSO-BP algorithm for calculation of surface subsidence coefficient

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,本文以线性下降惯性权重粒子群优化算法(LWPSO)为前处理器,优化BP网络的权值和阈值,利用实测资料数据,建立LWPSO-BP的地表下沉系数预计模型,并同普通BP模型预计结果对比。结果表明:LWPSO-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 disadvantages of BP neural network： low convergence rates, easily falling into the partial minimum point and so on, this article saw Particle Swarm Optimization based on linear decrease inertia weight（LWPSO） algorithm as a former proces- sor, optimized weights and thresholds of BP network. It used the actual material data, established the LWPSO-BP estimate model, and contrasted with ordinary BP model estimate result. The resuh indicated that the LWPSO-BP neural network could not only train in a fast speed, but also forecast in a distinctly enhanced precision, which would have certain application value in selecting the surface submersion coefficient.