论文部分内容阅读
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。
Aiming at the disadvantage that the BP neural network itself has a slow convergence speed and easily falls into a local minimum point, this paper uses the LWPSO as the preprocessor to optimize the weights and thresholds of the BP network. Using the measured data, The LWPSO-BP surface subsidence coefficient prediction model is established and compared with the predicted results of the ordinary BP model. The results show that the LWPSO-BP neural network not only has fast training speed but also significantly improves the prediction accuracy. The model has certain value for the selection of the surface subsidence coefficient.