BP神经网络在爆破振动中的研究与应用

为准确地预测爆破结果、减少爆破振动对建筑的损伤和保障工人的安全,利用具有处理非线性问题能力的BP神经网络预测爆破结果。选取合格的爆破结果作为网络模型的学习样本,经过一定次数的训练学习后通过神经网络的前馈特性确定各层阈值和误差,完成对BP神经网络的建立,发现预测结果与真实结果相比的误差在10%以内。再结合PAC算法、POS算法或者MATLAB软件等优化网络后甚至可以将误差控制在5%以内。通过建立BP神经网络预测可以减少爆破作业带来的危害,降低安全成本,指导爆破作业的施工。

Research and application of BP neural network to blasting vibration

In order to accurately predict the blasting results and reduce the damage of blasting vibration to the building and ensure the safety of workers, BP neural network with the ability to deal with nonlinear problems was used to forecast the blasting results. Qualified blasting results were selected as the learning samples of the network model, after a certain number of times of training and learning, the feed forward characteristics of the neural network were applied to determine the thresholds and errors of each layer, then the establishment of BP neural network was completed. Results indicated that the error between the prediction result and the real result was within 10%, which conformed to the engineering requirements. By combining PAC algorithm, POS algorithm or MATLAB to optimize the network, the error can even be controlled within 5%. The damage caused by blasting operation can be reduced by establishing BP neural network prediction, with safety cost reduced and guiding the construction of blasting operation.