基于PSO-ELM的建筑物爆破震动速度预测

 针对影响爆破震动速度因素之间复杂的非线性关系，利用粒子群算法（PSO）的全局搜索最优解原理和极限学习机（ELM）处理非线性关系能力，建立了爆破震动速度预测的PSO-ELM 模型。以某地区爆破震动实测数据为例，选取总药量、最大段药量、爆破点与监测点距离、建筑物所在地面震动速度和测点到地面的高度等5 个因素为输入变量，以建筑物震动速度为输出变量。结果表明，PSO-ELM 模型训练值与预测值，测试值与预测值的均方误差分别为0.18 和2.56，平均相对误差控制在6%以内，显示出该模型具有良好的训练精度和泛化能力。对比传统ELM 模型，PSO-ELM 模型不但提高了精度和泛化能力，而且降低了训练样本数和隐含层节点数变化对训练结果的影响，提高了模型的拟合能力，在类似预测工程中有一定的推广价值。

Prediction for Building Vibration Velocity Caused by Blasting Based on PSO-ELM

Aimed at the complicated nonlinear relation between the factors influencing the blasting vibration velocity, a blasting vibration velocity prediction model is built by using the particle swarm optimization (PSO) global search optimal solution principle and extreme learning machine (ELM) ability which can deal with the nonlinear relationship. Taking blasting vibration measured data in a certain area as an example, the total dose, the explosive charge, the distance between shot and monitoring point, the ground vibration velocity and the height of the monitoring point are selected as input variables and the building vibration velocity is chosen as the output variable. The result shows that the mean square errors between training value and predicted value and between test value and predicted value are 0.18 and 2.56, respectively, and the average relative error is controlled within 6%. It is proved that the model has good precision and generalization ability. Compared with the traditional ELM model, the PSO-ELM model not only improves the accuracy and generalization ability, but also reduces the influence on the result of training when the numbers of training samples and the hidden layer nodes change, thus the fitting ability of the model is improved. This model has great a promotional value in similar predictive engineering.