灌浆压力的非线性建模预测

Nonlinear Model Predictive Pressure for Grouting System

实际灌浆压力控制过程中,由于灌浆液的密度、粘度和地层等因素的影响,使得灌浆压力的变化具有不确定性、时变性和非线性特征。为了辨识、预测灌浆系统压力,提出了一种基于神经网络的多传感器数据融合技术。通过对灌浆工艺与机理分析得到该BP神经网络输入变量。该方法首先利用灌浆过程中采集的数据离线训练BP神经网络,获得一收敛的神经网络模型,然后用此神经网络模型实时预测所灌地层的灌浆压力。最后实验仿真结果表明,BP神经网络预测模型能够运用到灌浆系统中,模型的最大预测误差不超过15%,平均均方根误差仅为0.186。

In real grouting project, the changes of grouting pressure are uncertain, time-varied and nonlinear because the influences of grout cement ratio, viscosity and stratum, etc. The BP neural model based on multi-sensor data was proposed in order to predict of the nonlinear property of grouting pressure. The input variable of neural net was got through analyzing the grouting scheme and mechanism. The BP neural model was trained by off-line datum, and the grouting pressure was predicted by on-line data. At last, the simulation result proves that the predictive BP model can be applied to the real grouting system. The maximal error of BP model is less than 15%, and the mean square error is 0.186.