基于遗传算法的BP神经网络在轻质路基沉降预测中的应用

Application of BP neural network based on genetic algorithm in settlement prediction of light subgrade

为更好地掌握轻质路基施工过程中的沉降变形情况,选取宁芜保通线部分轻质路基沉降监测数据,在BP(back propagation)神经网络模型的基础上,采用遗传算法对其进行优化,并将优化后的模型应用于轻质路基沉降预测。结果表明:遗传算法优化的BP神经网络在全局搜索能力和收敛能力方面具有明显优势;在轻质路基沉降预测任务中,多数预测结果的相对误差集中在更低的范围内,监测点1和监测点2预测结果的模型评价指标MAE、RMSE、MAPE分别为0.017 mm、0.021 mm、0.679%和0.013 mm、0.016 mm、1.395%,预测结果拟合程度高,误差小,模型泛化能力强。因此,遗传算法优化的BP神经网络的沉降预测模型具有可靠的预测效果与预测精度,在实际工程中可行性较高,可作为轻质路基沉降预测和预警的一种辅助手段。

In order to better understand the settlement and deformation of light subgrade construction,a model based on BP neural network was established by selecting monitoring data of a section of the light subgrade settlement from Ning-Wu-Bao-Tong line and optimized by using genetic algorithm.Then the optimized model was used in the settlement prediction of light subgrade.The results showed that the BP neural network optimized by the genetic algorithm had obvious advantages in terms of global search ability and convergence ability.The majority of prediction results had a relative error within a lower range of errors.The evaluation indexes of MAE,RMSE and MAPE were 0.017 mm,0.021 mm and 0.679%respectively for monitoring point 1,and 0.013 mm,0.016 mm and 1.395%respectively for monitoring point 2.The prediction model had a high fitting degree,low error,and strong generalization ability.Therefore,the settlement prediction model optimized by genetic algorithm has reliable prediction effectiveness and accuracy,and is highly feasible in practical engineering.It can be used as auxiliary means of light subgrade settlement prediction and early warning.