人工神经网络在钢渣稳定土强度预测中的应用研究  

Research on Application of Artificial Neural Network in Prediction of Compressive Strength of Steel Slag Stabilized Soil

作　　者：李新明[1] 乐金朝[2] LI Xinming YUE Jinchao(School of Civil Engineering and Architecture, Zhongyuan University of Technology, Zhengzhou 450007, China School of Water Conservancy ＆ Environmental Engineering, Zhengzhou University, Zhengzhou 450003, China)

机构地区：[1]中原工学院建筑工程学院,郑州450007 [2]郑州大学水利与环境学院,郑州45003

出　　处：《路基工程》2017年第1期69-72,共4页Subgrade Engineering

摘　　要：在分析钢渣土强度影响因素基础上,选取钢渣龄期、钢渣细度、钢渣掺量3种主要因素作为人工神经网络的输入值,钢渣土7天无侧限抗压强度作为输出值,建立了钢渣土强度预测的BP网络模型。研究结果表明:训练BP神经网络时,17组自变量数据中无侧限抗压强度的网络拟合值与实测值基本重合,误差为-4.054%~3.214%。BP网络方法应用于钢渣土强度的预测方面具有较高的精度,预测与实测结果最大相差为0.02 MPa,最大误差为5.556%,可见,基于3参数的BP神经网络模型在钢渣稳定土新型路床材料7天无侧限抗压强度中的应用是可行的,可以满足工程应用需求。The BP network model is established for prediction of the strength of steel slag stabilized soil by using 3 factors of the steel slag such as ageing period, fineness and mixing amount as input parameters and 7 days unconfined compressive strength of steel slag as output parameter after the analysis of the influencing factors of the steel slag stabilized soil＇s strength. The results show that during cultivating the BP neural network, of the network fitting value of unconfined compressive strength of 17 groups of independent variables basically coincides with the measured value with relalive error -4. 054 % to 3. 214 %. the BP network method has higher accuracy when used to predict the strength of the steel slag, the maximum difference is 0. 02 MPa between the predicted result and the measured result with maximum error 5. 556 %. It is thus clear that application of the BP neural network model based on 3 parameters for predicting 7 days unconfined compressive strength of steel slag stabilized soil roadbed is feasible, which may meet the needs of engineering application.

关 键 词：钢渣稳定土 神经网络 预测 抗压强度 

分 类 号：TU449[建筑科学—岩土工程]