基于机器学习的钢纤维喷射混凝土强度预测研究

Strength Prediction of Steel Fiber Shotcrete Based on Machine Learning

为实现钢纤维喷射混凝土强度的精确预测,分析了钢纤维喷射混凝土配合比与抗压强度之间的关系,并构建了7 d抗压强度预测模型.基于32组配合比及抗压强度数据,利用主成分分析法(PCA)对8项影响因素指标(水、水泥、硅粉、细骨料、粗骨料、减水剂、速凝剂、钢纤维)进行降维处理.基于XGBoost模型并结合麻雀搜索算法(SSA)、贝叶斯优化算法(BO)、模拟退火优化算法(SAA)和粒子群优化算法(PSO)建立钢纤维喷射混凝土7 d抗压强度预测模型.研究结果表明:BO-XGBoost模型训练集与测试集的MSE分别为0.30、0.42,与7 d抗压强度拟合时R^(2)=0.99703,BO-XGBoost的误差最小,拟合效果最好,为最优模型.将该模型应用于滇中引水隧洞工程,对比试验结果做误差分析,得到MSE=0.1297,拟合得到R^(2)=0.98772,得出结论:BO-XGBoost模型具有良好的预测效果.

To achieve accurate prediction of the strength of steel fiber shotcrete,the relationship between the mix proportion of steel fiber shotcrete and compressive strength was analyzed,and a 7-day compressive strength prediction model was constructed.Based on 32 sets of mix proportion and compressive strength data,principal component analysis(PCA)was used to reduce the dimensionality of eight influencing factors(water,cement,silica fume,fine aggregate,coarse aggregate,water reducer,quick-setting agent,steel fiber).Based on the XGBoost model and combined with the Sparrow search algorithm(SSA),Bayesian optimization algorithm(BO),simulated annealing optimization algorithm(SAA),and particle swarm optimization algorithm(PSO),a 7-day compressive strength prediction model for steel fiber shotcrete was established.The research results showed that the BO-XGBoost model had the smallest mean squared error(MSE)of 0.30 for the training set and 0.42 for the test set,and the coefficient of determination(R^(2))for fitting the 7-day compressive strength was 0.99703.The BO-XGBoost model had the smallest error and the best fitting effect,making it the optimal model.Applying this model to the construction of the Central Yunnan diversion tunnel,error analysis was conducted by comparing the experimental results,resulting in an MSE of 0.1297 and an R^(2) of 0.98772,indicating that the BO-XGBoost model had good prediction performance.