基于机器学习的混凝土抗压强度预测：模型筛选与性能比较

doi:10.16552/j.cnki.issn1001-1625.2026.0140

摘要/Abstract

摘要：

为构建精准的混凝土抗压强度预测模型并对比不同机器学习模型的预测效果，本研究基于1 030组包含水胶比、龄期、骨胶比等8项特征的数据，对17种经典机器学习算法进行模型训练，并从中筛选出6种精度较高的模型。进一步将这6种模型分别与贝叶斯优化（BO）、麻雀搜索算法（SSA）、模拟退火算法（SA）、正余弦优化算法（SCA）及粒子群优化算法（PSO）5种超参数优化算法相结合，优化得到30个融合模型。通过对优化后的模型进行筛选，选取精度较高的模型进行外部验证与沙普利加法解释（SHAP）可解释分析。结果表明：经SSA优化的最小二乘提升模型（LSBoost-SSA）、经BO优化的极端梯度提升模型（XGBoost-BO），以及经SSA优化的轻量级梯度提升树模型（LGBM-SSA）性能表现最佳；在外部验证中，XGBoost-BO表现最优，其强度预测误差可控制在5 MPa以内；SHAP分析显示，水胶比、龄期和水泥用量是影响混凝土抗压强度的关键因素。

关键词: 混凝土, 抗压强度, 机器学习, 超参数优化, 预测模型

Abstract:

To develop an accurate prediction model for concrete compressive strength and compare the prediction performance of different machine learning methods， this study trained 17 machine learning models using a dataset consisting of 1 030 samples， which included eight features such as water-to-binder ratio， age， and aggregate-to-binder ratio. Six models with higher accuracy were selected from these. Subsequently， each of these six models was combined with five hyperparameter optimization algorithms， namely Bayesian optimization （BO）， sparrow search algorithm （SSA）， simulated annealing （SA）， sine cosine algorithm （SCA）， and particle swarm optimization （PSO）， resulting in 30 optimized hybrid models. The models with superior accuracy were further selected for external validation and SHapley Additive exPlanations （SHAP） interpretability analysis. The results show that the least squares boosting model optimized by SSA （LSBoost-SSA）， the eXtreme Gradient Boosting model optimized by BO （XGBoost-BO）， and the light gradient boosting machine model optimized by SSA （LGBM-SSA） exhibit the best performance； XGBoost-BO achieves the optimal performance in external validation， capable of controlling the prediction error of concrete strength within 5 MPa； SHAP analysis reveals that water-binder ratio， curing age， and cement content are the core factors influencing concrete compressive strength.

Key words: concrete, compressive strength, machine learning, hyperparameter optimization, prediction model

中图分类号:

TU528

田沈华, 李北星, 肖祥, 周骏康. 基于机器学习的混凝土抗压强度预测：模型筛选与性能比较[J]. 硅酸盐通报, 2026, 45(8): 2711-2723.

TIAN Shenhua, LI Beixing, XIAO Xiang, ZHOU Junkang. Machine Learning-Based Compressive Strength Prediction of Concrete: Model Screening and Performance Comparison[J]. BULLETIN OF THE CHINESE CERAMIC SOCIETY, 2026, 45(8): 2711-2723.

图/表 11

参考文献 33

[1]	NGUYEN N H， ABELLÁN-GARCÍA J， LEE S， et al. Efficient estimating compressive strength of ultra-high performance concrete using XGBoost model［J］. Journal of Building Engineering， 2022， 52： 104302.
[2]	罗广彬，洪成雨，程志良，等. 基于BP和GA-BP神经网络的混凝土抗压强度预测研究［J］. 混凝土， 2023（3）： 37-41.
	LUO G B， HONG C Y， CHENG Z L， et al. Study on prediction of concrete compressive strength based on BP and GA-BP neural network［J］. Concrete， 2023（3）： 37-41 （in Chinese）.
[3]	周继发，曾晓辉，谢友均，等. 基于HEMNG模型的混凝土抗压强度预测［J］. 铁道科学与工程学报， 2025， 22（2）： 875-886.
	ZHOU J F， ZENG X H， XIE Y J， et al. Predicting concrete compressive strength based on HEMNG model［J］. Journal of Railway Science and Engineering， 2025， 22（2）： 875-886 （in Chinese）.
[4]	JUBORI D S A， NABILAH A B， SAFIEE N A， et al. Machine learning technique for the prediction of blended concrete compressive strength［J］. KSCE Journal of Civil Engineering， 2024， 28（2）： 817-835.
[5]	M A AABD ELATY. Compressive strength prediction of Portland cement concrete with age using a new model［J］. HBRC Journal， 2014， 10（2）： 145-155.
[6]	ROSS C A. Moisture and strain rate effects on concrete strength［J］. ACI Materials Journal， 1996， 93 （3）： 293-300.
[7]	曹斐，周彧，王春晓，等. 一种改进的支持向量回归的混凝土强度预测方法［J］. 硅酸盐通报， 2021， 40（1）： 90-97.
	CAO F， ZHOU Y， WANG C X， et al. An improved support vector regression method for concrete strength prediction［J］. Bulletin of the Chinese Ceramic Society， 2021， 40（1）： 90-97 （in Chinese）.
[8]	YANG S H， SUN J S， XU Z F. Prediction on compressive strength of recycled aggregate self-compacting concrete by machine learning method［J］. Journal of Building Engineering， 2024， 88： 109055.
[9]	刘聪林，李盛，崔晓宁，等. 融合XGBoost和SHAP的混凝土抗压强度预测分析模型［J］. 水利水电技术（中英文）， 2025， 56（2）： 246-258.
	LIU C L， LI S， CUI X N， et al. Prediction and analysis of concrete compressive strength based on XGBoost and SHAP［J］. Water Resources and Hydropower Engineering， 2025， 56（2）： 246-258 （in Chinese）.
[10]	ABD A M， ABD S M. Modelling the strength of lightweight foamed concrete using support vector machine （SVM）［J］. Case Studies in Construction Materials， 2017， 6： 8-15.
[11]	RAY S， HAQUE M， RAHMAN M M， et al. Experimental investigation and SVM-based prediction of compressive and splitting tensile strength of ceramic waste aggregate concrete［J］. Journal of King Saud University-Engineering Sciences， 2024， 36（2）： 112-121.
[12]	RAY S， RAHMAN M M， HAQUE M， et al. Performance evaluation of SVM and GBM in predicting compressive and splitting tensile strength of concrete prepared with ceramic waste and nylon fiber［J］. Journal of King Saud University-Engineering Sciences， 2023， 35（2）： 92-100.
[13]	JIANG Y M， LI H Y， ZHOU Y S. Compressive strength prediction of fly ash concrete using machine learning techniques［J］. Buildings， 2022， 12（5）： 690.
[14]	LIU K X， ZHANG L Y， WANG W G， et al. Development of compressive strength prediction platform for concrete materials based on machine learning techniques［J］. Journal of Building Engineering， 2023， 80： 107977.
[15]	NGUYEN H， VU T， VO T P， et al. Efficient machine learning models for prediction of concrete strengths［J］. Construction and Building Materials， 2021， 266： 120950.
[16]	PAN P F， YANG W W， ZHANG Y， et al. Optimized strength modelling of foamed concrete using principal component analysis featurized regressors［J］. Structures， 2023， 48： 1730-1745.
[17]	MIAO X， ZHU J X， ZHU W B， et al. Intelligent prediction of comprehensive mechanical properties of recycled aggregate concrete with issueary cementitious materials using hybrid machine learning algorithms［J］. Case Studies in Construction Materials， 2024， 21： e03708.
[18]	LY H B， NGUYEN M H， PHAM B T. Metaheuristic optimization of Levenberg-Marquardt-based artificial neural network using particle swarm optimization for prediction of foamed concrete compressive strength［J］. Neural Computing and Applications， 2021， 33（24）： 17331-17351.
[19]	何印章，苏浩然，郝宁远，等. 基于机器学习和SHAP模型的超高性能混凝土性能预测与分析框架［J］. 中国公路学报， 2025.
	HE Y Z， SU H R， HAO N Y， et al. A predictive and analytical framework for ultra-high performance concrete properties using machine learning and SHAP modeling［J］. China Journal of Highway and Transport， 2025 （in Chinese）.
[20]	DHORABE P T， CHANDAK M A， BAHORIA B V， et al. Computational prediction of concrete strength via microstructure image analysis： a hybrid machine learning framework［J］. AI in Civil Engineering， 2025， 4（1）： 26.
[21]	ZVIAZHYNSKI B， WHITE C， LEES J M， et al. Predicting concrete strength using machine learning and fresh-state measurements［J］. Data-Centric Engineering， 2025， 6： e37.
[22]	YUAN X Z， TIAN Y Z， AHMAD W， et al. Machine learning prediction models to evaluate the strength of recycled aggregate concrete［J］. Materials， 2022， 15（8）： 2823.
[23]	GU J R， WEI C， MA S L， et al. Prediction and in-depth analysis of precast concrete strength by machine learning［J］. Innovative Infrastructure Solutions， 2025， 10（3）： 122.
[24]	ABDENNAJI T S， TIPU R K， ALASSAF Y. Predicting compressive and tensile strength of concrete with different sand types using machine learning［J］. Ain Shams Engineering Journal， 2025， 16（8）： 103474.
[25]	DUA D， GRAFF C. UCI machine learning repository［DB/OL］. （2007-08-02）［2026-01-15］.
[26]	WANG H J， ZENG J， DAI R B， et al. Understanding rejection mechanisms of trace organic contaminants by polyamide membranes via data-knowledge codriven machine learning［J］. Environmental Science & Technology， 2024， 58（13）： 5878-5888.
[27]	RAISSI M， PERDIKARIS P， KARNIADAKIS G E. Physics-informed neural networks： a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations［J］. Journal of Computational Physics， 2019， 378： 686-707.
[28]	KIM J， MCCOURT M， YOU T， et al. Bayesian optimization with approximate set kernels［J］. Machine Learning， 2021， 110（5）： 857-879.
[29]	XUE J K， SHEN B. A novel swarm intelligence optimization approach： sparrow search algorithm［J］. Systems Science & Control Engineering， 2020， 8（1）： 22-34.
[30]	MIRJALILI S. SCA： a Sine Cosine Algorithm for solving optimization problems［J］. Knowledge-Based Systems， 2016， 96： 120-133.
[31]	STEINBRUNN M， MOERKOTTE G， KEMPER A. Heuristic and randomized optimization for the join ordering problem［J］. The VLDB Journal， 1997， 6（3）： 191-208.
[32]	ZHAN Z H， ZHANG J， LI Y， et al. Adaptive particle swarm optimization［J］. IEEE Transactions on Systems， Man， and Cybernetics， Part B （Cybernetics）， 2009， 39（6）： 1362-1381.
[33]	李越，姚占全，李昊，等. 矿物掺合料对混凝土抗压强度的影响及模型预测［J］. 排灌机械工程学报， 2025， 43（4）： 395-403.
	LI Y， YAO Z Q， LI H， et al. Influence of mineral admixtures on compressive strength of concrete and predictive modeling［J］. Journal of Drainage and Irrigation Machinery Engineering， 2025， 43（4）： 395-403 （in Chinese）.

Mix ID	Materials dosage/（kg·m^-3）
Mix ID	CE	FA	GGBS	RS	LCS	Water	PCA
F0S0	480	0	0	712	1 114	154	5.28
F20S0	384	96	0	712	1 114	154	5.35
F15S5	384	72	24	712	1 114	154	5.28
F12S8	384	57.6	38.4	712	1 114	154	5.28
F10S10	384	48	48	712	1 114	154	5.28
F30S0	336	144	0	712	1 114	154	5.28
F22.5S7.5	336	108	36	712	1 114	154	5.28
F18S12	336	86.4	57.6	712	1 114	154	5.28
F15S15	336	72	72	712	1 114	154	5.28

Mix ID	Materials dosage/（kg·m^-3）
Mix ID	CE	FA	GGBS	RS	LCS	Water	PCA
F0S0	480	0	0	712	1 114	154	5.28
F20S0	384	96	0	712	1 114	154	5.35
F15S5	384	72	24	712	1 114	154	5.28
F12S8	384	57.6	38.4	712	1 114	154	5.28
F10S10	384	48	48	712	1 114	154	5.28
F30S0	336	144	0	712	1 114	154	5.28
F22.5S7.5	336	108	36	712	1 114	154	5.28
F18S12	336	86.4	57.6	712	1 114	154	5.28
F15S15	336	72	72	712	1 114	154	5.28

Model	Evaluation metrics	Train	Test	Score	Overall score	Model	Evaluation metrics	Train	Test	Score	Overall score
LMR	MAE	7.030 1	9.058 3	4	13	DT	MAE	2.012 4	4.719 3	11	44
	MAPE	0.294 3	0.369 9	3			MAPE	0.068 8	0.180 7	11
	RMSE	10.009 9	11.835 0	3			RMSE	3.044 8	6.527 9	11
	R²	0.637 35	0.514 92	3			R²	0.966 45	0.852 42	11
LASSO	MAE	7.940 2	9.061 2	3	9	RF	MAE	2.722 3	4.451 5	12	48
	MAPE	0.294 5	0.370 2	2			MAPE	0.100 9	0.183 7	12
	RMSE	10.010 0	11.836 2	2			RMSE	3.749 1	5.900 0	12
	R²	0.637 34	0.514 83	2			R²	0.949 13	0.879 45	12
MRR	MAE	7.956 3	9.071 8	1	12	SVM	MAE	7.766 3	9.065 4	2	9
	MAPE	0.297 3	0.371 8	1			MAPE	0.282 2	0.356 4	5
	RMSE	10.017 4	11.827 2	5			RMSE	10.386 6	12.488 4	1
	R²	0.636 81	0.515 56	5			R²	0.609 54	0.459 88	1
LSBoost	MAE	0.854 6	3.563 3	16	62	XGBoost	MAE	1.987 9	3.719 9	15	62
	MAPE	0.030 6	0.139 63	16			MAPE	0.068 1	0.144 5	15
	RMSE	1.609 2	5.055 4	15			RMSE	2.850 9	5.049 2	16
	R²	0.990 63	0.912 46	15			R²	0.970 58	0.911 71	16
GPR	MAE	2.999 2	4.074 9	13	53	LGBM	MAE	2.178 6	3.540 1	17	68
	MAPE	0.103 0	0.150 97	14			MAPE	0.080 3	0.129 0	17
	RMSE	4.072	5.413 6	13			RMSE	3.113 2	4.674 3	17
	R²	0.939 99	0.898 5	13			R²	0.964 92	0.924 33	17
GKR	MAE	4.859 0	6.191 7	8	32	LSTM	MAE	5.121 4	5.772 4	10	39
	MAPE	0.180 9	0.253 34	8			MAPE	0.180 2	0.219 9	9
	RMSE	6.605 5	8.256 2	8			RMSE	6.656 1	7.508 3	10
	R²	0.842 08	0.763 93	8			R²	0.839 65	0.804 77	10
ELM	MAE	7.939 1	9.058 3	5	17	GRU	MAE	5.141 7	5.832 6	9	37
	MAPE	0.294 3	0.369 88	4			MAPE	0.178 7	0.219 7	10
	RMSE	10.009 9	11.835 0	4			RMSE	6.673 4	7.547 6	9
	R²	0.637 35	0.514 92	4			R²	0.838 82	0.802 72	9
GRNN	MAE	7.615 7	8.476 6	6	24	CNN	MAE	6.884 7	7.774 4	7	28
	MAPE	0.311 4	0.356 3	6			MAPE	0.277 0	0.325 0	7
	RMSE	9.558 6	10.867 8	6			RMSE	8.859 6	10.047 0	7
	R²	0.669 31	0.590 97	6			R²	0.715 91	0.650 42	7
GAM	MAE	2.611 7	3.909 8	14	55
	MAPE	0.097 7	0.154 2	13
	RMSE	3.482 7	5.274 4	14
	R²	0.956 10	0.903 66	14

Model	Evaluation metrics	Train	Test	Score	Overall score	Model	Evaluation metrics	Train	Test	Score	Overall score
LMR	MAE	7.030 1	9.058 3	4	13	DT	MAE	2.012 4	4.719 3	11	44
	MAPE	0.294 3	0.369 9	3			MAPE	0.068 8	0.180 7	11
	RMSE	10.009 9	11.835 0	3			RMSE	3.044 8	6.527 9	11
	R²	0.637 35	0.514 92	3			R²	0.966 45	0.852 42	11
LASSO	MAE	7.940 2	9.061 2	3	9	RF	MAE	2.722 3	4.451 5	12	48
	MAPE	0.294 5	0.370 2	2			MAPE	0.100 9	0.183 7	12
	RMSE	10.010 0	11.836 2	2			RMSE	3.749 1	5.900 0	12
	R²	0.637 34	0.514 83	2			R²	0.949 13	0.879 45	12
MRR	MAE	7.956 3	9.071 8	1	12	SVM	MAE	7.766 3	9.065 4	2	9
	MAPE	0.297 3	0.371 8	1			MAPE	0.282 2	0.356 4	5
	RMSE	10.017 4	11.827 2	5			RMSE	10.386 6	12.488 4	1
	R²	0.636 81	0.515 56	5			R²	0.609 54	0.459 88	1
LSBoost	MAE	0.854 6	3.563 3	16	62	XGBoost	MAE	1.987 9	3.719 9	15	62
	MAPE	0.030 6	0.139 63	16			MAPE	0.068 1	0.144 5	15
	RMSE	1.609 2	5.055 4	15			RMSE	2.850 9	5.049 2	16
	R²	0.990 63	0.912 46	15			R²	0.970 58	0.911 71	16
GPR	MAE	2.999 2	4.074 9	13	53	LGBM	MAE	2.178 6	3.540 1	17	68
	MAPE	0.103 0	0.150 97	14			MAPE	0.080 3	0.129 0	17
	RMSE	4.072	5.413 6	13			RMSE	3.113 2	4.674 3	17
	R²	0.939 99	0.898 5	13			R²	0.964 92	0.924 33	17
GKR	MAE	4.859 0	6.191 7	8	32	LSTM	MAE	5.121 4	5.772 4	10	39
	MAPE	0.180 9	0.253 34	8			MAPE	0.180 2	0.219 9	9
	RMSE	6.605 5	8.256 2	8			RMSE	6.656 1	7.508 3	10
	R²	0.842 08	0.763 93	8			R²	0.839 65	0.804 77	10
ELM	MAE	7.939 1	9.058 3	5	17	GRU	MAE	5.141 7	5.832 6	9	37
	MAPE	0.294 3	0.369 88	4			MAPE	0.178 7	0.219 7	10
	RMSE	10.009 9	11.835 0	4			RMSE	6.673 4	7.547 6	9
	R²	0.637 35	0.514 92	4			R²	0.838 82	0.802 72	9
GRNN	MAE	7.615 7	8.476 6	6	24	CNN	MAE	6.884 7	7.774 4	7	28
	MAPE	0.311 4	0.356 3	6			MAPE	0.277 0	0.325 0	7
	RMSE	9.558 6	10.867 8	6			RMSE	8.859 6	10.047 0	7
	R²	0.669 31	0.590 97	6			R²	0.715 91	0.650 42	7
GAM	MAE	2.611 7	3.909 8	14	55
	MAPE	0.097 7	0.154 2	13
	RMSE	3.482 7	5.274 4	14
	R²	0.956 10	0.903 66	14

Model	Evaluation metrics	Train	Test	Score	Overall score	Model	Evaluation metrics	Train	Test	Score	Overall score
LSBoost-BO	MAE	0.899 1	2.616 9	29	116	RF-BO	MAE	1.329 4	3.518 3	14	57
	MAPE	0.032 1	0.096 1	29			MAPE	0.047 1	0.134 1	14
	RMSE	1.641 3	3.811 8	29			RMSE	2.099 0	4.835 5	15
	R²	0.990 25	0.950 23	29			R²	0.984 05	0.919 91	14
LSBoost-SSA	MAE	0.851 6	2.544 6	30	120	RF-SSA	MAE	1.612 8	3.577 9	12	47
	MAPE	0.030 3	0.093 9	30			MAPE	0.055 8	0.138 6	11
	RMSE	1.591 8	3.718 0	30			RMSE	2.427 1	4.924 1	12
	R²	0.990 83	0.952 65	30			R²	0.978 68	0.916 94	12
LSBoost-SCA	MAE	0.768 4	2.895 9	23	90	RF-SCA	MAE	1.628 5	3.802 4	10	42
	MAPE	0.027 4	0.108 5	21			MAPE	0.057 1	0.139 8	12
	RMSE	1.537 1	4.625 3	23			RMSE	2.585 9	5.308 0	10
	R²	0.991 45	0.926 72	23			R²	0.975 80	0.903 49	10
LSBoost-SA	MAE	0.765 3	2.651 8	28	109	RF-SA	MAE	1.424 4	3.518 2	15	59
	MAPE	0.027 3	0.102 6	26			MAPE	0.051 4	0.137 6	15
	RMSE	1.519 8	3.972 3	27			RMSE	2.186 6	4.888 5	14
	R²	0.991 64	0.945 95	28			R²	0.982 70	0.918 14	15
LSBoost-PSO	MAE	0.847 7	2.705 3	26	103	RF-PSO	MAE	1.439 9	3.692 6	13	49
	MAPE	0.030 4	0.097 8	27			MAPE	0.052 5	0.142 7	10
	RMSE	1.597 6	4.289 2	24			RMSE	2.211 4	4.999 4	13
	R²	0.990 76	0.936 98	26			R²	0.982 30	0.914 38	13
GPR-BO	MAE	2.666 0	4.169 5	8	33	XGBoost-BO	MAE	0.526 2	2.661 2	27	102
	MAPE	0.091 0	0.138 7	9			MAPE	0.017 8	0.108 0	22
	RMSE	3.688 3	5.677 3	8			RMSE	1.364 7	3.887 5	26
	R²	0.950 76	0.889 59	8			R²	0.993 26	0.948 23	27
GPR-SSA	MAE	2.999 2	4.420 3	5	20	XGBoost-SSA	MAE	0.981 1	3.133 8	16	61
	MAPE	0.103 0	0.150 4	5			MAPE	0.033 7	0.122 3	13
	RMSE	4.072 0	5.850 5	5			RMSE	1.724 0	4.534 0	16
	R²	0.939 99	0.882 75	5			R²	0.989 24	0.929 58	16
GPR-SCA	MAE	2.999 2	4.169 5	5	20	XGBoost-SCA	MAE	0.981 1	3.133 8	16	61
	MAPE	0.103 0	0.150 4	5			MAPE	0.033 7	0.122 3	13
	RMSE	4.072 0	5.850 5	5			RMSE	1.724 0	4.534 0	16
	R²	0.939 99	0.882 75	5			R²	0.989 24	0.929 58	16
GPR-SA	MAE	2.999 2	4.169 5	5	20	XGBoost-SA	MAE	0.981 1	3.133 8	16	61
	MAPE	0.103 0	0.150 4	5			MAPE	0.033 7	0.122 3	13
	RMSE	4.072 0	5.850 5	5			RMSE	1.724 0	4.534 0	16
	R²	0.939 99	0.882 75	5			R²	0.989 24	0.929 58	16