Abstract In order to build the algorithm model for carbon emission prediction during railway tunnel construction, a study has been conducted based on a railway tunnel project. First, the carbon emission during construction of the tunnel and the proportion of each subphase in the total emission are quantitatively determined by using the carbon emission calculation system. Then several different prediction models are created by using several machine learning algorithms, the parameters of the prediction models are optimized by using the animal optimization algorithm, the fitting results are compared and analyzed by using the assessment indicators including R2, MAE, MSE, RMSE, MAPE and SMAPE so as to select the optimum prediction model, and the importance degree of each parametric variable is determined by using the SHAP value. The results are stated as follows: The construction materials production stage has the greatest contribution to carbon emission in tunnel construction and the construction materials transportation stage has the least contribution; BP neural network generates better regression results than the algorithms such as random forest, LightGBM, SVR and extreme learning machine, and according to comparison of the regression prediction models optimized by the PSO, WOA and SSA algorithms, WOA-BP algorithm generates the best fitting results;based on the analysis by the SHAP algorithm, the rank of the parametric variables in terms of importance degree is as follows: excavation area > surrounding rock class > excavation method > buried depth.
Abstract：
In order to build the algorithm model for carbon emission prediction during railway tunnel construction, a study has been conducted based on a railway tunnel project. First, the carbon emission during construction of the tunnel and the proportion of each subphase in the total emission are quantitatively determined by using the carbon emission calculation system. Then several different prediction models are created by using several machine learning algorithms, the parameters of the prediction models are optimized by using the animal optimization algorithm, the fitting results are compared and analyzed by using the assessment indicators including R2, MAE, MSE, RMSE, MAPE and SMAPE so as to select the optimum prediction model, and the importance degree of each parametric variable is determined by using the SHAP value. The results are stated as follows: The construction materials production stage has the greatest contribution to carbon emission in tunnel construction and the construction materials transportation stage has the least contribution; BP neural network generates better regression results than the algorithms such as random forest, LightGBM, SVR and extreme learning machine, and according to comparison of the regression prediction models optimized by the PSO, WOA and SSA algorithms, WOA-BP algorithm generates the best fitting results;based on the analysis by the SHAP algorithm, the rank of the parametric variables in terms of importance degree is as follows: excavation area > surrounding rock class > excavation method > buried depth.
ZENG Hongrui1,
2 SUN Wenhao3 HE Wei3 GUO Yalin1,
2 GUO Chun1 etc
.Study on the Carbon Emission Prediction Model for Railway Tunnel Construction Based on Machine Learning[J] MODERN TUNNELLING TECHNOLOGY, 2023,V60(6): 29-39
2023, Vol. 60 