Tunnel construction carbon emission prediction method based on LCA and MLP neural network model

Abstract

The invention provides a tunnel construction carbon emission prediction method based on an LCA and MLP neural network model, and the method comprises the following steps: (1) dividing a tunnel construction process into material production, transportation and construction stages, quantifying the carbon emission of each stage based on a carbon emission factor method, and carrying out the accumulation to obtain the total carbon emission; (2) extracting initial influence factors from construction design parameters and geological conditions, screening significant variables through univariate regression analysis, and carrying out refined analysis on the surrounding rock grade in combination with SHAP value contribution degree quantification; (3) constructing an MLP neural network model by taking the I-level key index as an input variable, and optimizing the performance of the model through hyper-parameter tuning; and (4) generating a multivariable nonlinear empirical formula based on the optimized MLP model weight matrix and an SHAP value analysis result, and converting a model prediction result into a rapid estimation tool suitable for tunnel construction carbon emission by the empirical formula through combining a linear term, a nonlinear term and a variable interaction term.

Description

Technical Field

[0001] The present invention belongs to the technical fields such as engineering carbon emission prediction, and particularly relates to a tunnel construction carbon emission prediction method based on the LCA and MLP neural network models. Background Technique

[0002] With the aggravation of the global warming problem, reducing carbon emissions has become an important issue that needs to be solved urgently in the international community. In the field of infrastructure construction, tunnel engineering has become one of the important sources of carbon emissions due to its construction complexity and high energy consumption characteristics. How to scientifically, comprehensively and accurately predict and manage the carbon emissions during tunnel construction is an important technical problem for promoting green construction and achieving the carbon emission reduction goal.

[0003] At present, traditional carbon emission calculation methods mainly rely on empirical formulas and statistical data. Although these methods have certain advantages in dealing with simple linear relationships, they are often difficult to cope with complex non-linear problems and changing construction environments. On the one hand, traditional methods are difficult to capture the dynamic changes during the construction process; on the other hand, there are obvious limitations in feature selection and variable combination, resulting in insufficient reliability and accuracy of the prediction results. In addition, the existing carbon emission calculations during the construction stage based on BIM (Building Information Modeling) software have the following significant deficiencies:

[0004] 1. Insufficient data accuracy and integrity: Lack of sufficient measured data support, resulting in deviations in the model prediction results and being difficult to comprehensively reflect the actual carbon emissions during the construction process;

[0005] 2. Ignoring indirect carbon emissions: Most methods only focus on direct carbon emissions and do not fully consider the impacts of indirect carbon emissions such as transportation, energy consumption and the behaviors of construction workers;

[0006] 3. Lack of life cycle assessment: Existing methods mostly calculate for a single construction stage and fail to cover the entire life cycle of the project, making it difficult to comprehensively reflect the overall environmental impacts of construction activities. Summary of the Invention

[0007] Aiming at the defects and deficiencies existing in the prior art (such as weak non-linear processing ability of traditional carbon emission prediction methods, strong subjectivity in feature selection, and lack of rapid estimation tools), the present invention provides a tunnel construction carbon emission prediction method based on the integration of LCA and MLP. Its technical core and innovation points include:

[0008] LCA phased dynamic quantification and scenario adaptation: Decompose the tunnel construction carbon emissions into three stages: material production, transportation, and construction. Dynamically calculate the carbon emissions of each stage based on the IPCC standard formula, and adapt the carbon emission factors in combination with the tunnel engineering geological conditions (such as surrounding rock grade, buried depth) to solve the limitations of traditional LCA static calculations;

[0009] SHAP Value-Driven Feature Screening and Interpretability Optimization: Through a multi-level dimensionality reduction method that includes univariate regression pre-screening, Pearson correlation analysis, and SHAP value quantification of contribution degrees, 5 level-I key indicators (burial depth, surrounding rock grade, etc.) and 8 level-II geological refinement indicators (basic rock quality, rock mass integrity, etc.) are extracted from 13 initial indicators, achieving for the first time the scientific quantification of the impact of geological parameters on carbon emissions in tunnel construction scenarios;

[0010] MLP Model Optimization and Empirical Formula Generation: Construct a 4-layer MLP neural network model (neuron numbers 16-32-16-32, ReLU activation function, Adam optimizer). Through hyperparameter tuning and comparison and verification with multiple algorithms, its prediction error (MAE = 0.0503, R 2 = 0.8549) is significantly better than traditional models; Based on the MLP weight matrix and SHAP value contribution degrees, a multivariate empirical formula containing linear terms, non-linear terms, and interaction terms is generated, supporting rapid estimation and sensitivity analysis, with a prediction accuracy of 92.3%;

[0011] Enhanced Engineering Practicality: Verified through empirical data of tunnel projects in Zhejiang Province, taking into account complex geological conditions (such as dynamic changes in surrounding rock grade) and the optimization requirements of construction parameters, providing a precise quantification tool for green construction and carbon emission reduction.

[0012] The technical solution specifically adopted by the present invention to solve its technical problems is as follows:

[0013] A tunnel construction carbon emission prediction method based on LCA and MLP neural network model, comprising the following steps:

[0014] (1) Divide the tunnel construction process into material production, transportation, and construction stages, quantify the carbon emissions of each stage based on the carbon emission factor method, and accumulate to obtain the total carbon emissions;

[0015] (2) Extract initial influencing factors from construction design parameters and geological conditions, screen significant variables through univariate regression analysis, and combine with the SHAP value contribution degree to conduct a refined analysis of the surrounding rock grade, obtaining level-I key indicators and level-II geological refinement indicators;

[0016] (3) Use the level-I key indicators as input variables to construct an MLP neural network model, and optimize the model performance through hyperparameter tuning;

[0017] (4) Based on the optimized MLP model weight matrix and SHAP value analysis results, generate a multivariate non-linear empirical formula. The empirical formula combines linear terms, non-linear terms, and variable interaction terms to transform the model prediction results into a rapid estimation tool applicable to tunnel construction carbon emissions.

[0018] Further, in step (1):

[0019] The carbon emissions in the material production stage are calculated based on the material consumption and the carbon emission factor.

[0020] The carbon emissions in the material transportation stage are calculated based on the transportation distance, the type of transportation vehicle, and the corresponding carbon emission factor.

[0021] The carbon emissions in the construction stage are calculated based on the energy consumption and the carbon emission factor.

[0022] Further, in step (2):

[0023] The refined analysis of the surrounding rock grade includes: the weight ranking of the basic rock quality, the rock mass integrity, and the joint fissures; the weight ranking of the secondary indicators of the surrounding rock grade is: basic rock quality > rock mass integrity > joint fissures.

[0024] Further, in step (3), the parameters of the MLP neural network model are determined by collaborative optimization driven by the tunnel construction scenario, including: a 4-layer fully connected network structure for hierarchically extracting the non-linear interaction features between the carbon emissions in the construction stage and the geological conditions; a combination of the ReLU activation function and the Adam optimizer to suppress the vanishing gradient and adapt to the noise distribution of the construction data; a configuration of the learning rate of 0.001 and the number of iterations of 500 to balance the model convergence speed and the prediction accuracy.

[0025] In the actual implementation process, the optimal model can be selected through hyperparameter tuning and comparison verification with various machine learning algorithms; the comparison verification includes the performance comparison with the random forest, XGBoost, and stacked regression models, and the MLP neural network is selected as the optimal prediction model based on the MAE, MSE, RMSE, and R 2 index.

[0026] Further, the generation logic of the empirical formula in step (4) includes:

[0027] Extract the weight matrix of each input variable from the MLP neural network model;

[0028] Quantify the contribution degree of the primary and secondary indicators to the carbon emissions through SHAP value analysis;

[0029] Fit the weight matrix and the contribution degree to generate the formula coefficients, and form an expression including linear terms, non-linear terms, and variable interaction terms for quickly estimating the carbon emissions under different construction scenarios.

[0030] Further, the coefficients of the empirical formula are dynamically adapted to the changes in the geological conditions to support the sensitivity analysis of the key indicators of the surrounding rock.

[0031] Furthermore, the types of transportation tools include road, railway and waterway transportation, and the carbon emission factors are dynamically adjusted according to the types of transportation tools.

[0032] Furthermore, the categorical variables in the key indicators are processed by numerical coding, including the surrounding rock grade, construction method and lining type.

[0033] In addition, an electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the method described above are implemented.

[0034] A non-transitory computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor, the steps of the method described above are implemented.

[0035] Compared with the prior art, the present invention and its preferred solutions at least include the following beneficial effects:

[0036] Significantly improve the prediction accuracy and adaptability in complex scenarios: By integrating the phased dynamic quantification of LCA and the MLP neural network model, the non-linear carbon emission correlation characteristics in the material production, transportation and construction stages of tunnel construction are effectively captured, solving the problem of insufficient prediction accuracy of traditional linear models under complex geological conditions and dynamic construction parameters.

[0037] Enhance the scientificity and interpretability of feature screening: Based on multi-level statistical analysis and SHAP value contribution quantification, key indicators are screened from construction parameters and geological conditions and the weight ranking is refined, overcoming the defects of strong subjectivity in feature selection and rough processing of geological parameters in the prior art, and providing a clear basis for construction optimization.

[0038] Achieve the balance between engineering practicability and computational efficiency: Through the weight matrix of the MLP model and the SHAP value to generate a multi-variable empirical formula, the complex model is transformed into a rapid estimation tool, taking into account the requirements of high-precision prediction and on-site rapid decision-making, and making up for the shortcoming of the existing methods relying on complex models and insufficient practicability.

[0039] Improve the coverage and dynamic adaptation ability of the whole life cycle carbon emission analysis: The phased dynamic quantification method covers the whole process of material production, transportation and construction, combined with the dynamic factor adaptation mechanism such as the type of transportation tool and geological conditions, ensuring that the carbon emission calculation boundary is clear and adaptable to changing engineering scenarios.

[0040] Optimize data processing and model generalization performance: Through technical details such as categorical variable coding and residual distribution verification, improve the consistency and robustness of the model input data, support stable prediction under different geological environments and construction schemes, and expand the application scope of the technology.

[0041] This solution systematically solves problems such as low accuracy, weak interpretability, and insufficient practicality in carbon emission prediction through technological combination innovation (LCA + MLP + SHAP) and scenario adaptation optimization, providing a solution that combines theoretical rigor and engineering implementation for carbon emission reduction in tunnel construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The present invention will be further described in detail below in conjunction with the drawings and specific embodiments:

[0043] Figure 1 It is a flowchart of a carbon emission prediction method based on an LCA-MLP neural network model provided for an optional embodiment of the present invention;

[0044] Figure 2 It is a schematic diagram of carbon emission calculation based on life cycle assessment (LCA) for an optional embodiment of the present invention;

[0045] Figure 3 It is a schematic diagram of the process of an MLP multi-sensory neural network algorithm model provided for an optional embodiment of the present invention;

[0046] Figure 4 It is a comparison chart of four evaluation indicators of eight machine learning algorithms provided for an optional embodiment of the present invention; among them,

[0047] (a) Comparison of mean squared error (MSE) for model optimization;

[0048] (b) Comparison of root mean squared error (RMSE) for model optimization;

[0049] (c) Comparison of mean absolute error (MAE) for model optimization;

[0050] (d) Comparison of coefficient of determination (R 2 ) for model optimization.

[0051] Figure 5 It is the performance residual analysis and Q-Q plot of the MLP algorithm model provided for an optional embodiment of the present invention; among them, (a) Q-Q plot of residual analysis of the optimal prediction model; (b) residual distribution of the optimal prediction model.

[0052] Figure 6 It is a comparison chart of predicted values and actual values of eight machine learning algorithms for the test set provided for an optional embodiment of the present invention; among them, (a) MLR multiple linear regression model; (b) RF random forest model; (c) BAGGING model; (d) SVM model; (e) Adaboost model; (f) XGboost model; (g) stacked regression model; (h) MLP neural network model. SPECIFIC EMBODIMENTS

[0053] To make the features and advantages of this patent more obvious and understandable, specific embodiments are given below for detailed description as follows:

[0054] It should be noted that the following detailed description is illustrative and is intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs.

[0055] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to this application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should also be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0056] This method relates to the technical field of engineering carbon emissions prediction, and discloses a tunnel construction carbon emissions prediction method based on the whole-process life cycle assessment (LCA) and the multi-layer perceptron (MLP) neural network model. Its core process includes the following steps:

[0057] LCA phased carbon emissions quantification:

[0058] The tunnel construction process is divided into material production, material transportation, and construction stages. Based on the "Building Carbon Emission Calculation Standard" and IPCC carbon emission factors, the carbon emissions of each stage are calculated respectively and accumulated to obtain the total carbon emissions. For example: Material production stage: calculated based on building material consumption and carbon emission coefficients;

[0059] Material transportation stage: calculated in combination with transportation distance, tool type, and corresponding carbon emission factors;

[0060] Construction stage: calculated based on equipment energy consumption and energy carbon emission coefficients. Key index screening and dimensionality reduction:

[0061] Thirteen initial influencing factors are extracted from construction design parameters and geological conditions, and key indicators are screened through multi-level statistical analysis: Single-variable regression analysis (P value < 0.05) initially screens significant variables;

[0062] Pearson correlation analysis eliminates redundant variables;

[0063] SHAP value contribution quantification refines geological parameters (such as the secondary indicators of surrounding rock grade: rock mass integrity, weathering degree, etc.), and finally determines the burial depth, surrounding rock grade, excavation area, construction method, and lining type as the first-level key indicators. MLP neural network modeling and optimization:

[0064] Taking the Class-I key indicators as input variables, an MLP neural network model is constructed, and the prediction performance is improved through the following optimization strategies: Network structure: 4 fully connected layers (number of neurons: 16 - 32 - 16 - 32);

[0065] Activation function: ReLU;

[0066] Optimizer: Adam (learning rate 0.001, number of iterations 500);

[0067] In the training of the MLP model, GPU acceleration technology is used to optimize the calculation efficiency, and Dropout regularization is combined to suppress overfitting, ensuring the generalization performance of the model under complex geological data.

[0068] Comparison and verification: Compare with 8 algorithms such as random forest and XGBoost to screen the optimal model. Empirical formula generation and verification:

[0069] Based on the optimized MLP model weight matrix and the SHAP value analysis results, a multi-variable non-linear empirical formula is generated to support the rapid estimation of carbon emissions. The formula form includes linear terms, non-linear terms, and variable interaction terms.

[0070] During the generation of the empirical formula, the weight matrix of the MLP model can be extracted through the PyTorch framework, and the contribution degree of the input variables can be refined by combining the SHAP values, converting the complex non-linear mapping relationship inside the model into a multi-variable expression including linear terms, square terms, and interaction terms, realizing the transformation from a "black-box model" to an interpretable engineering tool.

[0071] The generation logic of the empirical formula includes the introduction of non-linear terms (such as square terms and logarithmic terms), and by adjusting the weight distribution ratio, it dynamically adapts to the carbon emission prediction requirements under different geological conditions (such as changes in surrounding rock grades), enhancing the engineering applicability of the formula.

[0072] To verify the universality and effectiveness of this method, the tunnel project from Jincun Fangxi to Qingtian Jizhai section of the Jincun section of the S218 Anji to Longgang Expressway in Zhejiang Province is taken as an example, and the implementation process is as follows:

[0073] Data collection: Carbon emission data covering the entire process of material production, transportation, and construction;

[0074] Model training: The data set is divided into an 80% training set and a 20% test set, and the Class-I key indicators are input;

[0075] Result verification:

[0076] The prediction error of the MLP model (MAE = 0.0503, R 2 = 0.8549) is significantly better than the comparison algorithms;

[0077] The matching accuracy between the predicted value of the empirical formula and the actual value reaches 92.3%.

[0078] Residual analysis (Q-Q plot) shows that the error distribution conforms to the normality assumption, and the model robustness is reliable.

[0079] For the convenience of understanding, the terms involved in the embodiments of the present application are explained below:

[0080] (1) LCA (Life Cycle Assessment, full-process life cycle assessment) is a systematic analysis method used to evaluate the environmental impact of a product or process throughout its life cycle. The purpose of LCA is to quantify the environmental impact of a product (such as carbon emissions, energy consumption, pollution, etc.) by analyzing all stages from raw material acquisition, production, use to waste treatment, so as to provide a basis for environmental management and decision-making.

[0081] (2) MLP (Multilayer Perceptron, multi-layer perceptron) is a typical feedforward artificial neural network, commonly used in tasks such as classification and regression, and has strong non-linear modeling capabilities. MLP is one of the basic structures of neural networks, composed of an input layer, hidden layers, and an output layer, belonging to a fully connected neural network (Fully Connected Neural Network). Neurons in each layer are connected to all neurons in the next layer.

[0082] The working principle of MLP can be divided into two stages: 1. Forward propagation: The input data is passed through the input layer to the hidden layer. Each neuron in the hidden layer processes the input signal through an activation function (such as ReLU, Sigmoid, Tanh) and outputs the result, which is then passed to the next layer. Finally, the output layer generates the prediction result of the model. 2. Backward propagation: By calculating the error between the output and the true value (such as mean squared error MSE or cross-entropy), the error is propagated backward to each layer. The gradient descent method is used to update the weights. The gradient is calculated through the chain rule, and the weights are gradually adjusted according to the learning rate, so that the error gradually decreases, thereby optimizing the model.

[0083] Figure 1 It is a flowchart of a carbon emission prediction method based on the LCA-MLP neural network model provided by an optional embodiment of the present invention. As Figure 1 shown, the method may include:

[0084] Taking 6 highway tunnel projects in Zhejiang Province as research cases, using the full-process life cycle assessment method (LCA), the tunnel construction is divided into three stages: building material production, material transportation, and construction, to obtain carbon emissions.

[0085] Select 13 carbon emission influencing factors from relevant construction units, the "Building Carbon Emission Calculation Standard", and the Intergovernmental Panel on Climate Change (IPCC). Use univariate regression analysis to reduce the dimension of the 13 indicators, divide them into 5 primary indicators and 8 secondary indicators, and analyze the importance of each parameter variable based on the Pearson correlation coefficient and SHAP value.

[0086] Next, use 8 machine learning algorithms to model and predict the carbon emissions per meter. Through the comparison of four error evaluation indicators with the predicted values and actual values, and at the same time perform hyperparameter tuning to comprehensively screen the best prediction model.

[0087] Figure 2 Schematic diagram of carbon emission calculation based on life cycle assessment (LCA) for an optional embodiment of the present invention. As Figure 2 shown, the specific calculation process is as follows:

[0088] (1) Direct carbon emissions from material production

[0089] Based on material consumption and the corresponding carbon emission coefficients, calculate the carbon emissions during the production stage of construction materials:

[0090]

[0091] Among them, E m represents the carbon emissions during the building material production stage, i is the type of building material, m i is the consumption of the i-th material, and f j is the carbon emission coefficient of the i-th material.

[0092] (2) Carbon emissions generated during material transportation

[0093] In order to move materials from one place to another, vehicles will generate carbon emissions during transportation. Therefore, the carbon emissions caused during the transportation stage:

[0094]

[0095] Among them, E t is the carbon emissions during material transportation, i is the type of building material, j is the type of transportation vehicle, m i is the consumption of the i-th material, d i is the transportation distance of the i-th building material, and f i is the carbon emission factor per unit weight and per unit transportation distance of the j-th transportation vehicle.

[0096] (3) Carbon emissions caused during on-site construction

[0097] The operation of construction machinery at the construction site requires gasoline, diesel, and electric energy as driving forces, and the food, clothing, housing, and transportation of construction workers also require corresponding energy. Therefore, the carbon emissions generated during the construction stage are as follows:

[0098]

[0099] Among them, E c represents the carbon emissions during the construction stage, i is the type of energy, and e i is the consumption of the i-th type of energy, and f i is the carbon emission coefficient of the i-th type of energy.

[0100] (4) Total carbon emissions

[0101] To sum up, by accumulating the carbon emissions in the material production, transportation, and construction stages, the total carbon emissions of the highway tunnel within the calculation boundary can be obtained: E T = E m + E t + E c

[0102] Among them: E T is the total carbon emissions of the highway tunnel; E m is the carbon emissions generated during the material production process; E t is the carbon emissions generated during the material transportation process; E C is the carbon emissions generated by machinery and workers during the construction process.

[0103]

[0104]

[0105] The total calculated carbon emissions and the proportion of each stage are shown in the following table:

[0106] Based on the construction design document and the review of relevant previous research literature, 13 factors affecting the carbon emissions of tunnel construction were selected. The univariate regression analysis method was used to analyze the relationship between 13 independent variables (surrounding rock grade, buried depth, construction method, lining type, excavation area, rock compressive strength, rock compressive strength, rock mass basic quality, rock mass integrity coefficient, rock quality, weathering degree, rock mass structure, joint fissures) and the target variable (carbon emissions), screening out variables that have a significant impact on carbon emissions, removing variables with a weak relationship with the target variable, reducing the feature dimension, improving the calculation efficiency of the model, and improving the prediction accuracy.

[0107] Among the 13 indicators, some are numerical variables and some are categorical variables. For categorical variables, data encoding is performed to make their representation methods the same to ensure the consistency of the model input data, improve the training efficiency and prediction stability. As shown in the following table:

[0108]

[0109]

[0110] Based on univariate regression analysis, the smaller the statistically significant P-Value, the greater the impact on carbon emissions. The results are shown in the following table:

[0111]

[0112] The data obtained from the LCA application of 6 highway tunnel projects in Zhejiang Province was segmented. 80% was used as the training set and 20% as the test set. Using the I-level indicators: surrounding rock grade, buried depth, construction method, lining type, and excavation area as the model inputs and carbon emissions as the model output, it was trained with eight machine learning algorithms. Figure 3 The process schematic diagram of the MLP multi-perceptron neural network algorithm model provided by an alternative embodiment of the present invention was used for hyperparameter tuning. Using the gradient descent method, the model parameters of the MLP neural network were determined to be: number of layers: 4, number of neurons in each layer: 16, 32, 16, 32, activation function: ReLU, learning rate: 0.001, optimization algorithm: Adam, and the prediction effect was the best when the number of iterations (epochs) was 500. The principles and model parameters of each algorithm are shown in the following table:

[0113]

[0114]

[0115] Figure 4 The comparison chart of the prediction capabilities of the MAE, MSE, RMSE, and R2 four indicators for the eight machine learning algorithms provided by an alternative embodiment of the present invention. It can be seen that the Adaboost model performs poorly in each indicator. Its MAE, MSE, and RMSE are 0.0876, 0.0173, and 0.1314 respectively, with a large error. The R2 is only 0.7395, which is the lowest among all models. The MLP neural network has the best performance in each evaluation indicator. Its MSE, RMSE, and MAE values are 0.0056, 0.0750, and 0.0503 respectively, all of which are the lowest among the model errors; R 2 is 0.8549, which is the best among all models. Therefore, the MLP neural network has the best model performance as a machine learning model for predicting carbon emissions and has strong generalization ability and higher prediction performance when compared with other models.

[0116] Figure 5 The comparison chart of the predicted values and actual values of the eight machine learning algorithms for the test set provided by an alternative embodiment of the present invention. As shown in the figure:

[0117] Multiple linear regression (MLR) has significant errors at some data points, with an amplitude of 0.1447, indicating that it performs poorly in capturing the complexity of the data. The prediction errors of the random forest (RF) are relatively scattered, with an amplitude of 0.4881, and its prediction performance needs to be improved, which is consistent with the fact that the random forest model performs poorly when there are too many input variables. Bagging has large errors at most data points, with an amplitude of 0.1775, and its prediction performance needs to be improved. The amplitude of SVM is 1.1469, and the matching degree between the prediction result and the actual value is not ideal enough. There is overfitting linearity, resulting in poor generalization ability for data processing. The amplitude of Adaboost is 1.12103, and the overall prediction error is relatively low, but there is a certain error volatility. Since the Adaboost algorithm often ignores the minority class when dealing with imbalanced data, the prediction performance is not ideal. The prediction errors of XGBoost are relatively scattered, with an amplitude of 0.161093, and it performs poorly when dealing with imbalanced data sets, usually performing poorly on the minority class. The stacking model has significant errors at individual data points. Although the prediction result of the stacking regression is close to the actual value, there are still slight differences, with an amplitude of 0.258246. The multi-layer perceptron (MLP) neural network model has relatively small prediction errors at all data points. The prediction result is highly consistent with the actual value, with an amplitude of only 0.112324, showing high precision and stability, and performing excellently in capturing the complexity of the data and improving the prediction accuracy.

[0118] Figure 5 The performance residual analysis and Q-Q plot of the MLP algorithm model provided by an optional embodiment of the present invention. To verify the reliability and practicality of the MLP model in predicting tunnel construction carbon emissions, the data of the S218 Anji to Longgang Expressway project in Zhejiang Province was used for in-depth analysis. The Q-Q plot shows that most residual data points are closely aligned with the theoretical normal distribution line, indicating that the distribution of model errors basically conforms to the normality assumption, thus supporting the robustness and reliability of its prediction of carbon emissions. Residual analysis further shows that the residuals of the MLP model are mainly distributed near zero, indicating that the model has high precision and stability in predicting tunnel construction carbon emissions. The excellent performance of the MLP model in practical engineering applications is verified. The above experimental results not only prove the superiority of the MLP model in predicting tunnel construction carbon emissions and verify the excellent performance of the MLP model in practical engineering applications, but also provide theoretical and practical bases for the popularization of related technologies.

[0119] To explore the contribution of each level of indicators to the prediction results, the weight matrix of each layer was extracted from the MLP neural network using PyTorch, and the Pearson coefficient was used to obtain the correlation coefficient between the primary indicators and the predicted carbon emissions. It can be seen that the correlation ranking of the primary indicators and the predicted carbon emissions is: surrounding rock grade > buried depth > construction method > lining type > excavation area. Among them, the correlation coefficients are 0.6380, 0.6028, 0.4850, 0.4779, and 0.1282 in sequence. Among the five primary indicators, the buried depth, construction method, lining type, and excavation area are all controlled by the design scheme, while the surrounding rock grade depends on natural characteristics such as weathering degree and rock mass integrity. Therefore, it was further refined into multiple secondary indicators, and the SHAP (SHapley Additive exPlanations) algorithm was used to obtain the weight proportion of each secondary indicator. The importance ranking of the secondary indicators refined from the surrounding rock grade is: basic rock quality (Bq) > rock mass integrity > joint fissures > rock quality > rock mass integrity coefficient (Kv) > rock mass structure > weathering degree > uniaxial compressive strength of rock (Rc). Among them, the weight rankings are 0.5516, 0.2134, 0.0815, 0.0675, 0.0482, 0.0190, 0.0114, and 0.0074 in sequence.

[0120] To verify the accuracy and usability of the empirical formula obtained from the LCA-MLP model in the prediction of tunnel construction carbon emissions, this embodiment used the data of the S218 Anji to Longgang Expressway project in Zhejiang Province for empirical analysis. By comprehensively modeling the input variables and their interaction effects, the following polynomial prediction formula was obtained:

[0121]

[0122] Under certain similar geological environment conditions, the empirical formula proposed in the practical application of this embodiment is:

[0123] E = 32.9393 - 6.6464*X1 + 2.9854*X2 + 14.9908*X3 - 2.0930*X4 - 0.4231*X5

[0124] Where:

[0125] X1 represents the buried depth

[0126] X2 represents the surrounding rock grade

[0127] X3 represents the construction method

[0128] X4 represents the lining type

[0129] X5 represents the excavation area

[0130] The actual values and theoretical values of the tunnel carbon emissions calculated by the empirical formula are shown in the following table:

[0131]

[0132] This expression illustrates how the model captures the complex relationships between input variables through linear terms, quadratic terms, and interaction terms. As shown in the table, the theoretical values calculated according to the empirical formula are very close to the actual carbon emissions, and the prediction accuracy is as high as 92.3%, which proves the accuracy of the empirical formula and can be well applied to the actual tunnel engineering field. However, when applying it to other projects, the similarity of geological and environmental conditions must be considered. Changes in geological characteristics and environmental factors may significantly affect the output results of the model, thereby affecting the accuracy of the prediction results. This model formula not only supports predictive analysis but also emphasizes the importance of considering specific situations.

[0133] In summary, the present invention combines the whole-process life cycle assessment (LCA) and the MLP neural network algorithm, and innovatively proposes an empirical formula for carbon emissions, which can comprehensively quantify the impact of each construction stage on carbon emissions, effectively predict the carbon emissions at each stage of tunnel construction, solve the defects of the prior art such as small application range, weak objectivity, and weak interpretability, and provide scientific construction management suggestions through the quantitative analysis of the model parameter characteristics, which is helpful to achieve green construction and carbon emission control.

[0134] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors, and a memory for storing one or more computer programs; the program includes program instructions, and the processor is used to execute the program instructions stored in the memory. The processor may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically used to load and execute one or more instructions in the computer storage medium to implement the above method.

[0135] It should be further noted that, based on the same inventive concept, the present invention also provides a computer storage medium, on which a computer program is stored, and when the computer program is run by a processor, it executes the above method. The storage medium may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may, for example, but not be limited to, an electrical, magnetic, optical, electro-magnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the computer-readable storage medium include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, the computer-readable storage medium may be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, apparatus, or device.

[0136] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should have the ordinary meanings understood by those with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar terms used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. Words such as "including" or "comprising" mean that the elements or objects appearing before this word cover the elements or objects listed after this word and their equivalents, without excluding other elements or objects. Words such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right", etc. are only used to represent relative position relationships, and when the absolute position of the object being described changes, the relative position relationship may also change accordingly.

[0137] As described above, these are only the preferred embodiments of the present invention, and the present invention is not limited to other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes. However, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the technical solution content of the present invention still fall within the protection scope of the technical solution of the present invention.

[0138] This patent is not limited to the above-mentioned best implementation mode. Anyone inspired by this patent can derive other various forms of a tunnel construction carbon emission prediction method based on the LCA and MLP neural network models. All equal changes and modifications made according to the scope of the patent application of the present invention shall fall within the scope covered by this patent.

Claims

1. A tunnel construction carbon emission prediction method based on the LCA and MLP neural network models, characterized in that, It includes the following steps: (1) Divide the tunnel construction process into material production, transportation, and construction stages. Quantify the carbon emissions of each stage based on the carbon emission factor method, and accumulate to obtain the total carbon emissions; (2) Extract the initial influencing factors from the construction design parameters and geological conditions. Screen the significant variables through univariate regression analysis, and combine the SHAP value contribution degree to conduct a refined analysis of the surrounding rock grade to obtain the Class I key indicators and Class II geological refinement indicators; (3) Take the Class I key indicators as input variables to construct an MLP neural network model, and optimize the model performance through hyperparameter tuning; (4) Based on the optimized MLP model weight matrix and SHAP value analysis results, generate a multivariable non-linear empirical formula. The empirical formula combines linear terms, non-linear terms, and variable interaction terms to transform the model prediction results into a rapid estimation tool applicable to tunnel construction carbon emissions.

2. A method for predicting tunnel construction carbon emissions based on the LCA and MLP neural network model according to claim 1, characterized in that: In step (1): The carbon emissions in the material production stage are calculated based on the material consumption and carbon emission factors; The carbon emissions in the material transportation stage are calculated based on the transportation distance, transportation tool type, and corresponding carbon emission factors; The carbon emissions in the construction stage are calculated based on the energy consumption and carbon emission factors.

3. A method for predicting tunnel construction carbon emissions based on the LCA and MLP neural network model according to claim 1, characterized in that: In step (2): The refined analysis of the surrounding rock grade includes: the weight ranking of the basic rock quality, rock mass integrity, and joint fissures; the weight ranking of the Class II indicators of the surrounding rock grade is: basic rock quality > rock mass integrity > joint fissures.

4. A tunnel construction carbon emission prediction method based on the LCA and MLP neural network model according to claim 1, characterized in that: The parameters of the MLP neural network model are determined by collaborative optimization driven by the tunnel construction scenario, including: a 4-layer fully connected network structure for hierarchically extracting the non-linear interaction characteristics between the carbon emissions in the construction stage and the geological conditions; a combination of the ReLU activation function and the Adam optimizer to suppress gradient disappearance and adapt to the noise distribution of the construction data; a configuration of a learning rate of 0.001 and an iteration number of 500 to balance the model convergence speed and prediction accuracy.

5. A method for predicting tunnel construction carbon emissions based on the LCA and MLP neural network model according to claim 1, characterized in that: The generation logic of the empirical formula in step (4) includes: Extract the weight matrix of each input variable from the MLP neural network model; Quantify the contribution degree of the Class I and Class II indicators to carbon emissions through SHAP value analysis; Fit the weight matrix and the contribution degree to generate formula coefficients, and form an expression containing linear terms, non-linear terms, and variable interaction terms for rapidly estimating the carbon emissions under different construction scenarios.

6. The tunnel construction carbon emission prediction method based on the LCA and MLP neural network model according to claim 5, characterized in that: The coefficients of the empirical formula are dynamically adapted to changes in geological conditions to support the sensitivity analysis of the key indicators of the surrounding rock.

7. A method for predicting carbon emissions in tunnel construction based on the LCA and MLP neural network model according to claim 2, characterized in that: The transportation tool types include road, railway, and waterway transportation, and the carbon emission factors are dynamically adjusted according to the transportation tool types.

8. A method for predicting tunnel construction carbon emissions based on the LCA and MLP neural network model according to claim 1, characterized in that: The categorical variables in the key indicators are processed through numerical coding, including the surrounding rock grade, construction method, and lining type.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the program, the steps of the method according to any one of claims 1-8 are implemented.

10. A non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1-8 are implemented.

Images