Road construction carbon emission monitoring method

Abstract

The invention provides a road construction carbon emission monitoring method, which belongs to the technical field of electric digital data processing, and comprises the following steps: establishing a monitoring network through a multi-source sensor, collecting equipment operation, material consumption and construction activity data by using an Internet of Things terminal, and calculating carbon emission in combination with a carbon emission factor database to form a fingerprint spectrum. Edge calculation is adopted to carry out time sequence analysis to recognize abnormal points, a neural unit jump algorithm and a multi-head attention architecture are applied to carry out fine adjustment on a pre-training model, and a jump threshold value is calculated through a specific function. And introducing a distributed gradient descent optimization equation to integrate data of multiple construction sites to protect privacy, and generating a monitoring report including total carbon emission, intensity and emission reduction suggestions. And by adaptively adjusting the monitoring time interval, the carbon emission monitoring precision and efficiency are improved. The pre-training model adopts a multi-layer time convolutional network structure, and the adaptability to a specific construction scene is improved through unsupervised pre-training and supervised fine tuning.

Description

Technical Field

[0001] The present invention belongs to the technical field of electrical digital data processing. Specifically, it relates to a method for monitoring carbon emissions during road construction. Background Art

[0002] A large amount of carbon emissions are generated during road construction. Traditional carbon emission monitoring methods mainly rely on manual data collection and calculation of static carbon emission factors, making it difficult to achieve real-time monitoring and dynamic analysis. A variety of technologies have been introduced into road construction carbon emission monitoring, such as fixed sensor networks, Internet of Things technology, and simple prediction models. However, these technologies are usually based on a single data source, with fixed carbon emission factors and unchanging monitoring frequencies.

[0003] Traditional road construction carbon emission monitoring technologies have obvious defects: First, the accuracy of monitoring data is insufficient, making it difficult to reflect the emission characteristics of different construction stages; second, it is unable to adaptively adjust the monitoring frequency for peak and stable carbon emission periods; third, prediction models generally lack the ability to effectively capture long-term carbon emission trends and mutation situations; fourth, it is difficult to effectively integrate data from multiple construction sites while protecting privacy.

[0004] Facing the complex and changeable road construction environment, how to construct a monitoring method that can adapt to the characteristics of different construction stages, accurately predict the change trend of carbon emissions, and at the same time ensure data privacy has become an urgent technical problem to be solved. Existing technologies are still difficult to achieve accurate identification and prediction of carbon emission patterns, especially during construction stage transitions and carbon emission mutations, where the prediction accuracy drops significantly. That is to say, there is a technical problem in the existing technologies that the monitoring and prediction accuracy of road construction carbon emissions is insufficient and difficult to adapt to the changes in carbon emission characteristics of different construction stages. Summary of the Invention

[0005] In view of this, the present invention provides a method for monitoring carbon emissions during road construction, which can solve the technical problem in the existing technologies that the monitoring and prediction accuracy of road construction carbon emissions is insufficient and difficult to adapt to the changes in carbon emission characteristics of different construction stages.

[0006] The present invention is implemented as follows: The present invention provides a method for monitoring carbon emissions during road construction, which includes the following steps: establishing a carbon emission monitoring network using multi-source sensors during road construction; collecting operation data of road construction equipment, material consumption data, and construction activity data using an Internet of Things terminal; establishing a carbon emission factor database for road construction; calculating the total carbon emissions by combining the collected data with the carbon emission factors to form a carbon emission fingerprint map; performing time series analysis on the carbon emission fingerprint map using edge computing technology to identify abnormal carbon emission points; fine-tuning and training a pre-trained carbon emission prediction model using a neuron jump algorithm and a multi-head attention calculation architecture, integrating carbon emission data from multiple construction sites using a distributed gradient descent optimization equation, and generating a carbon emission monitoring report; wherein the neuron jump threshold is calculated by a jump threshold determination function; wherein, the neuron jump algorithm refers to a calculation method for solving the long-term dependence problem of time series data in the pre-trained carbon emission prediction model, dynamically adjusting the neural network connection weights according to the importance of the carbon emission data at the current moment, and improving the prediction accuracy for carbon emission mutation situations.

[0007] Among them, when establishing a carbon emission monitoring network using multi-source sensors during road construction, the monitoring time interval is determined as the first time interval; adjusting the monitoring time interval and repeating the above steps, the optimized carbon emission monitoring method refers to adjusting the monitoring time interval to the second time interval, and repeating the steps of collecting operation data of road construction equipment, material consumption data, and construction activity data using an Internet of Things terminal until generating a carbon emission monitoring report.

[0008] Among them, the second time interval refers to determining a more accurate monitoring time interval according to the initial monitoring results, shortening the second time interval for the peak carbon emission period and lengthening the second time interval for the stable carbon emission period.

[0009] Among them, the carbon emission fingerprint map refers to a unique distribution pattern formed by the carbon emissions generated by different equipment, materials, and construction activities during road construction over time. The carbon emission characteristics and abnormal carbon emission situations during the construction stage are identified through the carbon emission fingerprint map.

[0010] Among them, the carbon emission factor refers to the calculation parameter for the carbon emissions generated per unit activity, including the carbon emission factor for fuel consumption, the carbon emission factor for electricity consumption, the carbon emission factor for material production, and the carbon emission factor for construction activities.

[0011] Among them, the carbon emission intensity refers to the carbon emissions generated per unit project volume, which is used to evaluate the carbon emission efficiency of road construction and formulate emission reduction targets.

[0012] Among them, the distributed gradient descent optimization equation allows multiple road construction sites to collaboratively train a carbon emission prediction model without sharing the original data, protecting the data privacy of construction units while improving the accuracy of the pre-trained carbon emission prediction model.

[0013] Among them, the multi-head attention calculation architecture refers to the mathematical structure in the pre-trained carbon emission prediction model that captures carbon emission patterns at different time scales. It processes the input data in parallel through multiple independent attention units. Each attention unit focuses on features with different time spans, and finally fuses the outputs of each unit to obtain a comprehensive representation.

[0014] Among them, the fluctuation range of the carbon emission fingerprint map refers to the degree of change in carbon emissions in the carbon emission fingerprint map within a continuous time window, which is calculated by dividing the difference between the maximum value and the minimum value by the average value.

[0015] Among them, the average value of the carbon emission factor refers to the weighted average of all carbon emission factors in the carbon emission factor database, and the weights are determined based on the usage frequency of various equipment, material consumption, and the proportion of construction activities in road construction.

[0016] Compared with the prior art, a method for monitoring carbon emissions in road construction provided by the present invention proposes a method for monitoring carbon emissions in road construction. By constructing a carbon emission monitoring network with multi-source sensors, and combining Internet of Things technology to collect data on the operation of construction equipment, material consumption, and construction activities, a carbon emission fingerprint map is generated. Then, the pre-trained model is fine-tuned by applying the neural unit jump algorithm and the multi-head attention calculation architecture to achieve accurate monitoring and prediction of carbon emissions.

[0017] This method solves the core defects of traditional technologies: First, it accurately identifies the carbon emission characteristics during the construction stage through the carbon emission fingerprint map; Second, it adaptively adjusts the monitoring frequency, shortening the time interval during the peak carbon emission period and extending the time interval during the stable period to improve the monitoring efficiency; Third, the neural unit jump algorithm and the multi-head attention architecture significantly enhance the prediction ability for long-term dependencies and mutation situations; Fourth, the distributed gradient descent optimization equation ensures the effective integration of data from multiple construction sites while protecting data privacy.

[0018] The present invention effectively solves the technical problem of insufficient accuracy in monitoring and predicting carbon emissions in road construction. It not only accurately captures the carbon emission characteristics of different construction stages, but also can adapt to changes in carbon emission patterns, improving the adaptability and accuracy of the prediction model in different construction environments, and providing reliable technical support for carbon emission reduction in road construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention.

[0021] As shown Figure 1 in the figure, it is a flowchart of a road construction carbon emission monitoring method provided by the present invention. This method includes the following steps:

[0022] S01. During the road construction process of this project, a multi-source sensor is used to establish a carbon emission monitoring network, and the monitoring time interval is determined as the first time interval;

[0023] S02. The Internet of Things terminal is used to collect the operation data of road construction equipment, material consumption data, and construction activity data of this project;

[0024] S03. Establish a road construction carbon emission factor database, which includes carbon emission factors corresponding to different construction equipment, materials, and construction activities;

[0025] S04. Calculate the total carbon emissions according to the data collected from this project in combination with the carbon emission factors, and form the carbon emission fingerprint map of this project;

[0026] S05. Use edge computing technology to perform time series analysis on the carbon emission fingerprint map of this project to identify carbon emission abnormal points;

[0027] S06. Use the neuron jump algorithm and the multi-head attention calculation architecture to fine-tune and train the pre-trained carbon emission prediction model according to the historical carbon emission data of this project, where the neuron jump threshold is calculated by the jump threshold determination function, and the input parameters of the jump threshold determination function are the first time interval, the fluctuation range of the carbon emission fingerprint map of this project, and the average value of the carbon emission factors;

[0028] S07. Use the distributed gradient descent optimization equation to integrate the carbon emission data of multiple construction sites;

[0029] S08. Generate the carbon emission monitoring report of this project, which includes the total carbon emissions, carbon emission intensity, and emission reduction suggestions;

[0030] Optionally, it further includes S09. Adjust the monitoring time interval of this project to the second time interval, and repeat steps S02 to S08 to optimize the carbon emission monitoring method.

[0031] Among them, the carbon emission fingerprint map refers to a unique distribution pattern formed by the carbon emissions generated by different equipment, materials, and construction activities during the road construction process over time. The carbon emission characteristics and abnormal carbon emission situations during the construction stage are identified through the carbon emission fingerprint map.

[0032] Among them, the carbon emission factor refers to the calculation parameter of the carbon emissions generated per unit activity amount, including the carbon emission factor of fuel consumption, the carbon emission factor of electricity consumption, the carbon emission factor of material production, and the carbon emission factor of construction activities.

[0033] Among them, the carbon emission intensity refers to the carbon emissions generated per unit of project volume, which is used to evaluate the carbon emission efficiency of road construction and formulate emission reduction targets.

[0034] Among them, the second time interval refers to a more precise monitoring time interval determined according to the initial monitoring results of this project. The second time interval is shortened for the peak carbon emission period of this project and extended for the stable carbon emission period of this project.

[0035] Among them, the distributed gradient descent optimization equation allows multiple road construction sites to collaboratively train the carbon emission prediction model without sharing the original data, protecting the data privacy of construction units while improving the accuracy of the pre-trained carbon emission prediction model.

[0036] Among them, the neuron jump algorithm refers to a calculation method for solving the long-term dependence problem of time series data in the pre-trained carbon emission prediction model. It dynamically adjusts the neural network connection weights according to the importance of carbon emission data at the current moment, improving the prediction accuracy for sudden changes in carbon emissions.

[0037] Among them, the multi-head attention calculation architecture refers to a mathematical structure in the pre-trained carbon emission prediction model for capturing carbon emission patterns at different time scales. It processes the input data in parallel through multiple independent attention units, with each attention unit focusing on features with different time spans, and finally fusing the outputs of each unit to obtain a comprehensive representation.

[0038] Among them, the fluctuation amplitude of the carbon emission fingerprint map refers to the degree of change in carbon emissions in the carbon emission fingerprint map within a continuous time window, which is calculated by dividing the difference between the maximum value and the minimum value by the average value.

[0039] Among them, the average value of carbon emission factors refers to the weighted average of all carbon emission factors in the carbon emission factor database, and the weights are determined based on the usage frequency of various equipment, material consumption, and the proportion of construction activities in road construction.

[0040] Among them, the jump threshold determination function refers to a function for calculating the neuron jump threshold through a system of polynomial equations. The inputs are the first time interval, the fluctuation amplitude of the carbon emission fingerprint map, and the average value of carbon emission factors, and the output is the neuron jump threshold. The system of polynomial equations includes first-order terms, second-order terms, and cross terms, and the coefficients are obtained through regression analysis of data from multiple historical road construction projects.

[0041] The specific structure of the pre-trained carbon emission prediction model is a multi-layer temporal convolutional network structure, which includes an input layer, multiple one-dimensional convolutional layers, multiple skip connection layers, a multi-head attention layer, and an output layer. The input layer receives the temporal data of the carbon emission fingerprint map, the one-dimensional convolutional layer extracts local temporal features, the skip connection layer alleviates the problem of gradient disappearance, the multi-head attention layer assigns different weights to features of different time scales, and the output layer generates the predicted value of carbon emissions in the future time window. The neuron jump threshold is calculated by the jump threshold determination function and is used to control the update frequency of the neuron state.

[0042] The pre-trained carbon emission prediction model is pre-trained based on the carbon emission data of multiple historical road construction projects.

[0043] The steps for establishing the training dataset during the pre-training process of the pre-trained carbon emission prediction model specifically include extracting the temporal carbon emission data from multiple historical road construction projects, cleaning the temporal carbon emission data to remove outliers and missing values, segmenting the temporal carbon emission data according to a fixed-length time window to form training samples, standardizing each training sample to eliminate the influence of dimensions, performing stratified sampling on the training samples according to different construction stages to ensure the balance of data distribution, dividing the processed dataset into a training set, a validation set, and a test set in chronological order, using data augmentation techniques to generate synthetic samples to supplement the training data in sparse data regions, and assigning different weights to the training set samples to make the pre-trained carbon emission prediction model pay more attention to the data patterns during the peak carbon emission period.

[0044] The pre-training steps of the pre-trained carbon emission prediction model specifically include initializing the model parameters using an initialization function to ensure stable gradient propagation. First, perform unsupervised pre-training on a large-scale comprehensive dataset of carbon emissions from multiple historical road construction projects to learn the internal representation of the data. Then, perform supervised fine-tuning on the data of specific construction types to improve the adaptability of the pre-trained carbon emission prediction model to specific construction scenarios. During the training process, use an early stopping strategy to avoid overfitting problems, use a learning rate decay strategy to improve the convergence stability of the pre-trained carbon emission prediction model, use an adversarial training method to improve the robustness of the pre-trained carbon emission prediction model to cope with input data noise interference, use knowledge transfer technology to transfer the knowledge of complex models to lightweight models to meet the deployment requirements of edge devices, and finally use a model integration method to fuse the prediction results of multiple trained models to further improve the prediction accuracy and stability.

[0045] The specific implementation manners of the above steps are described in detail below. The specific implementation manner of step S01 is to deploy a multi-source sensor network system in the initial stage of a road construction project. These sensors include carbon emission gas concentration sensors, equipment operation status sensors, material consumption record sensors, and construction activity monitoring sensors. First, determine the sensor layout positions according to the area and terrain characteristics of the construction site to ensure that the monitoring coverage rate reaches over 95%; then establish a sensor network topology using the ZigBee wireless communication protocol to achieve real-time data transmission; next, configure network edge nodes to collect data from each sensor and perform preliminary filtering; finally, set the initial monitoring time interval as the first time interval, which is usually selected as 15 minutes to ensure that the data collection frequency is sufficient to capture the carbon emission change trend. The function of this step is to build a data collection infrastructure to provide a data source for subsequent carbon emission monitoring.

[0046] The specific implementation manner of step S02 is to collect three types of key data by configuring Internet of Things terminals. First, install equipment operation status monitoring modules on various construction site equipment such as excavators, bulldozers, and rollers to record equipment operation time, power load, and fuel consumption; then install an intelligent weighing system and an RFID tag identification system in the material storage area to track the usage and flow of materials such as asphalt, cement, and steel bars; next, arrange activity monitoring cameras and sensors in the construction area, and combine computer vision technology to identify different construction activity types and durations; finally, all data is transmitted to the edge computing server through a secure encrypted channel and stored in JSON format. The purpose of this step is to comprehensively collect various data elements affecting carbon emissions to provide a data basis for carbon emission calculation.

[0047] The specific implementation manner of step S03 is to build a database containing multi-dimensional carbon emission factors. First, collect equipment carbon emission factor data in national and industry standards, including carbon emissions per unit fuel consumption and carbon emissions per unit power consumption of different models of construction machinery; then sort out material production carbon emission factors, covering the whole life cycle carbon emission data of commonly used road construction materials such as asphalt, concrete, and steel from raw material acquisition to transportation; next, establish a construction activity carbon emission factor library, including carbon emission characteristic parameters of typical construction activities such as subgrade treatment, paving, and compaction; finally, store these data using a relational database and establish an index structure for fast query. This database stores carbon emission factors under different regional and seasonal climate conditions differently to improve the calculation accuracy. The purpose of this step is to provide standardized parameter support for carbon emission calculation.

[0048] The specific implementation of step S04 is to calculate the total carbon emissions and generate a carbon emission fingerprint map using a multi-source data fusion method. First, multiply the equipment operation data by the corresponding equipment carbon emission factor to calculate the carbon emissions generated by equipment operation; then multiply the material consumption data by the corresponding material carbon emission factor to calculate the carbon emissions generated by material use; next, multiply the construction activity data by the activity carbon emission factor to calculate the carbon emissions generated by construction activities; finally, aggregate the above three types of carbon emissions in the time dimension to form a time-series carbon emission data stream, and generate a carbon emission fingerprint map through three-dimensional visualization technology. The x-axis of this map represents time, the y-axis represents different emission source categories, and the z-axis represents the carbon emissions. The purpose of this step is to quantify the carbon emissions during the road construction process and establish a unique carbon emission characteristic map for the project.

[0049] The specific implementation of step S05 is to analyze the time-series characteristics of the carbon emission fingerprint map using edge computing technology to identify abnormal emission situations. First, deploy a sliding window algorithm on the edge server, and set the window size to 8 time points; then apply Z-Score normalization processing to the data within the window to calculate the deviation degree of the carbon emissions at each time point from the window mean; next, set the abnormal threshold to 2.5 standard deviations, and a point exceeding this threshold is determined as an abnormal point; finally, use the DBSCAN density clustering algorithm to analyze the identified abnormal points to determine whether there are abnormal clusters. When abnormal points appear continuously and form a cluster, the system triggers an early warning mechanism. The purpose of this step is to promptly detect abnormal carbon emissions and provide decision-making support for on-site management personnel.

[0050] The specific implementation of step S06 is to perform customized fine-tuning on the pre-trained carbon emission prediction model based on the project's historical data. First, calculate the fluctuation range of the project's carbon emission fingerprint map, which is obtained by dividing the difference between the maximum value and the minimum value by the average value, and this value is usually between 0.3 and 1.5; then calculate the average value of the carbon emission factors, which is determined by weighting and averaging according to the usage frequencies of various types of equipment, the material consumption ratios, and the proportions of construction activities in the project; next, substitute the first time interval, the fluctuation range, and the factor average value into the jump threshold determination function, which is in the form of a second-order polynomial equation system, and the general formula composition includes first-order terms, second-order terms, and cross terms, and finally calculate the neuron jump threshold, and the typical value range of this threshold is from 0.15 to 0.35; finally, apply the neuron jump algorithm and the multi-head attention calculation architecture to fine-tune the pre-trained model, where the jump algorithm can adaptively adjust the neural network connection strength according to the importance of the current carbon emission data, and the multi-head attention architecture captures features at different time scales through 8 parallel attention units. The purpose of this step is to improve the accuracy of the model's carbon emission prediction for a specific project.

[0051] The specific implementation of step S07 is to integrate data from multiple construction sites using the distributed gradient descent optimization method. First, each construction site independently calculates the local data gradient instead of sharing the original data. Then, the calculated gradients are encrypted and transmitted to the central server. Next, the central server integrates all gradient information and updates the global model parameters. Finally, the updated model parameters are distributed to each construction site to complete one round of model training. This process uses homomorphic encryption technology to protect the gradient information, and the differential privacy noise threshold is set to 0.01 to ensure the optimization of the model performance without revealing the original data. The purpose of this step is to improve the generalization ability of the model by using data from multiple construction sites while protecting data privacy.

[0052] The specific implementation of step S08 is to generate a comprehensive and detailed carbon emission monitoring report. First, the total project carbon emissions are summarized, and the emissions from equipment, material-related emissions, and construction activities are calculated separately and summed up. Then, the carbon emission intensity is calculated by dividing the total carbon emissions by the completed project volume to obtain the carbon emission level per unit of project volume. Next, a comparative analysis is carried out with the industry benchmark value to evaluate the carbon emission efficiency of the project. Finally, based on the characteristics of the carbon emission fingerprint map and the analysis results of the abnormal points, targeted emission reduction suggestions are generated, including equipment optimization usage strategies, material selection optimization suggestions, and construction process improvement directions. The report uses visual charts to display key indicators, facilitating managers to quickly understand the carbon emission status of the project. The purpose of this step is to quantitatively evaluate the carbon emission performance of the project and provide emission reduction guidance.

[0053] The specific implementation of step S09 is to optimize the monitoring time interval strategy based on the initial monitoring results. First, analyze the carbon emission change rate in different time periods in the carbon emission fingerprint map. Then, for the peak carbon emission period, such as the period of concentrated equipment operation, shorten the monitoring time interval to the second time interval, usually 5 to 10 minutes. Next, for the stable carbon emission period, such as at night or during low-intensity construction stages, extend the monitoring time interval to 30 to 60 minutes. Finally, re-execute steps S02 to S08 according to the adjusted time interval strategy to optimize the overall monitoring efficiency. This adaptive monitoring strategy ensures high-precision data acquisition during critical periods while reducing system resource consumption. The purpose of this step is to improve the resource utilization efficiency of the monitoring system and ensure data quality.

[0054] The pre-trained carbon emission prediction model adopts a multi-layer temporal convolutional network structure, which is particularly suitable for processing time series data. The input layer receives the time series data of the standardized carbon emission fingerprint map, and its dimension is the product of the time length and the number of features; the one-dimensional convolutional layer contains 4 to 6 layers, and the size of each convolutional kernel is 3 to 7, with a stride of 1, and the ReLU activation function is used to extract local time features; the skip connection layer establishes a direct connection between every two convolutional layers to alleviate the problem of gradient disappearance; the multi-head attention layer contains 8 parallel attention units, and each unit focuses on features of different time scales, and the number of attention heads is usually set to 8; the output layer uses a fully connected layer to generate the predicted carbon emission values for the next 24 to 72 hours. This model is pre-trained with carbon emission data from multiple historical road construction projects and has the ability to model long-term dependence relationships in time series data. The neuron jump algorithm calculates the threshold value through the jump threshold determination function, and the typical range of this value is 0.15 to 0.35, which is used to control the update frequency of the neuron state and improve the prediction accuracy for sudden carbon emission situations.

[0055] The following details the mathematical models or calculation processes involved in the present invention.

[0056] In step S04, the process of calculating the total carbon emissions is specifically represented as follows:

[0057]

[0058] In the formula, E total is the total carbon emissions, with the unit of kilograms of carbon dioxide equivalent; E equip，i is the carbon emissions generated by the operation of the equipment in the i-th time interval; E mat，i is the carbon emissions generated by the material use in the i-th time interval; E act，i is the carbon emissions generated by the construction activities in the i-th time interval; n is the total number of time intervals within the monitoring period.

[0059] Among them, the calculation methods for each item of carbon emissions are as follows:

[0060]

[0061] In the formula, F j，i is the fuel consumption of the j-th equipment in the i-th time interval, with the unit of liters; EF fuel，j is the fuel carbon emission factor of the j-th equipment, with the unit of kilograms of carbon dioxide equivalent per liter; P j，i is the power of the j-th equipment in the i-th time interval, with the unit of kilowatts; T j，i is the operating time of the j-th equipment in the i-th time interval, with the unit of hours; EF elec is the electricity carbon emission factor, with the unit of kilograms of carbon dioxide equivalent per (kilowatt-hour); m is the total number of equipment.

[0062]

[0063] In the formula, M k，i is the usage amount of the k-th material in the i-th time interval, and the unit depends on the material type, such as tons or cubic meters; EF mat，k is the carbon emission factor of the k-th material, with the unit of kilograms of carbon dioxide equivalent per material unit; p is the total number of material types.

[0064]

[0065] In the formula, A l，i is the workload of the l-th type of construction activity in the i-th time interval, and the unit depends on the activity type; EF act，l is the carbon emission factor of the l-th type of construction activity, with the unit of kilograms of carbon dioxide equivalent per activity unit; q is the total number of construction activity types.

[0066] The above carbon emission calculation equation is based on the principle of mass conservation, and quantifies carbon emissions by multiplying activity data by the corresponding emission factors. This linear summation model can comprehensively capture the contributions of different emission sources and is convenient for decomposing and analyzing the proportion of each component.

[0067] In step S05, the Z-Score standardization method is used for outlier identification, which is specifically expressed as follows:

[0068]

[0069] In the formula, Z i is the Z-score at the i-th time point; E i is the carbon emission at the i-th time point; μ w is the mean value of carbon emissions within the sliding window; σ w is the standard deviation of carbon emissions within the sliding window.

[0070] The calculation method of the sliding window mean value is:

[0071]

[0072] In the formula, w is the window size, with a value of 8; E j is the carbon emission at the j-th time point.

[0073] The calculation method of the sliding window standard deviation is:

[0074]

[0075] In the formula, w is the window size, with a value of 8; E j is the carbon emission at the j-th time point; μ w is the mean value within the window.

[0076] The Z-Score method is based on the normal distribution theory. Through standardization, the data mean is set to 0 and the standard deviation is set to 1, which facilitates setting a unified anomaly threshold. The sliding window strategy can adapt to the local change trend of the data and improve the accuracy of anomaly detection. Points with a Z-score exceeding the threshold of 2.5 are determined as anomaly points. This threshold corresponds to the 99% confidence interval of the normal distribution, balancing the risks of false positives and false negatives.

[0077] In step S06, the jump threshold determination function uses a system of polynomial equations, which is specifically expressed as follows:

[0078]

[0079] In the formula, θ thresh is the jump threshold of the neuron, and its range is usually from 0.15 to 0.35; I t is the first time interval, with the unit of minute, and usually takes the value of 15 minutes; A f is the fluctuation amplitude of the carbon emission fingerprint spectrum, and its range is usually from 0.3 to 1.5, dimensionless; is the mean value of the carbon emission factor, with the unit of kg CO₂ equivalent / activity unit; α0 to α9 are polynomial coefficients obtained through regression analysis of historical data; ∈ is the error term, and its range is usually ±0.05.

[0080] The method for obtaining the polynomial coefficients α0 to α9 is through multiple regression analysis of data from multiple historical road construction projects. The specific steps include: collecting carbon emission monitoring data and the corresponding optimal neuron jump thresholds of at least 30 historical road construction projects; extracting the first time interval, the fluctuation amplitude of the carbon emission fingerprint spectrum, and the mean value of the carbon emission factor of each project as independent variables, and the optimal neuron jump threshold as the dependent variable; using the least squares method to fit the system of polynomial equations to determine the coefficient values; using the cross-validation method to verify the generalization performance of the model. Typically, α0 takes a value of about 0.2, and the absolute values of the remaining coefficients are usually less than 0.1.

[0081] The calculation method of the fluctuation amplitude of the carbon emission fingerprint spectrum is:

[0082]

[0083] In the formula, A f is the fluctuation amplitude, dimensionless; E max is the maximum value of the carbon emissions during the monitoring period; E min is the minimum value of the carbon emissions during the monitoring period; is the average value of the carbon emissions during the monitoring period.

[0084] The calculation method of the mean value of the carbon emission factor is:

[0085]

[0086] In the formula, is the average carbon emission factor; EF fuel，j , EF mat，k , EF act，l are the carbon emission factors of equipment fuel, materials, and construction activities respectively; w equip，j , w mat，k , w act，l are the corresponding weight coefficients, and satisfy

[0087] The jump threshold determination function adopts a second-order polynomial form, including first-order terms, second-order terms, and cross terms, which can capture the nonlinear relationship and interaction between variables. The first-order terms reflect the direct influence of each factor, the second-order terms capture the nonlinear effects of single factors, and the cross terms express the interaction between different factors. This complex function form is superior to the simple linear model and can more accurately describe the complex relationship between variables in actual engineering.

[0088] In step S07, the distributed gradient descent optimization equation is expressed as follows:

[0089]

[0090] In the formula, θ t is the parameter of the global model in the t-th round of iteration; θ t+1 is the parameter of the global model in the (t + 1)-th round of iteration; η is the learning rate, usually with a value range of 0.001 to 0.1; K is the number of construction sites participating in collaborative training; is the local gradient calculated by the k-th construction site in the t-th round of iteration; λ is the differential privacy strength parameter, with a value of 0.01; is a normal distribution noise with a mean of 0 and a variance of , usually with a value of 0.0001 to 0.001.

[0091] The local gradient is calculated as follows:

[0092]

[0093] In the formula, n k is the number of samples of the k-th construction site; is the i-th input sample of the k-th construction site; is the corresponding true label; is the model function with parameter θ t ; L is the loss function; represents the gradient with respect to the parameter θ.

[0094] The distributed gradient descent equation is based on the stochastic gradient descent algorithm and updates the global model by aggregating the local gradients of multiple construction sites. A differential privacy noise term is introduced to protect data privacy while maintaining the convergence of the algorithm. This equation achieves a balance between privacy protection and model performance, ensuring efficient model training without sharing the original data.

[0095] In step S08, the calculation method of carbon emission intensity is as follows:

[0096]

[0097] In the formula, I carbon is the carbon emission intensity, with the unit of kilograms of carbon dioxide equivalent per engineering quantity unit; E total is the total carbon emissions, with the unit of kilograms of carbon dioxide equivalent; Q const is the completed engineering quantity, and the unit depends on the type of project, such as square meters or kilometers.

[0098] Carbon emission intensity is a key indicator for evaluating the carbon emission efficiency of a project. By standardizing the processing, the influence of project scale differences is eliminated, facilitating horizontal comparison between different projects. This indicator provides a benchmark for formulating emission reduction targets and plays an important guiding role in promoting the low-carbon development of the industry.

[0099] In step S09, the second time interval adjustment strategy can be expressed as:

[0100]

[0101] In the formula, I t2 is the second time interval, with the unit of minutes; I t1 is the first time interval, with the unit of minutes, usually taking a value of 15 minutes; R e is the carbon emission change rate, with the unit of kilograms of carbon dioxide equivalent per hour; τ1 is the high change rate threshold, usually taking a value of 5 kilograms of carbon dioxide equivalent per hour; τ2 is the low change rate threshold, usually taking a value of 1 kilogram of carbon dioxide equivalent per hour; γ1 is the peak period shortening coefficient, usually taking a value between 0.33 and 0.67; γ2 is the stable period extension coefficient, usually taking a value between 2 and 4.

[0102] The calculation method of the carbon emission change rate is as follows:

[0103]

[0104] In the formula, R e is the carbon emission change rate, with the unit of kilograms of carbon dioxide equivalent per hour; E i is the carbon emissions at the i-th time point; E i-1 is the carbon emissions at the (i - 1)-th time point; I t1 is the first time interval, with the unit of minutes.

[0105] The dynamic time interval adjustment strategy is based on the principle of monitoring efficiency optimization and dynamically allocates monitoring resources according to the carbon emission change rate. The sampling interval is shortened during the period of drastic carbon emission changes to capture detailed changes, and the sampling interval is extended during the stable period to save system resources. This adaptive strategy balances data quality and system efficiency, improving the overall monitoring performance.

[0106] In the pre-trained carbon emission prediction model, the mathematical expression of the multi-head attention calculation architecture is as follows:

[0107]

[0108] In the formula, Q is the query matrix, with dimensions [seq_len, d k ; K is the key matrix, with dimensions [seq_len, d k ; V is the value matrix, with dimensions [seq_len, d v ; d k is the dimension of the key vector; d v is the dimension of the value vector; seq_len is the sequence length, corresponding to the number of time steps.

[0109] The multi-head attention mechanism is calculated as follows:

[0110] MultiHead(Q, K, V) = Concat(head1, head2,..., head h )W O ;

[0111] In the formula, head i = Attention(QE i Q , KW i K , VW i V ); W i Q , W i K , W i V are the parameter matrices of the i-th attention head, which map the original query, key, and value to lower dimensions respectively; W O is the output linear transformation matrix; h is the number of attention heads, with a value of 8.

[0112] The mathematical expression of the neuron jump algorithm is as follows:

[0113]

[0114] In the formula, h t is the hidden state at time t; ht-1 is the hidden state at time t-1; is the candidate hidden state; z t is the update gate, which determines the degree of hidden state update; ⊙ represents element-wise multiplication.

[0115] The update gate is calculated as follows:

[0116] z t = σ(W z · [x t , h t-1 + b z );

[0117] In the formula, z t is the update gate; σ is the sigmoid activation function; W z is the update gate weight matrix; x t is the input at time t; h t-1 is the hidden state at time t-1; b z is the bias term; [x t , h t-1 represents the concatenation of the input and the hidden state of the previous time.

[0118] In the neuron jump algorithm, the core of the jump mechanism lies in:

[0119]

[0120] In the formula, z t is the value of the update gate calculated currently; z t-1 is the value of the update gate at the previous time; θ thresh is the jump threshold, which is calculated by the jump threshold determination function.

[0121] Specifically, the principle of the present invention is: The core of the technical principle of the present invention lies in fusing multi-source data to construct a carbon emission fingerprint map and achieving accurate prediction based on this. First, a monitoring network is established through multi-source sensors, and the Internet of Things terminal collects data on equipment operation, material consumption, and construction activities. Combining with the carbon emission factor database, the total carbon emission is calculated to form a carbon emission fingerprint map. This map reflects the unique distribution pattern of carbon emissions generated by different equipment, materials, and construction activities over time, providing a data basis for subsequent analysis.

[0122] The key technology lies in a prediction model based on a neural unit jump algorithm and a multi-head attention calculation architecture. The neural unit jump algorithm solves the problem of long-term dependence in time series data. By dynamically adjusting the connection weights of the neural network, it improves the prediction accuracy for sudden changes in carbon emissions. The jump threshold in the algorithm is calculated by a jump threshold determination function, which takes the monitoring time interval, the fluctuation amplitude of the carbon emission map, and the average value of the carbon emission factor as input parameters, ensuring the algorithm's adaptability to the characteristics of different construction stages. The multi-head attention calculation architecture processes data in parallel through multiple independent attention units, capturing carbon emission patterns at different time scales and further enhancing the model's ability to recognize complex time series patterns.

[0123] The pre-trained carbon emission prediction model adopts a multi-layer temporal convolutional network structure, including an input layer, multiple one-dimensional convolutional layers, skip connection layers, multi-head attention layers, and an output layer. This structure design effectively solves the problem of gradient disappearance and enhances the ability to capture local time features and long-term dependence relationships. During the pre-training process, through a combination of unsupervised pre-training and supervised fine-tuning, the model learns the internal representation of the data and adapts to specific construction scenarios. At the same time, the application of techniques such as early stopping strategy, learning rate decay, and adversarial training improves the model's generalization ability, convergence stability, and anti-noise interference ability.

[0124] In addition, the present invention introduces a distributed gradient descent optimization equation to achieve collaborative training of data from multiple construction sites, improving the accuracy of the prediction model without sharing the original data and solving the problem of data privacy protection. The adaptive adjustment mechanism for the monitoring time interval shortens the interval during the carbon emission peak period and extends the interval during the stable period, further improving the monitoring efficiency and resource utilization rate.

[0125] The following provides a specific embodiment 1 of the present invention. The specific implementation methods of each step in this embodiment 1 are described in detail as follows.

[0126] The specific implementation methods of steps S01 - S03 are the same as those described above and will not be elaborated here.

[0127] The specific implementation method of step S04 is to use a multi-source data fusion method to calculate the total carbon emissions and generate a carbon emission fingerprint map. This step is based on the principle of mass conservation, and the total carbon emissions are quantified by multiplying the activity data by the corresponding emission factors. The formula for calculating the total carbon emissions is:

[0128]

[0129] In the formula, E total is the total carbon emissions, with the unit of kilograms of carbon dioxide equivalent; E equip，i is the carbon emissions generated by equipment operation within the i-th time interval; E mat，i is the carbon emissions generated by material use within the i-th time interval; Eact，i is the carbon emissions generated by construction activities during the \(i\)th time interval; \(n\) is the total number of time intervals within the monitoring period. The calculation formula for the carbon emissions generated by equipment operation is as follows:

[0130]

[0131] In the formula, \(F\) j，i is the fuel consumption of the \(j\)th equipment during the \(i\)th time interval, with the unit of liters; \(EF\) fuel，j is the fuel carbon emission factor of the \(j\)th equipment, with the unit of kilograms of carbon dioxide equivalent per liter; \(P\) j，i is the power of the \(j\)th equipment during the \(i\)th time interval, with the unit of kilowatts; \(T\) j，i is the operating time of the \(j\)th equipment during the \(i\)th time interval, with the unit of hours; \(EF\) elec is the electricity carbon emission factor, with the unit of kilograms of carbon dioxide equivalent per (kilowatt-hour); \(m\) is the total number of equipment. The calculation formula for the carbon emissions generated by material use is as follows:

[0132]

[0133] In the formula, \(M\) k，i is the usage amount of the \(k\)th material during the \(i\)th time interval, and the unit depends on the material type, such as tons or cubic meters; \(EF\) mat，k is the carbon emission factor of the \(k\)th material, with the unit of kilograms of carbon dioxide equivalent per material unit; \(p\) is the total number of material types. The calculation formula for the carbon emissions generated by construction activities is as follows:

[0134]

[0135] In the formula, \(A\) l，i is the workload of the \(l\)th type of construction activity during the \(i\)th time interval, and the unit depends on the activity type; \(EF\) act，k is the carbon emission factor of the \(l\)th type of construction activity, with the unit of kilograms of carbon dioxide equivalent per activity unit; \(q\) is the total number of construction activity types. Finally, the above three types of carbon emissions are aggregated in the time dimension to form a time-series carbon emission data stream, and a carbon emission fingerprint map is generated through three-dimensional visualization technology. The \(x\)-axis of this map represents time, the \(y\)-axis represents different emission source categories, and the \(z\)-axis represents the carbon emissions. The purpose of this step is to quantify the carbon emissions during the road construction process and establish a unique carbon emission characteristic map for the project.

[0136] The specific implementation method of step S05 is to analyze the time-series characteristics of the carbon emission fingerprint map using edge computing technology to identify abnormal emission situations. This step performs anomaly detection based on the Z-Score normalization method. First, a sliding window algorithm is deployed on the edge server, and the window size is set to 8 time points. The Z-Score normalization calculation formula is as follows:

[0137]

[0138] Wherein, Z i is the Z-score at the i-th time point; E i is the carbon emission at the i-th time point; μ w is the mean value of carbon emissions within the sliding window; σ w is the standard deviation of carbon emissions within the sliding window. The formula for calculating the sliding window mean is:

[0139]

[0140] Wherein, w is the window size, and the value is 8; E j is the carbon emission at the j-th time point. The formula for calculating the sliding window standard deviation is:

[0141]

[0142] Wherein, w is the window size, and the value is 8; E j is the carbon emission at the j-th time point; μ w is the mean value within the window. Then, an anomaly threshold of 2.5 standard deviations is set, and points exceeding this threshold are determined as anomaly points; finally, the DBSCAN density clustering algorithm is used to analyze the identified anomaly points to determine whether there are anomaly clusters. This threshold corresponds to the 99% confidence interval of the normal distribution, balancing the risks of false positives and false negatives. When anomaly points appear continuously and form a cluster, the system triggers an early warning mechanism. The purpose of this step is to timely detect abnormal carbon emission situations and provide decision-making support for on-site management personnel.

[0143] The specific implementation of step S06 is to perform customized fine-tuning on the pre-trained carbon emission prediction model based on project historical data. This step applies the neural unit jump algorithm and the multi-head attention calculation architecture to improve the model prediction accuracy. First, calculate the fluctuation amplitude of the project carbon emission fingerprint map, and the formula is:

[0144]

[0145] Wherein, A f is the fluctuation amplitude, dimensionless; E max is the maximum value of carbon emissions during the monitoring period; E min is the minimum value of carbon emissions during the monitoring period; is the average value of carbon emissions during the monitoring period. This value is usually between 0.3 and 1.5. Then, calculate the mean value of the carbon emission factor, and the formula is:

[0146]

[0147] Wherein, is the mean value of the carbon emission factor; EFfuel，j , EF mat，k , EF act，l are the carbon emission factors of the equipment fuel, materials, and construction activities respectively; w equip，j , w mat，k , w act，l are the corresponding weight coefficients respectively, and satisfy Next, substitute the first time interval, fluctuation amplitude, and factor mean into the jump threshold determination function. This function adopts the form of a second-order polynomial equation system, and the calculation formula is:

[0148]

[0149] In the formula, θ thresh is the neuron jump threshold, and the range is usually from 0.15 to 0.35; I t is the first time interval, with the unit of minute, and usually takes the value of 15 minutes; A f is the fluctuation amplitude of the carbon emission fingerprint spectrum; is the mean value of the carbon emission factor; α0 to α9 are polynomial coefficients obtained through historical data regression analysis; ∈ is the error term, and the range is usually ±0.05. Finally, apply the neuron jump algorithm and the multi-head attention calculation architecture to fine-tune the pre-trained model. The multi-head attention calculation formula is:

[0150]

[0151] In the formula, Q is the query matrix, with the dimension of [seq_len, d k ; K is the key matrix, with the dimension of [seq_len, d k ; V is the value matrix, with the dimension of [seq_len, d v ; d k is the dimension of the key vector; d v is the dimension of the value vector; seq_len is the sequence length, corresponding to the number of time steps. The multi-head attention mechanism calculation formula is:

[0152] MultiHead(Q, K, V) = Concat(head1, head2,..., head h )W O ;

[0153] In the formula, head i = Attention(QW i Q , KW i K , VW i V ); W i Q , W iK , W i V is the parameter matrix of the i-th attention head; W O is the output linear transformation matrix; h is the number of attention heads, with a value of 8. The calculation formula of the neuron jump algorithm is:

[0154]

[0155] In the formula, h t is the hidden state at time t; h t-1 is the hidden state at time t-1; is the candidate hidden state; z t is the update gate, which determines the degree of hidden state update; ⊙ represents element-wise multiplication. The calculation formula of the update gate is:

[0156] z t = σ(W z · [x t , h t-1 + b z );

[0157] In the formula, z t is the update gate; σ is the sigmoid activation function; W z is the update gate weight matrix; x t is the input at time t; h t-1 is the hidden state at time t-1; b z is the bias term; [x t , h t-1 represents the concatenation of the input and the hidden state at the previous moment. The core formula of the neuron jump mechanism is:

[0158]

[0159] In the formula, z t is the value of the update gate calculated currently; z t-1 is the value of the update gate at the previous moment; θ thresh is the jump threshold. The purpose of this step is to improve the accuracy of the model's carbon emission prediction for specific projects.

[0160] The specific implementation of step S07 is to use the distributed gradient descent optimization method to integrate data from multiple construction sites while protecting data privacy. The distributed gradient descent optimization equation is:

[0161]

[0162] In the formula, θ t is the parameter of the global model in the t-th round of iteration; θ t+1is the parameter of the global model at the (t + 1)-th iteration; η is the learning rate, usually ranging from 0.001 to 0.1; K is the number of construction sites participating in collaborative training; is the local gradient calculated by the k-th construction site at the t-th iteration; λ is the differential privacy intensity parameter, with a value of 0.01; is a normal distribution noise with a mean of 0 and a variance of ; usually ranges from 0.0001 to 0.001. The formula for calculating the local gradient is:

[0163]

[0164] In the formula, n k is the number of samples at the k-th construction site; is the i-th input sample at the k-th construction site; is the corresponding true label; is the model function with parameter θ t ; L is the loss function; denotes the gradient with respect to the parameter θ. First, each construction site independently calculates the local data gradient instead of sharing the original data; then, the calculated gradients are encrypted and transmitted to the central server; next, the central server integrates all gradient information and updates the global model parameters; finally, the updated model parameters are distributed to each construction site to complete one round of model training. This process uses homomorphic encryption technology to protect the gradient information, sets the differential privacy noise threshold to 0.01, and ensures the optimization of the model performance without revealing the original data. The purpose of this step is to improve the model generalization ability by using the data of multiple construction sites while protecting data privacy.

[0165] The specific implementation of step S08 is to generate a comprehensive and detailed carbon emission monitoring report. First, sum up the total project carbon emissions, calculate the equipment emissions, material-related emissions, and construction activity emissions separately and then sum them up; then calculate the carbon emission intensity, and the calculation formula is:

[0166]

[0167] In the formula, I carbon is the carbon emission intensity, with the unit of kg CO₂ equivalent / project quantity unit; E total is the total carbon emissions, with the unit of kg CO₂ equivalent; Q constTo complete the engineering quantity, the unit depends on the project type, such as square meters or kilometers. Then, a comparative analysis is conducted with the industry benchmark value to evaluate the project's carbon emission efficiency. Finally, based on the characteristics of the carbon emission fingerprint and the analysis results of the abnormal points, targeted emission reduction suggestions are generated, including optimized equipment usage strategies, optimized material selection suggestions, and directions for improving construction techniques. This report uses visual charts to display key indicators, facilitating project managers to quickly understand the project's carbon emission status. The purpose of this step is to quantitatively evaluate the project's carbon emission performance and provide emission reduction guidance.

[0168] The specific implementation of step S09 is to optimize the monitoring time interval strategy based on the initial monitoring results. The formula for the second time interval adjustment strategy is:

[0169]

[0170] In the formula, I t2 is the second time interval, in minutes; I t1 is the first time interval, in minutes, usually taking a value of 15 minutes; R e is the carbon emission change rate, in kilograms of carbon dioxide equivalent per hour; τ1 is the high change rate threshold, usually taking a value of 5 kilograms of carbon dioxide equivalent per hour; τ2 is the low change rate threshold, usually taking a value of 1 kilogram of carbon dioxide equivalent per hour; γ1 is the peak period shortening coefficient, usually taking a value between 0.33 and 0.67; γ2 is the stable period extension coefficient, usually taking a value between 2 and 4. The formula for calculating the carbon emission change rate is:

[0171]

[0172] In the formula, R e is the carbon emission change rate, in kilograms of carbon dioxide equivalent per hour; E i is the carbon emission at the i-th time point; E i-1 is the carbon emission at the (i - 1)-th time point; I t1 is the first time interval, in minutes. First, analyze the carbon emission change rate in different time periods in the carbon emission fingerprint. Then, for the peak carbon emission period, such as when equipment is operating intensively, shorten the monitoring time interval to the second time interval, usually 5 to 10 minutes. Next, for the stable carbon emission period, such as at night or during low-intensity construction stages, extend the monitoring time interval to 30 to 60 minutes. Finally, re-execute steps S02 to S08 according to the adjusted time interval strategy to optimize the overall monitoring efficiency. This adaptive monitoring strategy ensures high-precision data acquisition during critical periods while reducing system resource consumption. The purpose of this step is to improve the resource utilization efficiency of the monitoring system and ensure data quality.

[0173] The pre-trained carbon emission prediction model adopts a multi-layer temporal convolutional network structure, which is particularly suitable for processing time series data. The input layer receives the time series data of the standardized carbon emission fingerprint map, and the dimension is the time length multiplied by the number of features; the one-dimensional convolutional layer contains 4 to 6 layers, the size of each convolutional kernel is 3 to 7, the stride is 1, and the ReLU activation function is used to extract local time features; the skip connection layer establishes a direct connection between every two convolutional layers to alleviate the problem of gradient disappearance; the multi-head attention layer contains 8 parallel attention units, each unit focuses on features of different time scales, and the number of attention heads is usually set to 8; the output layer uses a fully connected layer to generate the carbon emission prediction values for the next 24 to 72 hours. The model is pre-trained with the carbon emission data of multiple historical road construction projects and has the ability to model the long-term dependence relationship of time series data. The neuron jump algorithm calculates the threshold value through the jump threshold determination function, and the typical range of this value is 0.15 to 0.35, which is used to control the neuron state update frequency and improve the prediction accuracy of sudden carbon emission situations.

[0174] Through the above specific implementation manners, the carbon emission monitoring method for road construction realizes the full-process precise monitoring from data collection, processing, analysis to prediction, can effectively identify abnormal carbon emission situations, predict future emission trends, and provide targeted emission reduction suggestions. This method combines Internet of Things technology, edge computing, deep learning and distributed optimization technology, improves the model prediction accuracy while protecting data privacy, and is applicable to the carbon emission monitoring and management of various road construction projects.

[0175] To better understand and implement the present invention, the following provides Embodiment 2 of a specific application scenario of the present invention: A certain mountainous expressway project is 58 kilometers long and passes through multiple ecologically sensitive areas. In order to meet the local environmental protection requirements, the project leader needs to accurately monitor the carbon emissions during the construction process and adjust the construction plan in real time. The researchers decided to adopt the carbon emission monitoring method for road construction to monitor the carbon emissions throughout the process.

[0176] First, at the project startup stage, the researchers deployed a multi-source sensor network according to the characteristics of the construction sections. A set of gas monitoring stations was installed every 800 meters in the main construction areas, with a total of 72 monitoring points. Each monitoring station was equipped with a carbon emission gas concentration sensor, an environmental parameter sensor, etc. At the same time, equipment operation state sensors were installed on 28 excavators, 15 bulldozers, and 10 rollers, which are the main equipment, and an intelligent weighing system was arranged in 3 material yards. All sensors were networked using the ZigBee protocol, and the data was collected through 12 edge computing nodes. The initial data collection time interval was set to 15 minutes, and the network coverage rate reached 97.8%.

[0177] After the data acquisition system was built, various types of data began to be collected. Taking a certain construction day as an example, the equipment operation data collected during 8:00 - 18:00 is shown in Table 1:

[0178] Table 1 Main Equipment Operation Data (Partial)

[0179] Equipment ID Equipment type Power (kW) Operation time (h) Fuel consumption (L) Time period E001 Excavator 125 3.5 42.8 8:00-11:30 E002 Excavator 132 4.2 56.9 8:00-12:12 E008 Bulldozer 98 6.8 65.3 8:30-15:18 E015 Roller 76 5.5 38.6 10:00-15:30 E022 Asphalt paver 145 4.3 58.7 13:00-17:18

[0180] Meanwhile, material consumption data was collected, including the usage amounts of asphalt, cement, steel, etc., as shown in Table 2:

[0181] Table 2 Main Material Consumption Data (Partial)

[0182] Material type Usage amount (t) Usage section Time period Asphalt mixture 128.5 K25+600 - K26+200 13:00-17:00 Cement 86.3 K24+800 - K25+300 9:00-12:00 Steel bars 12.8 K26+000 - K26+500 8:30-11:30 Crushed stones 325.6 K24+500 - K26+500 8:00-16:00

[0183] The researchers established a carbon emission factor database based on national standards and industry specifications, covering carbon emission factors for different equipment, materials, and construction activities. Part of the data is shown in Table 3:

[0184] Table 3 Main Carbon Emission Factor Data

[0185] Category Project <![CDATA[Carbon emission factor (kgCO2e / unit)]]> Unit Fuel Diesel (excavator) 2.68 L Fuel Diesel (bulldozer) 2.72 L Fuel Diesel (roller) 2.65 L Electricity Grid power supply 0.785 kWh Material Asphalt mixture 118.5 t Material Cement 842.3 t Material Steel bars 1652.8 t Construction activity Earth excavation 6.8 <![CDATA[m 3 > Construction activity Subgrade filling 4.2 <![CDATA[m 3 > Construction activity Asphalt paving 12.4 <![CDATA[m 2 >

[0186] Based on the collected data, the researchers calculated the total carbon emissions. Taking the above data as an example, the carbon emissions from equipment operation on that day were calculated as follows: E1 = 42.8×2.68 + 56.9×2.68 + 65.3×2.72 + 38.6×2.65 + 58.7×2.68 = 704.64 kgCO2e. The carbon emissions from material usage: E2 = 128.5×118.5 + 86.3×842.3 + 12.8×1652.8 = 103175.23 kgCO2e. The carbon emissions from construction activities: E3 = 680×6.8 + 520×4.2 + 3600×12.4 = 52984 kgCO2e. The total carbon emissions on that day: E1 + E2 + E3 = 156863.87 kgCO2e.

[0187] The researchers integrated the monitoring data for 30 consecutive days to generate a carbon emission fingerprint map and performed time series analysis through edge computing technology. The sliding window Z - Score method was used to identify abnormal points, and the abnormal threshold was set at 2.5 standard deviations. During the analysis, it was found that there was an abnormally high emission during the period from 10:30 to 12:00 in the morning on the 12th day, with a Z - score of 3.2, exceeding the threshold. Further analysis found that the gravel transport fleet had idle waiting due to a temporary road blockage during this period, resulting in low fuel efficiency.

[0188] To improve the prediction accuracy, the researchers fine-tuned the pre-trained carbon emission prediction model. First, calculate the fluctuation amplitude of the project carbon emission fingerprint map \(A_f=(185642 - 95328) / 133975 = 0.675\). Then calculate the average value of the carbon emission factor. According to the equipment usage frequency, material consumption ratio, and construction activity ratio, the weight coefficients are determined to be 0.15, 0.65, and 0.2 respectively, and the calculated average value is 582.4 kgCO2e. Substitute the first time interval (15 minutes), the fluctuation amplitude (0.675), and the factor average value (582.4) into the jump threshold determination function, and calculate that the neuron jump threshold is 0.23.

[0189] When fine-tuning the model using the neuron jump algorithm and the multi-head attention calculation architecture, the multi-head attention mechanism is used, and the number of attention heads is set to 8. Each attention head focuses on features at different time scales. The neuron jump algorithm dynamically adjusts the neural network connection weights according to the importance of the current carbon emission data, and updates the state when the change in the update gate value is greater than the jump threshold of 0.23.

[0190] To integrate data from multiple construction sites while protecting data privacy, the researchers adopted the distributed gradient descent optimization method. This project combined 3 similar road construction projects in the vicinity to jointly train the model. Each project independently calculates the local gradient and then encrypts and transmits it to the central server. Set the differential privacy strength parameter \(\lambda = 0.01\) and the noise variance \(\sigma_p\) 2 \(= 0.0005\) to ensure data privacy protection. After 50 rounds of iteration, the prediction accuracy of the global model increased by 18.6%.

[0191] Based on the collected data and analysis results, the researchers generated a carbon emission monitoring report. The carbon emission intensity was calculated in the report. Taking the road surface paving as an example, the carbon emission intensity \(I = 156863.87\div8500 = 18.45\) kgCO2e / m 2 , which is 13.4% lower than the industry benchmark value of 21.3 kgCO2e / m 2 . According to the characteristics of the carbon emission fingerprint map, targeted emission reduction suggestions were put forward, including optimizing equipment idle management, adjusting the material supply chain, and improving the construction process.

[0192] Based on the initial monitoring results, the researchers optimized the monitoring time interval. For the peak carbon emission periods from 8:00 to 10:00 in the morning and from 14:00 to 16:00 in the afternoon, the carbon emission change rate \(R_e = |162.5 - 128.3|\div(15 / 60)=136.8\ kgCO_2e / h\), which exceeds the high change rate threshold of \(5\ kgCO_2e / h\), so the monitoring interval was shortened to 5 minutes; for the stable carbon emission period from 22:00 to 6:00 at night, the carbon emission change rate \(R_e = |12.8 - 12.5|\div(15 / 60)=1.2\ kgCO_2e / h\), which is lower than the low change rate threshold of \(1\ kgCO_2e / h\), so the monitoring interval was extended to 45 minutes.

[0193] Traditional carbon emission monitoring in road construction mainly uses static emission factors and regular sampling detection, which has problems such as strong data lag, low accuracy, and inability to provide real-time feedback. Usually, emission calculations are carried out once a week or a month, and the calculation method is simple, unable to capture the emission characteristics and anomalies during the construction process. In contrast, the present invention realizes real-time data collection through a multi-source sensor network, rapid identification of abnormal points through edge computing technology, improved prediction accuracy through a neural unit jump algorithm and a multi-head attention calculation architecture, and protection of data privacy through a distributed gradient descent optimization method. During the implementation process, the present invention has achieved a 68.5% improvement in monitoring accuracy compared with traditional methods, the abnormal identification time has been shortened from an average of 24 hours to 15 minutes, the prediction accuracy has been increased by 25.7%, and the emission reduction potential identification efficiency has been increased by 42.3%. In addition, the strategy of dynamically adjusting the monitoring time interval saves 38.6% of computing resources and 34.2% of data storage space compared with fixed-interval monitoring, while ensuring the data accuracy during key periods, reflecting the significant progress of the present invention in the field of carbon emission monitoring in road construction.

[0194] It should be noted that the detailed explanations of the variables involved in the present invention are shown in Tables 4 and 5 below.

[0195] Table 4 Variable Explanation Table (First Part)

[0196]

[0197]

[0198] Table 5 Variable Explanation Table (Second Part)

[0199]

[0200] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention.

Claims

1. A road construction carbon emission monitoring method, characterized in that: The following steps are involved: During the road construction process, multi-source sensors are used to establish a carbon emission monitoring network; IoT terminals are used to collect road construction equipment operation data, material consumption data and construction activity data; a road construction carbon emission factor database is established; the total carbon emissions are calculated based on the collected data combined with the carbon emission factors to form a carbon emission fingerprint map; edge computing technology is used to perform time series analysis on the carbon emission fingerprint map to identify carbon emission anomalies; the neural unit jump algorithm and multi-head attention computing architecture are used to fine-tune the pre-trained carbon emission prediction model based on historical carbon emission data, and the distributed gradient descent optimization equation is used to integrate the carbon emission data of multiple construction sites and generate a carbon emission monitoring report; The neural unit jump threshold is calculated by the jump threshold determination function; wherein the neural unit jump algorithm refers to a calculation method for solving the long-term dependency problem of time series data in the pre-trained carbon emission prediction model, and dynamically adjusts the neural network connection weights according to the importance of carbon emission data at the current moment to improve the prediction accuracy of carbon emission mutations.

2. The road construction carbon emission monitoring method according to claim 1, characterized in that: The method of optimizing carbon emission monitoring comprises adopting multi-source sensors to establish a carbon emission monitoring network during road construction and determining the monitoring time interval as a first time interval; adjusting the monitoring time interval and repeating the above steps refer to adjusting the monitoring time interval to a second time interval and repeating the steps of using the Internet of Things terminal to collect road construction equipment operation data, material consumption data and construction activity data to generate a carbon emission monitoring report.

3. The road construction carbon emission monitoring method according to claim 2 is characterized in that: The second time interval refers to a more accurate monitoring time interval determined based on the initial monitoring results. The second time interval is shortened during the peak period of carbon emissions, and the second time interval is extended during the stable period of carbon emissions.

4. The road construction carbon emission monitoring method according to claim 3 is characterized in that: The carbon emission fingerprint refers to the unique distribution pattern of carbon emissions generated by different equipment, materials and construction activities during road construction that changes over time. The carbon emission fingerprint is used to identify carbon emission characteristics and abnormal carbon emissions during the construction phase.

5. The road construction carbon emission monitoring method according to claim 4 is characterized in that: The carbon emission factor refers to the calculation parameters of carbon emissions generated per unit of activity, including the carbon emission factor of fuel consumption, the carbon emission factor of electricity consumption, the carbon emission factor of material production and the carbon emission factor of construction activities.

6. The road construction carbon emission monitoring method according to claim 5 is characterized in that: The carbon emission intensity refers to the carbon emissions generated per unit of engineering work, which is used to evaluate the carbon emission efficiency of road construction and set emission reduction targets.

7. The road construction carbon emission monitoring method according to claim 6 is characterized in that: The distributed gradient descent optimization equation allows multiple road construction sites to collaboratively train carbon emission prediction models without sharing original data, protecting the data privacy of construction units while improving the accuracy of pre-trained carbon emission prediction models.

8. The road construction carbon emission monitoring method according to claim 7 is characterized in that: The multi-head attention computing architecture refers to a mathematical structure in a pre-trained carbon emission prediction model that captures carbon emission patterns at different time scales. It processes input data in parallel through multiple independent attention units. Each attention unit focuses on different time span features, and finally integrates the outputs of each unit to obtain a comprehensive representation.

9. The road construction carbon emission monitoring method according to claim 8, characterized in that: The fluctuation range of the carbon emission fingerprint spectrum refers to the degree of change of carbon emissions in the carbon emission fingerprint spectrum within a continuous time window, which is calculated by dividing the difference between the maximum and minimum values ​​by the average value.

10. The road construction carbon emission monitoring method according to claim 9, characterized in that: The carbon emission factor mean refers to the weighted average of all carbon emission factors in the carbon emission factor database, and the weights are determined based on the frequency of use of various equipment in road construction, material consumption and the proportion of construction activities.

Images