Highway engineering slope greening carbon sink quantification method, device, equipment and medium

Abstract

The invention relates to a highway engineering slope greening carbon sink quantification method and device, equipment and a medium. The method comprises the following steps: obtaining highway engineering slope multi-source spatio-temporal data to form a data set, dividing data credibility classification according to precision, integrity and timeliness, constructing a dynamic weight matrix based on the result, importing the data set to obtain an assimilated spatio-temporal data set, calibrating carbon sink related parameters by combining slope landform features and ecological partitions, and obtaining a slope model. And inputting the calibrated parameter set into a pre-trained random forest model to obtain the comprehensive carbon sink quantity. By adopting the method, the parameter uncertainty can be systematically eliminated, the carbon sink quantification precision and generalization ability are improved, the effective response to the dynamic change of the carbon sink is realized, and the low-carbon evaluation and long-term monitoring requirements of the highway engineering are met.

Description

Technical Field

[0001] This invention belongs to the field of highway engineering technology, and in particular relates to a method, device, equipment and medium for quantifying carbon sequestration in highway slope greening. Background Technology

[0002] With the advancement of low-carbon highway engineering, the carbon sequestration quantification technology for slope greening has gradually developed, forming a quantification method based on vegetation surveys and soil data collection, combined with simple biomass models, providing preliminary support for the low-carbon assessment of engineering projects.

[0003] In traditional quantitative methods, information such as vegetation type and density is often obtained through manual field surveys. Soil sampling is combined with laboratory analysis to determine soil organic carbon content and bulk density, and then parameter formulas are used for calculation. Furthermore, different sub-items such as roadbed slopes and borrow pit slopes are often treated with uniform parameters and logic, lacking differentiated processing.

[0004] However, traditional methods have significant limitations: First, the data foundation is fragile. The technology relies heavily on idealized assumptions about input parameters but lacks a systematic mechanism to eliminate parameter uncertainties. In practice, there are problems such as model simplification and subjective assumptions about probability distributions, which leads to a nonlinear decay in the reliability of carbon sequestration results with parameter errors. Second, the dynamic response capability is weak. In slope operation, vegetation growth cycles, pests and diseases, extreme weather, and maintenance pruning can change carbon sequestration capacity. However, traditional methods use static data and fixed models for calculation, which cannot reflect the dynamic changes in carbon sequestration, resulting in a large deviation from actual benefits and making it difficult to meet the needs of long-term monitoring. Summary of the Invention

[0005] Based on this, it is necessary to provide a method, device, equipment, and medium for quantifying carbon sequestration in highway engineering slope greening by systematically eliminating parameter uncertainties and achieving dynamic response through credibility classification and dynamic weight assimilation of multi-source spatiotemporal data, combined with terrain ecological calibration and random forest model calculation.

[0006] Firstly, this application provides a method for quantifying carbon sequestration in highway engineering slope greening, including:

[0007] Multi-source spatiotemporal data of highway engineering slopes are acquired to obtain a multi-source spatiotemporal dataset;

[0008] The data credibility levels of multi-source spatiotemporal datasets are classified according to data accuracy, completeness and timeliness to obtain data quality classification results;

[0009] Construct a dynamic weight matrix based on the data quality grading results;

[0010] The multi-source spatiotemporal datasets are imported into the dynamic weight matrix according to the corresponding mapping relationship to obtain the assimilated spatiotemporal datasets;

[0011] Based on the topographic features and ecological zoning of the slope, the carbon sink-related parameters in the assimilated spatiotemporal dataset are calibrated to obtain the calibrated carbon sink parameter set.

[0012] The calibrated carbon sink parameter set is input into a pre-trained random forest model to obtain the comprehensive carbon sink amount.

[0013] In one embodiment, the pre-trained random forest model is obtained through the following method:

[0014] To obtain multi-period historical monitoring data of highway engineering slopes, the multi-period historical monitoring data includes vegetation data of multiple growth cycles, soil parameter data of multiple soil layers, and record data of all types of disturbance events, thus obtaining a multi-period historical dataset.

[0015] Preprocess the multi-period historical dataset to obtain the preprocessed historical dataset;

[0016] The preprocessed historical dataset is divided into training dataset and test dataset according to the spatiotemporal dimension.

[0017] The training set is input into the initial random forest model for training, resulting in the trained random forest model;

[0018] The test dataset is input into the trained random forest model, and the weight allocation is adjusted based on the carbon sink contribution error to obtain the pre-trained random forest model.

[0019] In one embodiment, the random forest model includes:

[0020] The LSTM feature extraction module is used to retrieve the preprocessed historical dataset, extract the time-series feature data of the carbon sink parameters, and associate the time-series feature data with the corresponding divided training dataset to output a joint training dataset containing time-series features.

[0021] The model adaptation and adjustment module is used to input the joint training dataset containing time-series features into the initial LSTM model, adjust the LSTM model parameters by combining the carbon sink contribution error calculation method of the random forest model, and output a pre-trained LSTM model adapted to the pre-trained random forest model.

[0022] The temporal evolution analysis module is used to extract real-time time-series data of carbon sink-related parameters from the assimilated spatiotemporal dataset. The real-time time-series data is input into a pre-trained LSTM model, which is used to analyze the temporal evolution law of carbon sink parameters under dynamic disturbances and output the carbon sink temporal evolution results.

[0023] The feature fusion module is used to incorporate the carbon sink time-series evolution results as supplementary features into the calibrated carbon sink parameter set and output a joint feature parameter set containing time-series information.

[0024] The joint computation module is used to input the joint feature parameter set containing time-series information into the pre-trained random forest model, recalculate the carbon sink, and output the optimized comprehensive carbon sink that takes into account both spatiotemporal characteristics and dynamic evolution laws.

[0025] In one embodiment, a multi-source spatiotemporal dataset is imported into a dynamic weight matrix according to a corresponding mapping relationship to obtain an assimilated spatiotemporal dataset, including:

[0026] Spatiotemporal coordinate alignment is performed based on multi-source spatiotemporal datasets to obtain a spatiotemporally aligned dataset;

[0027] Based on the dynamic weight matrix, the spatiotemporally aligned dataset is weighted to obtain weighted fused data.

[0028] The weighted fused data is corrected for errors using the Kalman filter algorithm to obtain the assimilated spatiotemporal dataset.

[0029] In one embodiment, the multi-source spatiotemporal dataset is classified into data reliability levels according to data accuracy, completeness, and timeliness to obtain data quality classification results, including:

[0030] Statistical features are extracted from multi-period historical monitoring data to obtain a set of feature parameters;

[0031] By analyzing the spatial variation and temporal stability of the feature parameter set, a hierarchical benchmark matrix is ​​obtained;

[0032] Quality features are extracted from multi-source spatiotemporal datasets to obtain a set of quality evaluation indicators;

[0033] The quality grading index is obtained by matching the quality evaluation index set with the grading benchmark matrix using the following formula:

[0034] QI=ω p ˙μ p (P) + ω c ˙μ c (C) + ω t ˙μ t (T) + λ˙

[0035] Where QI is the quality grading index, ω p For precision weights, ω c For integrity weight, ω t For timeliness weighting, μ p (P) is the precision function, μ c (C) is the integrity function, μ t (T) is the timeliness function, λ is the dynamic correction coefficient, k is the attenuation coefficient, t0 is the current timestamp, and t is the data acquisition timestamp;

[0036] The quality grading index is dynamically normalized to obtain the data quality grading results.

[0037] In one embodiment, based on the slope topography and ecological zoning, carbon sink-related parameters in the assimilated spatiotemporal dataset are calibrated to obtain a calibrated carbon sink parameter set, including:

[0038] Geomorphological features are extracted from the topographic features of the slope to obtain a slope zoning feature dataset.

[0039] By fusing the slope zoning feature dataset with the ecological zoning dataset, the ecological zoning results of the slope are obtained;

[0040] Based on the ecological zoning results of slopes, the assimilated spatiotemporal data are divided and classified to obtain basic data for slope zoning.

[0041] Carbon sink-related parameters are extracted from the assimilated spatiotemporal dataset to obtain the set of carbon sink parameters to be calibrated;

[0042] Based on the slope ecological zoning results, the set of carbon sink parameters to be calibrated is classified to obtain the parameters to be calibrated corresponding to each zoning.

[0043] Based on multi-period historical monitoring data, the historical reasonable range of carbon sink parameters for each ecological zone is extracted to obtain the historical range of zone parameters;

[0044] Compare the parameters to be calibrated for each partition with the historical range of the partition parameters, adjust the parameters that are outside the historical range of the partition parameters, and obtain the preliminary calibration carbon sink parameter set;

[0045] Calculate the temporal fluctuation range of parameters within the same ecological zone in the preliminary calibration carbon sink parameter set, correct parameters whose fluctuation range exceeds the preset range, and obtain the calibrated carbon sink parameter set.

[0046] In one embodiment, the calibrated carbon sink parameter set is input into a pre-trained random forest model to obtain the comprehensive carbon sink amount, including:

[0047] Extract the spatial distribution matrix and time series tensor of the calibrated carbon sink parameter set to obtain the spatiotemporal feature tensor of the carbon sink;

[0048] Spatiotemporal dimension feature encoding is performed on the calibrated carbon sink parameter set to obtain a joint feature matrix that fuses the spatial distribution matrix and the time series tensor;

[0049] The joint feature matrix is ​​input into a pre-trained random forest model, and the comprehensive carbon sink is obtained through parallel computation using multiple decision trees.

[0050] Secondly, this application also provides a quantification device for carbon sequestration in highway engineering slope greening, comprising:

[0051] The data acquisition module is used to acquire multi-source spatiotemporal data of highway engineering slopes to obtain a multi-source spatiotemporal dataset.

[0052] The grading module is used to classify the data credibility of multi-source spatiotemporal datasets according to data accuracy, completeness and timeliness, and obtain data quality grading results.

[0053] The matrix construction module is used to construct a dynamic weight matrix based on the data quality classification results.

[0054] The data assimilation module is used to input multi-source spatiotemporal datasets into a weight matrix to obtain assimilated spatiotemporal datasets;

[0055] The calibration module is used to calibrate carbon sink-related parameters in the assimilated spatiotemporal dataset based on the topographic features and ecological zoning of the slope, and to obtain the calibrated carbon sink parameter set.

[0056] The integrated carbon sink calculation module is used to input the calibrated carbon sink parameter set into a pre-trained random forest model to obtain the integrated carbon sink.

[0057] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods in the first aspect of this application.

[0058] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods in the first aspect of this application.

[0059] This application provides a quantitative method for carbon sequestration in highway engineering slope greening. Through multi-source spatiotemporal data fusion and a reliability grading mechanism, a dynamic weight matrix is ​​systematically constructed to achieve data assimilation, effectively eliminating the reliability defects caused by nonlinear decay due to parameter uncertainty in traditional methods. Carbon sequestration parameters are calibrated by combining slope topographic features and ecological zoning, overcoming the lack of differentiation caused by a unified processing mode. Utilizing the parallel computing capabilities of a pre-trained random forest model with multiple decision trees, the nonlinear correlation characteristics between spatiotemporal data are fully explored, significantly improving the accuracy and generalization ability of carbon sequestration quantification under complex conditions. Simultaneously, this method establishes a dynamic response framework for long-term monitoring, capable of continuously adapting to spatiotemporal evolution factors such as vegetation growth cycles and extreme climates, overcoming the limitations of static data and fixed models, and providing a highly robust and adaptable technical solution for assessing carbon sequestration in highway engineering slope greening. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 A schematic diagram of an implementation environment provided for one embodiment of the present invention;

[0062] Figure 2 This is a flowchart illustrating a method for quantifying carbon sequestration in highway engineering slope greening, as described in one embodiment of the present invention.

[0063] Figure 3 This is a schematic diagram of the structure of a carbon sequestration quantification device for highway engineering slope greening in one embodiment of the present invention. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0065] First, a brief introduction to the terms used in the embodiments of this application will be given.

[0066] Random Forest model: It is an ensemble learning algorithm that combines prediction results by constructing a large number of independent decision trees. For classification tasks, a voting method is used, and for regression tasks, the average value is taken, thereby improving the model's accuracy and controlling overfitting. At the same time, it uses bootstrap sampling and random feature selection to enhance generalization ability and robustness.

[0067] LSTM (Long Short-Term Memory) feature extraction: It uses long short-term memory networks to process time-series data step by step, and automatically captures long-term dependencies and dynamic change patterns through gating mechanisms, transforming the input sequence into a feature representation containing contextual semantics.

[0068] The method for quantifying carbon sequestration in highway engineering slope greening provided in this application embodiment can be applied to, for example... Figure 1 The application environment shown. The method for quantifying carbon sequestration in highway engineering slope greening provided in this application embodiment can be applied to, for example, Figure 1In the application environment shown, sensor 101 and camera 102 are connected to terminal 100 via a network. Terminal 100 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. Sensors can be, but are not limited to, vegetation growth monitoring sensors, soil parameter monitoring sensors, environmental and disturbance factor monitoring sensors, spatial location sensors, and terrain sensors.

[0069] In one exemplary embodiment, such as Figure 2 As shown, a quantitative method for carbon sequestration in highway engineering slope greening is provided, and this method is applied to... Figure 1 Taking terminal 100 as an example, the method includes:

[0070] S101: Obtain multi-source spatiotemporal data of highway engineering slopes to obtain a multi-source spatiotemporal dataset.

[0071] The multi-source spatiotemporal data can be obtained through devices such as sensors 101 and cameras 102, including multi-growth-cycle vegetation data, multi-soil-layer soil parameter data, all types of disturbance event records, and topographic and geomorphological feature data of highway engineering slopes. These data collectively constitute the basic dataset for quantifying carbon sinks in the spatiotemporal dimension. For example, by deploying various types of devices such as vegetation growth monitoring sensors, soil parameter monitoring sensors, environmental and disturbance factor recording devices, topographic sensors, and cameras in the highway engineering slope area, vegetation data such as vegetation type, density, and growth status in multiple growth cycles, soil parameter data such as soil organic carbon content and bulk density in multiple soil layers, records of the occurrence time, impact range, and degree of all types of disturbance events such as pests and diseases, extreme climate, and maintenance pruning, as well as topographic and geomorphological feature data such as slope gradient, aspect, and elevation, these data covering different dimensions and interrelated in time and space are summarized and organized. After removing obviously invalid data, a complete multi-source spatiotemporal dataset is formed.

[0072] S102: The multi-source spatiotemporal dataset is classified into data credibility levels according to data accuracy, completeness and timeliness to obtain data quality classification results.

[0073] For example, based on a multi-source spatiotemporal dataset, Terminal 100 extracts key statistical features related to data quality. Combining the spatial distribution differences and temporal stability patterns of these features, a grading benchmark system for judging data quality is constructed. For the currently acquired multi-source spatiotemporal dataset, quality features are extracted from three core dimensions: accuracy focuses on evaluating the degree of agreement between the data and the true values ​​and internal consistency; completeness focuses on the data coverage and overall missing data; and timeliness focuses on the interval between the data collection time and the current analysis time, as well as the data update frequency. This multi-dimensional extraction forms a comprehensive set of quality evaluation information. Based on the aforementioned grading benchmark system, the quality evaluation information of the current data is systematically compared and analyzed with the benchmark. Simultaneously, considering the influence weights of different quality dimensions on the carbon sequestration quantification results, the credibility level of various types of data is comprehensively determined, initially classifying them into high, medium, and low credibility levels. The initially classified credibility levels are dynamically adjusted and normalized to eliminate differences in quality evaluation standards under different data sources and spatiotemporal dimensions, resulting in a data quality grading result.

[0074] S103: Construct a dynamic weight matrix based on the data quality classification results.

[0075] The dynamic weight matrix is ​​a mathematical tool that dynamically assigns weights based on data quality and performs error correction. For example, based on the data quality grading results, terminal 100 assigns differentiated weights to data of different confidence levels, with higher confidence data assigned higher weights and lower confidence data assigned lower weights. At the same time, the weight ratios are adjusted in conjunction with the correlation between data type and carbon sequestration, thus constructing a dynamic weight matrix that can reflect the differences in data quality and the dynamic changes in time and space.

[0076] S104: Import the multi-source spatiotemporal dataset into the dynamic weight matrix according to the corresponding mapping relationship to obtain the assimilated spatiotemporal dataset.

[0077] For example, a unified spatiotemporal benchmark is achieved for various data from different acquisition devices and time points in a multi-source spatiotemporal dataset. By matching spatial reference frames and synchronizing time recording standards, spatiotemporal misalignment caused by differences in data sources is eliminated, forming a basic dataset with consistent spatiotemporal dimensions. Based on the constructed dynamic weight matrix, and according to the weight correlations of each data source in the quality grading, multi-dimensional integration processing is performed on the spatiotemporally coordinated basic dataset, allowing data of different quality levels to be effectively fused according to their weight proportions. A systematic deviation calibration is then performed on the fused dataset to specifically eliminate errors generated during data acquisition and transmission, improving the overall consistency and reliability of the data, resulting in an assimilated spatiotemporal dataset.

[0078] S105: Based on the topographic features and ecological zoning of the slope, calibrate the carbon sink-related parameters in the assimilated spatiotemporal dataset to obtain the calibrated carbon sink parameter set.

[0079] Carbon sequestration-related parameters refer to core variables extracted from the assimilated spatiotemporal dataset for directly calculating the carbon sequestration capacity of slope greening. These mainly include multi-cycle vegetation data, multi-soil layer soil parameter data, and dynamic change indicators affected by all types of disturbance events. These parameters need to be calibrated according to the slope's topographic and geomorphological characteristics and ecological zoning to eliminate spatial heterogeneity and abnormal fluctuations. For example, by combining the geographical characteristics, geomorphological features, and regional ecological classification system of highway engineering slopes, the ecological attributes and geographical adaptability of different slope areas are clarified, forming a slope ecological geographical classification result. Based on this classification result, the assimilated spatiotemporal data are categorized and organized according to the corresponding ecological geographical regions. Simultaneously, various parameters related to carbon sequestration capacity are screened from the assimilated spatiotemporal data. Referring to multi-cycle historical monitoring data or the reference range of carbon sequestration parameters for similar slopes in the region, and considering the characteristics of different ecological geographical regions, the selected carbon sequestration-related parameters are preliminarily validated for rationality, and parameters exceeding the reasonable range are adjusted. Further analysis of the time-dimensional variation trends of parameters within the same ecological and geographical region, and correction of parameters that deviate from the normal fluctuation range based on the spatiotemporal variation patterns of the parameters, yields a calibrated carbon sink parameter set.

[0080] S106: Input the calibrated carbon sink parameter set into the pre-trained random forest model to obtain the comprehensive carbon sink amount.

[0081] For example, the calibrated carbon sink parameter set undergoes multi-dimensional feature extraction, encompassing the spatiotemporal distribution patterns of parameters, the correlation characteristics and ecological attributes between different parameters, and the matching characteristics of climatic conditions, forming a feature set that comprehensively covers the influencing factors of carbon sink. This feature set is imported into a carbon sink quantification model pre-trained and optimized with multi-scenario samples. These multi-scenario samples include carbon sink monitoring data from different climate zones, slope types, and vegetation systems, ensuring the model can adapt to the carbon sink calculation needs of various highway engineering slopes. During model operation, the model combines the long-term ecological evolution patterns of slopes and regional carbon cycle characteristics to comprehensively analyze the direct and indirect impacts of each parameter on carbon sink, avoiding calculation biases caused by single factors or static logic. Through the model's systematic calculation and integration of multi-dimensional results, a comprehensive carbon sink quantity that meets the needs of different assessment scenarios is output.

[0082] The technical solution provided in this application includes the following technical effects: This application provides a quantitative method for carbon sequestration in highway engineering slope greening. Through multi-source spatiotemporal data fusion and a reliability grading mechanism, a dynamic weight matrix is ​​systematically constructed to achieve data assimilation, effectively eliminating the reliability defects caused by nonlinear decay due to parameter uncertainty in traditional methods. Carbon sequestration parameters are calibrated by combining slope topographic features and ecological zoning, overcoming the problem of lack of differentiation caused by a unified processing mode. Utilizing the parallel computing capabilities of a pre-trained random forest model with multiple decision trees, the nonlinear correlation characteristics between spatiotemporal data are fully explored, significantly improving the accuracy and generalization ability of carbon sequestration quantification under complex working conditions. Simultaneously, this method establishes a dynamic response framework for long-term monitoring, capable of continuously adapting to spatiotemporal evolution factors such as vegetation growth cycles and extreme climates, breaking through the limitations of static data and fixed models, and providing a highly robust and adaptable technical solution for assessing carbon sequestration in highway engineering slope greening.

[0083] In one embodiment of the present invention, the pre-trained random forest model is obtained through the following method:

[0084] Step 201: Obtain multi-cycle historical monitoring data of highway engineering slopes. The multi-cycle historical monitoring data includes vegetation data of multiple growth cycles, soil parameter data of multiple soil layers, and record data of all types of disturbance events to obtain a multi-cycle historical dataset.

[0085] The multi-period historical monitoring data can include vegetation data from multiple growth cycles, soil parameter data from multiple soil layers, and records of all types of disturbance events. This data can be obtained by deploying various types of equipment, such as vegetation growth monitoring sensors, soil parameter monitoring sensors, environmental and disturbance factor recording devices, topographic sensors, and cameras, in the slope area of ​​highway engineering projects, combined with regular manual ground quadrat surveys. For example, obtaining multi-period historical monitoring data for highway engineering slopes requires coverage of multiple complete vegetation growing seasons to reflect carbon sequestration characteristics at different growth stages. Vegetation data is obtained through a combination of vegetation sensors 101, cameras 102, and ground quadrat surveys. The vegetation sensors monitor plant height and growth rate in real time, while the cameras periodically capture vegetation morphology to identify species, canopy coverage, and phenological stages. Ground quadrats are laid out in layers according to slope gradient, and fresh weight and oven-dried weight are collected, with biomass calculated using the allometric growth equation. Soil parameters are measured using stratified soil sensors to determine organic carbon content and bulk density, combined with laboratory analysis of porosity. Disturbances are monitored using temperature and humidity sensors and rainfall sensors to detect pests and diseases and extreme weather events. Security cameras record maintenance and pruning, and vehicle collisions. Regular manual inspections supplement the event duration and impact range. All data are integrated into a GIS database according to spatiotemporal coordinates, linked to monitoring point IDs and timestamps, forming a multi-period historical dataset.

[0086] Step 202: Preprocess the multi-period historical dataset to obtain the preprocessed historical dataset.

[0087] For example, the preprocessing of multi-period historical datasets is divided into three stages. The first step is data cleaning. Random missing data refers to single data points missing due to temporary sensor malfunctions, which are filled using linear interpolation of data from adjacent time points. Systematic missing data refers to data from monitoring points that have been missing for a period of time due to sensor malfunctions; the scope of the malfunction is then removed. Outliers are filtered according to statistical criteria and processed in conjunction with field records. Measurement errors, such as those caused by sampling contamination, are replaced with the median of parameters from the same region and time period. Genuine anomalies, such as parameter mutations caused by extreme weather, are retained and their causes are noted. The second step is parameter standardization, using common methods to unify the units to a specific interval. The third step is format conversion: spatial data is converted to raster, temporal data is normalized to the mean, and categorical data is one-hot encoded, resulting in a preprocessed historical dataset with a unified structure.

[0088] Step 203: Divide the preprocessed historical dataset into training dataset and test dataset according to the spatiotemporal dimension.

[0089] For example, the preprocessed data is divided according to the spatiotemporal dimensions. The temporal dimension is based on the vegetation growing season, with each growing season subdivided into the initial, vigorous, and declining stages. Training and test sets are then divided in a reasonable ratio, with the test set's time units lagging behind the training set. For instance, data from the first few growing seasons plus some units from the last season are included in the training set, while the remaining units from the last season are included in the test set. The spatial dimension is divided according to slope ecological zones, including sunny herbaceous areas, shady shrub areas, tree areas at the top of the slope, and mixed shrub-grass areas at the bottom of the slope. Data from a portion of monitoring points in each zone are randomly sampled and included in the training set, with the remainder included in the test set, ensuring data distribution across various microhabitats. After the division, the distribution of key parameters in the two sets of data is verified. If the difference exceeds an acceptable range, the sampling ratio is adjusted to obtain training and test datasets with comprehensive spatiotemporal coverage.

[0090] Step 204: Input the training set into the initial random forest model for training to obtain the trained random forest model.

[0091] For example, the initial random forest training first determines the basic parameters, and the number of decision trees is determined through pre-experiments to ensure stable accuracy and control computational costs. The number of features selected for each tree is calculated according to the feature dimension rule. The splitting criterion adopts a type suitable for regression calculation, which meets the needs of continuous carbon sink value calculation. A minimum number of samples per leaf node is set to avoid overfitting, and the maximum tree depth is not limited. Training uses a bootstrap sampling method, where each tree draws samples with replacement from the training set. The out-of-bag samples that are not drawn are used to evaluate the performance of a single tree. Samples are input in batches, and each tree minimizes the number of split nodes according to the splitting criterion. The out-of-bag error is monitored, and training is stopped if there is no decrease in multiple consecutive batches. After all trees are trained, the predicted mean is output, and the mean absolute error is calculated using the reserved validation samples that are not used in training. If the error exceeds the range, the parameters are adjusted and retrained until the error is reasonable, resulting in the trained random forest model.

[0092] Step 205: Input the test dataset into the trained random forest model, adjust the weight allocation based on the carbon sink contribution error, and obtain the pre-trained random forest model.

[0093] For example, the test dataset is input into the trained random forest model to obtain the predicted carbon sink for each test sample, while simultaneously retrieving the actual carbon sink corresponding to the test sample. The actual carbon sink is calculated through field measurements, where vegetation carbon storage is calculated by multiplying vegetation biomass by a default carbon content coefficient, and soil carbon storage is calculated by multiplying soil organic carbon content by soil bulk density, soil layer thickness, and sample plot area. The sum of these two values ​​is the actual comprehensive carbon sink. The carbon sink contribution error for each sample is calculated and categorized into three levels: low error, medium error, and high error. The contribution of each decision tree to the prediction result of the test sample is calculated using the Shapley value, and the weights of the decision trees are adjusted according to the error level: for low-error samples, the weights of their corresponding high-contribution decision trees are appropriately increased; for high-error samples, the weights of their corresponding high-contribution decision trees are appropriately decreased, ensuring that the sum of the weights of all decision trees is 1. After adjusting the weights, the average error of the model is recalculated. If the error meets the preset standard, the current weight scheme is saved; if the error does not meet the standard, the cause needs to be analyzed back. For example, if there is insufficient training data for a certain ecological zone, the decision tree is retrained after supplementing the data for that region, and then the weights are adjusted again. This adjustment and verification process is repeated until the average carbon sink contribution error of the model stabilizes within a reasonable range and changes very little. The final determined weight allocation is then combined with the trained random forest model to obtain the pre-trained random forest model.

[0094] The carbon sequestration quantification method based on highway engineering slope greening provided in this application utilizes a pre-trained random forest model with high carbon sequestration quantification accuracy. Through multi-dimensional data support and weight optimization, it accurately captures the dynamic changes in carbon sequestration on highway engineering slopes, effectively reducing prediction bias caused by parameter uncertainty. Its generalization ability and adaptability to slope scenarios are strong, reliably handling the complex influences of different growth cycles, soil conditions, and disturbance events. It solves the problems of weak dynamic response and insufficient reliability of results in traditional models, providing accurate and stable technical support for slope ecological carbon sequestration assessment.

[0095] Based on the above embodiments, the random forest model includes:

[0096] The LSTM feature extraction module is used to retrieve the preprocessed historical dataset, extract the time-series feature data of the carbon sink parameters, and associate the time-series feature data with the corresponding divided training dataset to output a joint training dataset containing time-series features.

[0097] For example, the LSTM feature extraction module retrieves the preprocessed historical dataset through a standardized API interface, constructing a three-dimensional index structure based on monitoring point ID, parameter type, and timestamp to ensure the uniqueness of data association. The sliding time window length is set according to the vegetation growth pattern of highway engineering slopes, taking the value as a complete sub-cycle of growth, with the window step size matching the regular collection frequency of soil and vegetation parameters. Within the window range, a linear regression algorithm is used to obtain the trend characteristic coefficients of carbon sink parameters, and a fast Fourier transform is used to capture the main cycle characteristics of parameter fluctuations. A threshold is established based on the 3σ criterion to identify parameter mutation feature points, simultaneously associating the correspondence between interference events and parameter responses. The extracted features are bound to static features such as spatial coordinates, soil type, and vegetation type of the training dataset samples by sample ID, expanding the feature dimension for each training sample, and outputting a unified training dataset with dynamic features and consistent format.

[0098] The model adaptation and adjustment module is used to input the joint training dataset containing time-series features into the initial LSTM model, adjust the parameters of the LSTM model by combining the carbon sink contribution error calculation method of the random forest model, and output a pre-trained LSTM model adapted to the pre-trained random forest model.

[0099] For example, the model adaptation module initializes the LSTM model structure, keeping the input layer dimension consistent with the total number of features in the joint training dataset, setting three hidden layers, configuring the number of neurons in each layer to be 1.5 times the input layer dimension, setting the initial forget gate threshold to 0.6 to enhance information retention, and using a linear activation function in the output layer to adapt to feature regression requirements. The joint training dataset containing dynamic features is divided into an adaptation training set and an adaptation validation set in an 8:2 ratio, inputting 64 samples per batch into the model, using the Adam optimizer to update parameters, with the initial learning rate set to 0.001 and dynamically decaying with each training epoch. The carbon sink contribution error calculation method of the random forest model is called, using the adaptation validation set error value as the optimization objective, and adjusting the weight matrix and bias terms of the LSTM model through the error backpropagation mechanism. Training is terminated when the adaptation validation set error decreases by less than 0.5% for five consecutive epochs, and the output is the pre-trained random forest model.

[0100] The temporal evolution analysis module is used to extract real-time time-series data of carbon sink-related parameters from the assimilated spatiotemporal dataset, input the real-time time-series data into the pre-trained LSTM model, and the pre-trained LSTM model is used to analyze the temporal evolution law of carbon sink parameters under dynamic disturbances and output the carbon sink temporal evolution results.

[0101] For example, the temporal evolution analysis module extracts carbon sink-related parameters from the assimilated spatiotemporal dataset hourly via a real-time data acquisition interface. These parameters cover indicators such as vegetation growth rate, soil organic carbon content, and microbial activity. Combined with sensor logs from the slope monitoring system and camera recognition results, the module labels the occurrence information of dynamic disturbance events, including the time type and affected area coordinates of events such as rainstorms, pests and diseases, and maintenance operations. The extracted real-time data undergoes Z-score standardization to eliminate dimensional differences. After being input into a pre-trained LSTM model, the model filters valid data through an input gate, stores parameter change patterns using memory units, and outputs parameter change gradients through an output gate. Based on the disturbance event labeling information, the module analyzes the response characteristics of parameters at each stage of disturbance occurrence and outputs carbon sink evolution results containing the predicted rate of change of carbon sink parameters and the degree of disturbance impact for the next 72 hours.

[0102] The feature fusion module is used to incorporate the carbon sink time-series evolution results as supplementary features into the calibrated carbon sink parameter set and output a joint feature parameter set containing time-series information.

[0103] For example, the feature fusion module employs a lateral feature stitching strategy to extract core information from the carbon sink evolution results, including key indicators such as the mean trend change coefficient of future parameter predictions, the peak value of the disturbance response, and the stabilization period. This information is then aligned with the static parameters of the corresponding samples in the calibrated carbon sink parameter set by sample ID. Static parameters cover soil layer thickness, vegetation carbon content coefficient, and slope gradient. To balance the weighting of dynamic and static features, a Min-Max normalization method is used to map all feature values ​​to the 0-1 range. Redundant features are removed through Pearson correlation coefficient analysis, retaining only one feature with an absolute correlation coefficient greater than 0.85. This results in a joint feature parameter set containing dynamic information, with reduced dimensionality, complete information, and synergistic effects between the two types of features.

[0104] The joint calculation module is used to input the joint feature parameter set containing time-series information into the pre-trained random forest model, recalculate the carbon sink, and output the optimized comprehensive carbon sink that takes into account both spatiotemporal characteristics and dynamic evolution laws.

[0105] The joint computation module inputs a set of joint feature parameters containing dynamic information into a pre-trained random forest model. The model assigns a weight of 15% to 20% to dynamic features, a proportion verified in previous experiments to maximize the contribution of dynamic features to carbon sequestration prediction. Each decision tree simultaneously receives both dynamic and static features, and uses node splitting criteria to uncover the intrinsic correlation between these two types of features and carbon sequestration. For example, by combining the dynamic trend of vegetation growth with static soil properties, the carbon sequestration capacity at different growth stages can be accurately calculated. The prediction results of all decision trees are weighted and averaged to obtain the final carbon sequestration value. The weight values ​​are determined based on the accuracy of each tree in the test set, and the output includes the optimized comprehensive carbon sequestration value of the entire slope.

[0106] The carbon sequestration quantification method based on highway engineering slope greening provided in this application deeply integrates dynamic and static features to accurately capture the variation patterns of carbon sequestration parameters with disturbances, significantly improving the accuracy of carbon sequestration quantification and greatly reducing prediction bias caused by parameter fluctuations and disturbance events. The pre-trained model adapts to different slope ecological scenarios, and the output results take into account regional carbon sequestration contribution and short-term change trends, effectively compensating for the shortcomings of traditional models such as insufficient dynamic response and one-sided evaluation results, providing high-precision and comprehensive technical support for the dynamic assessment of carbon sequestration on highway engineering slopes.

[0107] In an exemplary embodiment, a multi-source spatiotemporal dataset is imported into a dynamic weight matrix according to a corresponding mapping relationship to obtain an assimilated spatiotemporal dataset, including:

[0108] Step 301: Perform spatiotemporal coordinate alignment based on the multi-source spatiotemporal dataset to obtain the spatiotemporally aligned dataset;

[0109] For all data points from different monitoring devices in the multi-source spatiotemporal dataset, the data points are parsed and associated based on their associated device IDs, spatial location codes, and acquisition timestamps. The spatial locations of all data are transformed and mapped using a unified slope engineering coordinate system to ensure that data from different sources have a definite correspondence on the same spatial grid. A time synchronization mechanism is used to uniformly convert the acquisition timestamps of all data to a relative time series based on the start time of the data assimilation task, eliminating the time scale differences caused by different sampling frequencies of devices. Through spatial interpolation and time alignment algorithms, data values ​​from all sources are filled into each unified spatiotemporal unit, forming a spatiotemporally aligned dataset that is aligned in both spatial and temporal dimensions and can be directly used for matrix operations.

[0110] Step 302: Based on the dynamic weight matrix, perform weighted calculations on the spatiotemporal aligned dataset to obtain weighted fused data;

[0111] The dynamic weight matrix constructed based on the data quality grading results is parsed into a weight tensor that perfectly corresponds to the structure of the spatiotemporally aligned dataset. Each weight value is directly mapped from the data's credibility grading results in terms of accuracy, completeness, and timeliness. Data units with high credibility levels are assigned higher weight values, and data units with low credibility levels are assigned lower weight values. A parallel weighted calculation process is initiated, multiplying each data value in the spatiotemporally aligned dataset element-wise with the corresponding weight value in the weight tensor. This process ensures that high-quality data plays a dominant role in the fusion result, while the influence of low-quality data is significantly suppressed. All weighted results from different data sources under the same spatiotemporal unit are summed to obtain a weighted fused data that initially integrates multi-source information and highlights high-quality data.

[0112] Step 303: Use the Kalman filter algorithm to correct errors in the weighted fused data to obtain the assimilated spatiotemporal dataset.

[0113] The weighted fused data field is constructed as the observation vector of the Kalman filter algorithm. Based on the spatiotemporal evolution law of carbon sink parameters obtained from multi-period historical monitoring data, the system's state transition equation is established to predict the parameter state at the next moment. The filtering process is initialized, the prior estimate of the state vector and its error covariance matrix are set, and the observed value and the predicted value are compared at each step. The Kalman gain is calculated using the observation noise covariance matrix determined according to the data quality classification. The prior predicted value is corrected by the Kalman gain to obtain a posterior state estimate with smaller bias and reduced uncertainty, and the estimated error covariance matrix is ​​updated to prepare for the next filtering step. This prediction-correction loop is recursively performed on the entire spatiotemporal sequence, outputting an assimilated spatiotemporal dataset with higher internal consistency after systematic error correction.

[0114] The quantification method for carbon sequestration based on slope greening in highway engineering provided in this application effectively eliminates the differences and uncertainties in spatiotemporal benchmarks and quality of multi-source heterogeneous data, significantly improving the internal consistency and reliability of the data. The resulting high-quality assimilated spatiotemporal dataset provides an accurate and stable data foundation for subsequent parameter calibration and model calculation.

[0115] In an exemplary embodiment, the multi-source spatiotemporal dataset is classified into data reliability levels according to data accuracy, completeness, and timeliness to obtain data quality classification results, including:

[0116] Step 401: Extract statistical features from multi-period historical monitoring data to obtain a set of feature parameters.

[0117] Core data subsets related to data quality were selected from multi-period historical monitoring data, covering various types of data such as vegetation growth parameters, soil physicochemical indicators, and disturbance event records. Statistical analysis methods were used to extract key features, including descriptive statistics such as the mean, variance, extreme values, median, and standard deviation of each parameter. At the same time, reliability indicators such as the internal consistency coefficient and repeated observation error were calculated. These indicators, which cover data distribution characteristics, dispersion, and reliability, were classified and integrated according to data type and monitoring period to form a well-structured feature parameter set.

[0118] Step 402: Analyze the spatial variation and temporal stability of the feature parameter set to obtain the hierarchical benchmark matrix.

[0119] Spatial interpolation and coefficient of variation analysis were used to analyze the spatial distribution differences of the feature parameter set. By calculating the degree of variation of parameters among different monitoring areas, the distribution pattern of spatial heterogeneity of the data was clarified. Time series trend analysis and sliding window fluctuation amplitude calculation were used to evaluate the stability of feature parameters in long-term monitoring, quantifying the fluctuation range and trend strength of parameters over time. Combining the spatial variation results and the conclusions of temporal stability analysis, a hierarchical benchmark matrix was constructed according to data type and monitoring dimension. The matrix includes the reasonable fluctuation range of each statistical feature, the spatial variation threshold, and the temporal stability judgment criteria.

[0120] Step 403: Extract quality features from the multi-source spatiotemporal dataset to obtain a set of quality evaluation indicators.

[0121] For multi-source spatiotemporal datasets, quality features are extracted from three core dimensions: accuracy, completeness, and timeliness. The accuracy dimension obtains accuracy-related indicators by comparing the deviation between the data and high-precision verification data and calculating the consistency of repeated observations. The completeness dimension calculates the percentage of valid data for each parameter, the data missing rate during key monitoring periods, and spatial coverage completeness indicators. The timeliness dimension calculates the interval between data acquisition time and the current analysis time, data update frequency, and lag indicators. The extraction results from these three dimensions are integrated in a unified format to form a comprehensive quality evaluation indicator set reflecting the data quality status.

[0122] Step 404: Using the following formula, match and calculate the quality grading index by matching the quality evaluation index set with the grading benchmark matrix:

[0123] QI=ω p ˙μ p (P) + ω c ˙μ c (C) + ω t ˙μ t (T) + λ˙

[0124] Where QI is the quality grading index, ω p For precision weights, ω c For integrity weight, ω t For timeliness weighting, μ p (P) is the precision function, μ c (C) is the integrity function, μ t (T) is the timeliness function, λ is the dynamic correction coefficient, k is the attenuation coefficient, t0 is the current timestamp, and t is the data acquisition timestamp.

[0125] The values ​​of accuracy weight ωp, integrity weight ωc, and timeliness weight ωt are determined by combining entropy weighting with expert experience to ensure that the weight allocation aligns with the requirements of carbon sequestration for each quality dimension. The accuracy function μp(P) is defined as the ratio of the actual measurement error to the accuracy threshold in the grading benchmark matrix; the integrity function μc(C) is the normalized result of the effective data ratio and the benchmark integrity standard; and the timeliness function μt(T) is constructed by the ratio of the data time interval to the benchmark timeliness threshold. The dynamic correction coefficient λ ranges from 0.1 to 0.3, the attenuation coefficient k is set from 1.5 to 3.0 according to the data type, t0 uses the unified timestamp of the current system, and t extracts the original data collection timestamp. Substituting each indicator of the quality evaluation index set into the formula and matching it with the corresponding threshold in the grading benchmark matrix, the quality grading index QI of each data source and each spatiotemporal node is obtained.

[0126] Step 405: Perform dynamic normalization on the quality grading index to obtain the data quality grading results.

[0127] The min-max normalization method was used to process all quality grading indices, mapping QI values ​​to the [0,1] interval. During normalization, extreme value parameters were dynamically adjusted, taking into account the distribution characteristics of data quality across different spatiotemporal dimensions to avoid the influence of a single extreme value on the overall result. Based on the distribution of normalized QI values ​​and the actual data quality requirements for carbon sequestration quantification, grading thresholds were set: QI values ​​≥ 0.8 were classified as high confidence level, QI values ​​≤ 0.5 < 0.8 as medium confidence level, and QI values ​​< 0.5 as low confidence level, forming the final data quality grading result.

[0128] The carbon sequestration quantification method based on highway engineering slope greening provided in this application's embodiments accurately matches the objective benchmark established by historical data characteristics with the multi-dimensional quality characteristics of real-time data. Combined with dynamic correction, it ensures that the grading results meet the carbon sequestration quantification requirements, enabling data quality from different data sources and different spatiotemporal dimensions to have a unified and comparable quantification standard. This provides a scientific and reliable quality basis for subsequent dynamic weight matrix construction and data assimilation.

[0129] In an exemplary embodiment, based on the slope topography and ecological zoning, carbon sink-related parameters in the assimilated spatiotemporal dataset are calibrated to obtain a calibrated carbon sink parameter set, including:

[0130] Step 501: Extract geomorphic features from the slope topography to obtain a slope zoning feature dataset.

[0131] For example, by using sensors 101 deployed on the slope of a highway engineering project, slope gradient, aspect, and elevation data are collected. Combined with information on the distribution of exposed rock areas and soil texture stratification obtained from manual field surveys, and the initial vegetation cover density obtained from the analysis of images captured by camera 102, core feature indicators reflecting the topographic undulation, surface material composition, and initial cover status of the slope are selected from multi-source spatiotemporal data. These indicators are then integrated into a unified spatial grid unit to eliminate spatial coordinate differences caused by different acquisition methods, ensuring that the topographic feature data within each grid unit is complete and consistent, resulting in a slope zoning feature dataset containing slope gradient, aspect, elevation, exposed rock rate, soil texture, and initial vegetation density.

[0132] Step 502: Integrate the slope zoning feature dataset with the ecological zoning dataset to obtain the slope ecological zoning results.

[0133] For example, based on regional ecological zoning data and slope zoning feature dataset, the spatial overlay analysis method is used to achieve correlation matching between the two. The similarity of slope topographic features and the consistency of functional attributes of ecological zoning are used as dual matching criteria. For overlapping zoning boundaries, the buffer analysis method is used to achieve smooth fusion of transition areas to avoid boundary data breakage. The matching results of different topographic features and ecological attributes are verified to ensure that each fused zoning conforms to the integrity of the topographic unit and has the unity of ecological function, thus obtaining the slope ecological zoning results.

[0134] Step 503: Based on the slope ecological zoning results, the assimilated spatiotemporal data are divided and classified to obtain the basic data of slope zoning.

[0135] For example, a spatial index for slope ecological zoning is established based on the results. Each data unit in the assimilated spatiotemporal dataset includes data on vegetation, soil, and disturbance events. Following the spatiotemporal alignment logic of data assimilation in the document, spatiotemporal coordinate alignment technology is used to correlate and match with the spatial index to determine the ecological zoning to which each data unit belongs. A batch classification algorithm is used to automatically aggregate the assimilated spatiotemporal data, dividing data from the same ecological zoning into corresponding subsets. Integrity checks are performed on each subset of data. For missing marginal data, neighborhood data interpolation using the error correction interpolation approach in the reference document is used to supplement and improve the data, ultimately forming basic slope zoning data that is independent for each zoning and has a unified data structure.

[0136] Step 504: Extract carbon sink-related parameters from the assimilated spatiotemporal dataset to obtain the carbon sink parameter set to be calibrated.

[0137] For example, parameters directly related to carbon sinks are selected from the assimilated spatiotemporal dataset, including vegetation biomass, vegetation cover, litter volume across multiple growth cycles, soil organic carbon content and soil bulk density across multiple soil layers, and microbial activity data related to carbon sinks. Data filtering algorithms are used to remove invalid data caused by sensor malfunctions or acquisition errors, such as soil organic carbon content values ​​exceeding normal physical ranges. The rationality of the extracted parameters is ensured through parameter correlation verification, such as verifying the logical matching between vegetation biomass and cover. The selected parameters are then standardized in a consistent format to obtain the set of carbon sink parameters to be calibrated.

[0138] Step 505: According to the slope ecological zoning results, classify the set of carbon sink parameters to be calibrated to obtain the parameters to be calibrated corresponding to each zoning.

[0139] For example, based on the spatial attribute information of the slope ecological zoning results, the spatial coordinates of each parameter in the carbon sink parameter set to be calibrated are traversed. A spatial association algorithm is used to compare the parameter coordinates with the ecological zoning boundaries to determine the ecological zoning category to which each parameter belongs. A classification storage strategy is adopted, storing the carbon sink parameters of different ecological zonings into corresponding data containers. A zoning identifier is added to each parameter, such as adding a zoning identifier and category label like "sunny slope herbaceous area - vegetation biomass," to ensure that the correspondence between parameters and zonings is unique and accurate. Batch verification is performed on the classified parameters to correct zoning mismatches caused by coordinate deviations, obtaining the parameters to be calibrated corresponding to each zoning.

[0140] Step 506: Based on multi-period historical monitoring data, extract the historical reasonable range of carbon sink parameters for each ecological zone to obtain the historical range of zone parameters.

[0141] For example, historical carbon sink parameter data corresponding to each ecological zone are extracted from multi-period historical monitoring data. Statistical analysis methods are used to process each carbon sink parameter within each zone, and the historical reasonable range of the carbon sink parameters is obtained using the following formula:

[0142] PR=μ±α×σ×[1-I×e^(-t / τ)]

[0143] Wherein, PR represents the historical reasonable range of a specific carbon sink parameter within an ecological zone; μ is the historical mean of the carbon sink parameter, obtained through statistical calculation of multi-period historical monitoring data; σ is the historical standard deviation of the carbon sink parameter, used to reflect the dispersion of historical data; α is the safety factor, ranging from 1.3 to 1.6, with the specific value dynamically adjusted based on the parameter's importance in the carbon sink quantification process and the stability of historical data; I is the interference event impact coefficient, its value being positively correlated with the intensity of the interference event's impact on the carbon sink parameter; e is the natural constant, with a value of approximately 2.718; t is the interval between the current monitoring period and the end time of the historical interference event, in days; τ is the time decay coefficient, its value matching the natural recovery cycle of vegetation within the corresponding ecological zone, with a smaller τ value indicating faster vegetation recovery. This formula allows for the precise definition of the historical reasonable fluctuation range of different carbon sink parameters in each ecological zone, yielding the historical range of the zone's parameters.

[0144] Step 507: Compare the parameters to be calibrated for each partition with the historical range of the partition parameters, adjust the parameters that exceed the historical range of the partition parameters, and obtain the preliminary calibration carbon sink parameter set.

[0145] For example, the parameters to be calibrated for each partition are compared with the historical range of the partition parameters, and parameters exceeding the range are identified by threshold judgment. For parameters below the lower limit of the historical range, a weighted adjustment method of "historical lower limit × 0.6 + mean of neighboring parameters in the same partition and period × 0.4" is used, with the weight set based on the data accuracy classification results. The weight of high-precision neighboring data can be increased to 0.5. For parameters above the upper limit of the historical range, a weighted adjustment method of "historical upper limit × 0.7 + mean of parameters under similar soil conditions at the same growth stage × 0.3" is used. During the adjustment process, the original time-series trend of the parameters is preserved, such as the seasonal growth trend of vegetation biomass, to avoid trend distortion caused by over-correction, thus obtaining a preliminary calibration carbon sink parameter set.

[0146] Step 508: Calculate the temporal fluctuation range of parameters within the same ecological zone in the preliminary calibration carbon sink parameter set, correct parameters whose fluctuation range exceeds the preset range, and obtain the calibrated carbon sink parameter set.

[0147] For example, based on the growth cycle of different vegetation types on the slope and the data collection frequency, an appropriate sliding time window is set. For each parameter in the same ecological zone within the preliminary calibration carbon sink parameter set, a time-series fluctuation analysis is conducted across the entire growth stage and monitoring cycle. The fluctuation amplitude is obtained by calculating the difference between the maximum and minimum values ​​of the parameters within the window. The preset range is determined comprehensively by combining the long-term historical parameter fluctuation statistics of the zone with the regional ecological background to ensure coverage of natural fluctuations. For parameters exceeding the range, when correcting using linear interpolation, adjacent normal data, trends of similar microenvironmental parameters in the same zone, and the overall pattern of regional carbon sinks are referenced to avoid bias. After correction, the parameter time series is verified to be smooth and consistent with the evolution logic of the zoned carbon sink, resulting in the calibrated carbon sink parameter set.

[0148] The quantification method for carbon sequestration based on greening of highway slopes provided in this application effectively achieves accurate matching of carbon sequestration parameters with different ecological scenarios of slopes, avoids the problem of parameters deviating from the actual environment caused by uniform processing, significantly improves the accuracy and temporal stability of carbon sequestration parameters, eliminates abnormal data and corrects unreasonable fluctuations, ensures that parameters conform to the laws of ecological evolution and natural change trends, provides high-quality data support for the quantification of carbon sequestration in greening of highway slopes, significantly enhances the rationality and credibility of subsequent carbon sequestration assessment results, and can be adapted to diverse carbon sequestration calculation scenarios such as different ecological zones and long-term monitoring.

[0149] In an exemplary embodiment, the calibrated carbon sink parameter set is input into a pre-trained random forest model to obtain the comprehensive carbon sink amount, including:

[0150] Step 601: Extract the spatial distribution matrix and time series tensor of the calibrated carbon sink parameter set to obtain the spatiotemporal feature tensor of the carbon sink.

[0151] For example, parameter data containing spatial location information is selected from the calibrated carbon sink parameter set, and a spatial distribution matrix is ​​constructed according to a unified spatial grid division rule. The matrix dimension corresponds to the spatial range of the slope ecological zoning, and the matrix elements are the carbon sink parameter values ​​within each grid unit. At the same time, the sequence data of parameter changes over time are extracted and organized into a time series tensor according to the monitoring cycle. The tensor dimension corresponds to the monitoring duration and parameter type. The spatial dimension information of the spatial distribution matrix and the time dimension information of the time series tensor are concatenated to ensure that the spatial attribute and time attribute of each parameter correspond one-to-one, forming a carbon sink spatiotemporal feature tensor containing spatial distribution characteristics and temporal evolution characteristics.

[0152] Step 602: Perform spatiotemporal dimension feature encoding on the calibrated carbon sink parameter set to obtain a joint feature matrix that fuses the spatial distribution matrix and the time series tensor.

[0153] For example, for the spatial dimension features of the calibrated carbon sink parameter set, a grid index encoding method is used to convert the spatial distribution matrix into a computable spatial feature vector, with each vector element associated with the parameter value and spatial location weight of the corresponding grid; for the temporal dimension features, a time step encoding method is used to convert the time series tensor into a temporal feature vector, with the vector elements containing the parameter values ​​and time decay coefficients of each monitoring period; the spatial feature vector and the temporal feature vector are made to have the same dimension through feature dimension alignment technology, and the two are integrated according to the spatial influence weight and the temporal contribution weight using a weighted fusion algorithm to eliminate the feature bias caused by the difference between the spatiotemporal dimensions, and obtain a joint feature matrix that is deeply fused with the spatial distribution matrix and the time series tensor.

[0154] Step 603: Input the joint feature matrix into the pre-trained random forest model, and obtain the comprehensive carbon sink through parallel computation of multiple decision trees.

[0155] For example, the joint feature matrix is ​​adapted to the input format requirements of the pre-trained random forest model to ensure that the matrix dimension matches the model input layer dimension. The model calls the decision tree weight allocation rules determined in the pre-training stage to start the multi-decision tree parallel computing process. Each decision tree independently analyzes the spatial and temporal features in the joint feature matrix based on its own feature splitting rules and outputs the carbon sink prediction value of a single tree. Through the model's built-in result fusion mechanism, the prediction values ​​of all decision trees are weighted and summed. The weight values ​​refer to the results adjusted based on the carbon sink contribution error during pre-training to obtain a comprehensive carbon sink that takes into account both spatiotemporal characteristics and dynamic evolution laws.

[0156] The quantitative method for carbon sinks based on greening of highway engineering slopes provided in this application effectively integrates the spatial distribution and temporal evolution characteristics of carbon sink parameters, avoids quantitative bias caused by the lack of single-dimensional information, and fully captures the spatiotemporal correlation of parameters; it efficiently mines the complex nonlinear relationships between parameters, significantly improves the accuracy and generalization ability of carbon sink quantification under different ecological zones and different monitoring periods of highway engineering slopes; it accurately presents the spatial heterogeneity of slope carbon sinks, clearly reflects the dynamic evolution trend of carbon sinks over time, and meets the core needs of long-term monitoring of carbon sinks in greening of slopes and low-carbon assessment of engineering projects.

[0157] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0158] Based on the same inventive concept, this application also provides a device for quantifying carbon sequestration in highway engineering slope greening, which implements the aforementioned method for quantifying carbon sequestration in highway engineering slope greening. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the device for quantifying carbon sequestration in highway engineering slope greening provided below can be found in the limitations of the method for quantifying carbon sequestration in highway engineering slope greening described above, and will not be repeated here.

[0159] In one exemplary embodiment, such as Figure 3 As shown, a quantification device 700 for carbon sequestration in highway engineering slope greening is provided, comprising:

[0160] Data acquisition module 701 is used to acquire multi-source spatiotemporal data of highway engineering slopes to obtain a multi-source spatiotemporal dataset;

[0161] The grading module 702 is used to classify the data credibility of multi-source spatiotemporal datasets according to data accuracy, completeness and timeliness, and obtain data quality grading results.

[0162] Matrix construction module 703 is used to construct a dynamic weight matrix based on data quality classification results;

[0163] The data assimilation module 704 is used to input the multi-source spatiotemporal dataset into the weight matrix to obtain the assimilated spatiotemporal dataset.

[0164] The calibration module 705 is used to calibrate carbon sink-related parameters in the assimilated spatiotemporal dataset based on the topographic features and ecological zoning of the slope, and to obtain the calibrated carbon sink parameter set.

[0165] The integrated carbon sink calculation module 706 is used to input the calibrated carbon sink parameter set into a pre-trained random forest model to obtain the integrated carbon sink.

[0166] In one embodiment of the present invention, the pre-trained random forest model is obtained through the following method:

[0167] To obtain multi-period historical monitoring data of highway engineering slopes, the multi-period historical monitoring data includes vegetation data of multiple growth cycles, soil parameter data of multiple soil layers, and record data of all types of disturbance events, thus obtaining a multi-period historical dataset.

[0168] Preprocess the multi-period historical dataset to obtain the preprocessed historical dataset;

[0169] The preprocessed historical dataset is divided into training dataset and test dataset according to the spatiotemporal dimension.

[0170] The training set is input into the initial random forest model for training, resulting in the trained random forest model;

[0171] The test dataset is input into the trained random forest model, and the weight allocation is adjusted based on the carbon sink contribution error to obtain the pre-trained random forest model.

[0172] In one embodiment of the present invention, the random forest model includes:

[0173] The LSTM feature extraction module is used to retrieve the preprocessed historical dataset, extract the time-series feature data of the carbon sink parameters, and associate the time-series feature data with the corresponding divided training dataset to output a joint training dataset containing time-series features.

[0174] The model adaptation and adjustment module is used to input the joint training dataset containing time-series features into the initial LSTM model, adjust the LSTM model parameters by combining the carbon sink contribution error calculation method of the random forest model, and output a pre-trained LSTM model adapted to the pre-trained random forest model.

[0175] The temporal evolution analysis module is used to extract real-time time-series data of carbon sink-related parameters from the assimilated spatiotemporal dataset. The real-time time-series data is input into a pre-trained LSTM model, which is used to analyze the temporal evolution law of carbon sink parameters under dynamic disturbances and output the carbon sink temporal evolution results.

[0176] The feature fusion module is used to incorporate the carbon sink time-series evolution results as supplementary features into the calibrated carbon sink parameter set and output a joint feature parameter set containing time-series information.

[0177] The joint computation module is used to input the joint feature parameter set containing time-series information into the pre-trained random forest model, recalculate the carbon sink, and output the optimized comprehensive carbon sink that takes into account both spatiotemporal characteristics and dynamic evolution laws.

[0178] In one embodiment of the present invention, the data acquisition module is further configured to:

[0179] Spatiotemporal coordinate alignment is performed based on multi-source spatiotemporal datasets to obtain a spatiotemporally aligned dataset;

[0180] Based on the dynamic weight matrix, the spatiotemporally aligned dataset is weighted to obtain weighted fused data.

[0181] The weighted fused data is corrected for errors using the Kalman filter algorithm to obtain the assimilated spatiotemporal dataset.

[0182] In one embodiment of the present invention, the hierarchical module is further configured to:

[0183] Statistical features are extracted from multi-period historical monitoring data to obtain a set of feature parameters;

[0184] By analyzing the spatial variation and temporal stability of the feature parameter set, a hierarchical benchmark matrix is ​​obtained;

[0185] Quality features are extracted from multi-source spatiotemporal datasets to obtain a set of quality evaluation indicators;

[0186] The quality grading index is obtained by matching the quality evaluation index set with the grading benchmark matrix using the following formula:

[0187] QI=ω p ˙μ p (P) + ω c ˙μ c (C) + ω t ˙μ t (T) + λ˙

[0188] Where QI is the quality grading index, ω p For precision weights, ω c For integrity weight, ω t For timeliness weighting, μ p (P) is the precision function, μ c (C) is the integrity function, μ t (T) is the timeliness function, λ is the dynamic correction coefficient, k is the attenuation coefficient, t0 is the current timestamp, and t is the data acquisition timestamp;

[0189] The quality grading index is dynamically normalized to obtain the data quality grading results.

[0190] In one embodiment of the present invention, the calibration module is further configured to: extract geomorphic features from the slope topography and geomorphic features to obtain a slope zoning feature dataset;

[0191] By fusing the slope zoning feature dataset with the ecological zoning dataset, the ecological zoning results of the slope are obtained;

[0192] Based on the ecological zoning results of slopes, the assimilated spatiotemporal data are divided and classified to obtain basic data for slope zoning.

[0193] Carbon sink-related parameters are extracted from the assimilated spatiotemporal dataset to obtain the set of carbon sink parameters to be calibrated;

[0194] Based on the slope ecological zoning results, the set of carbon sink parameters to be calibrated is classified to obtain the parameters to be calibrated corresponding to each zoning.

[0195] Based on multi-period historical monitoring data, the historical reasonable range of carbon sink parameters for each ecological zone is extracted to obtain the historical range of zone parameters;

[0196] Compare the parameters to be calibrated for each partition with the historical range of the partition parameters, adjust the parameters that are outside the historical range of the partition parameters, and obtain the preliminary calibration carbon sink parameter set;

[0197] Calculate the temporal fluctuation range of parameters within the same ecological zone in the preliminary calibration carbon sink parameter set, correct parameters whose fluctuation range exceeds the preset range, and obtain the calibrated carbon sink parameter set.

[0198] In one embodiment of the present invention, the comprehensive carbon sink calculation module is further used for:

[0199] Extract the spatial distribution matrix and time series tensor of the calibrated carbon sink parameter set to obtain the spatiotemporal feature tensor of the carbon sink;

[0200] Spatiotemporal dimension feature encoding is performed on the calibrated carbon sink parameter set to obtain a joint feature matrix that fuses the spatial distribution matrix and the time series tensor;

[0201] The joint feature matrix is ​​input into a pre-trained random forest model, and the comprehensive carbon sink is obtained through parallel computation using multiple decision trees.

[0202] In one embodiment, this application provides a computer device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0203] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0204] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0205] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for quantifying carbon sequestration of highway engineering slope greening, characterized in that, The method comprises: obtaining multi-source spatio-temporal data of a highway engineering slope to obtain a multi-source spatio-temporal data set; dividing data credibility levels according to data accuracy, completeness and timeliness of the multi-source spatio-temporal data set to obtain a data quality grading result; constructing a dynamic weight matrix based on the data quality grading result; importing the multi-source spatio-temporal data set into the dynamic weight matrix according to a corresponding mapping relationship to obtain an assimilated spatio-temporal data set; calibrating carbon sink related parameters in the assimilated spatio-temporal data set according to the topographic and geomorphic features and ecological zoning of the slope to obtain a calibrated carbon sink parameter set; inputting the calibrated carbon sink parameter set into a pre-trained random forest model to obtain a comprehensive carbon sink amount.

2. The method of claim 1, wherein, The pre-trained random forest model is obtained by the following method: obtaining multi-period historical monitoring data of a highway engineering slope, the multi-period historical monitoring data including multi-growth-period vegetation data, multi-soil-layer soil parameter data and full-type interference event record data to obtain a multi-period historical data set; preprocessing the multi-period historical data set to obtain a preprocessed historical data set; dividing the preprocessed historical data set into a training data set and a test data set according to spatio-temporal dimensions; inputting the training set into an initial random forest model for training to obtain a trained random forest model; inputting the test data set into the trained random forest model, combining carbon sink contribution error adjustment weight distribution to obtain the pre-trained random forest model.

3. The method of claim 2, wherein, The random forest model comprises: an LSTM feature extraction module for retrieving the preprocessed historical data set, extracting time sequence feature data of carbon sink parameters therein, and correlating the time sequence feature data with the divided training data set to output a joint training data set containing time sequence features; a model adaptation adjustment module for inputting the joint training data set containing time sequence features into an initial LSTM model, adjusting LSTM model parameters in combination with a carbon sink contribution error calculation method of the random forest model, and outputting a pre-trained LSTM model adapted to the pre-trained random forest model; a time sequence evolution analysis module for extracting real-time time sequence data of carbon sink related parameters from the assimilated spatio-temporal data set, inputting the real-time time sequence data into the pre-trained LSTM model, the pre-trained LSTM model being used to analyze time sequence evolution rules of carbon sink parameters under dynamic interference, and outputting carbon sink time sequence evolution results; a feature fusion module for fusing the carbon sink time sequence evolution results as supplementary features into the calibrated carbon sink parameter set to output a joint feature parameter set containing time sequence information; a joint calculation module for inputting the joint feature parameter set containing time sequence information into the pre-trained random forest model, recalculating carbon sink amounts, and outputting optimized comprehensive carbon sink amounts taking into account spatio-temporal features and dynamic evolution rules.

4. The method of claim 1, wherein, The importing of the multi-source spatio-temporal data set into the dynamic weight matrix according to a corresponding mapping relationship to obtain an assimilated spatio-temporal data set comprises: aligning spatio-temporal coordinates according to the multi-source spatio-temporal data set to obtain a spatio-temporal aligned data set; performing weighted calculation on the spatio-temporal aligned data set based on the dynamic weight matrix to obtain weighted fusion data; The Kalman filtering algorithm is used for error correction on the weighted fusion data, to obtain a spatio-temporal data set after assimilation.

5. The method of claim 1, wherein, The multi-source spatio-temporal data set is divided according to data accuracy, integrity and timeliness to obtain data quality grading results, including: Statistical characteristics of the multi-period historical monitoring data are extracted to obtain a characteristic parameter set; Spatial variation and time sequence stability of the characteristic parameter set are analyzed to obtain a grading reference matrix; Quality characteristic extraction is performed on the multi-source spatio-temporal data set to obtain a quality evaluation index set; The quality evaluation index set and the grading reference matrix are matched and calculated using the following formula to obtain a quality grading index: QI = ω p μ p (P) + ω c μ c (C) + ω t μ t (T) + λ wherein QI is a quality ranking index, ω p is an accuracy weight, ω c is a completeness weight, ω t is a timeliness weight, μ p (P) is an accuracy function, μ c (C) is a completeness function, μ t (T) is a timeliness function, λ is a dynamic correction coefficient, k is a decay coefficient, t0 is a current timestamp, and t is a data acquisition timestamp. The quality grading index is dynamically normalized to obtain the data quality grading results.

6. The method of claim 1, wherein, The carbon sink related parameters in the spatio-temporal data set after assimilation are calibrated according to the slope topography and ecological zoning to obtain a calibrated carbon sink parameter set, including: Topographic feature extraction is performed on the slope topography and geomorphology to obtain a slope zoning feature data set; The slope zoning feature data set is fused with the ecological zoning to obtain a slope ecological zoning result; The spatio-temporal data after assimilation are classified according to the slope ecological zoning result to obtain a slope zoning basic data; Carbon sink related parameters are extracted from the spatio-temporal data set after assimilation to obtain a set of carbon sink parameters to be calibrated; The set of carbon sink parameters to be calibrated is classified according to the slope ecological zoning result to obtain a set of zoned corresponding parameters to be calibrated; According to the multi-period historical monitoring data, the historical reasonable range of carbon sink parameters in each ecological zone is extracted to obtain a zoned parameter historical range; The zoned corresponding parameters to be calibrated are compared with the zoned parameter historical range, and the parameters that exceed the zoned parameter historical range are adjusted to obtain a set of preliminary calibrated carbon sink parameters; The time sequence fluctuation amplitude of parameters in the same ecological zone in the set of preliminary calibrated carbon sink parameters is calculated, and the parameters whose fluctuation amplitude exceeds a preset range are corrected to obtain a set of calibrated carbon sink parameters.

7. The method of claim 1, wherein, The set of calibrated carbon sink parameters is input into a pre-trained random forest model to obtain a comprehensive carbon sink amount, including: The spatial distribution matrix and time sequence tensor of the set of calibrated carbon sink parameters are extracted to obtain a carbon sink spatio-temporal feature tensor; The set of calibrated carbon sink parameters is spatio-temporally dimensionally encoded to obtain a joint feature matrix fused from the spatial distribution matrix and the time sequence tensor; The joint feature matrix is input into a pre-trained random forest model, and a comprehensive carbon sink amount is obtained through parallel calculation of multiple decision trees.

8. A device for quantifying carbon sequestration of highway engineering slope greening, characterized in that, The device comprises: A data acquisition module is configured to acquire multi-source spatio-temporal data of a highway engineering slope to obtain a multi-source spatio-temporal data set; A grading module is configured to divide the multi-source spatio-temporal data set according to data accuracy, integrity and timeliness to obtain data quality grading results; A matrix construction module is configured to construct a dynamic weight matrix based on the data quality grading results; A data assimilation module is configured to input the multi-source spatio-temporal data set into the weight matrix to obtain a spatio-temporal data set after assimilation. A calibration module is configured to calibrate carbon sink related parameters in the assimilated spatio-temporal data set according to the slope topography and the ecological division, to obtain a calibrated carbon sink parameter set; A comprehensive carbon sink amount calculation module is configured to input the calibrated carbon sink parameter set into a pre-trained random forest model to obtain a comprehensive carbon sink amount. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8. The processor executes the computer program to implement the steps of the method of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 7.

Images