Method and system for estimating carbon dioxide emissions of off-road mobile machinery

Abstract

This invention relates to the field of carbon dioxide emission estimation technology, and in particular to a method and system for estimating carbon dioxide emissions from non-road mobile machinery. The method is based on the NONROAD non-road mobile source emission model framework and combines it with the carbon balance method to construct an initial calculation model for carbon dioxide emission factors. It identifies potential influencing factors of carbon dioxide emissions from non-road mobile machinery and filters key influencing factors through correlation analysis and rough set theory to form a sample dataset. Furthermore, it constructs a deep extreme learning machine model based on an autoencoder, and after training on a training set and verification on a test set, it achieves the estimation of carbon dioxide emissions from non-road mobile machinery. By combining mechanistic analysis with a data-driven approach, it can improve the accuracy, stability, and adaptability of emission estimation, better meeting the actual needs of refined accounting of carbon dioxide emissions from non-road mobile machinery under complex operating conditions.

Description

Technical Field

[0001] This invention relates to the field of carbon dioxide emission estimation technology, and in particular to a method and system for estimating carbon dioxide emissions from non-road mobile machinery. Background Technology

[0002] Non-road mobile machinery generally refers to various types of mechanical equipment that do not travel on public roads and are driven by engines or power units. It mainly includes construction machinery, agricultural machinery, port loading and unloading equipment, and airport ground equipment. It is characterized by its wide distribution, complex operating environment, and variable operating conditions. During operation, this type of equipment uses fuel combustion as its main energy source and produces a variety of emissions, including carbon dioxide. As a major greenhouse gas, the accurate estimation of carbon dioxide emissions is of great significance for carbon emission accounting, energy conservation and emission reduction, and related policy formulation. The so-called carbon dioxide emission estimation method for non-road mobile machinery generally refers to the technical means of calculating or predicting the carbon dioxide emissions generated by the equipment under specific time or operating conditions based on the equipment's activity level, fuel consumption characteristics, and emission patterns. Its core lies in the reasonable determination of emission factors and the ability to adapt to complex operating conditions.

[0003] In existing technologies, the estimation of CO2 emissions from non-road mobile machinery mainly employs methods based on emission models or empirical factors. For example, estimations are made using the NONROAD model framework combined with equipment parameters and activity data, or by directly calculating emissions based on fuel consumption and carbon content using the carbon balance method. Some methods also correct emission factors through simple statistical analysis. However, these methods generally rely on empirical parameters or average emission factors, which are difficult to accurately reflect emission characteristics under conditions of multi-factor coupling and significant changes in operating conditions. At the same time, when introducing data-driven methods, feature selection often relies on single statistical means, making it difficult to effectively extract key influencing factors. Furthermore, the models used have limited ability to characterize complex nonlinear relationships. Therefore, the main drawback of existing technologies is that emission factors are not accurately characterized under complex operating conditions, resulting in insufficient accuracy in CO2 emission estimation.

[0004] The information disclosed in this background section is intended only to enhance the understanding of the general background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] This invention provides a method and system for estimating carbon dioxide emissions from non-road mobile machinery, thereby effectively solving the problems in the background art.

[0006] To achieve the above objectives, the technical solution adopted by this invention is: a method for estimating carbon dioxide emissions from non-road mobile machinery, comprising the following steps: Based on the NONROAD non-road mobile source emission model framework and combined with the carbon balance method, an initial calculation model for carbon dioxide emissions from non-road mobile machinery is constructed to calculate the initial value of the carbon dioxide emission factor. Identify potential influencing factors of carbon dioxide emissions from non-road mobile machinery, screen and quantify key influencing factors through correlation analysis and rough set theory, and form a sample dataset. Construct a deep extreme learning machine model with the extreme learning machine autoencoder as the basic unit; preprocess the sample dataset and divide it into training set and test set; Using the key influencing factors and the initial value of the carbon dioxide emission factor as model input, and the measured value of the carbon dioxide emission factor as model output, the deep extreme learning machine model is trained using the training set to determine the model parameters; The test set is input into the trained deep extreme learning machine model, and the deep extreme learning machine model is validated by performance evaluation metrics. The validated deep extreme learning machine model is then used to estimate the carbon dioxide emissions of non-road mobile machinery.

[0007] Furthermore, an initial calculation model for carbon dioxide emissions from non-road mobile machinery is constructed, including: Based on the NONROAD non-road mobile source emission model framework, the basic values ​​of hydrocarbon emission factors and carbon monoxide emission factors of non-road mobile machinery are obtained, and the emission factors are corrected. Based on the carbon balance method, a model is established to show the relationship between fuel consumption and pollutant emissions of non-road mobile machinery equipped with gasoline engines and diesel engines. Based on the aforementioned relationship model, an initial calculation model for the carbon dioxide emission factors of non-road mobile machinery equipped with gasoline engines and non-road mobile machinery equipped with diesel engines is determined.

[0008] Furthermore, initial calculation models for carbon dioxide emission factors of non-road mobile machinery equipped with gasoline engines and non-road mobile machinery equipped with diesel engines are determined, including: The CO2 emission factor model for gasoline engines is as follows: ; The diesel engine CO2 emission factor model is as follows: ; In the formula: This refers to gasoline consumption. This refers to diesel fuel consumption. , and CO and CO2 emissions.

[0009] Furthermore, potential influencing factors of CO2 emissions from non-road mobile machinery were identified. Key influencing factors were screened and quantified using correlation analysis and rough set theory, forming a sample dataset, including: From three dimensions—machine parameters, operating parameters, and environmental factors—we identify potential influencing factors of carbon dioxide emissions from non-road mobile machinery. Based on measured emissions data from non-road mobile machinery, Spearman correlation coefficient was used to conduct correlation analysis between the potential influencing factors and the measured values ​​of carbon dioxide emission factors, and redundant influencing factors were eliminated. The screened influencing factors are discretized, and the attribute dependence and importance of each influencing factor are calculated based on rough set theory. A decision table is constructed and the key influencing factors are identified. The key influencing factors, the initial calculated values ​​of carbon dioxide emission factors, and the measured values ​​of carbon dioxide emission factors were compiled to construct a sample dataset.

[0010] Furthermore, the key influencing factors include: type of non-road mobile machinery, emission stage, rated power, and service life.

[0011] Furthermore, the screened influencing factors are discretized, including: By using a self-organizing feature mapping neural network to perform cluster analysis on the experimental data, continuous influencing factors are automatically divided into a preset number of discrete levels according to the data distribution characteristics.

[0012] Furthermore, a deep extreme learning machine model is constructed based on the extreme learning machine autoencoder as the basic unit, including: The Extreme Learning Machine (ELM) is used as the basic unit for unsupervised learning. The output weight matrix is ​​obtained by the least squares method, and a deep Extreme Learning Machine is constructed in a stack manner.

[0013] Further, the deep extreme learning machine model is trained using the training set to determine the model parameters, including: The training set data is trained unsupervised using an extreme learning machine autoencoder to obtain the corresponding weight parameters, and a multi-layer stacked deep extreme learning machine model is constructed based on the weight parameters. The training set data is input into the first layer of the Extreme Learning Machine Autoencoder of the Deep Extreme Learning Machine model to obtain the hidden layer feature vectors. The hidden layer feature vectors are then used as the input of the next layer of the Extreme Learning Machine Autoencoder. Training is performed layer by layer until all hidden layers have been trained. The output layer of the deep extreme learning machine model is trained based on the measured values ​​of carbon dioxide emission factors corresponding to the training set, and the parameters of the deep extreme learning machine model are determined by adjusting the model regularization coefficient.

[0014] Furthermore, the deep extreme learning machine model is validated using performance evaluation metrics, including: The test set is input into the trained deep extreme learning machine model to obtain the predicted value of carbon dioxide emission factor for each test sample; Based on the predicted and measured values ​​of the carbon dioxide emission factor, the preset performance evaluation index values ​​are calculated. The verification result of the deep extreme learning machine model is determined based on the relationship between the performance evaluation index value and the preset threshold.

[0015] The present invention also includes a carbon dioxide emission estimation system for non-road mobile machinery, the system comprising: The initial model building module is used to construct an initial calculation model for carbon dioxide emissions from non-road mobile machinery based on the NONROAD non-road mobile source emission model framework and combined with the carbon balance method, and is used to calculate the initial value of the carbon dioxide emission factor. The influencing factor screening module is used to identify potential influencing factors of carbon dioxide emissions from non-road mobile machinery. It screens and quantifies key influencing factors through correlation analysis and rough set theory to form a sample dataset. The Deep Extreme Learning Machine Model Building Module is used to build a Deep Extreme Learning Machine model based on the Extreme Learning Machine Autoencoder as the basic unit. The data preprocessing module is used to preprocess the sample dataset and divide it into a training set and a test set. The model training module is used to train the deep extreme learning machine model using the key influencing factors and the initial value of the carbon dioxide emission factor as model input, and the measured value of the carbon dioxide emission factor as model output, and to determine the model parameters using the training set. The model validation and estimation module is used to input the test set into the trained deep extreme learning machine model, validate the deep extreme learning machine model through performance evaluation indicators, and estimate the carbon dioxide emissions of non-road mobile machinery using the validated deep extreme learning machine model.

[0016] The beneficial effects of this invention are as follows: By constructing an initial calculation model for carbon dioxide emission factors based on the NONROAD non-road mobile source emission model framework and combined with the carbon balance method, while retaining the characteristics of clear physical meaning and strong interpretability of the mechanism model, a key influencing factor screening mechanism combining potential influencing factor identification, correlation analysis and rough set theory is introduced. Core factors with significant impact on carbon dioxide emissions are extracted from a large number of candidate variables. Furthermore, by combining a deep extreme learning machine model based on an extreme learning machine autoencoder as the basic unit, the nonlinear emission law under complex working conditions of non-road mobile machinery is modeled and corrected. This effectively overcomes the problems of existing technologies that rely on empirical factors or average emission factors and are difficult to accurately reflect the coupled effects of multiple factors and dynamic changes in working conditions.

[0017] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating the method for estimating carbon dioxide emissions from non-road mobile machinery; Figure 2 Heatmap for correlation analysis of emissions data from non-road mobile machinery; Figure 3 A comparison chart of measured and predicted values ​​of CO2 emission factors; Figure 4 This is a graph showing the relative absolute value error analysis. Figure 5 This is a histogram of the relative absolute value error distribution; Figure 6 This is a schematic diagram of a system for estimating carbon dioxide emissions from non-road mobile machinery. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Example 1:

[0022] like Figures 1 to 5 As shown, this application provides a method for estimating carbon dioxide emissions from non-road mobile machinery, the method comprising: S10: Based on the NONROAD non-road mobile source emission model framework and combined with the carbon balance method, construct an initial calculation model for carbon dioxide emissions from non-road mobile machinery to calculate the initial value of carbon dioxide emission factor. S20: Identify potential influencing factors of carbon dioxide emissions from non-road mobile machinery, screen and quantify key influencing factors through correlation analysis and rough set theory, and form a sample dataset; S30: Construct a deep extreme learning machine model based on the extreme learning machine autoencoder as the basic unit; preprocess the sample dataset and divide it into training and test sets; S40: Using the initial values ​​of key influencing factors and carbon dioxide emission factors as model inputs, and the measured values ​​of carbon dioxide emission factors as model outputs, the deep extreme learning machine model is trained using the training set to determine the model parameters. S50: Input the test set into the trained deep extreme learning machine model, validate the deep extreme learning machine model through performance evaluation metrics, and use the validated deep extreme learning machine model to estimate the carbon dioxide emissions of non-road mobile machinery.

[0023] Specifically, based on the NONROAD emission model framework for non-road mobile sources and combined with the carbon balance method, an initial calculation model for carbon dioxide emissions is constructed. This involves obtaining parameters such as fuel consumption, fuel type, and carbon content of non-road mobile machinery, and combining these with the carbon oxidation rate and the carbon-to-carbon dioxide conversion relationship to calculate initial values ​​of carbon dioxide emission factors with clear physical meaning. Simultaneously, parameters such as equipment type, service life, and load factor are integrated to improve the rationality of the initial estimate. Building upon this, multidimensional potential factors affecting carbon dioxide emissions from non-road mobile machinery are further identified, including equipment power, load rate, operating conditions, environmental conditions, and maintenance status. Correlation analysis is used to preliminarily screen the correlation between each factor and the measured emission factors, eliminating variables with low correlation. Subsequently, rough set theory is introduced to construct a decision table and perform attribute reduction. By calculating attribute dependencies and employing heuristic or intelligent optimization algorithms, a subset of key influencing factors is obtained, thereby achieving effective dimensionality reduction and quantification of high-dimensional variables, forming a model that includes key influencing factors and initial emission factors. The model uses a sample dataset of emission factors. Further, a deep extreme learning machine (LEM) model is constructed, using an LEM autoencoder as the basic unit. High-order feature representations of the data are extracted through unsupervised training of multiple layers of autoencoders. The weights of each layer are randomly initialized and the output weights are solved analytically. Before model training, the sample data undergoes missing value processing, outlier removal, and normalization. The training and test sets are divided according to a preset ratio. During model training, key influencing factors and initial emission factors are used as input variables, and measured carbon dioxide emission factors are used as output variables. The autoencoders are trained layer by layer, and the nonlinear mapping relationship between input and output is established and model parameters are determined by selecting activation functions and setting regularization parameters. In the model validation phase, the test set is input into the trained LEM model. The model performance is evaluated using metrics such as mean squared error, mean absolute error, and coefficient of determination. After meeting the accuracy requirements, the model is used to estimate carbon dioxide emissions from non-road mobile machinery under unknown operating conditions.

[0024] By constructing an initial calculation model for carbon dioxide emission factors based on the NONROAD non-road mobile source emission model framework and combining it with the carbon balance method, this paper introduces a key influencing factor screening mechanism that combines potential influencing factor identification, correlation analysis, and rough set theory. This mechanism extracts core factors with significant impact on carbon dioxide emissions from a large number of candidate variables. Furthermore, it combines a deep extreme learning machine model based on an automatic encoder to model and correct the nonlinear emission patterns of non-road mobile machinery under complex operating conditions. This effectively overcomes the problems of existing technologies that rely on empirical factors or average emission factors and cannot accurately reflect the coupled effects of multiple factors and dynamic changes in operating conditions. By organically combining mechanistic analysis with data-driven methods, this paper not only improves the accuracy and stability of carbon dioxide emission factor estimation for non-road mobile machinery but also enhances the model's adaptability to emission differences under different machinery types, operating states, and environmental conditions. Therefore, it can better meet the needs of refined accounting and practical application of carbon dioxide emissions from non-road mobile machinery.

[0025] As a preferred embodiment of the above, in step S10, an initial calculation model for the carbon dioxide emissions of non-road mobile machinery is constructed, including: S11: Based on the NONROAD non-road mobile source emission model framework, obtain the basic values ​​of hydrocarbon emission factors and carbon monoxide emission factors of non-road mobile machinery, and perform correction processing on the emission factors. Based on the pollutant emission factor calculation method and correction coefficient system (including transient condition correction coefficient TAF, deterioration coefficient DF, temperature correction coefficient TCF, etc.) of the NONROAD model, the basic values ​​of HC and CO emission factors of non-road mobile machinery are obtained, and the correction calculation is completed in combination with the actual operating conditions of the machinery to obtain HC and CO emission factor values ​​applicable to actual scenarios. S12: Based on the carbon balance method, establish a model for the relationship between fuel consumption and pollutant emissions of non-road mobile machinery equipped with gasoline engines and diesel engines; For vehicles equipped with gasoline engines, fuel consumption is determined using the "Test Method for Automobile Fuel Consumption" (GBT 12545-1990). ; In the formula: Gasoline consumption, L / (kW·h); , and , and Emissions, g / (kW·h); The density of gasoline at 20℃ is 747 kg / m³. 3 .Right now: ; Fuel consumption was determined using the carbon balance calculation method for diesel-powered machinery as specified in the "Methods for Measuring Fuel Consumption of Heavy Commercial Vehicles" (CB / T 27840-2011). ; In the formula: This represents diesel fuel consumption, expressed in L / (kW·h). , and , and Emissions, g / (kW·h); The density of diesel fuel at 20℃ is 845 kg / m³. 3 .Right now: ; S13: Based on the relational model, determine the initial calculation model for the carbon dioxide emission factor of non-road mobile machinery equipped with gasoline engines and non-road mobile machinery equipped with diesel engines.

[0026] S14: Verify the accuracy of the initial model using measured data; Measured emission data from four typical non-road mobile machinery types—forklifts, loaders, excavators, and bulldozers—were selected and substituted into the initial calculation model to obtain calculated CO2 emission factor values. The calculated values ​​were compared with the measured values ​​to analyze the relative error. For example, using 283 sets of measured data from non-road mobile machinery for verification, the results showed that the average absolute relative error between the model's calculated values ​​and the measured values ​​was 13.58%, the median was 14.64%, and the maximum absolute relative error reached 25.88%. Only 2.12% of the data had errors controlled within 2%, indicating that the initial model had systematic biases and needed further optimization considering influencing factors.

[0027] In this embodiment, step S13, determining the initial calculation model for the carbon dioxide emission factors of non-road mobile machinery equipped with gasoline engines and non-road mobile machinery equipped with diesel engines, includes: The CO2 emission factor model for gasoline engines is as follows: ; The diesel engine CO2 emission factor model is as follows: ; In the formula: This refers to gasoline consumption. This represents diesel fuel consumption, expressed in L / (kW·h). , and CO and CO2 emissions, g / (kW·h).

[0028] First, based on the NONROAD emission model framework for non-road mobile sources, baseline values ​​for hydrocarbon and carbon monoxide emission factors are obtained according to parameters such as equipment type, rated power range, emission standard stage, service life, and fuel type of non-road mobile machinery. These baseline values ​​are then corrected based on actual operating conditions. Correction processes include linear or non-linear corrections based on load rate, corrections based on ambient temperature and altitude, and corrections based on equipment aging, resulting in corrected values ​​for hydrocarbon and carbon monoxide emission factors that better reflect actual operating conditions. On this basis, a carbon balance method is used to establish the relationship between fuel consumption and pollutant emissions for non-road mobile machinery equipped with gasoline engines and those equipped with diesel engines. The carbon balance method, based on the principle of carbon conservation during fuel combustion, correlates the carbon input in the fuel with the carbon output in the emissions. For non-road mobile machinery equipped with gasoline engines… For non-road mobile machinery equipped with gasoline engines, the proportion of incompletely burned carbon can be determined by combining the carbon mass fraction of gasoline with the emissions of hydrocarbons and carbon monoxide, thereby deriving the amount of carbon dioxide generated. For non-road mobile machinery equipped with diesel engines, the carbon distribution relationship is adjusted by combining the carbon mass fraction of diesel, the lower hydrocarbon emission characteristics, and the corresponding carbon monoxide emission characteristics to establish a relationship model suitable for the emission characteristics of diesel engines. Subsequently, the modified hydrocarbon emission factors and carbon monoxide emission factors are substituted into the above relationship model to determine the initial calculation model of carbon dioxide emission factors for non-road mobile machinery equipped with gasoline engines and diesel engines, respectively. Specifically, the carbon dioxide emission under unit fuel consumption conditions can be normalized to obtain the initial emission factor in the form of carbon dioxide emission corresponding to unit fuel consumption, and can be further converted into an expression based on unit power output or unit workload as needed.

[0029] As a preferred embodiment of the above, in step S20, potential influencing factors of carbon dioxide emissions from non-road mobile machinery are identified, and key influencing factors are screened and quantified through correlation analysis and rough set theory to form a sample dataset, including: S21: Identify potential influencing factors of carbon dioxide emissions from non-road mobile machinery from three dimensions: machinery parameters, operating parameters, and environmental factors. Based on three dimensions—machinery parameters, operating parameters, and environmental factors—eight potential CO2 emission influencing factors were identified, including non-road mobile machinery type, emission stage, service life, cumulative service time, rated power, operating conditions, load, and machinery mass. S22: Based on the measured emission data of non-road mobile machinery, Spearman correlation coefficient was used to conduct correlation analysis between potential influencing factors and measured values ​​of carbon dioxide emission factors, and redundant influencing factors were eliminated. Based on 283 sets of measured emission data of non-road mobile machinery obtained from PEMS tests, Spearman correlation coefficient was used to conduct univariate correlation analysis on the measured values ​​of CO2 emission factors and eight potential influencing factors. The results showed that the correlation coefficient between service life and cumulative service time was 0.823, which is a strongly correlated redundant factor. Considering the availability of data, service life was retained to replace cumulative service time.

[0030] For example, correlation analysis showed that the measured values ​​of CO2 emission factors were strongly negatively correlated with rated power (r=-0.79, p<0.001), strongly positively correlated with service life (r=0.82, p<0.001), and strongly negatively correlated with emission stage (r=-0.83, p<0.001), verifying the significant impact of each factor on carbon emissions.

[0031] Correlation analysis was conducted on non-road mobile machinery across three dimensions: machinery parameters, operating parameters, and environmental factors. Parameters such as type of non-road mobile machinery, emission stage, service life, cumulative service time, rated power, operating conditions, load, and machinery mass were identified and correlated with emission factors such as CO, HC, and CO2. The correlation analysis results are shown in the correlation heatmap. Figure 2 As shown.

[0032] S23: Discretize the screened influencing factors, calculate the attribute dependence and importance of each influencing factor based on rough set theory, construct a decision table, and determine the key influencing factors; Rough set theory was used to calculate the attribute dependence and importance of the seven influencing factors after screening. First, the experimental data was discretized by a self-organizing feature map (SOM) neural network (e.g., the service life was divided into very new (1 year), new (3 years), general (5 years), and old (7 years), and the emission stage was divided into National I (1), National II (2), National III (3), and National IV (4)). Then, a decision table was constructed and the core attributes were obtained. Finally, the non-road mobile machinery type, emission stage, rated power, and service life were determined to be the key factors affecting CO2 emissions. According to the calculation, the weight of the four core factors exceeded 80%, of which the weight of rated power was 0.2802, emission stage was 0.2591, non-road mobile machinery type was 0.1907, and service life was 0.1437, which provided a basis for the selection of model input parameters.

[0033] S24: Organize the key influencing factors, the initial calculated values ​​of carbon dioxide emission factors, and the measured values ​​of carbon dioxide emission factors to construct a sample dataset.

[0034] By integrating three types of data—four key influencing factors, initial calculated values ​​of CO2 emission factors, and measured values ​​of CO2 emission factors—a model sample dataset containing 283 sets of measured emission data from non-road mobile machinery was formed. Each sample in the dataset contains six attributes, five of which are input attributes for the subsequent model and one is an output attribute.

[0035] For a specific example: Partial sample data of the dataset is shown in Table 1: Table 1. Partial sample data from the dataset Specifically, the process begins by identifying potential influencing factors from three dimensions: machine parameters, operating parameters, and environmental factors. Machine parameters include equipment type, engine type, rated power, displacement, service life, and emission stage. Operating parameters include load rate, speed, operating time, fuel consumption rate, and operating conditions. Environmental factors include ambient temperature, humidity, altitude, and atmospheric pressure. This multi-dimensional identification forms a set of candidate influencing factors covering equipment attributes, operating status, and the external environment. Corresponding data is obtained through equipment records, sensor data collection, and on-site monitoring. Subsequently, based on measured emissions data from non-road mobile machinery, the Spearman correlation coefficient is used to analyze the correlation between each potential influencing factor and the measured values ​​of carbon dioxide emission factors. By performing rank transformation on the sample data and calculating the correlation coefficient, influencing factors with a significant monotonic relationship with emission factors are screened out. At the same time, redundant variables with low correlation or duplicate information are removed, thus completing the initial dimensionality reduction. The process involves discretizing the selected influencing factors. Discretization can be achieved using equal-width binning, equal-frequency binning, or segmentation based on empirical thresholds to transform continuous variables into discrete attributes. A decision table is constructed using the discretized influencing factors as conditional attributes and the measured values ​​or levels of carbon dioxide emission factors as decision attributes. The dependence and importance of each influencing factor on the decision attributes are calculated based on rough set theory. The contribution of each attribute is determined by comparing the changes in classification ability before and after attribute reduction, thereby identifying a set of key influencing factors that significantly affect emission factors. Finally, the key influencing factors, the initial calculated values ​​of carbon dioxide emission factors obtained in step S10, and the corresponding measured values ​​of carbon dioxide emission factors are uniformly organized to construct a structured sample dataset. Each sample record contains the values ​​of key influencing factors and their corresponding initial calculated and measured values. Further data cleaning, consistency checks, and missing value processing are performed to ensure data quality.

[0036] Key influencing factors include: type of non-road mobile machinery, emission stage, rated power, and service life.

[0037] As a preferred embodiment of the above, in step S23, the screened influencing factors are discretized, including: Cluster analysis of experimental data is performed using a self-organizing feature map neural network (SAMR) to automatically classify continuous influencing factors into a preset number of discrete levels based on data distribution characteristics. Specifically, continuous influencing factors may include one or more variables such as rated power, service life, load rate, fuel consumption rate, operating time, engine speed, ambient temperature, relative humidity, and altitude. Before inputting the SAMR into the network, the original experimental data undergoes preprocessing, including missing value completion, outlier removal, and normalization or standardization, to reduce the impact of differences in the dimensions and value ranges of different variables on the clustering results. Subsequently, the processed continuous influencing factor samples are input into the SAMR, where the number of nodes in the input layer corresponds to the dimension of the sample features, and the competing layer consists of several neurons. For different cluster centers, the best matching unit is determined by calculating the distance between the input sample and the weight vector of each competing layer neuron. The weights of the best matching unit and its neighboring neurons are iteratively updated by combining the neighborhood function and the learning rate, so that the competing layer neurons gradually reflect the distribution characteristics and similarity relationships of the sample data. After the network training is completed, based on the allocation results of the best matching unit corresponding to each sample and the location of the cluster center, the continuous influencing factors are automatically divided into a preset number of discrete levels. The preset number can be set to three, four, five or other levels according to the sample size, variable fluctuation characteristics and subsequent decision table construction requirements, thereby forming discrete attribute encoding values.

[0038] As a preferred embodiment of the above, in step S30, constructing a deep extreme learning machine model based on an extreme learning machine autoencoder as the basic unit includes: The Extreme Learning Machine (ELM) is used as the basic unit for unsupervised learning. The output weight matrix is ​​obtained by the least squares method, and a deep Extreme Learning Machine is constructed in a stack manner.

[0039] The construction of a deep extreme learning machine model includes: Set the number of hidden layers in the deep extreme learning machine model, with each layer corresponding to an extreme learning machine autoencoder; The hidden layer output of the previous Extreme Learning Machine (ELM) autoencoder is used as the input of the next ELM autoencoder. Each ELM autoencoder is trained independently and the training process is independent of each other, and there is no need for backpropagation of error.

[0040] This step involves building a Deep Extreme Learning Machine (DELM) model based on the Extreme Learning Machine-Autoencoder (ELM-AE) as the basic unit, and completing the standardization of sample data and dataset partitioning. Specifically, it consists of the following sub-steps: S31: Construct the DELM model and determine the input and output parameters A DELM model based on ELM-AE is constructed. The model input consists of the initial calculated value of CO2 emission factor plus four core influencing factors (non-road mobile machinery type, rated power, emission stage, and service life). The model output is the measured value of CO2 emission factor, realizing the mapping prediction from the initial calculated value to the high-precision measured value.

[0041] S32: Standardization of Sample Datasets The 283 sets of sample datasets were standardized to eliminate the model training bias caused by the differences in the dimensions of various indicators (such as rated power in kW, service life in years, and emission factor in g / (kW・h)). The standardized sample datasets were obtained, which improved the accuracy and convergence speed of model training.

[0042] S33: Holdout method for partitioning training and test sets The standardized sample dataset was divided into training and test sets in a ratio of 80%:20% using the hold-out method. 226 sets of data were used as the training set for training the parameters of the DELM model, and 57 sets of data were used as the test set for verifying the accuracy of the model. The partitioning process ensured the randomness and uniformity of the data distribution.

[0043] Specifically, firstly, the input variables in the preprocessed sample dataset are used as model inputs. These input variables include key influencing factors selected through correlation analysis and rough set theory, as well as the initial calculated value of the carbon dioxide emission factor obtained in step S10. Based on this, an Extreme Learning Machine (ELM) autoencoder is used as the basic unit for unsupervised learning to construct the model hierarchy. Each ELM autoencoder includes an input layer, a hidden layer, and an output layer. The input layer receives the input data for the current layer, the hidden layer extracts the latent features of the input data through nonlinear mapping, and the output layer reconstructs or expresses the features of the input data. Unlike traditional autoencoders that rely on backpropagation for iterative optimization, in this implementation, the connection weights from the input layer to the hidden layer and the hidden layer biases can be determined randomly. The output weight matrix from the hidden layer to the output layer is solved analytically using the least squares method. That is, after the input weight matrix and biases are determined, the hidden layer output matrix is ​​first calculated based on the input samples. Then, using the input samples themselves as the target output, the least squares method is used to find the output weight matrix that minimizes the sum of squared errors, thus completing the unsupervised training of the single-layer extreme learning machine autoencoder. The hidden layer output matrix can be obtained by linearly combining the input samples and the input weight matrix and then applying the least squares method to Si. The gmoid function, hyperbolic tangent function, radial basis function, ReLU function, or other nonlinear activation functions are used for mapping. After training a single-layer extreme learning machine (ELM) autoencoder, a deep extreme learning machine is constructed using a stacked approach. This involves connecting multiple ELM autoencoders sequentially in hierarchical order, using the hidden layer output features of the previous autoencoder as the input to the next, thus forming a multi-layered, progressively abstract deep network structure. Specifically, the original input samples are first input into the first-layer ELM autoencoder to obtain the first-layer hidden feature representation, and then this first-layer hidden feature representation is input into the second-layer ELM autoencoder. The machine learning autoencoder extracts higher-level features, and then continues to use the output of the second layer as the input of the third layer or higher layers, stacking sequentially until the entire deep extreme learning machine model is completed. In some implementations, the number of nodes in each hidden layer can be set according to the dimension of the input variables, the sample size, and the nonlinear complexity of the data. It can be configured by using the same number of nodes, increasing the number of nodes layer by layer, decreasing the number of nodes layer by layer, or increasing first and then decreasing, to adapt to different modeling needs. At the same time, a regularization term can be introduced when solving the output weight matrix by the least squares method to improve the solution stability and reduce the risk of overfitting caused by sample noise or a large number of hidden layer nodes.

[0044] In step S30, the sample dataset is preprocessed and divided into a training set and a test set, including: Standardize the sample dataset to eliminate differences in the dimensions of each indicator; The standardized sample dataset is divided into training and test sets according to a preset ratio using the hold-out method.

[0045] In this embodiment, in step S40, the deep extreme learning machine model is trained using the training set to determine the model parameters, including: S41: Use the Extreme Learning Machine autoencoder to perform unsupervised training on the training set data to obtain the corresponding weight parameters, and build a multi-layer stacked deep Extreme Learning Machine model based on the weight parameters. The formula for calculating the output layer weights of the basic unit of ELM-AE is as follows: ; In the formula, It is the identity matrix; Here, H is the regularization coefficient; H is the hidden layer output matrix. For input and output; S42: Input the training set data into the first layer of the Extreme Learning Machine autoencoder of the Deep Extreme Learning Machine model to obtain the hidden layer feature vectors, and use the hidden layer feature vectors as the input of the next layer of the Extreme Learning Machine autoencoder. Train layer by layer until the training of all hidden layers is completed. The ELM-AE is used as the unsupervised learning unit to train the input data of the training set. The output weight matrix of the ELM-AE obtained by the least squares method is saved. Based on this, the stack is constructed to build a deep extreme learning machine (DELM) model to extract deep features of the data. S43: The output layer of the deep extreme learning machine model is trained based on the measured values ​​of carbon dioxide emission factors corresponding to the training set, and the parameters of the deep extreme learning machine model are determined by adjusting the model regularization coefficient.

[0046] The training set input data is fed into the first layer ELM-AE of the DELM model to obtain the feature vector of the first hidden layer, which is then used as the input of the next layer ELM-AE. This process is repeated layer by layer until all hidden layers are trained, and the input weight matrix and hidden layer feature vector of each layer are obtained, thus realizing multi-layer mapping of the input data features. S44: Iterative adjustment to determine the optimal parameters of the model The output layer of the DELM model is trained based on the output data (measured values ​​of CO2 emission factors) of the training set. By iteratively adjusting core parameters such as the regularization coefficient C, the deviation between the model's predicted values ​​and the measured values ​​is minimized, and the optimal parameters of the model are finally determined.

[0047] For a specific example: Model training is implemented using MATLAB 2019a software. During the iteration process, the root mean square error (RMSE) between the predicted and measured values ​​is used as the deviation judgment index. When the RMSE drops to 23.39 g / (kW・h) and tends to stabilize, the regularization coefficient at this time is determined to be the optimal parameter of the model.

[0048] As a preferred embodiment of the above, step S50 verifies the deep extreme learning machine model using performance evaluation metrics, including: S51: Input the test set into the trained deep extreme learning machine model to obtain the predicted value of carbon dioxide emission factor for each test sample; The input parameters (initial calculated values ​​of CO2 emission factors + 4 core influencing factors) of 57 test sets were fed into the trained DELM model to obtain the predicted values ​​of CO2 emission factors for each sample.

[0049] S52: Calculate the preset performance evaluation index value based on the predicted and measured values ​​of carbon dioxide emission factors; Among them, the mean absolute percentage error (MAPE), mean absolute error (MAE), and root mean square error (RMSE) are selected as the model performance evaluation indicators, and the values ​​of each indicator are calculated respectively. The calculation formulas are as follows: ; ; ; In the formula: This represents the measured value of carbon dioxide emissions; Values ​​for predicting carbon dioxide emissions; The number of test samples (n=57 in this example); The performance evaluation results of the DELM model in this embodiment are as follows: MAPE=2.84%, MAE=23.36g / (kW·h), RMSE=23.39g / (kW·h), and the maximum relative absolute value error is only 2.92%.

[0050] For a specific example: Table 2 shows the model error indices for different types of machinery. The errors for each type are at a low level, verifying the model's universality. Table 2. Model error indices for different types of machinery The trained model and the test set input are used for prediction. Finally, the prediction model is analyzed using model performance evaluation metrics. The results are as follows: Figures 3 to 5 As shown.

[0051] S53: Determine the validation result of the deep extreme learning machine model based on the relationship between the performance evaluation index value and the preset threshold.

[0052] Based on the evaluation index calculation results, the MAPE of the DELM model in this embodiment is far below the industry accuracy threshold of 5%, the MAE and RMSE values ​​are small and similar, the error distribution is uniform, and there is no significant difference in the prediction error of different types of non-road mobile machinery. It is determined that the model has good prediction accuracy and stability and can be used for high-precision estimation of CO2 emission factors of non-road mobile machinery.

[0053] Specifically, firstly, the test set is input into the trained deep extreme learning machine model to obtain the predicted carbon dioxide emission factor value for each test sample. The test set is not used in model training but only for independent verification of the model's generalization ability and prediction accuracy. The input variables of the test set are consistent with those of the training set, including key influencing factors and the initial calculated value of the carbon dioxide emission factor. The output corresponds to the measured value of the carbon dioxide emission factor for each test sample. Before inputting into the model, the test set undergoes the same data processing procedure as the training phase, including the same encoding method, normalization parameters, discretization rules, or standardization rules, to ensure consistency between the test data and training data in the feature space. Then, the processed test samples are input into the trained deep extreme learning machine model one by one or in batches. After feature mapping in each hidden layer and calculation in the output layer, the predicted value of the carbon dioxide emission factor for each test sample is obtained. Next, based on the predicted and measured values ​​of the carbon dioxide emission factor, the predicted value is calculated... The system calculates preset performance evaluation index values, which may include mean squared error, mean absolute error, and coefficient of determination. Mean squared error characterizes the average level of the squared deviation between predicted and measured values; mean absolute error characterizes the average level of the absolute deviation between predicted and measured values; and the coefficient of determination characterizes the model's fit to the trend of measured values. In some implementations, root mean square error, mean absolute percentage error, mean relative error, deviation coefficient, or residual distribution statistics can be further calculated to comprehensively evaluate model performance from multiple aspects, including absolute error, relative error, and fit consistency. Specifically, the difference between the predicted and measured values ​​for each test sample is calculated, and then all test samples are statistically summarized according to the corresponding index formulas to obtain the values ​​of each performance evaluation index. Finally, the validation result of the deep extreme learning machine model is determined based on the relationship between the performance evaluation index values ​​and preset thresholds. Corresponding thresholds can be set for different performance evaluation indices. Example 2:

[0054] According to another aspect of the present invention, a carbon dioxide emission estimation system for non-road mobile machinery is provided, employing the carbon dioxide emission estimation method for non-road mobile machinery of Embodiment 1. The system has a modular architecture, including an initial model building module, an influencing factor screening module, a DELM model building module, a model training and validation module, and an emission estimation module. These modules work together to achieve automated and high-precision carbon emission estimation. The specific functions of each module are as follows: The present invention also includes a carbon dioxide emission estimation system for non-road mobile machinery, such as Figure 6 As shown, the system includes: The initial model building module is used to construct an initial calculation model for carbon dioxide emissions from non-road mobile machinery based on the NONROAD non-road mobile source emission model framework and combined with the carbon balance method, and is used to calculate the initial value of the carbon dioxide emission factor. The influencing factor screening module is used to identify potential influencing factors of carbon dioxide emissions from non-road mobile machinery. It screens and quantifies key influencing factors through correlation analysis and rough set theory to form a sample dataset. The Deep Extreme Learning Machine Model Building Module is used to build a Deep Extreme Learning Machine model based on the Extreme Learning Machine Autoencoder as the basic unit. The data preprocessing module is used to preprocess the sample dataset and divide it into training and test sets; The model training module is used to train the deep extreme learning machine model using the training set with the initial values ​​of key influencing factors and carbon dioxide emission factors as model inputs and the measured values ​​of carbon dioxide emission factors as model outputs, thereby determining the model parameters. The model validation and estimation module is used to input the test set into the trained deep extreme learning machine model, validate the deep extreme learning machine model through performance evaluation metrics, and estimate the carbon dioxide emissions of non-road mobile machinery using the validated deep extreme learning machine model.

[0055] Initial model building module: This project aims to construct and preliminarily validate an initial calculation model for CO2 emissions from non-road mobile machinery, based on the US Environmental Protection Agency's (EPA) Nonroad model framework and carbon balance method. Specifically, it involves: obtaining and correcting the baseline values ​​of HC and CO emission factors using the correction coefficient system of the Nonroad model; deriving the correlation formulas between fuel consumption and pollutant emissions for gasoline and diesel engines respectively, and inverting to obtain the initial calculation formula for CO2 emission factors; validating the initial model using 283 sets of measured data, analyzing the relative error between calculated and measured values, and clarifying the accuracy characteristics of the initial model.

[0056] Influencing Factor Screening Module: To identify, screen, and quantify key influencing factors of CO2 emissions from non-road mobile machinery, a model sample dataset was ultimately compiled. Specifically, the following steps were taken: potential influencing factors were identified from three dimensions: machinery parameters, operating parameters, and environmental factors; Spearman correlation coefficient analysis was used to eliminate redundant factors, and the service life was retained to replace the cumulative service time; through rough set theory combined with SOM neural network discretization, machinery type, emission stage, rated power, and service life were determined as core influencing factors; and the core influencing factors, initial calculated values ​​of CO2 emission factors, and measured values ​​were integrated to form a sample dataset containing 283 sets of data.

[0057] DELM model building module: This method is used to construct a Deep Extreme Learning Machine (DELM) model, completing the preprocessing of the sample dataset and the partitioning of the training and test sets. Specifically, the DELM model is constructed using ELM-AE as the basic unit. The model input is defined as the initial calculated value of the CO2 emission factor plus four core influencing factors, and the output is the measured value of the CO2 emission factor. The sample dataset is standardized to eliminate dimensional differences. Using the hold-out method, the standardized data is divided into 226 training sets and 57 test sets in an 80%:20% ratio.

[0058] Model training and validation module: This method is used to train and verify the prediction accuracy and stability of the DELM model, and to determine the optimal parameters of the model. Specifically, a DELM model is constructed based on the ELM-AE output layer weight formula stack. The model is trained layer by layer on the training set data to obtain the input weight matrix of each layer and the feature vector of the hidden layer. Through iterative adjustments of parameters such as the regularization coefficient, the deviation between the model's predicted values ​​and the measured values ​​is minimized, thus determining the optimal parameters. The test set is then input into the trained DELM model to obtain predicted values. Three evaluation metrics—MAPE, MAE, and RMSE—are calculated to verify the model's prediction accuracy and stability.

[0059] Emissions estimation module: This method enables high-precision prediction of CO2 emission factors for non-road mobile machinery, providing data support for carbon emission control. Specifically, it involves: collecting relevant parameters of the machinery to be estimated (machine type, rated power, emission stage, service life, fuel consumption, etc.); obtaining initial calculated values ​​of CO2 emission factors through the initial calculation model in the initial model building module; standardizing the initial calculated values ​​and the core influencing factors of the machinery; and inputting the standardized parameters into the validated DELM model to finally obtain a high-precision predicted value of the CO2 emission factor for the machinery.

[0060] The adjustment system described above in this invention can effectively realize the method for estimating carbon dioxide emissions from non-road mobile machinery, and the technical effects it can achieve are as described in the above embodiments, and will not be repeated here.

[0061] Similarly, the above-mentioned optimization schemes for the system can also achieve the optimization effects corresponding to the methods in Embodiment 1, which will not be repeated here.

[0062] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and accompanying drawings are merely exemplary illustrations of the application as defined herein, and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Thus, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for estimating carbon dioxide emissions from non-road mobile machinery, characterized in that, The method includes: Based on the NONROAD non-road mobile source emission model framework and combined with the carbon balance method, an initial calculation model for carbon dioxide emissions from non-road mobile machinery is constructed to calculate the initial value of the carbon dioxide emission factor. Identify potential influencing factors of carbon dioxide emissions from non-road mobile machinery, screen and quantify key influencing factors through correlation analysis and rough set theory, and form a sample dataset. Construct a deep extreme learning machine model with the extreme learning machine autoencoder as the basic unit; preprocess the sample dataset and divide it into training set and test set; Using the key influencing factors and the initial value of the carbon dioxide emission factor as model input, and the measured value of the carbon dioxide emission factor as model output, the deep extreme learning machine model is trained using the training set to determine the model parameters; The test set is input into the trained deep extreme learning machine model, and the deep extreme learning machine model is validated by performance evaluation metrics. The validated deep extreme learning machine model is then used to estimate the carbon dioxide emissions of non-road mobile machinery.

2. The method for estimating carbon dioxide emissions from non-road mobile machinery according to claim 1, characterized in that, An initial calculation model for carbon dioxide emissions from non-road mobile machinery is constructed, including: Based on the NONROAD non-road mobile source emission model framework, the basic values ​​of hydrocarbon emission factors and carbon monoxide emission factors of non-road mobile machinery are obtained, and the emission factors are corrected. Based on the carbon balance method, a model is established to show the relationship between fuel consumption and pollutant emissions of non-road mobile machinery equipped with gasoline engines and diesel engines. Based on the aforementioned relationship model, an initial calculation model for the carbon dioxide emission factors of non-road mobile machinery equipped with gasoline engines and non-road mobile machinery equipped with diesel engines is determined.

3. The method for estimating carbon dioxide emissions from non-road mobile machinery according to claim 2, characterized in that, Initial calculation models for carbon dioxide emission factors of non-road mobile machinery equipped with gasoline engines and non-road mobile machinery equipped with diesel engines were determined. include: The CO2 emission factor model for gasoline engines is as follows: ； The diesel engine CO2 emission factor model is as follows: ； In the formula: This refers to gasoline consumption. This refers to diesel fuel consumption. , and CO and CO2 emissions.

4. The method for estimating carbon dioxide emissions from non-road mobile machinery according to claim 1, characterized in that, Potential influencing factors of CO2 emissions from non-road mobile machinery were identified. Key influencing factors were screened and quantified using correlation analysis and rough set theory, resulting in a sample dataset, including: From three dimensions—machine parameters, operating parameters, and environmental factors—we identify potential influencing factors of carbon dioxide emissions from non-road mobile machinery. Based on measured emissions data from non-road mobile machinery, Spearman correlation coefficient was used to conduct correlation analysis between the potential influencing factors and the measured values ​​of carbon dioxide emission factors, and redundant influencing factors were eliminated. The screened influencing factors are discretized, and the attribute dependence and importance of each influencing factor are calculated based on rough set theory. A decision table is constructed and the key influencing factors are identified. The key influencing factors, the initial calculated values ​​of the carbon dioxide emission factor, and the measured values ​​of the carbon dioxide emission factor were compiled to construct a sample dataset.

5. The method for estimating carbon dioxide emissions from non-road mobile machinery according to claim 4, characterized in that, The key influencing factors include: type of non-road mobile machinery, emission stage, rated power, and service life.

6. The method for estimating carbon dioxide emissions from non-road mobile machinery according to claim 4, characterized in that, The screened influencing factors were discretized, including: By using a self-organizing feature mapping neural network to perform cluster analysis on experimental data, continuous influencing factors are automatically divided into a preset number of discrete levels according to the data distribution characteristics.

7. The method for estimating carbon dioxide emissions from non-road mobile machinery according to claim 1, characterized in that, Constructing a deep extreme learning machine model based on the extreme learning machine autoencoder as the basic unit includes: The Extreme Learning Machine (ELM) is used as the basic unit for unsupervised learning. The output weight matrix is ​​obtained by the least squares method, and a deep Extreme Learning Machine is constructed in a stack manner.

8. The method for estimating carbon dioxide emissions from non-road mobile machinery according to claim 1, characterized in that, The deep extreme learning machine model is trained using the training set to determine the model parameters, including: The training set data is trained unsupervised using an extreme learning machine autoencoder to obtain the corresponding weight parameters, and a multi-layer stacked deep extreme learning machine model is constructed based on the weight parameters. The training set data is input into the first layer of the Extreme Learning Machine Autoencoder of the Deep Extreme Learning Machine model to obtain the hidden layer feature vectors. The hidden layer feature vectors are then used as the input of the next layer of the Extreme Learning Machine Autoencoder. Training is performed layer by layer until all hidden layers have been trained. The output layer of the deep extreme learning machine model is trained based on the measured values ​​of carbon dioxide emission factors corresponding to the training set, and the parameters of the deep extreme learning machine model are determined by adjusting the model regularization coefficient.

9. The method for estimating carbon dioxide emissions from non-road mobile machinery according to claim 1, characterized in that, The deep extreme learning machine model was validated using performance evaluation metrics, including: The test set is input into the trained deep extreme learning machine model to obtain the predicted value of carbon dioxide emission factor for each test sample; Based on the predicted and measured values ​​of the carbon dioxide emission factor, the preset performance evaluation index values ​​are calculated. The verification result of the deep extreme learning machine model is determined based on the relationship between the performance evaluation index value and the preset threshold.

10. A carbon dioxide emission estimation system for non-road mobile machinery, characterized in that, The system includes: The initial model building module is used to construct an initial calculation model for carbon dioxide emissions from non-road mobile machinery based on the NONROAD non-road mobile source emission model framework and combined with the carbon balance method, and is used to calculate the initial value of the carbon dioxide emission factor. The influencing factor screening module is used to identify potential influencing factors of carbon dioxide emissions from non-road mobile machinery. It screens and quantifies key influencing factors through correlation analysis and rough set theory to form a sample dataset. The Deep Extreme Learning Machine Model Building Module is used to build a Deep Extreme Learning Machine model based on the Extreme Learning Machine Autoencoder as the basic unit. The data preprocessing module is used to preprocess the sample dataset and divide it into a training set and a test set. The model training module is used to train the deep extreme learning machine model using the key influencing factors and the initial value of the carbon dioxide emission factor as model input, and the measured value of the carbon dioxide emission factor as model output, and to determine the model parameters using the training set. The model validation and estimation module is used to input the test set into the trained deep extreme learning machine model, validate the deep extreme learning machine model through performance evaluation indicators, and estimate the carbon dioxide emissions of non-road mobile machinery using the validated deep extreme learning machine model.

Images