Carbon emission prediction method, system, device and storage medium for high energy consumption enterprise

Abstract

The application relates to the technical field of carbon emission prediction, and discloses a carbon emission prediction method, system, device and storage medium for high-energy-consumption enterprises, which comprises the following steps: establishing a carbon emission intensity coupling model according to carbon emission key indicators; adopting kernel density estimation and Bayesian inference to establish a marginal probability density estimation model of the carbon emission key indicators; adopting a Copula structure to establish a carbon emission scenario coupling model; sampling the carbon emission scenario coupling model and performing inverse mapping to obtain an initial panoramic coupling scenario set; performing scenario reduction on the initial panoramic coupling scenario set to obtain a representative scenario set; calculating the carbon emission prediction value of each representative scenario according to the carbon emission intensity coupling model, and performing weighted summation to obtain the total carbon emission prediction value of the enterprise to be predicted. The application introduces the whole reduction theory, improves the representation ability and prediction robustness of the model for a complex production system, and thus improves the accuracy of carbon emission prediction for high-energy-consumption enterprises.

Description

Technical Field

[0001] This invention relates to the field of carbon emission prediction technology, and in particular to a method, system, device and storage medium for predicting carbon emissions of high-energy-consuming enterprises. Background Technology

[0002] Currently, high-energy-consuming enterprises such as steel, chemical, power and cement industries are facing enormous pressure to reduce carbon emissions and manage carbon assets. Accurately predicting future carbon emissions is not only the basis for enterprises to formulate production plans and fulfill their carbon quota trading obligations, but also a key basis for regulatory authorities to implement precise control.

[0003] Existing carbon emission prediction methods are mainly divided into two categories: deterministic calculation models based on emission factor methods and black-box prediction models based on time series analysis and machine learning. Among them, deterministic calculation models based on emission factor methods are currently the most commonly used calculation methods. This method is essentially a post-event calculation rather than a pre-event prediction method. It assumes that emission factors are static and constant. However, the production process of high-energy-consuming enterprises is a typical complex system with deep coupling between people, machines, and the environment. The system complexity of carbon emission generation mechanisms means that the carbon emissions of enterprises are not only affected by equipment energy consumption, but also by the nonlinear effects of multi-dimensional factors such as process flow arrangement and personnel operating habits. This post-event calculation method not only ignores the dynamic changes such as equipment operating condition drift and process parameter fluctuations during the production process, but also fails to reflect the potential emission reduction effects brought about by management optimization or personnel behavior improvement. This results in overly rigid prediction results and a lack of ability to extrapolate future scenarios.

[0004] Black-box prediction models based on time series analysis and machine learning mostly utilize algorithms such as Long Short-Term Memory Networks and Support Vector Machines to perform regression predictions based on historical energy consumption data. However, data-driven black-box prediction methods rely excessively on the physical characteristics of historical data and lack an understanding of the internal mechanisms of the system, thus resulting in poor model robustness. Summary of the Invention

[0005] This invention provides a method, system, device, and storage medium for predicting carbon emissions of high-energy-consuming enterprises, which can solve the above-mentioned technical problems and achieve the technical effect of significantly improving the accuracy of carbon emission prediction for high-energy-consuming enterprises.

[0006] In a first aspect, the present invention provides a method for predicting carbon emissions from high-energy-consuming enterprises, the method comprising: Based on the overall reduction theory, the key carbon emission indicators of high-energy-consuming enterprises are determined, and a carbon emission intensity coupling model of high-energy-consuming enterprises is established. Based on the historical data of the enterprises to be predicted, a marginal probability density estimation model for the key carbon emission indicators is established using the kernel density estimation algorithm and the Bayesian inference algorithm. Based on the marginal probability density estimation model and using the Copula structure, a carbon emission scenario coupling model for the enterprises to be predicted is established. Scene sampling is performed on the carbon emission scene coupling model, and probability space inverse mapping is performed according to the marginal probability density estimation model to obtain an initial panoramic coupled scene set; The initial panoramic coupled scene set is reduced using a preset reduction algorithm to obtain a representative scene set; Based on the carbon emission intensity coupling model, the predicted carbon emission value for each representative scenario in the representative scenario set is calculated, and the predicted carbon emission values ​​for each representative scenario are weighted and summed to obtain the total predicted carbon emission value for the enterprise to be predicted.

[0007] Furthermore, the steps of determining the key carbon emission indicators of high-energy-consuming enterprises based on the overall reduction theory and establishing a carbon emission intensity coupling model for high-energy-consuming enterprises include: Based on the holistic reduction theory, the key carbon emission indicators for high-energy-consuming enterprises are determined. These key carbon emission indicators include physical indicators, theoretical indicators, and human-related indicators. Based on the physical indicators, a theoretical physical baseline emission intensity model for high-energy-consuming enterprises is established; based on the theoretical indicators, a process fluctuation correction coefficient is calculated; and based on the humanistic indicators, a behavioral intervention coefficient is calculated. Based on the process fluctuation correction coefficient and the behavioral intervention coefficient, the theoretical physical baseline emission intensity model is modified to obtain the carbon emission intensity coupled model.

[0008] Furthermore, the step of establishing a marginal probability density estimation model for the key carbon emission indicators based on historical data of the enterprise to be predicted, using kernel density estimation algorithm and Bayesian inference algorithm, includes: Based on the historical data of the enterprise to be predicted, the first probability density function of the physical index is constructed using an adaptive bandwidth kernel density estimation algorithm, and the second probability density function of the rational index and the third probability density function of the human index are constructed using a Bayesian inference method. By combining the first probability density function, the second probability density function, and the third probability density function, a marginal probability density estimation model for the key carbon emission indicators of the enterprise to be predicted is obtained.

[0009] Furthermore, the step of establishing a coupled model of the carbon emission scenario for the enterprise to be predicted based on the marginal probability density estimation model and using a Copula structure includes: Based on tail correlation analysis, the family of Pair-Copula functions among various key carbon emission indicators is determined, and the optimal parameters of the family of Pair-Copula functions are calculated using the maximum likelihood estimation algorithm based on the historical data. Based on the marginal probability density estimation model and the calculated Pair-Copula function family, a joint probability density function is constructed using the C-Vine structure to obtain a coupled model of the carbon emission scenario for the enterprise to be predicted.

[0010] Further, the step of sampling the carbon emission scene coupling model and performing probability space inverse mapping based on the marginal probability density estimation model to obtain the initial panoramic coupled scene set includes: Independent uniform random numbers are generated using the Latin hypercube sampling method. Based on the joint probability density function of the carbon emission scenario coupling model, the independent uniform random numbers are subjected to inverse Rosenblatt transformation to obtain the coupling correlation variables. Based on the inverse cumulative distribution function of the marginal probability density estimation model, the coupling correlation variables are subjected to probability space inverse mapping to obtain the initial panoramic coupling scene set.

[0011] Furthermore, the step of using a preset reduction algorithm to reduce the initial panoramic coupled scene set to obtain a representative scene set includes: Using a pre-defined surrogate model, the characteristic importance of each key carbon emission indicator to the total carbon emission is calculated, and the sensitivity factor of each key carbon emission indicator is obtained. The surrogate model is constructed based on a machine learning algorithm and trained using historical data of the enterprise to be predicted. Based on the variable distribution status of each key carbon emission indicator in each coupled scenario in the initial panoramic coupled scenario set, determine the indicator anomaly penalty item for each key carbon emission indicator in each coupled scenario. Using the sensitivity factor and the index anomaly penalty term as index weights, the Kantorovich distance between coupled scenarios is weighted and corrected to obtain the improved Kantorovich distance. Based on the improved Kantorovich distance, the synchronous back-substitution elimination method is used to reduce the initial panoramic coupled scene set to obtain a representative scene set.

[0012] Furthermore, the steps of establishing a theoretical physical baseline emission intensity model for high-energy-consuming enterprises based on the physical indicators, calculating process fluctuation correction coefficients based on the rational indicators, and calculating behavioral intervention coefficients based on the human indicators include: The direct carbon emissions of the fuel are calculated based on fuel activity data, the lower heating value of the fuel and the carbon content per unit of the fuel in the aforementioned physical indicators. Based on the operating load rate of energy-consuming equipment, aging attenuation coefficient and physical operating efficiency of waste gas treatment equipment in the physical indicators, calculate the comprehensive energy efficiency coefficient of the equipment. The direct carbon emissions are corrected based on the comprehensive energy efficiency coefficient of the equipment to obtain the theoretical physical baseline emission intensity model for high-energy-consuming enterprises; Based on the production status, energy medium scheduling coupling degree, and production scheduling compactness index in the aforementioned indicators, the process fluctuation correction coefficient is obtained by weighted summation. The behavioral intervention coefficient is obtained by performing nonlinear calculations on the personnel proficiency, personnel fatigue, and personnel operational reliability in the aforementioned human factors indicators.

[0013] Secondly, the present invention provides a carbon emission prediction system for high-energy-consuming enterprises, the system comprising: The first model construction module 10 is used to determine the key carbon emission indicators of high-energy-consuming enterprises based on the overall reduction theory, and to establish a carbon emission intensity coupling model for high-energy-consuming enterprises. The second model construction module 20 is used to establish a marginal probability density estimation model of the key carbon emission indicators based on the historical data of the enterprise to be predicted, using a kernel density estimation algorithm and a Bayesian inference algorithm. Based on the marginal probability density estimation model, and using a Copula structure, a carbon emission scenario coupling model of the enterprise to be predicted is established. The scene generation module 30 is used to sample the carbon emission scene coupling model and perform probability space inverse mapping according to the marginal probability density estimation model to obtain an initial panoramic coupled scene set. Scene reduction module 40 is used to reduce the initial panoramic coupled scene set using a preset reduction algorithm to obtain a representative scene set; The carbon emission prediction module 50 is used to calculate the carbon emission prediction value of each representative scenario in the representative scenario set according to the carbon emission intensity coupling model, and to perform a weighted summation of the carbon emission prediction values ​​of each representative scenario to obtain the total carbon emission prediction value of the enterprise to be predicted.

[0014] Thirdly, embodiments of the present invention also provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0015] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0016] This invention provides a method, system, device, and storage medium for predicting carbon emissions from high-energy-consuming enterprises. By introducing the WRT (Warshall Reduction Theory) approach, it enhances the model's comprehensive representation of the complex production systems of high-energy-consuming enterprises and improves its predictive robustness. Through joint probabilistic modeling using C-Vine Copula, it accurately captures the complex nonlinear coupling risks among multidimensional factors, improving the ability to predict carbon emission surge events. By improving the Kantorovich distance synchronous back-substitution elimination algorithm, it ensures that massive scenarios are reduced to a small number of representative scenarios while still preserving key high-emission risk characteristics with high fidelity. This invention effectively solves the problems of traditional methods being single-dimensional, lacking interpretability, and unable to cope with dynamic coupling risks, significantly improving the accuracy of carbon emission prediction for high-energy-consuming enterprises. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the carbon emission prediction method for high-energy-consuming enterprises in an embodiment of the present invention. Figure 2 This is a schematic diagram of the carbon emission prediction system for high-energy-consuming enterprises in an embodiment of the present invention; Figure 3 This is an internal structural diagram of the computer device in an embodiment of the present invention.

[0018] Figure label: 10. First model building module; 20. Second model building module; 30. Scene generation module; 40. Scene reduction module; 50. Carbon emission prediction module. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 The first embodiment of the present invention proposes a carbon emission prediction method for high-energy-consuming enterprises, including steps S10 to S50: Step S10: Based on the overall reduction theory, determine the key carbon emission indicators of high-energy-consuming enterprises and establish a carbon emission intensity coupling model for high-energy-consuming enterprises. Step S20: Based on the historical data of the enterprise to be predicted, a marginal probability density estimation model of the key carbon emission indicators is established using the kernel density estimation algorithm and the Bayesian inference algorithm. Based on the marginal probability density estimation model and using the Copula structure, a carbon emission scenario coupling model of the enterprise to be predicted is established. Step S30: Sampling is performed on the carbon emission scenario coupling model, and probability space inverse mapping is performed according to the marginal probability density estimation model to obtain an initial panoramic coupling scenario set; Step S40: Use a preset reduction algorithm to reduce the initial panoramic coupled scene set to obtain a representative scene set; Step S50: Based on the carbon emission intensity coupling model, calculate the carbon emission prediction value for each representative scenario in the representative scenario set, and perform a weighted summation of the carbon emission prediction values ​​for each representative scenario to obtain the total carbon emission prediction value for the enterprise to be predicted.

[0021] This embodiment provides a carbon emission prediction method for high-energy-consuming enterprises based on the Wuli-Shili-Renli System Approach (WRT). Before explaining this solution, the Wuli-Shili-Renli System Approach must first be explained. The core of the Wuli-Shili-Renli System Approach is to comprehensively analyze and solve complex system problems from three dimensions: physical, logical, and human. Its core logic is that any complex system simultaneously contains an objectively existing material basis, orderly operating rules and processes, and human factors involved in it. These three are interconnected and inseparable, and need to be considered holistically. Among them, physical factors are the objective material basis of the system, referring to the objectively existing, quantifiable, and verifiable natural laws and material entities in the system, following the basic laws of natural science, and are the premise for the existence of the system; logical factors are the system's structural framework and process logic, referring to the methods, processes, systems, and tools for dealing with physical problems; and human factors are the human factors and relationships in the system. The analytical framework of reductionism preserves the behavior of the overall system's mathematical model under perturbation scenarios in the system's time response curve (perturbed trajectory) through numerical simulation. This transforms the problem of understanding complex systems into extracting analytical solutions from the mathematical model and the perturbed trajectory. Specifically, for a highly uncertain, high-dimensional, time-varying, and nonlinear complex system, its highest level exhibits uncertainty complexity, while the lower levels exhibit deterministic complexity. In the analysis, the overall system is reconstructed starting from the trajectory of the high-dimensional system, mapping the complex system step by step into relatively simple subsystems, and ultimately decomposing it into a unit system that can be solved analytically, or a "reducible unit system." The system composed of these unit systems can reveal the mechanisms of deterministic complex systems. Based on the analysis of deterministic complexity, the impact of high-dimensional uncertainties on the overall system can be transformed into a series of WRT (Warranty-Reducible Complexity) case studies under deterministic scenarios.

[0022] This embodiment applies reduction theory to the field of carbon emission prediction for high-energy-consuming enterprises. It establishes a carbon emission intensity coupling model using three-dimensional indicators of physics, logic, and human factors to represent the deterministic complexity of carbon emissions for high-energy-consuming enterprises. Then, based on the production and operation characteristics of different enterprises, it constructs a probability density function in three-dimensional space to comprehensively represent the uncertainty of the system. It utilizes scenario generation and reduction techniques to retain the overall coupling characteristics of the system while reducing the dimensionality by extracting representative scenarios. The system is decomposed into WRT examples under deterministic scenarios. Finally, by solving for carbon emissions in representative scenario units, the basic carbon emission prediction value of the enterprise is restored.

[0023] In this embodiment, firstly, based on the overall reduction theory, physical, theoretical, and human-related three-dimensional indicators related to carbon emissions of high-energy-consuming enterprises are selected. Then, based on these key influencing factors of carbon emissions, a coupled model of carbon emission intensity for high-energy-consuming enterprises is established. The specific steps include: Based on the holistic reduction theory, the key carbon emission indicators for high-energy-consuming enterprises are determined. These key carbon emission indicators include physical indicators, theoretical indicators, and human-related indicators. Based on the physical indicators, a theoretical physical baseline emission intensity model for high-energy-consuming enterprises is established; based on the theoretical indicators, a process fluctuation correction coefficient is calculated; and based on the humanistic indicators, a behavioral intervention coefficient is calculated. Based on the process fluctuation correction coefficient and the behavioral intervention coefficient, the theoretical physical baseline emission intensity model is modified to obtain the carbon emission intensity coupled model.

[0024] In this embodiment, the key carbon emission indicators based on the holistic reduction theory include physical indicators, event-based indicators, and human-based indicators. Among them, physical indicators include physical operation-related parameters of energy-consuming equipment, which mainly characterize the baseline carbon emissions of enterprise production; event-based indicators include production management status-related parameters, which mainly characterize the disturbance of the enterprise's production organization logic and management process on the physical system; and human-based indicators include personnel status-related parameters, which mainly characterize the disturbance of the operator's status on the physical system.

[0025] These key carbon emission indicators can be determined manually or based on historical data analysis. In a preferred embodiment, a holographic candidate indicator library is constructed, and key carbon emission indicators are found through dimensionality reduction screening. First, based on the theory of reduction and reorganization, all objectively measurable parameters that may be related to carbon emissions in the production system of high-energy-consuming enterprises are identified to form an initial candidate set, namely the holographic candidate indicator library. This set covers raw data in three dimensions. For example, the physical dimension candidate subset includes physical and chemical parameters such as rated power of equipment, real-time current, furnace temperature, flue gas oxygen content, raw material calorific value, and volatile matter content; the process dimension candidate subset includes process parameters such as production plan output, start-up and shutdown instructions, preventive maintenance work order status, and scheduling records of energy media (water, electricity, gas); for the human dimension candidate subset, to ensure the objectivity and accessibility of the data, automatically collectable indicators are selected as variables of human status, such as the average length of service of key personnel (years), continuous working time of on-duty personnel (hours), number of historical violation alarms (times), monthly energy consumption performance assessment weight (%), and team staffing ratio (%).

[0026] Then, a dual filtering mechanism is used to screen the indicators in the indicator library. The first filtering uses an expert scoring mechanism for qualitative screening. This stage aims to quickly eliminate pseudo-correlated indicators with extremely low theoretical relevance or extremely difficult-to-obtain data, retaining indicators with physical or logical causal relationships. Specifically, multiple domain experts score the indicators to establish an evaluation matrix, and a weighted sum is performed based on preset expert weights to obtain the comprehensive relevance score of each indicator. A screening threshold is set according to a threshold truncation criterion. If the comprehensive relevance score is less than the threshold, the indicator is eliminated, resulting in the initial set of indicators. The second filtering uses maximum relevance and minimum redundancy based on mutual information for quantitative screening. This stage uses historical measured data to solve the nonlinear coupling and redundancy problems between the three-dimensional indicators. In this embodiment, mutual information (MI) is used to measure the complex dependencies between variables. Specifically, the historical indicator data of all high-energy-consuming enterprises are first discretized, then the mutual information between candidate indicators and actual carbon emissions is calculated, and the maximum relevance and minimum redundancy algorithm is used for feature screening to obtain a preset number of feature subsets, which are the key carbon emission indicators.

[0027] In a preferred embodiment, the preset quantity is set to 11, and the key carbon emission indicators are obtained through index screening, consisting of 5 physical indicators, 3 theoretical indicators, and 3 humanistic indicators. Specifically, the physical state vector composed of the physical indicators is... ,in Real-time operating load rate (%) of major energy-consuming equipment; The lower heating value of the fuel (GJ / t or GJ / Nm3). The carbon content per unit of fuel (tCO2 / t); The physical operating efficiency (%) of end-of-pipe carbon capture or waste gas treatment equipment. This represents the aging and degradation coefficient of energy-consuming equipment (calculated based on the total operating time of the equipment). The principle state vector, composed of principle indicators, is... ,in, The production batch switching or start-up / shutdown status is the production status (0 for steady state, 1 for start-up, -1 for shutdown). The start-up / shutdown process is usually accompanied by incomplete combustion, which leads to a surge in carbon emissions. The coupling degree of energy medium scheduling (e.g., whether the waste heat recovery system matches the production rhythm, matching is 1, mismatch is 0); This represents the compactness index of production scheduling (i.e., the percentage of idling and waiting time). The human resource state vector, composed of human resource indicators, is... ,in, The operator's proficiency is used to characterize the operator's level of experience and skill proficiency. It is preferred to use the average length of service (in years) to represent it. The longer the length of service, the higher the operational reliability is usually. To measure operator fatigue, the preferred method is to use the continuous working time (in hours) to quantify the degree of fatigue. If the continuous working time exceeds a certain threshold, the probability of misoperation increases non-linearly. To assess the reliability of personnel operations, the preferred method is to use the number of instances of illegal commands or misoperations during the same historical period to characterize the degree of compliance with personnel's behavior.

[0028] Based on the aforementioned key carbon emission indicators, and using physical indicators as the physical foundation and rational and human-related indicators as perturbation factors, the specific steps for establishing a coupled model of carbon emission intensity for high-energy-consuming enterprises include: The direct carbon emissions of the fuel are calculated based on fuel activity data, the lower heating value of the fuel and the carbon content per unit of the fuel in the aforementioned physical indicators. Based on the operating load rate of energy-consuming equipment, aging attenuation coefficient and physical operating efficiency of waste gas treatment equipment in the physical indicators, calculate the comprehensive energy efficiency coefficient of the equipment. The direct carbon emissions are corrected based on the comprehensive energy efficiency coefficient of the equipment to obtain the theoretical physical baseline emission intensity model for high-energy-consuming enterprises; Based on the production status, energy medium scheduling coupling degree, and production scheduling compactness index in the aforementioned indicators, the process fluctuation correction coefficient is obtained by weighted summation. The behavioral intervention coefficient is obtained by performing nonlinear calculations on the personnel proficiency, personnel fatigue, and personnel operational reliability in the aforementioned human factors indicators.

[0029] In this embodiment, a theoretical physical baseline emission intensity model for the enterprise is first established based on physical indicators. Specifically, based on the carbon emission accounting principle of enterprise greenhouse gas emissions, the carbon dioxide emissions generated by fuel combustion are equal to the product of fuel consumption, lower heating value, unit carbon content of the fuel, carbon oxidation rate, and 44 / 12, where 44 / 12 is the carbon-to-carbon dioxide conversion coefficient. Therefore, based on the determined activity level input of the g-th type of fuel energy, combined with the measured lower heating value and unit carbon content of the fuel mentioned above, the carbon dioxide emissions of this type of fuel can be calculated, and this is taken as the direct carbon emissions.

[0030] To accurately describe the nonlinear impact of equipment load rate and equipment aging on overall energy efficiency, this embodiment preferably uses the Sigmoid function to construct the overall energy efficiency function, the expression of which is: In the formula, and These are the preset sensitivity parameters; and These represent the load rate under standard operating conditions and the equipment aging status, respectively. and These represent the upper and lower limits of the equipment's overall energy efficiency, respectively. This represents the overall energy efficiency function of the equipment.

[0031] Preferred selection , Therefore: And due to waste disposal efficiency It takes values ​​in the range [0.8, 1], therefore: The above design ensures that the model always remains positive and that the denominator does not tend to zero. At the same time, it can better reflect the diminishing marginal effect caused by equipment load and aging.

[0032] Generally, the lower the load rate or the more severe the aging, the lower the unit energy efficiency and the higher the emission intensity. Using the overall energy efficiency of the equipment as the denominator and the physical operating efficiency of the waste gas treatment equipment as the numerator, we obtain the overall energy efficiency coefficient of the equipment. Then, we multiply the direct carbon emissions by the overall energy efficiency coefficient of the equipment to obtain the corrected carbon emissions. Finally, we add the corrected carbon emissions of various fuels used by the enterprise to obtain the theoretical physical baseline emission intensity model for high-energy-consuming enterprises. The formula is expressed as follows: In the formula, Let be the theoretical physical baseline emission intensity at time t. Total fuel energy; Based on measured values and The calculated direct carbon emissions of the g-th fuel energy source; This is the overall energy efficiency function of the equipment. It is the equipment operating load rate at time t. It represents the physical operating efficiency of the waste gas treatment equipment at time t. It is the aging and degradation coefficient of the energy-consuming equipment at time t.

[0033] Since the event-based indicators can alter carbon emissions by affecting the stability of the physical system, the impact coefficient of process fluctuations on carbon emissions can be calculated based on these indicators. Considering the strong correlation among some candidate event-based indicators, after screening by maximum correlation and minimum redundancy, only three representative indicators—production status, energy medium scheduling coupling degree, and production scheduling compactness—are retained to avoid variable redundancy and improve model stability. The specific expression is as follows: In the formula, This is the process fluctuation correction factor for the enterprise at time t. The emission penalty factor for start-up and shutdown is a preset value that is greater than 0; This is the indicator function, which takes the value 1 when the state is unsteady, and 0 otherwise. The production status at time t; The energy efficiency loss coefficient caused by energy dispatch mismatch; Let be the coupling degree of energy medium scheduling at time t; This represents the energy waste coefficient caused by idling and waiting. Let be the compactness index of the production schedule at time t. This formula characterizes the situation where the production schedule is chaotic (e.g., frequent starts and stops, scheduling mismatches). This will be significantly greater than 1, resulting in actual emissions being higher than the physical baseline value.

[0034] Human behavior indicators optimize or worsen system performance by fine-tuning operational parameters. Therefore, through nonlinear calculations of human behavior indicators, the impact coefficient of operational behavior on corporate carbon emissions, i.e., the behavioral intervention coefficient, can be obtained. The specific expression is as follows: In the formula, Let be the behavioral intervention coefficient at time t. It is an exponential function used to simulate nonlinear effects; The preset skill gain coefficient represents individuals who are proficient in their skills and in good condition (i.e., High and Low energy consumption can make the system energy-efficient, thereby reducing carbon emissions; This is the penalty coefficient for violations; if the number of violations in the same period in history... A higher value will lead to increased carbon emissions. , and These represent the personnel proficiency, personnel fatigue, and personnel operational reliability at time t, respectively.

[0035] Based on the above models, a holographic key factor state matrix of the carbon emission system of a high-energy-consuming enterprise at time t is defined. : Here, variable t represents the prediction time step. Considering the inertia of production scheduling in high-energy-consuming enterprises and the periodicity of personnel status changes, the time step is taken as 1 hour, and all state variables in the matrix refer to the mean or state within the t-th time step.

[0036] Based on the aforementioned state matrix, a coupled model of carbon emission intensity for high-energy-consuming enterprises at time t is constructed. : In the formula, This is the preset residual term. The formula clarifies the inherent coupling logic of physics as the cardinality, reasoning as the multiplier correction, and human reasoning as the exponential fine-tuning, providing a priori physical constraints for the subsequent system decomposition process.

[0037] Based on the aforementioned three-dimensional indicators of physics, logic, and human reasoning, and using historical data of the enterprise to be predicted, a probability density function for each indicator variable is constructed to quantify the randomness and uncertainty of the system input. The probability density function of the indicator variables can be obtained using a kernel density estimation algorithm. By performing kernel density estimation on historical data of different variables, a marginal probability density estimation model for each variable is obtained. To improve the accuracy of the model representation, in a preferred embodiment, different estimation methods are used to construct the marginal probability density estimation model based on different indicator types. Specific steps include: Based on the historical data of the enterprise to be predicted, the first probability density function of the physical index is constructed using an adaptive bandwidth kernel density estimation algorithm, and the second probability density function of the rational index and the third probability density function of the human index are constructed using a Bayesian inference method. By combining the first probability density function, the second probability density function, and the third probability density function, a marginal probability density estimation model for the key carbon emission indicators of the enterprise to be predicted is obtained.

[0038] In this embodiment, physical indicators are monitored by a massive amount of data and often exhibit a non-normal, multi-modal distribution (e.g., the equipment operates most frequently at low and full load conditions). Traditional normal distribution assumptions cannot fit this characteristic; therefore, non-parametric kernel density estimation is employed. Specifically, based on historical sample data of the physical indicators, a Gaussian kernel density estimation function is used to define the probability density function of the physical indicator variable. To avoid over-smoothing or under-smoothing due to fixed bandwidth, an adaptive bandwidth calculation formula based on Silverman's Rule of Thumb is used to adaptively adjust the bandwidth in the Gaussian kernel density estimation function, thereby ensuring that the accuracy of the probability density function automatically improves as historical physical data accumulates.

[0039] Since the data for rational and human indicators are usually discrete and small-sample, direct frequency statistics can lead to significant bias. Therefore, Bayesian inference is used to integrate prior knowledge with observational data. Specifically, for the indicator variables in rational and human indicators, their prior distribution is defined to follow a Beta distribution. Then, based on the historical data of the indicator variable, the likelihood function of its binomial distribution is determined. According to Bayes' theorem, since the Beta distribution is the conjugate prior of the binomial distribution, the posterior distribution is still a Beta distribution. The probability density distribution of the indicator variable is obtained through parameter updates.

[0040] Based on the above calculations, a marginal probability density model for the enterprise is constructed. This model is a set of marginal probability density functions for various indicator variables, including the probability density function obtained from kernel density estimation and the probability density function obtained from Bayesian inference. Then, based on the marginal probability density model, a joint distribution function among physical, event-rational, and human-rational state variables is established, and multiple high-dimensional coupled scenario samples are generated based on this distribution.

[0041] First, a coupled model of the carbon emission scenario for the enterprise to be predicted is established using a Copula structure. Since the carbon emissions of high-energy-consuming enterprises are mainly driven by physical energy consumption, with objective and human factors playing a corrective role, the system has a clear central variable. Therefore, the C-Vine structure is chosen for modeling because C-Vine presents a star topology, which is suitable for describing this kind of dependency structure with a core-periphery relationship.

[0042] Specifically, we first define the joint probability density formula for C-Vine, assuming the vector representation of key carbon emission indicators is as follows: Where d is the total number of indicators, The core variable selected as the physical indicator (i.e., the operating load rate of energy-consuming equipment). According to Sklar's theorem and Vine decomposition, its joint probability density function... It can be decomposed into a series of products of marginal density and Pair-Copula density: In the formula, the first term of the product It is the product of the independent marginal probabilities of each physical, logical, and humanistic index obtained in the above embodiments. The k-th index x k The independent marginal probabilities; the second term of the product is the density function of all pairs of Copula. The product describes the entanglement relationship between variables; It is the conditional cumulative distribution function; v is the set of variables whose indices precede j. The first tree (Tree 1) of C-Vine connects the dominant variables. This accurately simulates the physical state, along with all other variables. It constrains the actual operating conditions of other variables.

[0043] For each pair of Copula density functions in the formula, a Gaussian Copula function can be used, or a family of functions can be selected based on data characteristics. Different families of functions can be determined by the correlation of the data tails. Specifically, tail correlation analysis is performed on the historical data of the indicator variable. For symmetric dependence (such as normal fluctuations), the Frank Copula function is used; for upper tail dependence (such as high physical load leading to high human error), the Gumbel Copula function is used; and for lower tail dependence, the Clayton Copula function is used. The specific correlation analysis process can be referred to the conventional analysis steps, and will not be elaborated here.

[0044] After determining the family of Pair-Copula functions among the key carbon emission indicators, and using historical data, a log-likelihood function is constructed to find the optimal parameters for each pair of Copulas, thereby obtaining the joint probability density function of the enterprise to be predicted, which is the carbon emission scenario coupling model of the enterprise to be predicted.

[0045] Based on the above carbon emission scenario coupling model, the coupling scenarios are determined through sampling modeling. The specific steps include: Independent uniform random numbers are generated using the Latin hypercube sampling method. Based on the joint probability density function of the carbon emission scenario coupling model, the independent uniform random numbers are subjected to inverse Rosenblatt transformation to obtain the coupling correlation variables. Based on the inverse cumulative distribution function of the marginal probability density estimation model, the coupling correlation variables are subjected to probability space inverse mapping to obtain the initial panoramic coupling scene set.

[0046] In this embodiment, to improve the coverage of the sample space, the Latin Hypercube Sampling (LHS) method is first used to generate K groups in... Independent random vectors, uniformly distributed and non-overlapping in space, are generated, where d represents the total number of indicators, i.e., the indicator dimension. Then, based on the C-Vine structure of the carbon emission scene coupling model, the independent random vectors undergo an inverse Rosenblatt transformation in sequence. The inverse function used in the inverse transformation is the inverse function of the conditional distribution function obtained by integrating the joint probability density function in the carbon emission scene coupling model. This inverse transformation converts the independent random numbers into variables with coupling correlation. Then, using the inverse function of the marginal probability density estimation model's marginal probability distribution function (i.e., the cumulative distribution function), the variables with coupling correlation in the interval [0,1] are restored to actual physical quantities, thereby generating an initial panoramic coupled scene set containing K scenes.

[0047] Each scenario in the initial panoramic coupled scenario set is a d-dimensional vector, representing a combination of specific states of all key indicators across the three dimensions of physics, logic, and human factors in a given simulation. Thanks to the use of C-Vine Copula, this set not only includes high-probability stable operation scenarios but also scientifically generates extreme scenarios with multiple overlapping risks that conform to statistical laws (such as inferior raw materials + maintenance delays + new personnel on duty). This perfectly embodies the comprehensive capture of system complexity emphasized in WRT's holistic reduction theory.

[0048] Based on Latin hypercube sampling, an initial panoramic coupled scene set is generated. Combined with scene similarity measurement, an appropriate scene reduction strategy is adopted to reduce the scene, such as clustering algorithm or greedy algorithm, in order to extract representative scenes and obtain a representative scene set. The specific scene reduction strategy can be flexibly selected according to the actual situation, and no specific limitation is imposed here.

[0049] In a preferred embodiment, to more accurately extract extreme scenes highly sensitive to carbon emissions, this embodiment employs a synchronous back-substitution elimination method based on an improved Kantorovich distance to reduce the initial panoramic coupled scene set. Specific steps include: Using a pre-defined surrogate model, the characteristic importance of each key carbon emission indicator to the total carbon emission is calculated, and the sensitivity factor of each key carbon emission indicator is obtained. The surrogate model is constructed based on a machine learning algorithm and trained using historical data of the enterprise to be predicted. Based on the variable distribution status of each key carbon emission indicator in each coupled scenario in the initial panoramic coupled scenario set, determine the indicator anomaly penalty item for each key carbon emission indicator in each coupled scenario. Using the sensitivity factor and the index anomaly penalty term as index weights, the Kantorovich distance between coupled scenarios is weighted and corrected to obtain the improved Kantorovich distance. Based on the improved Kantorovich distance, the synchronous back-substitution elimination method is used to reduce the initial panoramic coupled scene set to obtain a representative scene set.

[0050] In this embodiment, the iterative steps of scene reduction using the synchronous back-substitution elimination method based on Kantorovich distance include: 1) Initialization: Initialize the current holdout set to the initial panoramic coupled scene set, with a probability value of 1 / K for each scene, and calculate the Kantorovich distance between all scene pairs; 2) Elimination scenario determination: For each scenario k, select the scenario m with the smallest Kantorovich distance to that scenario, and calculate the product of the Kantorovich distance and the probability value of scenario k as the potential loss value of scenario k. Calculate the potential loss value of each scenario, and select the scenario with the smallest potential loss value as the elimination scenario n. 3) Removal and probability redistribution: Remove the removed scene n from the current set of retention, select the scene n' that is closest to the removed scene n, and update the probability of the nearest scene n' according to the probability value of the removed scene n. The updated probability value of the nearest scene n' is equal to the probability value before the update plus the probability value of the removed scene n. 4) Repeat the above steps iteratively until the target number of clusters for the preset clustering scenario is reached. The final output is a representative scene set, which includes... A representative scenario and the probability value corresponding to the scenario.

[0051] Building upon the aforementioned reduction steps, to maximize the preservation of information from extreme scenarios highly sensitive to carbon emissions, this embodiment employs an improved Kantorovich distance. The original Kantorovich distance is weighted and corrected based on carbon sensitivity. Specifically, a lightweight surrogate model is first trained using historical data from the companies to be predicted. This surrogate model then calculates the feature importance of each key indicator to the total carbon emissions, denoted as the sensitivity factor. Preferably, the surrogate model is constructed using a machine learning algorithm, such as XGBoost. Its input is the key carbon emission indicator for each scenario, and its output is the total carbon emissions. The model is trained using historical data from the companies to be predicted. The trained XGBoost model automatically calculates feature importance, and the importance score can be obtained through the member variable `feature_importances_`. By normalizing the importance scores of these key indicators, the sensitivity factor for each key indicator is obtained.

[0052] In addition to sensitivity factors, penalty items for abnormal indicators can be set based on the distribution of various indicator variables in the scenario. For each key indicator in each scenario, a certain variable in scenario a can be used as a basis for setting penalty items for abnormal indicators. For example, if the value of the indicator variable is in the extreme region of the variable's distribution, such as being greater than μ+2σ or less than μ-2σ, where μ is the mean of the variable and σ is the standard deviation of the variable, then an indicator anomaly penalty term is defined for this scenario. The value of this abnormality penalty term is positively correlated with the distribution of the variable; that is, the further away from the central region (the closer to the extreme region), the larger the value of this penalty term; otherwise... This anomaly penalty ensures that scenes containing abnormal states are amplified in distance calculations, making them appear unique and difficult to remove.

[0053] Then, the Kantorovich distance is weighted and corrected based on the sensitivity factor and the indicator anomaly penalty term to obtain the carbon sensitivity weighted distance between the two scenarios, which is expressed by the following formula: In the formula, Representing a scene a and scene b The improved Kantorovich distance between them, where d represents the indicator dimension. The sensitivity factor for the h-th key carbon emission indicator is... Representing a scene a The h-th key carbon emission indicator, Representing a scene b The h-th key carbon emission indicator, Representing a scene aThe penalty item for abnormality of the h-th key carbon emission indicator. Representing a scene b The penalty item for abnormality of the h-th key carbon emission indicator. This ensures that the distance between two scenarios is increased as long as one of them is in an extreme state.

[0054] Then, using the improved Kantorovich distance and the aforementioned synchronous back-substitution elimination method, scenario reduction is performed to obtain a representative scenario set with high carbon sensitivity. Carbon emission prediction is then calculated based on this representative scenario set. During the calculation, the representative scenario set is mapped to specific carbon emission prediction values, and the final probabilistic prediction results are synthesized using statistical methods. Specific steps include: 1) Deterministic calculation of carbon emission intensity in a single scenario Each representative scenario is treated as a snapshot of a specific operating condition. The state vector of the key carbon emission indicators in that scenario is used as input. The carbon emission intensity coupling model constructed in the previous steps is called to calculate the predicted carbon emission value in that specific scenario. 2) Global carbon emission probability prediction synthesis The carbon emission prediction values ​​of each scenario in the representative scenario set (i.e., the scenario reconstruction probability) are used as the scenario weights. The carbon emission prediction values ​​of each representative scenario are weighted and summed to obtain the total carbon emission prediction value of the enterprise.

[0055] The total carbon emission forecast calculated by the carbon emission forecasting method provided in this embodiment can be used as the expected forecast value of corporate carbon emissions, and can be used as a benchmark for corporate daily quota management, so as to improve the corporate carbon emission management level.

[0056] This embodiment provides a carbon emission prediction method for high-energy-consuming enterprises. This method breaks through the limitation of traditional carbon emission prediction that only focuses on single physical energy consumption data. By introducing the WRT holistic reduction theory, it constructs a three-dimensional holographic index system of physical-event-human (WSR), incorporating the logical fluctuations of production scheduling (event) and the subjective uncertainties of personnel operation (human) into the quantitative model. This effectively solves the prediction bias problem caused by the neglect of "soft" management and behavioral factors in existing technologies, and significantly improves the model's comprehensive representation ability and prediction robustness for the complex production systems of high-energy-consuming enterprises. By using the C-Vine Copula function to replace the traditional independent hypothesis sampling method for joint probability modeling, it can simulate multiple risk superposition extreme working conditions that traditional methods cannot generate, improve the ability to predict carbon emission surge events, and achieve accurate capture of complex nonlinear coupling risks between multi-dimensional factors. By adopting an improved Kantorovich distance synchronous back-substitution elimination algorithm that introduces carbon emission sensitivity weights, it solves the defect of traditional scenario reduction technology that easily removes low-probability but high-impact carbon peak scenarios as noise. While ensuring that a large number of scenarios are reproduced into a small number of representative scenarios, the key high-emission risk characteristics are still preserved with high fidelity.

[0057] Please see Figure 2 Based on the same inventive concept, the second embodiment of this invention proposes a carbon emission prediction system for high-energy-consuming enterprises, comprising: The first model construction module 10 is used to determine the key carbon emission indicators of high-energy-consuming enterprises based on the overall reduction theory, and to establish a carbon emission intensity coupling model for high-energy-consuming enterprises. The second model construction module 20 is used to establish a marginal probability density estimation model of the key carbon emission indicators based on the historical data of the enterprise to be predicted, using a kernel density estimation algorithm and a Bayesian inference algorithm. Based on the marginal probability density estimation model, and using a Copula structure, a carbon emission scenario coupling model of the enterprise to be predicted is established. The scene generation module 30 is used to sample the carbon emission scene coupling model and perform probability space inverse mapping according to the marginal probability density estimation model to obtain an initial panoramic coupled scene set. Scene reduction module 40 is used to reduce the initial panoramic coupled scene set using a preset reduction algorithm to obtain a representative scene set; The carbon emission prediction module 50 is used to calculate the carbon emission prediction value of each representative scenario in the representative scenario set according to the carbon emission intensity coupling model, and to perform a weighted summation of the carbon emission prediction values ​​of each representative scenario to obtain the total carbon emission prediction value of the enterprise to be predicted.

[0058] The technical features and effects of the carbon emission prediction system for high-energy-consuming enterprises proposed in this embodiment of the invention are the same as those of the method proposed in this embodiment of the invention, and will not be repeated here. Each module in the aforementioned carbon emission prediction system for high-energy-consuming enterprises can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in or independent of the processor in a computer device in hardware form, or it can be stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0059] Furthermore, embodiments of the present invention also propose a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0060] Please see Figure 3 The diagram illustrates the internal structure of a computer device in one embodiment. This computer device can specifically be a terminal or a server. The computer device includes a processor, memory, network interface, display, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the computer device is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a carbon emission prediction method for high-energy-consuming enterprises. The display screen of the computer device can be a liquid crystal display (LCD) or an e-ink display. The input devices of the computer device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse, etc.

[0061] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computing devices may include more or fewer components than those shown in the figure, or combine certain components, or have the same component arrangement.

[0062] Furthermore, embodiments of the present invention also propose a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.

[0063] In summary, the embodiments of this invention propose a method, system, device, and storage medium for predicting carbon emissions from high-energy-consuming enterprises. The method includes: determining key carbon emission indicators for high-energy-consuming enterprises based on global reductionism and establishing a carbon emission intensity coupling model for high-energy-consuming enterprises; establishing a marginal probability density estimation model for the key carbon emission indicators based on historical data of the enterprise to be predicted, using a kernel density estimation algorithm and a Bayesian inference algorithm; establishing a carbon emission scenario coupling model for the enterprise to be predicted based on the marginal probability density estimation model and using a Copula structure; sampling scenarios in the carbon emission scenario coupling model and performing probability space inverse mapping based on the marginal probability density estimation model to obtain an initial panoramic coupled scenario set; reducing the initial panoramic coupled scenario set using a preset reduction algorithm to obtain a representative scenario set; calculating the predicted carbon emission value for each representative scenario in the representative scenario set based on the carbon emission intensity coupling model, and performing a weighted summation of the predicted carbon emission values ​​for each representative scenario to obtain the total predicted carbon emission value for the enterprise to be predicted. This invention significantly improves the model's comprehensive representation and predictive robustness for complex production systems of high-energy-consuming enterprises by introducing the WRT (Wide Impact Theory) approach. Through joint probabilistic modeling using C-Vine Copula, it accurately captures the complex nonlinear coupling risks among multidimensional factors, enhancing the ability to predict carbon emission surges. By improving the Kantorovich distance synchronous back-substitution elimination algorithm, it ensures that while reducing massive scenarios to a small number of representative scenarios, it still retains key high-emission risk characteristics with high fidelity. This invention effectively solves the problems of traditional methods being single-dimensional, lacking interpretability, and unable to handle dynamic coupling risks, significantly improving the accuracy of carbon emission prediction for high-energy-consuming enterprises.

[0064] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0065] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the technical principles of this invention, and these improvements and substitutions should also be considered within the scope of protection of this application.

Claims

1. A method for predicting carbon emissions from high-energy-consuming enterprises, characterized in that, include: Based on the overall reduction theory, the key carbon emission indicators of high-energy-consuming enterprises are determined, and a carbon emission intensity coupling model of high-energy-consuming enterprises is established. Based on the historical data of the enterprises to be predicted, a marginal probability density estimation model for the key carbon emission indicators is established using the kernel density estimation algorithm and the Bayesian inference algorithm. Based on the marginal probability density estimation model and using the Copula structure, a carbon emission scenario coupling model for the enterprises to be predicted is established. Scene sampling is performed on the carbon emission scene coupling model, and probability space inverse mapping is performed according to the marginal probability density estimation model to obtain an initial panoramic coupled scene set; The initial panoramic coupled scene set is reduced using a preset reduction algorithm to obtain a representative scene set; Based on the carbon emission intensity coupling model, the predicted carbon emission value for each representative scenario in the representative scenario set is calculated, and the predicted carbon emission values ​​for each representative scenario are weighted and summed to obtain the total predicted carbon emission value for the enterprise to be predicted.

2. The carbon emission prediction method for high-energy-consuming enterprises according to claim 1, characterized in that, The steps for determining key carbon emission indicators for high-energy-consuming enterprises and establishing a coupled model of carbon emission intensity for high-energy-consuming enterprises based on the overall reduction theory include: Based on the holistic reduction theory, the key carbon emission indicators for high-energy-consuming enterprises are determined. These key carbon emission indicators include physical indicators, theoretical indicators, and human-related indicators. Based on the physical indicators, a theoretical physical baseline emission intensity model for high-energy-consuming enterprises is established; based on the theoretical indicators, a process fluctuation correction coefficient is calculated; and based on the humanistic indicators, a behavioral intervention coefficient is calculated. Based on the process fluctuation correction coefficient and the behavioral intervention coefficient, the theoretical physical baseline emission intensity model is modified to obtain the carbon emission intensity coupled model.

3. The carbon emission prediction method for high-energy-consuming enterprises according to claim 2, characterized in that, The steps for establishing a marginal probability density estimation model for the key carbon emission indicators based on historical data of the enterprises to be predicted, using kernel density estimation algorithm and Bayesian inference algorithm, include: Based on the historical data of the enterprise to be predicted, the first probability density function of the physical index is constructed using an adaptive bandwidth kernel density estimation algorithm, and the second probability density function of the rational index and the third probability density function of the human index are constructed using a Bayesian inference method. By combining the first probability density function, the second probability density function, and the third probability density function, a marginal probability density estimation model for the key carbon emission indicators of the enterprise to be predicted is obtained.

4. The carbon emission prediction method for high-energy-consuming enterprises according to claim 1, characterized in that, The steps of establishing a coupled model of carbon emission scenarios for the enterprise to be predicted based on the marginal probability density estimation model and using the Copula structure include: Based on tail correlation analysis, the family of Pair-Copula functions among various key carbon emission indicators is determined, and the optimal parameters of the family of Pair-Copula functions are calculated using the maximum likelihood estimation algorithm based on the historical data. Based on the marginal probability density estimation model and the calculated Pair-Copula function family, a joint probability density function is constructed using the C-Vine structure to obtain a coupled model of the carbon emission scenario for the enterprise to be predicted.

5. The carbon emission prediction method for high-energy-consuming enterprises according to claim 1, characterized in that, The steps of sampling the carbon emission scene coupling model and performing probability space inverse mapping based on the marginal probability density estimation model to obtain the initial panoramic coupled scene set include: Independent uniform random numbers are generated using the Latin hypercube sampling method. Based on the joint probability density function of the carbon emission scenario coupling model, the independent uniform random numbers are subjected to inverse Rosenblatt transformation to obtain the coupling correlation variables. Based on the inverse cumulative distribution function of the marginal probability density estimation model, the coupling correlation variables are subjected to probability space inverse mapping to obtain the initial panoramic coupling scene set.

6. The carbon emission prediction method for high-energy-consuming enterprises according to claim 1, characterized in that, The step of using a preset reduction algorithm to reduce the initial panoramic coupled scene set to obtain a representative scene set includes: Using a pre-defined surrogate model, the characteristic importance of each key carbon emission indicator to the total carbon emission is calculated, and the sensitivity factor of each key carbon emission indicator is obtained. The surrogate model is constructed based on a machine learning algorithm and trained using historical data of the enterprise to be predicted. Based on the variable distribution status of each key carbon emission indicator in each coupled scenario in the initial panoramic coupled scenario set, determine the indicator anomaly penalty item for each key carbon emission indicator in each coupled scenario. Using the sensitivity factor and the index anomaly penalty term as index weights, the Kantorovich distance between coupled scenarios is weighted and corrected to obtain the improved Kantorovich distance. Based on the improved Kantorovich distance, the synchronous back-substitution elimination method is used to reduce the initial panoramic coupled scene set to obtain a representative scene set.

7. The carbon emission prediction method for high-energy-consuming enterprises according to claim 2, characterized in that, The steps of establishing a theoretical physical baseline emission intensity model for high-energy-consuming enterprises based on the physical indicators, calculating process fluctuation correction coefficients based on the rational indicators, and calculating behavioral intervention coefficients based on the human indicators include: The direct carbon emissions of the fuel are calculated based on fuel activity data, the lower heating value of the fuel and the carbon content per unit of the fuel in the aforementioned physical indicators. Based on the operating load rate of energy-consuming equipment, aging attenuation coefficient and physical operating efficiency of waste gas treatment equipment in the physical indicators, calculate the comprehensive energy efficiency coefficient of the equipment. The direct carbon emissions are corrected based on the comprehensive energy efficiency coefficient of the equipment to obtain the theoretical physical baseline emission intensity model for high-energy-consuming enterprises; Based on the production status, energy medium scheduling coupling degree, and production scheduling compactness index in the aforementioned indicators, the process fluctuation correction coefficient is obtained by weighted summation. The behavioral intervention coefficient is obtained by performing nonlinear calculations on the personnel proficiency, personnel fatigue, and personnel operational reliability in the aforementioned human factors indicators.

8. A carbon emission prediction system for high-energy-consuming enterprises, characterized in that, include: The first model construction module is used to determine the key carbon emission indicators of high-energy-consuming enterprises based on the overall reduction theory, and to establish a carbon emission intensity coupling model for high-energy-consuming enterprises. The second model building module is used to establish a marginal probability density estimation model of the key carbon emission indicators based on the historical data of the enterprise to be predicted, using kernel density estimation algorithm and Bayesian inference algorithm. Based on the marginal probability density estimation model, and using Copula structure, a carbon emission scenario coupling model of the enterprise to be predicted is established. The scene generation module is used to sample the carbon emission scene coupling model and perform probability space inverse mapping according to the marginal probability density estimation model to obtain an initial panoramic coupled scene set. The scene reduction module is used to reduce the initial panoramic coupled scene set using a preset reduction algorithm to obtain a representative scene set; The carbon emission prediction module is used to calculate the carbon emission prediction value for each representative scenario in the representative scenario set according to the carbon emission intensity coupling model, and to perform a weighted summation of the carbon emission prediction values ​​of each representative scenario to obtain the total carbon emission prediction value of the enterprise to be predicted.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

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

Images