Carbon footprint uncertainty calculation method and electronic device

Abstract

The present application relates to the technical field of carbon footprint uncertainty analysis, and particularly relates to a carbon footprint uncertainty calculation method and electronic equipment, the method comprising: automatically generating a plurality of rules from unstructured text comprising carbon footprint historical data; fuzzy weighted synthesis of the plurality of rules in a scenario to generate a credibility distribution; extracting a global range and a scenario range according to a credibility threshold, and truncating the probability density function of the obtained truncated normal distribution on the global range to obtain a background distribution; equally dividing the scenario range into three scenario sub-intervals, and calculating the scenario weights using the background distribution; conditionally truncating sampling for each scenario sub-interval, and independently running Monte Carlo simulation to output the corresponding mean and variance; and weighting and synthesizing the Monte Carlo simulation results of each scenario through the corresponding scenario weights to obtain the uncertainty of the carbon footprint. Compared with the prior art, the present application has the advantages of efficient and accurate analysis of carbon footprint uncertainty, etc.

Description

Technical Field

[0001] This invention relates to the field of carbon footprint assessment and uncertainty analysis technology, and in particular to a carbon footprint uncertainty calculation method and electronic device based on truncated normal background and fuzzy interval weighting. Background Technology

[0002] Traditional carbon footprint assessment methods typically require calculating carbon emissions at each stage, including raw materials, production, transportation, use, and recycling. This is not only labor-intensive but also cumbersome and time-consuming.

[0003] Quantifying the uncertainty of carbon footprint requires taking into account the complex characteristics of multi-stage coupling in the life cycle, random fluctuations in emission factors, and supply chain ambiguity (such as process experience judgment). Existing technologies mainly focus on four major directions: life cycle assessment (LCA) framework, probability and statistics, fuzzy mathematics, and data-driven models.

[0004] Uncertainty propagation in Life Cycle Inventory (LCI): Based on the standardized LCA process, the uncertainty of activity levels (such as raw material usage and energy consumption) and emission factors (such as carbon emissions per unit of energy) at each stage is first quantified. Then, the uncertainty range of the product-level carbon footprint is derived through error propagation formulas (variance-covariance method, Monte Carlo simulation). For example, when an automobile company calculates its carbon footprint, it needs to model the emissions of dozens of links such as "battery lithium mining → cell manufacturing → vehicle assembly → end use" one by one, and then superimpose the statistical fluctuations of parameters at each link.

[0005] Probability and Statistics and Bayesian Inference: Frequentist Approach (Monte Carlo Simulation): Simulates the probability distribution of uncertain variables through random sampling (e.g., assuming emission factors follow a normal distribution), and iteratively calculates the output distribution of the carbon footprint (e.g., 95% confidence interval). Bayesian Method: Combines "prior knowledge" (e.g., industry experience distribution) with "observational data" (e.g., actual emissions from enterprises) to update the posterior probability distribution of parameters, reducing the subjectivity of purely empirical assumptions.

[0006] Fuzzy mathematics and interval analysis: For “fuzzy uncertainty” in the supply chain (such as subjective judgment of process fluctuations), fuzzy numbers (triangular fuzzy, trapezoidal fuzzy) or interval numbers are used to describe parameters, and uncertainty is transmitted through fuzzy rules or interval arithmetic.

[0007] Data-driven and machine learning: Using historical carbon emission data to train models (such as random forests and neural networks), fitting the nonlinear relationship between "emission factors → process / energy / material characteristics", and combining Bootstrap or Bayesian neural networks to estimate the prediction interval, thereby achieving "data-driven" uncertainty quantification.

[0008] However, existing methods often struggle to comprehensively consider fuzzy uncertainties in complex systems with multiple stages and factors, and are unable to efficiently find the most similar cases from a large number of already accounted products, leading to a dilemma of low efficiency, poor accuracy, and difficulty in reuse in carbon footprint assessment.

[0009] Therefore, there is an urgent need for an efficient and scalable similarity measurement technology to enable more accurate and faster carbon emission assessments across different product lifecycles.

[0010] A search revealed Chinese invention patent application publication number CN118840128A, which discloses a method and apparatus for carbon footprint uncertainty analysis, as well as a computer-readable storage medium. The method includes: using a global sensitivity analysis method to determine a first index value for each of a plurality of reference variables used for carbon footprint accounting of a target product, where the first index value characterizes the importance of each reference variable to the carbon footprint accounting; selecting at least one reference variable from the plurality of reference variables whose first index value satisfies a preset condition as at least one target variable; determining a second index value for each target variable to the carbon footprint accounting result based on the correlation between the at least one target variable and the carbon footprint accounting result, where the second index value characterizes the degree of influence of each target variable on the carbon footprint accounting result; and assessing the uncertainty of the carbon footprint accounting result for the target product based on the second index value of the at least one target variable. This existing patent application suffers from low efficiency and low accuracy in carbon footprint uncertainty assessment.

[0011] Improving the efficiency and accuracy of carbon footprint uncertainty calculations has become a technical problem that needs to be solved. Summary of the Invention

[0012] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a method and electronic device for calculating carbon footprint uncertainty.

[0013] The objective of this invention can be achieved through the following technical solutions:

[0014] According to one aspect of the present invention, a method for calculating carbon footprint uncertainty is provided, the method comprising:

[0015] Evidence distillation process: Automatically generate multiple rules from unstructured text including historical carbon footprint data;

[0016] Fuzzy synthesis process: Multiple rules are fuzzily weighted and synthesized in a scenario to generate a confidence distribution;

[0017] Interval extraction process: Extract the global range and scenario range based on the confidence threshold, and truncate the probability density function of the sampleable truncated normal distribution on the global range to obtain the background distribution;

[0018] Scenario weighting and Monte Carlo simulation process: Divide the scenario range into three equal scenario sub-intervals, calculate the scenario weight of each scenario sub-interval using the background distribution; perform conditional truncation sampling on each scenario sub-interval, and run Monte Carlo simulation independently to output the corresponding mean and variance;

[0019] Weighted synthesis uncertainty process: The Monte Carlo simulation results of each scenario sub-interval are combined by weighting the corresponding scenario weights to obtain the uncertainty of carbon footprint.

[0020] Preferably, the evidence distillation process is based on an LLM model and includes: extracting "rule candidates" from unstructured text, extracting the parameter range and evidence strength of the context through a fixed prompt template, and automatically converting each piece of data into a rule unit, wherein the rule unit includes rule conditions, support set, typical band and evidence strength, and the unstructured text includes the company's historical carbon footprint data, industry reports, papers and expert texts.

[0021] Preferably, the fuzzy synthesis process includes: fuzzifying the parameter range of the rule into membership degree; and, based on the rule activation degree, evidence strength, and membership degree, performing fuzzy weighted synthesis of multiple rules in a scenario using the max-min operator to generate a credibility distribution.

[0022] More preferably, similarity is calculated for each scene dimension, and activation is calculated based on the similarity rules;

[0023] Methods for calculating rule activation include selecting the minimum value and selecting the product.

[0024] More preferably, the confidence distribution is specifically represented as follows:

[0025] ,

[0026] in, : No. i The activation degree of a rule in the current scenario c represents the applicability or influence of that rule; : No. i The weight of evidence for each rule indicates the credibility of that rule; : No. i The membership degree of a rule indicates the scope of support for that rule; : No. i The weighted activation degree of a rule is the combined effect of the rule's applicability and the strength of evidence.

[0027] Preferably, the global range includes all value intervals that cause the confidence distribution to be greater than the first confidence threshold;

[0028] The process of extracting the scenario range includes: selecting a coverage rate, solving for the second confidence threshold that satisfies the confidence distribution being greater than the candidate threshold variable and has the smallest interval length that satisfies the coverage rate, and then determining the smallest interval that includes all variables whose confidence distribution is greater than the second confidence threshold.

[0029] Preferably, the calculation process of the scenario weight includes: dividing the scenario range into three scenario sub-intervals to represent low, medium and high scenarios; directly defining the probability quality of the scenario sub-interval by the area of ​​the background distribution under each scenario sub-interval; and then normalizing within the scenario interval to obtain the scenario weight corresponding to the scenario sub-interval.

[0030] Preferably, the weighted synthesis uncertainty process includes: synthesizing the uncertainty of the carbon footprint based on the global estimate, the scenario weights of each scenario sub-interval, and the mean and variance of the Monte Carlo simulation output under the corresponding scenario, specifically:

[0031] ,

[0032] in, and Each represents a scenario i Variance and mean of the Nemont Carlo simulation Representing discrete terms between scenarios, This is a global estimate; Context weights.

[0033] More preferably, the global estimate is obtained by weighting and summing the scenario weights of each scenario sub-interval with the mean of the Monte Carlo simulation output under the corresponding scenario.

[0034] According to another aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described thereon.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1) This invention forms a rule base for evaluating the uncertainty of carbon footprint through a three-stage rule reasoning system of evidence distillation, fuzzy synthesis, and interval extraction. The evidence distillation stage automatically generates multiple rules from unstructured text, and the fuzzy synthesis stage performs fuzzy weighted synthesis of multiple rules to generate a parameter confidence distribution. The interval extraction stage extracts the global range based on a confidence threshold. With high-confidence scenario range The scenario range is divided into three sub-intervals: low, medium, and high. The fuzzy results are mapped to a sampleable truncated normal distribution. Monte Carlo simulations are performed independently in each scenario subspace. The Monte Carlo simulation results are weighted and synthesized to obtain the uncertainty of the carbon footprint. This method can realize the dynamic fusion of multi-source evidence and fuzzy reasoning in complex scenarios, providing a highly reliable parameter uncertainty assessment system for carbon footprint accounting.

[0037] 2) The scenario range of this invention is obtained by comprehensively considering the confidence threshold, coverage, and confidence distribution. It is a highly reliable working interval, which can help the parameter values ​​to be concentrated in the most likely value range. Subsequent Monte Carlo simulations are performed based on the highly reliable scenario range, and the uncertainty is weighted and synthesized, thereby improving the accuracy and stability of uncertainty calculation.

[0038] 3) This invention utilizes LLM-based automatic evidence distillation and activation-driven rule pruning to only include highly fit rules in fuzzy synthesis; then, based on the confidence distribution... Quick Extraction and By strictly limiting integration and sampling to a narrow domain and directly obtaining scenario weights using closed-form CDF differences with truncated normal distributions, the computational cost of numerical integration is reduced. Subsequently, conditional truncation and parallel sampling are performed on low, medium, and high scenarios, and rapid convergence is achieved with a unified random seed and variance monitoring. When evidence changes, only local incremental updates are required, avoiding repeated runs of the entire Monte Carlo method. The efficiency is highlighted by "shrinking before calculation, replacing numerical with analytical methods, and promoting convergence through parallelism," thereby significantly reducing the overall computational cost and improving computational efficiency while ensuring credibility and interpretability. Attached Figure Description

[0039] Figure 1 This is a flowchart illustrating the carbon footprint uncertainty calculation method in this invention;

[0040] Figure 2 This is a schematic diagram of the rule base construction process in this invention. Detailed Implementation

[0041] 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, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0042] This invention addresses the problems of traditional carbon footprint calculations, where input data only yields approximate values ​​or empirical ranges, and black-box algorithms lack interpretability. It proposes a transparent reasoning framework that embeds fuzzy logic within the computational graph. This method can distinguish between the global boundary and reasonable fluctuation ranges related to specific scenarios, and naturally maps fuzzy language such as "low-medium-high" into weighted functions, which are then transmitted to the final carbon emission result. This carbon footprint uncertainty calculation method can be widely applied in fields such as life cycle assessment (LCA), manufacturing process carbon management, and energy system optimization, providing enterprises with a high-precision, interpretable quantification scheme for carbon emission uncertainty. Furthermore, this invention can be used to compare and decide on different scenario schemes during product carbon footprint accounting and carbon label evaluation. It can also be applied to supply chain carbon emission monitoring, evaluation of the effects of low-carbon process transformation, and risk analysis and benefit assessment of emission reduction investment projects, providing quantitative support for enterprises' carbon peaking, carbon neutrality path planning, and policy response.

[0043] Example 1

[0044] This embodiment relates to a method for calculating carbon footprint uncertainty based on a truncated normal background and fuzzy interval weighting, including:

[0045] By constructing a rule-based reasoning system consisting of three stages—"evidence distillation—fuzzy synthesis—interval extraction"—structured rule units are automatically generated using multi-source data and language models, forming a rule base with hierarchical weights.

[0046] In the fuzzy synthesis stage, the activation degree of the rules is calculated based on scene similarity, and multiple rules are fuzzily weighted and synthesized using the max-min operator to generate a parameter confidence distribution. ;

[0047] Finally, the global interval is extracted based on the confidence threshold. High confidence interval The fuzzy results are mapped to a sampleable truncated normal distribution for Monte Carlo simulation and uncertainty propagation analysis.

[0048] This method can achieve dynamic fusion and fuzzy reasoning of multi-source evidence in complex scenarios, providing a highly reliable parameter uncertainty assessment system for carbon footprint accounting.

[0049] like Figure 1 The method includes:

[0050] (S1) Construct an indicator system, including a primary indicator set and a secondary indicator set.

[0051] Primary indicator set:

[0052] ,

[0053] The set of secondary indicators is shown in Table 1.

[0054] Table 1

[0055]

[0056] (S2) Rule base construction, including evidence distillation (based on LLM), fuzzy synthesis (based on max-min), interval extraction and coverage strategies, such as Figure 2 .

[0057] (21) The process of evidence distillation is as follows:

[0058] 21-1) Data Ingestion and Standardization: In this invention, historical carbon footprint data of enterprises (such as energy consumption, electricity usage, etc.) are first acquired, and combined with unstructured texts such as industry reports, academic papers, and expert demonstration materials, the above multi-source data are uniformly organized into structured candidate data with time, space, and process scenario labels. Specifically, based on a pre-established variable dictionary, energy consumption and emissions from different sources and units are converted into unified standard units, while retaining the original values, original units, and conversion coefficients used to ensure traceability; records that obviously violate physical laws or exceed reasonable engineering boundaries are directly eliminated or given a weight reduction label; for observations that are within a reasonable range but show extreme deviations, "winsoring" processing is performed according to preset quantiles or thresholds, truncating values ​​exceeding the threshold to the threshold itself, and recording the corresponding processing flags and rule parameters. For unstructured expressions in yearbooks, papers, and expert reports, the "object-value / interval-unit" and their applicable conditions are extracted. Vague descriptions such as "approximately," "less than," and "between" are standardized into point values ​​or intervals before unit conversion. After this standardization process, the original multi-source heterogeneous data is unified into structured candidate data with consistent units, compliant boundaries, and accompanying source and quality labels. This provides a standardized and reliable input foundation for subsequent rule unit construction and carbon footprint parameter uncertainty modeling.

[0059] 21-2) Data extraction and normalization based on the LLM model: Extracting "rule candidates" from unstructured text, that is, converting each data point into a rule unit. By extracting the parameter range and evidence strength of the context using fixed prompt templates, rules can be automatically generated and standardized in expression.

[0060] ,

[0061] A rule unit is the basic carrier of rules; each rule extracted from the text corresponds to one rule unit.

[0062] : Scenario preconditions (e.g., "process = electric arc furnace, region = EU, scale = medium-high"); that is, rule conditions, which are used to describe the business scenario that triggers the rule, and it is necessary to clarify which preconditions are met for the rule to take effect.

[0063] Business parameters to be constrained;

[0064] : Support set (allowed range); indicates when When satisfied, A reasonable range of values ​​is .

[0065] Typical characteristics of rules; The typical band (mode / high-frequency band) indicates that in In the scenario, The most frequently occurring intervals. The support set is the allowed range, and the typical band is the most common range.

[0066] The strength of evidence for the rule It represents the confidence level / weight of the rule, which is related to the authority of the source, timeliness, consistency with similar evidence, and rule score. Specifically, it is expressed as follows:

[0067] ,

[0068] The order of authority of the sources is: national / international standards > industry yearbooks / large databases > peer-reviewed papers > company records > expert testimonials.

[0069] (22) The process of fuzzy synthesis (max-min) is as follows:

[0070] 22-1) Scene Similarity and Rule Activation Library: Quantifies the degree of adaptation of rules in the current scene c.

[0071] For each scene dimension Define similarity If the current scenario and the preconditions of the rules are exactly the same, the scenario similarity is... =1, similarity can be measured using distance metrics. For example, an exponential decay function (or Gaussian decay) can be used to express the difference between the rule conditions and the current scenario.

[0072] Exponential decay form:

[0073] ,

[0074] Gaussian distribution form:

[0075] ,

[0076] in, For scene dimension d Similarity; For the current scene in dimension d The possible values ​​of ; For rules i In dimensions d The anchor point value; For dimension d The attenuation coefficient.

[0077] There are two ways to calculate rule activation:

[0078] 1) Method for selecting the minimum value: Select the similarity of each scene dimension. The minimum value in the range is used as the rule activation degree:

[0079] ,

[0080] This ensures that the rule will only be activated when all dimensions are sufficiently similar.

[0081] 2) Choosing the product method: These are the weights of each dimension, representing the degree of influence of different dimensions on rule activation. This method balances the influence of multiple dimensions. The rule activation can be obtained using the above method. Its value, ranging from 0 to 1, indicates the applicability of the rule. The closer the value is to 1, the more applicable the rule is to the current scenario; the closer the value is to 0, the less applicable the rule is. When calculating activation, sometimes dimensions may be missing or mismatched, which can lead to unreasonable activation results. To avoid this, for dimensions that are missing or cannot be matched in the scenario, "weighting" can be used (e.g., setting the similarity value to 0 or a smaller value) to prevent unreasonable activation by the rule.

[0082] 22-2) Core Interval and Data Weighting: This transforms interval values ​​or typical values ​​into membership functions, allowing fuzzy sets to be used in subsequent rule synthesis. Through fuzzification, the membership degree of a parameter within a certain interval is obtained, thus better reflecting the "degree to which the parameter belongs to that interval" in rule synthesis. Trapezoid or trigonometric functions are used to represent the "numerical rationality" of the membership function. That is, by fuzzifying the relationship between parameter values ​​and the defined interval, a specific value is transformed into a membership value within the interval [0,1], representing the applicability of that value within that interval.

[0083] Trapezoid function This is one of the most commonly used membership functions; it describes an interval. Membership degree.

[0084] Trapezoid function The definition of is:

[0085] ,

[0086] in: x The input parameter represents the range or typical value to be evaluated; and These are the lower and upper bounds of the interval (support set); and These are the lower and upper limits of the typical band, usually representing the most common or typical range of values ​​(e.g., the average value). scope).

[0087] meaning:

[0088] when Less than or greater than When the membership degree is 0, it means that the interval is not belonged to at all.

[0089] when lie in or When the membership degree is between these two values, it changes linearly (either increasing or decreasing linearly).

[0090] when Located in the interval When the membership degree is 1, it means that the value completely fits the interval.

[0091] Use trigonometric functions when there is no typical zone: if no typical zone is defined (i.e. (), can use a triangular membership function, which is typically:

[0092] ,

[0093] in this case, Typical values ​​represent the most common or most likely values. Typical bands Used to describe the region where values ​​are most concentrated, where the membership degree of a parameter is 1, indicating that it best fits that interval. Triangular membership functions are suitable for cases without a clear typical band, indicating that the entire interval is symmetrical and that values ​​are concentrated from a certain typical value (…). It begins to change linearly.

[0094] 22-3) Rule synthesis yields the confidence distribution. : Combine multiple data points into a single confidence curve under scenario c, using max-min synthesis with weights, to obtain... ∈[0,1] xThe function is max-min synthesis, which is a method of combining rules by maximizing the minimum value to generate a confidence distribution. That is, it selects the minimum confidence value for each rule from multiple rules and then takes the maximum value among all rules.

[0095] ,

[0096] : No. i The activation level of a rule in the current scenario c represents the applicability or influence of that rule.

[0097] : No. i The evidence weight of a rule indicates its credibility; rules with higher evidence weights have a greater impact on the final result.

[0098] : No. i The membership degree of a rule indicates the scope of support for that rule, for example, given a certain input value. x At that time, the "similarity" or "applicability" of this rule.

[0099] : No. i The weighted activation degree of a rule is the combined effect of the rule's applicability and the strength of evidence.

[0100] Credibility distribution π ( x The final output is the result of the input. x When the input meets all the rule conditions, it is the probability or likelihood that the input meets all the rule conditions.

[0101] (23) Interval extraction and coverage strategy

[0102] 23-1) Extracting the global range [A, B]: This range represents the possible value interval of a parameter under all evidence and scenario conditions. To obtain the global range, it is necessary to determine the minimum global range [A, B], also known as the global interval, which contains all values ​​that make the confidence distribution π( x The value of ϵ is defined as follows: within a certain interval, the confidence or probability of the parameter value is greater than the first confidence threshold ϵ. That is, this interval should cover the most likely range of values ​​and eliminate extreme or unlikely values.

[0103] ,

[0104] in, : Credibility or probability distribution function, representing parameters x Credibility;

[0105] The first confidence threshold is typically set to a value between [value missing]. arrive The interval between these values ​​indicates that we want to filter out values ​​with lower probabilities.

[0106] For all confidence levels greater than the first confidence threshold ϵ x A set of values.

[0107] 23-2) Extracting the scenario range [L,H]: Given a working interval (high confidence), determine a working interval. This indicates that the parameter values ​​within this interval have high confidence; this interval is also called the scenario interval. This interval helps the parameter values ​​to concentrate within the most probable range, thereby improving the accuracy and stability of the model. First, a coverage ρ needs to be selected, commonly 0.90 or 0.95, indicating that the scenario interval [L,H] is expected to contain 90% or 95% of the probability. Next, the second confidence threshold α* is calculated to minimize the interval length that satisfies the following conditions:

[0108] ,

[0109] in:

[0110] It is a parameter The probability density function (or confidence function);

[0111] It is to filter out all variables with a confidence level greater than the candidate threshold. of Value; candidate threshold variable This is used to enumerate different confidence thresholds, and the second confidence threshold α* is selected from these candidate thresholds based on the minimum interval length and coverage. The optimal threshold is obtained by using the criterion ρ.

[0112] It is an interval The length.

[0113] After finding the minimum a*, calculate the scenario interval [L, H] such that this interval contains all cases satisfying π( x )≥a* Value. Specifically, It includes all cases that satisfy π(x) > a*. The minimum range of values ​​is determined to ensure that the range contains at least ρ confidence.

[0114] ,

[0115] Where, minimal interval covering represents the minimum coverage interval, which contains the range that satisfies... All values ​​ensured that the confidence level was greater than the threshold. If π(x) has multiple peaks, a local circumscribed interval can be used to preserve information.

[0116] 23-3) Probability Distribution Mapping: This transforms the fuzzy results into a sampleable parameter distribution for Monte Carlo calculations. A coverage ρ is chosen; commonly used values ​​are 0.90 or 0.95, indicating a desired parameter coverage of at least 90% or 95%. Based on this coverage value, a threshold a* is calculated such that the parameters within the scenario interval [L,H] satisfy π(x)>a*, and the interval coverage is at least ρ. Next, the mean and standard deviation are calculated. The mean μ is approximately the center of the scenario interval [L,H], and the standard deviation σ is calculated as a proportion of the interval length. The probability density function (PDF) of the truncated normal distribution is then used.

[0117] ,

[0118] in:

[0119] It is the probability density function (PDF) of the standard normal distribution.

[0120] It is the cumulative distribution function (CDF) of the standard normal distribution.

[0121] and These are the lower and upper limits of the global scope, respectively.

[0122] The final rule base outputs a global scope for each parameter. and the scope of the scenario .

[0123] (S3) Setting the truncated normal background distribution

[0124] make This is the current input value. Select a coverage area. (default ), to work area Consider the confidence interval of this normal distribution Approximately, the estimate of the standard deviation can be obtained as follows:

[0125] ,(like =0.95, z 0.975 ≈1.96),

[0126] Where σ is the standard deviation, It is the critical value of the standard normal distribution.

[0127] Then the normal distribution is applied globally. By truncating the top part, the background distribution can be obtained. .

[0128] (S4) Scenario Subintervals and Fuzzy Weights

[0129] To align the low, medium, and high scenarios with probability meanings, we first define the scenario intervals. Divided into three equal sub-intervals of the scenario ( They are I1, I2, I3, o=1,2,3, and the length of each segment is Δ); then the background distribution is used. The area under each segment is directly defined as the probability mass of that segment. Finally, in the scenario range Internal normalization yields the scenario weights Its physical meaning is: the known parameters fall within When the condition falls into the low / medium / high scenario subinterval:

[0130] ,

[0131] ,

[0132] ,

[0133] Area within a segment as probability mass :

[0134] ,

[0135] Context weights as conditional probabilities (Normalize within the window):

[0136] ,

[0137] Right now ,therefore .

[0138] (S5) Conditional truncation sampling and Monte Carlo simulation

[0139] For each Conditional truncation sampling (sampling only within this segment) will directly execute simulation calculations within the integrated Monte Carlo module. By running Monte Carlo simulations independently within each scenario sub-interval, the corresponding output mean can be obtained. and variance To ensure the stability and comparability of the simulation results, the same random seed and sampling size were used in the simulation process to ensure that the parameter differences under different scenarios only originated from interval characteristics rather than random fluctuations.

[0140] (S6) Weighted composition and uncertainty output

[0141] The Monte Carlo simulation results for each scenario sub-interval are combined using a weighted method to obtain a global estimate and its uncertainty.

[0142] The global estimate is:

[0143] ,

[0144] in, Assigning weights to each scenario, This formula represents the output mean under the corresponding scenario, reflecting the weighted contribution of different confidence intervals to the overall result. The combined uncertainty is calculated as follows:

[0145] ,

[0146] in, This represents the variance term of the Monte Carlo simulation within the scenario. This represents the discrete terms between scenarios, reflecting the differences between different confidence intervals. Through the above synthesis process, both intra-scenario uncertainty (internal fluctuations) and inter-scenario uncertainty (structural differences) can be considered simultaneously, thus obtaining the comprehensive variance of the overall output. Based on the synthesis uncertainty... The results can be used to calculate confidence intervals or provide key quantiles of the output distribution (such as 5%, 50%, and 95% points) for uncertainty visualization and risk assessment.

[0147] (S7) Interpretation and Visualization of Results

[0148] Output the final estimated value And simultaneously report the weights of the three scenarios. This is to enhance the interpretability and transparency of the model results.

[0149] Example 2

[0150] The electronic device of this invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in read-only memory (ROM) or loaded from a storage unit into random access memory (RAM). The RAM may also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0151] Multiple components in the device are connected to the I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0152] The processing unit performs the various methods and processes described above. For example, in some embodiments, the methods may be implemented as computer software programs tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute the methods by any other suitable means (e.g., by means of firmware).

[0153] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0154] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0155] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0156] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for calculating carbon footprint uncertainty, characterized in that, The method includes: Evidence distillation process: Automatically generate multiple rules from unstructured text including historical carbon footprint data; Fuzzy synthesis process: Multiple rules are fuzzily weighted and synthesized in a scenario to generate a confidence distribution; Interval extraction process: Extract the global range and scenario range based on the confidence threshold, and truncate the probability density function of the sampleable truncated normal distribution on the global range to obtain the background distribution; wherein the global range includes all value intervals that make the confidence distribution greater than the first confidence threshold; the scenario range extraction process includes: selecting the coverage rate, solving for the second confidence threshold that satisfies the confidence distribution being greater than the candidate threshold variable and has the smallest interval length that satisfies the coverage rate, and then determining the smallest interval that contains all the confidence distributions that satisfy the second confidence threshold. Scenario weighting and Monte Carlo simulation process: Divide the scenario range into three equal scenario sub-intervals, calculate the scenario weight of each scenario sub-interval using the background distribution; perform conditional truncation sampling on each scenario sub-interval, and run Monte Carlo simulation independently to output the corresponding mean and variance; Weighted synthesis uncertainty process: The Monte Carlo simulation results of each scenario sub-interval are combined by weighting the corresponding scenario weights to obtain the uncertainty of carbon footprint.

2. The method for calculating carbon footprint uncertainty according to claim 1, characterized in that, The evidence distillation process is based on an LLM model and includes: extracting rule candidates from unstructured text, extracting the parameter range and evidence strength of the context through a fixed prompt template, and automatically converting each piece of data into a rule unit. The rule unit includes rule conditions, support set, typical band, and evidence strength. The unstructured text includes historical carbon footprint data of enterprises, industry reports, papers, and expert texts.

3. The method for calculating carbon footprint uncertainty according to claim 1, characterized in that, The fuzzy synthesis process includes: fuzzifying the parameter range of the rules into membership degrees; and using the max-min operator to perform fuzzy weighted synthesis of multiple rules in a scenario based on rule activation degree, evidence strength, and membership degree to generate a credibility distribution.

4. The method for calculating carbon footprint uncertainty according to claim 3, characterized in that, Calculate similarity for each scene dimension, and calculate rule activation based on similarity. Methods for calculating rule activation include selecting the minimum value and selecting the product.

5. The method for calculating carbon footprint uncertainty according to claim 3, characterized in that, The confidence distribution is specifically represented as follows: ， in, : No. i The activation degree of a rule in the current scenario c represents the applicability or influence of that rule; : No. i The weight of evidence for each rule indicates the credibility of that rule; : No. i The membership degree of a rule indicates the scope of support for that rule; : No. i The weighted activation degree of a rule is the combined effect of the rule's applicability and the strength of evidence.

6. The method for calculating carbon footprint uncertainty according to claim 1, characterized in that, The calculation process of the scenario weight includes: dividing the scenario range into three scenario sub-intervals to represent low, medium and high scenarios; directly defining the probability quality of the scenario sub-interval by the area of ​​the background distribution under each scenario sub-interval; and then normalizing within the scenario interval to obtain the scenario weight corresponding to the scenario sub-interval.

7. The method for calculating carbon footprint uncertainty according to claim 1, characterized in that, The weighted synthesis uncertainty process includes: synthesizing the uncertainty of the carbon footprint based on the global estimate, the scenario weights of each scenario sub-interval, and the mean and variance of the Monte Carlo simulation output under the corresponding scenario, specifically: ， in, and Each represents a scenario i Variance and mean of the Nemont Carlo simulation Representing discrete terms between scenarios, This is a global estimate; Context weights.

8. The method for calculating carbon footprint uncertainty according to claim 7, characterized in that, The global estimate is obtained by weighting and summing the scenario weights of each scenario sub-interval with the mean of the Monte Carlo simulation output under the corresponding scenario.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 8.

Images