Carbon footprint full-link acquisition and trusted accounting system based on artificial intelligence

Abstract

The invention discloses a carbon footprint full-link acquisition and trusted accounting system based on artificial intelligence, which relates to the technical field of carbon emission data processing, and comprises the steps of collecting original records from multi-source data such as resource planning, manufacturing execution, energy consumption metering, logistics and computing power operation, cleaning and normalizing, uniformly generating an accounting event, and merging the accounting event into an additional event account book; the evidence file enters a content addressing evidence library, and a consistency score write-back event is calculated based on cross-source constraint; solidifying a rule-factor-algorithm-parameter combined snapshot number, calculating a risk score, arranging, checking and sampling, carrying out closed loop on a minimum evidence list and a problem, and completing snapshot accounting; and constructing a full-amount commitment root, generating a verifiable minimum disclosure evidence packet according to sampling, and performing additional correction and versioning updating according to check feedback. Therefore, results can be traced, recalculated and checked, the evidence obtaining disclosure range is reduced, and mismatching and tampering risks are reduced.

Description

Technical Field

[0001] This invention relates to the field of carbon emission data processing technology, specifically to an artificial intelligence-based system for the full-chain collection and reliable accounting of carbon footprint. Background Technology

[0002] Manufacturing enterprises and computing power service providers are required to generate carbon footprint accounting results by product, batch, order, or artificial intelligence training and inference when delivering products, bidding, export compliance, supplier access, and information disclosure, and accept third-party verification when necessary.

[0003] This type of accounting typically requires data from enterprise resource planning, manufacturing, energy management and sub-metering, warehousing and transportation, testing and inspection, invoices and billing, as well as cluster scheduling and resource metering. This data flows across departments, systems, and time windows, with different business entities constantly being split and merged in procurement, production, logistics, and computing power scheduling. The accounting standards need to remain consistent under constraints such as system boundaries, allocation rules, and emission factor versions. The auditing party must also be able to find key original vouchers and measurement evidence. Within the industry, there are technical pathways such as carbon accounting platforms, energy and carbon management systems, and product carbon footprint data exchange standards for connecting with business systems, collecting activity data, and outputting results.

[0004] Current technologies lack direct constraints and common constraints in verification practices: inconsistencies in the granularity and semantics of multi-source heterogeneous data can lead to gaps or inconsistencies in time alignment, unit conversion, object attribution, and allocation mapping for the same business object's energy consumption, output, working hours, invoices, logistics, and computing power measurement; emission factors and methodological configurations iterate with policy and database changes, and the recalculation of historical results and interpretation of discrepancies require evidence of traceable version information. Verifications are mainly based on sampling and key tracing. In the absence of a structured mapping relationship from results to original evidence, enterprises need to temporarily collect a large amount of sensitive business data to meet evidence requirements. This increases the cost of cross-departmental collaboration and repeated reconciliation, while also posing a risk of exposing trade secrets and data security. If consistency verification and evidence collection cannot be completed within the specified time limit, it will affect customer acceptance, certification conclusions, and transaction access, while also reducing the availability and comparability of accounting results in the supply chain.

[0005] Therefore, the current technical problem is: Given the dispersed nature of multi-source heterogeneous business and measurement data, the changing accounting standards and factor versions, and the need for sampling and verification, how can we ensure that the carbon footprint accounting results can be verified by third parties through traceable, recalculated, and verifiable evidence without requiring excessive disclosure of sensitive data? Summary of the Invention

[0006] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides an AI-based end-to-end carbon footprint data collection and reliable accounting system. It uses evidence files to enter a content-addressable evidence library, calculates consistency scores based on cross-source constraints, and writes back events. It solidifies the combination of rules, factors, algorithms, and parameters into snapshot numbers, calculates risk scores, and arranges verification sampling, a minimum evidence list, and issue closure loops to complete snapshot accounting. It constructs a full commitment root and generates verifiable minimum disclosure evidence packages based on sampling, adding corrections and updating versions according to verification feedback. This makes the results traceable, recalculated, and verifiable, solving the technical problems described in the background section.

[0007] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: The AI-based carbon footprint full-chain collection and trusted accounting system includes: collecting raw records, cleaning and normalizing them to a unit time window, associating them with business affiliation, generating accountable events and calculating event fingerprint values ​​to write them into an append-only event ledger; writing evidence files into a content-addressed evidence library to form evidence citations; and establishing constraint sets to calculate cross-source consistency scores and write back accountable events. Generate a combined snapshot number, calculate a risk score based on cross-source consistency score, data quality and emission contribution under constraints, generate a verification sampling plan, a minimum evidence list template and a problem closure list, and aggregate key results summary based on the combined snapshot number; Based on calculable events, evidence citations, combined snapshot numbers, and key result summaries, a commitment structure is constructed to generate a full commitment root and form a commitment layer evidence package; for events that hit the verification sampling plan, a sampling disclosure layer evidence package is generated, which includes member proof, minimal evidence object or desensitized fragment, and recalculated formula.

[0008] Furthermore, the field sequence of the original record is standardized according to the event mode version, and the unit mapping, occurrence time window opening and closing intervals, numerical fixed precision and null value placeholder processing are performed to form a standardized event string. The event fingerprint value is generated by concatenating the standardized event string with the domain separation label and then performing hash mapping.

[0009] Furthermore, before being written into the content-addressable evidence database, the evidence files undergo byte normalization and archiving encoding processing according to the evidence type, generating evidence fingerprints and using the evidence fingerprints as object keys; the append-only event ledger records the ordered sequence of evidence fingerprints in the entries corresponding to the event fingerprint values ​​as evidence references.

[0010] Furthermore, the constraint set consists of constraints related to energy consumption and operating conditions, output and cycle time, billing cycle and measurement, logistics mileage and load, computing power usage and energy consumption model; the first step is to calculate the normalized residuals of the constraint terms and generate residual summaries, and write the residual summaries and cross-source consistency scores back to the accountable events.

[0011] Furthermore, the combined snapshot number is assembled from rule entries, factor entries, algorithm entries, and parameter entries, and each entry includes a version number and scope of effectiveness; the combined snapshot number is written into the accounting task record, and a unique mapping table between the combined snapshot number and the entry is established for recalculation retrieval.

[0012] Furthermore, the risk score is obtained by fusing the data quality score, anomaly score, contribution score, change score, and cross-source consistency score; the data quality score is generated by mapping data quality labels to evidence citation completeness, and the contribution score is generated by normalizing the event emission estimates under the combined snapshot number.

[0013] Furthermore, the verification sampling plan includes a mandatory verification set and a sample verification set; the mandatory verification set includes calculable events whose risk scores meet the mandatory verification threshold and calculable events whose cross-source consistency scores meet the consistency threshold, and the sample verification set is generated by combining the contribution scores of the remaining calculable events according to activity type.

[0014] Furthermore, the minimum evidence list template associates the required evidence types with the alternative evidence types according to the activity type, and verifies the completeness of the evidence citations by the evidence type corresponding to the evidence fingerprint in the evidence citation; the problem closed-loop list generates gap identification entries, supplementary evidence request entries and recalculation verification entries by indexing the event fingerprint value.

[0015] Furthermore, the commitment structure includes a commitment tree, the leaf summary of which is calculated by combining the snapshot number, event fingerprint value and the ordered sequence of evidence fingerprints in the evidence citation; the commitment root is calculated from bottom to top according to the leaf summary and a fixed splicing direction rule, and the commitment root and the splicing direction rule version are written into the commitment layer evidence package.

[0016] Furthermore, for each hit event, the sampling disclosure layer evidence package includes the corresponding member proof, the desensitized fragment of the minimum evidence object, and the recalculation formula, and associates the event fingerprint value with the issue closed-loop list; when the issue closed-loop list generates a supplementary evidence request, the correction entry is added to the appended event ledger and the evidence reference is updated, and the commitment layer evidence package and the sampling disclosure layer evidence package that establish a reference relationship with the previous version are regenerated.

[0017] (III) Beneficial Effects This invention provides an artificial intelligence-based system for end-to-end carbon footprint data collection and reliable accounting, which has the following advantages: By unifying raw records from resource planning, manufacturing execution, energy consumption metering, logistics, and computing power into accountable events based on business object identifiers and occurrence time windows, and writing event fingerprints into an append-only event ledger and evidence fingerprints into a content-addressable evidence database, stable backtracking of accounting results to corresponding vouchers and metering documents can be achieved without cross-system reconciliation or manual retrieval. A cross-source constraint set is established for related data of the same business object, calculating cross-source consistency scores and residuals and writing them back to accountable events. The risks of mismatch, missing data, and substitution are defined at the event level, providing directly referable clues for risk stratification and sampling review, and reducing ambiguity in verification communication.

[0018] By generating snapshot numbers that combine rules, factors, algorithms, and parameters, and calculating risk scores under the constraints of these snapshot numbers, the system generates verification sampling plans, minimum evidence list templates, and problem loop lists. It also outputs event-level contribution decomposition and difference location clues, enabling the scope of evidence collection, supplementary evidence paths, and recalculation scope to be reproduced and connected under the same caliber.

[0019] By constructing a full set of commitment roots based on a set of calculable events, a set of evidence citations, combined snapshot numbers, and key result summaries, a commitment layer evidence package is formed. This generates a sampling disclosure layer evidence package containing member proof, minimal evidence object or desensitized fragments, and recalculated formulas for sampled hit events. This enables the auditor to independently verify the source of disclosure items and limit the disclosure content to the constraints of the minimal evidence list template. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the structure of the artificial intelligence-based carbon footprint full-link collection and trusted accounting system of the present invention. Detailed Implementation

[0021] 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, and 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.

[0022] Please see Figure 1 This invention provides an AI-based system for the full-chain collection and reliable accounting of carbon footprints. Existing technologies typically aggregate records and calculate emission results through interface connections or offline import. Due to the dispersed nature of record sources and varying semantic granularity, the same business object is often presented with different identifiers, time windows, and measurement standards, leading to unclear attribution, difficulty in verifying allocation criteria, and a lack of stable mapping between evidence documents and calculation results.

[0023] There are cross-source constraints between energy metering and billing cycles, logistics mileage and load, and computing power usage and node energy consumption models. However, in the summary accounting stage, manual comparison is often used during the verification phase. This leads to the need for temporary retrieval of a large amount of sensitive data and repeated reconciliation during sampling verification, and it is difficult to form a calculable representation of the risks of mismatch, missing, or replaced data. Step one is used to complete standardization, evidence linking, and cross-source consistency proof at the event level, providing a data foundation for subsequent risk stratification and sampling verification.

[0024] Step 1: Convert the original records into measurable events with unique business attribution and recalculation reference, and solidify cross-source consistency proof at the measurable event level, so that subsequent risk stratification and sampling verification can directly call the same event and evidence reference without repeated reconciliation.

[0025] To address the ambiguity arising from the presentation of the same business fact across different systems using varying field arrangements, units, and time representations, a unified event representation must be established at the entry point. Differences in field order, null value handling, and unit conversion can prevent the same event from obtaining a stable reference fingerprint, making it difficult for sampling checks to locate the original vouchers corresponding to emission contributions.

[0026] Therefore, the raw records are constrained to a fixed, calculable event pattern, and a normalized event string is generated using deterministic rules, from which a summary fingerprint is then generated. After the data acquisition terminal obtains the raw records, the data retrieval path can be either by capturing incremental transaction records through database changes or by subscribing to business events through a message queue. For metering, the data retrieval path can be achieved by reading registers through an industrial internet gateway and encapsulating them into raw records according to the sampling period.

[0027] First, perform field alignment: write the business object identifier into a field with the same name. The business object identifier can be composed of product identifier, batch identifier, process identifier, order identifier, and job identifier, and the combination order is fixed. Then, perform measurement caliber alignment: unify physical quantities into physical quantity values ​​+ physical quantity units, and units can only come from the unit mapping table. Next, perform time window alignment: unify the occurrence time window into a closed-open interval representation of start time and end time and fix the time zone source. Then, perform null value normalization: missing fields are written into fixed placeholders but do not participate in numerical conversion. Finally, concatenate the pre-set field sequence into a normalized event string. Numerical fields adopt fixed precision rules to avoid non-business differences at the string level.

[0028] Furthermore, a normalized event string is generated and a domain separation tag is attached. The domain separation tag identifies the event mode version and activity type. A hash mapping is performed on the domain separation tag + normalized event string to obtain a digest fingerprint of the verifiable event, which is used for ledger writing and evidence citation binding. The calculation form of the digest fingerprint is as follows:

[0029] Where: event fingerprint value : A verifiable event digest fingerprint, with a value of length [value missing]. A binary sequence of bits; a hash mapping function : Deterministic hash mapping function, taking values ​​from the Chinese Cryptographic Standard No. 3 hash algorithm or an equivalent security strength hash algorithm, used to suppress collisions and tampering; Domain separation label : Domain separation tag, with a value of length 1 to Fixed-length strings of bytes; normalized event strings : Normalized event string, whose value is a byte string concatenated according to the field sequence; Evidence fingerprint value A unique index used for the content addressing evidence repository, and associated with the event fingerprint value. A binding is formed.

[0030] Among them, the evidence byte normalization rules are as follows: electronic invoices and structured reports are parsed into a set of fields, and the archived document byte sequence is re-rendered according to a fixed field sequence; photos / scanned documents are first normalized in direction and size, and the archived image byte sequence is output according to fixed compression parameters; log fragments are first normalized to a unified encoded byte sequence according to a fixed line terminator and encoding conversion.

[0031] Evidence fingerprint value format:

[0032] in The normalized sequence of evidence bytes. Separate labels for the evidence domain.

[0033] Evidence citation writing: In the append-only event ledger, evidence citations are uniformly recorded as evidence fingerprint values. The sequence, and sorted according to a fixed rule based on evidence type (e.g., by evidence type number, generation time, ...). ).

[0034] When used, the normalized event string fixes the unit, time window and field sequence, so that the same business fact can obtain the same event representation under different source systems; the summary fingerprint provides a stable anchor point for ledger writing and evidence citation, so that the verification and location can be directly indexed to the corresponding event by fingerprint.

[0035] To ensure consistency in the mapping between events and evidence during the accounting and verification periods, and to avoid directly exposing large volumes of evidence files in the ledger carrier, an append-only event ledger is used to carry event fingerprints and evidence references, and evidence files are stored using content addressing. If the ledger only stores descriptive text and not verifiable references, verification points cannot be established when evidence is replaced or uploaded repeatedly.

[0036] The ledger writer obtains the event fingerprint value Ledger entries are then generated, which include event fingerprints, business object identifiers, occurrence time windows, activity types, physical quantity units, allocation keys, data quality labels, and permission labels.

[0037] Evidence files related to the event are subject to evidence unification: For scanned documents, rotation correction and grayscale conversion are performed before generating a unified compression format; electronic invoices and test reports are converted to archived document format with fixed font embedding rules to avoid byte differences in the same content under different export methods. Scanned documents, electronic invoices, test reports, meter original files, and work log fragments are uniformly encoded and paged to ensure consistent byte sequences for the same file under different upload methods; evidence fingerprints are generated when evidence objects are added to the database, and these fingerprints are used as content addressing addresses in the content-addressed evidence database; ledger entries only store evidence fingerprints and evidence types. Finally, the ledger adopts an append-only writing strategy, with subsequent corrections achieved by adding correction entries. These correction entries reference the original event fingerprint and the original evidence fingerprint to establish a version relationship.

[0038] First, the ledger entries are constructed and added to the append-only event ledger. The ledger uses sequence numbers and timestamps to fix the order of the entries. Then, evidence is linked: the evidence fingerprint is written back to the ledger entries and forms a binding pair with the event fingerprint. During verification, the event fingerprint is used to locate the entries, and then the evidence fingerprint is used to retrieve the evidence file or de-identified fragment from the content-addressed evidence database.

[0039] For example, on an electronic assembly line, on-site personnel upload photos of the electricity meter readings and work order completion records from the metering terminal and manufacturing execution terminal, respectively. The data acquisition terminal uniformly encodes the photos to generate evidence fingerprints, which are then written into the content-addressable evidence database. Simultaneously, it reads the work order number, batch number, completion time, and the sub-item electricity consumption within the same time window to generate calculable events and calculate event fingerprints. The ledger writer writes the event fingerprints, batch numbers, time windows, sub-item electricity units, and allocation keys into an append-only event ledger, and includes the photo evidence fingerprint as an evidence reference in the same entry. Accounting personnel can retrieve the entry by batch number to obtain a fixed correspondence between events and evidence references; verification personnel retrieve photo files by sampling using event fingerprints and verify the read content.

[0040] When in use, the append-only event ledger solidifies event entries in sequence, and evidence objects are stored in a content-addressable manner, enabling evidence replacement to be explicitly discovered through fingerprint verification; the correction entry mechanism keeps the correction process traceable without destroying historical entries.

[0041] To bring the inherent physical and business constraints across systems forward to the event level and form a computable and consistent representation, in manufacturing scenarios, there are verifiable relationships between sub-item energy consumption and equipment operating conditions, output and cycle time, billing cycle and metering cycle, logistics mileage and load, and computing power usage and node energy consumption models. If manual comparison is only performed during the verification stage, it is difficult to locate specific events in a timely manner when mismatches, omissions, or replacements occur, and it is even more difficult to provide a unified measurement for subsequent risk stratification.

[0042] Therefore, related events are aggregated around the same business object, a constraint set is constructed, and normalized residuals are calculated.

[0043] The consistency constraint generator aggregates event entries into event clusters based on business object identifiers and generates constraint terms with a fixed form for each event cluster. For energy consumption and operating condition constraints, it generates estimated energy consumption using the rated power of the equipment or the energy consumption model parameters of the work node and forms a difference with the metered electricity consumption. For billing cycle and metering constraints, if an event crosses the settlement boundary, it splits the time window according to the billing boundary and uses piecewise linear interpolation to project the metered value onto the sub-time window. The piecewise linear interpolation uses the endpoint of the original time window of the metering event and the endpoint of the target sub-time window as nodes to allocate the metered value to the target sub-time window according to the time ratio. When the metering record is a discrete reading point, it first forms a line segment with adjacent reading points, and then calculates the value at both ends of the target sub-time window and takes the difference.

[0044] For logistics mileage and load constraints, the estimated fuel consumption is generated using a vehicle fuel consumption coefficient table and compared with refueling receipts to form a difference. For computing power occupancy and energy consumption model constraints, the estimated node energy consumption is generated using graphics processor occupancy time and node energy consumption model parameters and compared with converted electricity to form a difference. The above differences are normalized to residuals using a normalization scale so that constraint terms with different dimensions can be integrated in the same cross-source consistency score.

[0045] Define and deduce the family of quantitative generating functions The constraints can be categorized into the following different forms based on their type (examples can be used for one implementation): Energy consumption - operating condition constraints: Estimated energy consumption is obtained from the equipment's rated power and operating time:

[0046] Rated power Obtained from equipment ledger, runtime It is obtained from the intersection length of the work order time window and the equipment start / stop record;

[0047] Logistics constraints: Estimated fuel consumption is obtained from mileage and vehicle fuel consumption coefficient: mileage Fuel consumption coefficient obtained from the transportation management system It is obtained from the vehicle fuel consumption coefficient table (the version is fixed according to the rules).

[0048] Computing power consumption - node energy consumption model constraints: The estimated node energy consumption is obtained from the GPU occupancy time and energy consumption model parameters:

[0049] The graphics processor takes up time. Model parameters from cluster scheduling logs From the nodal energy consumption model parameter table (with combined snapshot number) (Cure).

[0050] First, estimate generation is performed: necessary fields are extracted from the event cluster and combined with the baseline table or energy consumption model parameter table to generate estimate quantities; when time windows are misaligned, piecewise linear interpolation is used to complete the projection, and then residual normalization is performed: the difference between the measured value and the estimate is normalized according to the normalization scale to obtain dimensionless residuals. The residual calculation form is as follows:

[0051] Where: residual value : No. The normalized residuals of each constraint term take values ​​of _____. Actual measurement : No. Each constraint item is a measured, non-negative number, derived from meters, invoices, or vouchers; the estimated quantity... : No. Each constraint term is a non-negative value, derived from the event cluster field and parameters of the benchmark table or energy consumption model; normalized scale. : No. Each constraint term is a normalized scale, and its value is... This is derived from the measurement resolution, allowable interval width, or contract tolerance. Obtaining normalized residuals Then, the residual sequence and total amount of punishment In the input consistency detail database, the index keys are the business object and the event cluster; append-only event ledger entries only write the cross-source consistency score. And residual summary fingerprint (consistency check), residual details are not subject to access control, and are only used in internal risk calculation and verification recalculation processes.

[0052] Abnormal score The generation is a reproducible quantile mapping: within a historical window of the same activity type and the same location identifier, the total penalty is taken. The empirical distribution is used to calculate the current... quantile values ​​and mapping to .

[0053] Normalization scale Selection priority: If the measuring equipment provides a resolution or accuracy class, then the measurement resolution or accuracy will be converted to [the appropriate value]. If the contract or verification plan specifies an allowable range, then the allowable range width shall be used. If the above is not available, then the quantile interval of historical similar event clusters is used to form the default scale, and the default source is written into the consistency details database.

[0054] Estimated Quantity The minimum feasible path is as follows: Energy consumption and operating condition constraints: the rated power of the equipment and the running time form the estimated energy consumption; Billing cycle and metering constraints: piecewise linear interpolation metering is projected onto the billing sub-time window; Computing power occupation and energy consumption model constraints: the graphics processor occupation time and node energy consumption model parameters form the estimated node energy consumption, which is converted into electricity.

[0055] When used, constraint sets transform cross-source associations into event-level computable objects, residual normalization enables deviations of different dimensions to be uniformly measured, and piecewise linear interpolation aligns cross-period events to the same time window, avoiding spurious differences caused by bill boundaries.

[0056] Piecewise linear interpolation: when adjacent reading points exist. and ,right The interpolation value is:

[0057] Supplementary boundary: When At that time, take directly This situation should be recorded as an abnormal data retrieval flag in the consistency details database.

[0058] To fuse residuals from multiple constraints into a single cross-source consistency score, and to solidify this score along with the residual summary into the event entry, subsequent steps can determine whether there is a risk of mismatch that should be prioritized for verification without exposing the detailed residual information. If only the fused score is saved without saving the verifiable summary, it would be difficult to prove that the set of residuals corresponding to the score was not subsequently replaced during dispute verification. Therefore, a smoothing penalty function is used to fuse the residuals, and simultaneously solidify the residual summary fingerprint and the cross-source consistency score.

[0059] The cross-source consistency score calculator reads the residual sequence of the same event cluster and the constraint weight table. The constraint weight table is bound to the activity type and is fixed with the event mode version. The source of the constraint weight table can be preset by the accountant according to industry rules, or it can be specified by the contract or verification plan and fixed in the event mode version to ensure consistent weights during subsequent recalculations.

[0060] The penalty function employs a log-exponential smoothing form, mapping each residual to a monotonically increasing penalty and then summing them with weights to obtain the total penalty; subsequently, the total penalty is mapped to... The cross-source consistency score of the interval. Before writing back, the residual sequence generates a residual summary fingerprint. The residual summary fingerprint and the cross-source consistency score are written together into the extended field of the event entry, so that any sampled event carries the same consistency proof entry.

[0061] Furthermore, the total penalty amount should be calculated first:

[0062] Where: Total penalty Total residual fusion penalty, with possible values. Used to compress multi-constraint residuals into a single scalar while maintaining sensitivity to large residuals; number of constraints : Number of constraints, ranging from positive integers, representing the number of constraint terms participating in the fusion; Weight : No. The weights of each constraint term, with values ​​ranging from [value range missing]. And satisfy This is used to demonstrate the importance of constraints; residual value : No. The residuals of each constraint term have a range of values. As a penalty input; smoothing scale : Smoothing scale, with a value range of , used to adjust the steepness of the penalty function transition; Furthermore, perform cross-source consistency score mapping and write back:

[0063] Where: Cross-source consistency score Cross-source consistency score, value range This is used to characterize the degree of cross-source data consistency of event clusters and to provide subsequent risk stratification calls; Total amount of punishment Total penalty amount, with possible values. , as the input for the exponential decay of cross-source consistency score; When used, the smoothing penalty function remains continuous when the residuals are close to zero and remains sensitive when the residuals increase. The cross-source consistency score has both slight bias and serious mismatch. The residual summary fingerprint and cross-source consistency score write-back can help sampling verification quickly locate event clusters that need to be reviewed on an event-by-event basis.

[0064] Step 2, Step 1 is used to generate calculable events and solidify cross-source consistency scores. Event fingerprint value Step two is used to combine snapshot numbers. Risk scores are formed under constraints The sampling plan is then developed and verified. Step three is used to construct a full commitment root and deliver a verifiable minimum disclosure evidence package.

[0065] In the same accounting task, rules, factors, algorithms, and parameters may come from different versions of configuration libraries and external factor libraries, and may differ due to product category, location, and time window. If the above elements are temporarily selected or replaced multiple times during the calculation process, the accounting results are difficult to recalculate, and it is also difficult to explain the source of the discrepancies to the auditor.

[0066] Using the rule-factor-algorithm-parameter quadruple as the archive object, it is assembled into a combined snapshot and fixed as a combined snapshot number. This ensures that subsequent calculations can only reference the content corresponding to that number.

[0067] To pre-determine the selection criteria and write them into the searchable entries, the assembly process reads the activity type, occurrence time window, location identifier, and allocation key of the measurable event: the activity type is used to select the algorithm entry, the occurrence time window and location identifier are used to constrain the applicable period and region of the factor entry, and the allocation key is used to bind the allocation rule entry; for computing power occupancy events, the energy consumption model parameter version and the source of the data center power usage efficiency coefficient are also written into the parameter entry. To avoid using multiple versions of factors for the same activity type, the assembly process performs intersection verification on the candidate factor entries and determines a unique version according to a preset priority table, writing the unused version into the snapshot description field for review.

[0068] First, the snapshot component is normalized: rule entries, factor entries, algorithm entries, and parameter entries are transcribed into a fixed field sequence, and the version number, scope of effect, source identifier, and selection reason are explicitly written in. During snapshot sealing and reference writing: a snapshot fingerprint is generated in a deterministic splicing order and used as the combined snapshot number. Combine snapshot numbers Write the accounting task record and output summary, and at the same time store the snapshot component description in the immutable object store or relational database append table so that the same number corresponds to the same content.

[0069] When using, combine snapshot numbers. The selection of fixed calibers allows for recalculation of accounting results under the same number; snapshot component descriptions include version and reason fields, enabling verification and review to directly reference entries without relying on verbal explanations; combined snapshot numbers... Event fingerprint value Parallel existence avoids finding the breakpoint where the event is found but the corresponding caliber cannot be found.

[0070] Verification sampling needs to cover clusters of high-risk and high-contribution events, given the constraints of evidentiary costs. If sampling is based solely on empirical ranking, it is easy to miss cross-source consistency scores. The low risk of mismatch also makes it easy to overlook systemic changes caused by changes in statistical caliber. (Regarding the combined snapshot number...) After solidification, a risk score is generated for each measurable event. and assign risk scores After being broken down into interpretable sub-items, the event extension field is written back to make the sampling basis replayable.

[0071] To calculate the cross-source consistency score in step one The degree of traceability of evidence is included in the sampling criteria. First, data quality labels are mapped to data quality scores. The mapping rules take the integrity of evidence citation as the entry point: Events with original measurement documents and valid supporting documentation received higher scores. Events involving only summary bills or manually entered vouchers received lower scores. Regenerate abnormal scores Abnormal scores Starting with the event cluster residual summary, the quantile positions of the residuals are read and mapped. ; and then in the combined snapshot number The emission estimates are obtained by performing a preliminary calculation on the physical quantities of the event, and the contribution score is formed by the proportion of the estimated values. Then, using the business object identifier, locate the cluster of similar events from the previous period, which can be mapped to the combined snapshot number. Calculate the differences from historical snapshots and map them to change scores. Finally, , , , Cross-source consistency score The risk score is then fused using a smoothing penalty function. And retain the sub-items for explanation.

[0072] The smallest form of the emissions estimate: for each event, in the combined snapshot number Under constraints, emission factors are selected and event emission estimates are calculated; for multi-physical quantity events, corresponding factors are selected according to activity type.

[0073] Contribution points Generation: The event emission estimates are summed and normalized over all events to obtain the result. .

[0074] Change score Generation: Using the same type of business objects from the previous cycle as a reference, the same combination of snapshot numbers is used. The method of recalculating the event cluster of the previous period is used to eliminate the difference in caliber, and then the difference is mapped to... The mapping function can be in the form of exponential saturation to avoid discontinuous jumps caused by extreme differences.

[0075] First, the component quantity generation and write-back are performed: generating data quality scores. Abnormal scores Contribution score Change of score and cross-source consistency score Write back to the calculable event extension field as well; Then perform smoothing penalty fusion: use a differentiable function to ensure that changes in component quantities are continuously transmitted to the risk score. This facilitates explaining the sensitivity during the review process. Risk Score The calculation form is:

[0076] Where: Risk score Risk score, value range This is used to measure the priority of calculable events entering the mandatory or sampled set; data quality score. Data quality score, values It is used to characterize the completeness of evidence citation and the reliability of data entry; anomaly score : Abnormal score, value , used to characterize the degree of deviation of residual quantiles; Contribution points Contribution score, value range This is used to characterize the proportion of the emission estimate within the accounting scope; change score. Change the score, select a value It is used to characterize the magnitude of change of similar events relative to the previous period; cross-source consistency score Cross-source consistency score, value range Used to characterize the degree of self-consistency of cross-source constraints; smoothing scale : Smoothing scale, value This is used to adjust the steepness of the penalty transition and avoid abrupt changes in risk scores; Risk weight Risk weight, value And the sum of the other risk weights is Risk weights Risk weight, value And the sum of the other risk weights is Risk weights Risk weight, value And the sum of the other risk weights is Risk weights Risk weight, value And the sum of the other risk weights is Risk weights Risk weight, value And the sum of the other risk weights is This is used to characterize the impact of cross-source consistency scores; When using it, the risk score Data quality, anomalies, contributions, changes, and cross-source consistency are unified on the same scale, and sampling criteria can be interpreted item by item; risk scores. Cross-source consistency score It is differentiable, and during the review, the sensitivity can be calculated using an automatic differentiation process and the effects of changes in sub-items on priority can be explained; the sub-item quantity write-back event extension field allows the same event item to be reused in subsequent sampling plans and problem closure lists.

[0077] The verification sampling should cover both risk scores and High-value event clusters should also cover contribution scores. High event clusters, but limited by evidence collection costs and verification cycles. Based on event fingerprint values. As the unique location key for the sampling item, the mandatory core set and the sampled core set are formed in the same sampling plan, and the triggering reason is written into the reason field of the item, so that the sampling process can be replayed.

[0078] To transform sampling from an experience list into an interpretable selection process, the process first filters according to a pre-defined mandatory criteria: when the risk score... Exceeding the mandatory core threshold or when the cross-source consistency score is... When events fall below the consistency threshold, they are included in the mandatory core set, and the triggering reason is recorded. The remaining events are then stratified by activity type, and within each stratum, candidate sequences are formed based on contribution priority followed by risk. An evidence cost boundary is then introduced, with the evidence cost determined by the evidence citation type and acquisition path. The generation of the core set employs a two-stage approach: first, a greedy approach is used to add candidate sequences until the coverage lower bound or cost upper bound is reached; then, near the boundary, an integer programming solver is used for local replacement, ensuring that events with a significant increase in coverage contribution are included in the core set without violating the mandatory core constraint. A mixed integer programming solver is commonly used in engineering; if the environment is limited, only a greedy approach is used, and the reason for insufficient coverage lower bound is written into the sampling plan description field.

[0079] First, the mandatory core set is locked and written to the sampling plan: the mandatory core events are assigned event fingerprint values. Write the sampling plan entries and attach risk scores. Cross-source consistency score And the triggering reason.

[0080] Then, perform a constrained solution for the sample set and sort and distribute it: generate the sample set under the cost boundary and sort it by production line or computer room segment, so that on-site certification can be executed continuously according to geographical location; the equivalent implementation path can sort the intra-layer ranking from contribution score. Switch to change score To adapt to scenarios with frequent changes in operating conditions.

[0081] The accounting personnel select a specific batch of electricity and steam event clusters, as well as a specific day's computing power occupancy event cluster for a particular model training job, on the accounting task interface. Then, they read the combined snapshot number. Cross-source consistency score in event extension fields Risk score A sampling plan is generated, and steam meter events and graphics processor occupancy events are written into the mandatory core set. On-site personnel photograph and upload the seals and readings from the meter housing according to the sampling plan. Maintenance personnel export node power curve segments from the energy consumption monitoring system and upload them. Accounting personnel see that the evidence references for the corresponding event fingerprint values ​​have been completed in the problem loop list, and then generate local recalculation entries.

[0082] When used, the mandatory core set locking pre-includes high-risk and low-consistency events in the verification scope; the constrained solution ensures that the sampled core set covers the main contribution paths within the evidence cost boundary; the sampling plan uses event fingerprint values. Bind entries and record reasons; sampling basis can be replayed.

[0083] After the sampling plan is completed, in the combined snapshot number The accounting process is executed and output to the business object dimension, transforming verification feedback into traceable supplementary verification and recalculation tasks. By linking accounting execution, contribution mapping, and the issue closure list, discrepancies in verification feedback can be located at the event level and partially recalculated without introducing caliber drift. This avoids pulling full data during the verification phase. Accounting execution reads rules and factor entries under the combined snapshot number and aggregates and allocates them according to activity type and allocation key: public energy consumption derives entries according to the allocation rules, and the derived entries inherit the original event fingerprint value. References are made and derived link identifiers are written. The computing power occupancy event is converted into electricity based on the energy consumption model parameters and the data center power usage efficiency coefficient and aggregated into the job identifier.

[0084] The accounting output also generates an event-level contribution map, listing which event fingerprint values ​​each business object has. The contribution is then presented, and a priori order of contributions is provided. Subsequently, the problem loop generator checks the completeness of evidence citations for mandatory and sampled events according to the sampling plan and evidence list template. For gaps, supplementary evidence request entries are generated. After supplementary evidence is completed, a local recalculation is performed using event cluster spectrum boundaries, numbering only the affected event clusters within the same combination snapshot. The version relationship is then recalculated and written.

[0085] First, the accounting execution and contribution alignment are performed: outputting business object-level results and event fingerprint values. The contribution mapping is performed, and the combined snapshot number is entered in the accounting task record. Next, a closed-loop driven local recalculation is performed: supplementary evidence entries are generated according to the problem closed-loop list of gap events. After the evidence references are completed, a local recalculation is performed. The local recalculation is performed through work queue scheduling, and there is no version relationship between supplementary evidence entries, recalculation entries, and update result entries. The equivalent path can transform the local recalculation slice from the event cluster boundary into a business object level slice, which is suitable for situations where batch aggregation is stronger.

[0086] Calculate and output event fingerprint values The contribution mapping and verification feedback pinpoint specific evidence citations and derivation links; local recalculation is performed on the combined snapshot number. The update proceeds without changing the scope; the version relationship traceability ensures that step three can only select the same set of events and result summary when establishing the commitment layer evidence package.

[0087] Step 3, Step 1 is used to generate the carrier and The calculable events, step two is used to... Lock caliber and The sampling plan is formulated, and step three is used to analyze the full set of calculable events, the set of evidence citations, and... The system generates a commitment layer evidence package from key results summaries and a sample disclosure layer evidence package for sampled events that can be verified with minimal disclosure, thereby forming a verification feedback loop.

[0088] To meet the requirement of recalculation at the verification site, the event entries involved in this verification output are transcribed into a set of repeatable commitment objects. The set of commitment objects, bounded by business objects, is used to prune unique, valid events from the original-correction entry version chain of the append-only event ledger, and its order is determined by a fixed sequence key, ensuring that different operators use the same snapshot number combination. The same set and the same order are obtained.

[0089] The commitment layer evidence package must allow the verifier to recalculate and verify membership relationships; if the selection or sorting rules for valid events are implicit in the implementation details, the same business object may correspond to different event fingerprint values. This causes the commitment root to become misaligned. The system first reads the combined snapshot number from the accounting task record. Then, based on the contribution mapping output in step two, locate the event fingerprint values ​​that participated in the output. List, and trace each along the version relationship of the append-only event ledger. The correction chain.

[0090] Furthermore, valid events are determined based on version priority: when a correction entry contains a replacement physical quantity field, the old event fingerprint value generated by the correction entry replaces the new one. If the correction only supplements the evidence cited, the old entry is retained. And expand the evidence citation set. Then, based on the business object identifier, the start point of the occurrence time window, the activity type sequence number, and the event fingerprint value... A fixed ordinal key is used to sort valid events; for derived entries, the derived entries inherit from the original key. It also adds a derived link identifier, and appends the derived link identifier to the end of the sequence key to maintain replayability.

[0091] First, normalize the commitment object entries: extract each valid event. Event fingerprint value Evidence citation fingerprint sequence, key field summary, cross-source consistency score Summary and Risk Score The summary is concatenated into commitment object entries according to a fixed field sequence; the key field summary only covers activity type, physical quantity and unit, allocation key and occurrence time window, excluding supplier name and unit price fields. Then, the leaf summary is fixed: domain separation labels are added to the commitment object entries. Then, a hash mapping function is used. Generate a leaf digest and write it to the leaf digest-event fingerprint value. Mapping. Domain separation label Used to distinguish the commitment leaf from the event fingerprint computation domain of step one; hash mapping function This step reuses the deterministic hash mapping function already disclosed in step one to ensure consistency.

[0092] When used, the unique event pruning and fixed ordering key remove version chain ambiguity, allowing the set of committed objects to be replayed at the verification site; committed object entries are summarized and stored. and This information allows for risk identification during subsequent sampling without revealing business details. Leaf Summary and The mappings remaining in the leaves are used to generate member proofs that are directly located in the event entries without needing to look up the ledger.

[0093] After the leaf digests are fixed, a commitment tree is constructed to form the commitment root, making the commitment root the unique verification anchor point of the entire set. The commitment tree also handles the generation of membership proof paths; therefore, in addition to the commitment root, it is necessary to solidify the parent-child indexes and splicing direction rules, and link these rules with the combined snapshot number. Bind and write the evidence package to the commitment layer.

[0094] To compress the entire set into verifiable identifiers and provide verifiable membership proofs for minimal disclosure, the system uses leaf summary sequences as the first-level nodes; when the number of nodes is odd, a padding node is appended to the end, with the padding node consisting of a domain-separated label. This method uses a fixed padding string to avoid different implementations using different end-node copying strategies.

[0095] The parent node summary is generated by concatenating adjacent node summaries pairwise in a bottom-up manner. The concatenation direction is determined by the node's position in the sequence, and the parent node summary is generated by a hash mapping function. Generate; iterate until a unique commitment root is obtained. After the commitment root is generated, the system encapsulates the commitment layer evidence package, which contains the commitment root and the combined snapshot number. The system calculates the output range summary, sampling plan summary, number of leaf summaries, commitment tree construction rule version, and parent-child index positions. Then, it generates a signature value using the enterprise signature key, writes the signature value into the header of the evidence package, writes the body of the evidence package once and reads it multiple times, and appends the evidence package publication entry to the append-only event ledger, referencing the commitment root and the evidence package position.

[0096] Commitment tree generation and index solidification: Generate commitment root and solidify parent and child index tables and splicing direction rules. The parent and child index tables are written to key-value storage for path backtracking.

[0097] The commitment layer evidence package is signed and a publication entry is appended to make the commitment root have a traceable publication time and publication subject; the signature algorithm can adopt the national cryptographic elliptic curve signature algorithm, and the corresponding public key certificate is used for verification.

[0098] When using the commitment root, it becomes a verification anchor point using the entire set, with sampled disclosed member proofs serving as a reference. The rules for parent-child indexes and concatenation directions are fixed, allowing the verifier to recalculate paths and verify membership relationships under the same rules. The signing and publishing process links the commitment root with... The binding is written into the ledger publication entry, enabling subsequent closed-loop updates to form a version relationship.

[0099] The sampling plan from step two is then translated into deliverables, namely, the sampling disclosure layer evidence package. This package uses event fingerprint values. Using the location key, the commitment root as the verification anchor point, and the combined snapshot number as the location key. As an anchor point for recalculation, and following the minimum evidence list template to limit the scope of disclosure, the verifying party can verify that the disclosed items belong to the commitment set and complete the recalculation comparison according to the recalculation formula.

[0100] To reduce the exposure surface for evidence collection and avoid disclosure boundary drift caused by manual selection, the system reads sampling plan entries and sorts them according to the event fingerprint values ​​within those entries. Locate leaf summary - The mapping is performed, and the leaf nodes are backtracked to the commitment root in the parent-child index table. The summary of each adjacent node and the splicing direction sequence are extracted to form the membership proof.

[0101] Evidence objects are retrieved from the content-addressable evidence database according to the minimum evidence list template, and fixed anonymization rules are applied: For invoice images, the invoice number, invoice date, product name, and quantity fields are retained, while the payee account and unit price fields are obscured; for test report images, the report number, sample number, test items, and conclusion fields are retained, while the contact person and telephone number fields are obscured; for work log fragments, the work identifier, start and end times, graphics processor usage time, and node identifier fields are retained, while the path field and parameter table field are obscured. The rule version number of the anonymization action is written into the disclosure entry.

[0102] Two feasible paths for de-identification and localization: Structured documents (electronic invoices, structured test reports): After parsing the fields, re-render them, retain them according to the whitelist of fields, leave the blacklist fields empty or cover them, and record the version number of the de-identification rule.

[0103] Scanned images (photos, scanned documents): First, perform text recognition to generate text box coordinates, then mask the field area according to template anchor points (such as fixed relative positions near the invoice number field) or keywords; the coordinates of the masked area are written into the desensitization record, and the desensitization record is signed along with the disclosure entry.

[0104] Furthermore, generate a recalculation recipe, which includes... Fingerprint value of the event being verified List, physical quantities and units, allocation keys and aggregation caliber summary; the disclosed items consist of three parts: member certification, desensitized evidence object, and recalculated formula, and are signed by the enterprise signing key to prevent midway replacement.

[0105] First, calculate the member proof path: backtrack based on the parent-child index table to generate adjacent node summary sequences and splicing direction sequences, and write them into the header of the disclosure entry; then perform minimum disclosure evidence assembly: extract evidence objects according to the minimum evidence list template and execute fixed desensitization rules, and write the desensitized evidence objects and recalculation formulas into the disclosure entry; the member proof path backtracking can use a graph traversal algorithm, and the desensitization can use a document processing library and write the coordinates of the masked area into the desensitization record for verification.

[0106] As an example, the inspectors selected three event fingerprint values ​​from the sampling plan in the meeting room. These correspond to steam metering events, transportation events, and model training operation events, respectively. Enterprise personnel select the published commitment layer evidence package on the accounting terminal. The system generates a sample disclosure layer evidence package and exports it as an archived document and a structured text file. Verification personnel import the commitment root and member certifications from the disclosure items into the verification terminal. After completing member verification, they check the anonymized invoice images and test report images to verify the numbers and quantities. Then, based on the recalculated formula, they recalculate the emission contributions of the three items on the verification terminal and compare them with the accounting output range summary. Finally, they write the discovered evidence gaps into the problem closure list and submit it.

[0107] In use, membership proofs allow disclosed items to be independently verified as members of the commitment set, eliminating the need for the auditing party to retrieve all events and evidence; the minimum evidence checklist template and fixed anonymization rules limit the scope of disclosed fields, ensuring that sensitive fields do not drift with manual selection; recalculated formulas... and Using it as an anchor point, the verification and recalculation are consistent with the accounting standards and can pinpoint specific event items.

[0108] The verification feedback is converged to the event level and forms a traceable change chain. The change chain is triggered by a closed-loop issue list, solidified by an append-based event ledger, and uses combined snapshot numbers. To constrain the recalculation criteria, an updated evidence package with version relationships is generated. This process requires that supplementary evidence and recalculation be conducted within the same timeframe. The process is completed, and a version relationship is established between the new commitment root and the previous commitment root, enabling the verifier to identify the source of the differences before and after the update.

[0109] To avoid caliber drift and record breakpoints caused by re-pulling the full data after supplementary certification, the system reads gap entries from the problem loop list. These gap entries contain event fingerprint values. Gap types and evidence types. Company personnel supplement evidence documents according to the gap entries; these documents are written into the content-addressed evidence database and new evidence citation fingerprints are generated; correction entries are added to the append-only event ledger, and these correction entries cite the original... It points to a new evidence reference fingerprint, and simultaneously writes the reason for correction and the submitter's identification. If the supplementary evidence results in a change to the physical quantity field, the correction entry generates a new event fingerprint value and replaces the old one. Establish a version relationship with the new event fingerprint value; if only supplementary evidence is cited, retain the version. And expand the set of evidence citations.

[0110] In the combined snapshot number Local recalculation is triggered at the bottom: the scope of the local recalculation slice is determined by the contribution mapping in step two. Only the event clusters and derived links affected by the corrected entries are recalculated, and the key result summary of the recalculation output forms the updated summary entry. Subsequently, the system replaces the affected leaf summaries and recalculates the parent node summaries on the affected paths from bottom to top along the parent-child index table to obtain a new commitment root. The new commitment root is encapsulated into a new commitment layer evidence package, and the header of the evidence package is written with a reference to the preceding commitment root to form a version chain. For the sampled disclosure layer evidence package, the system regenerates the membership proof and recalculation recipe for the affected disclosure entries and writes a reference to the preceding disclosure entry into the disclosure entry, so that the verifier can trace back to the original disclosure entry.

[0111] The process involves correcting additions and maintaining consistency: supplementary evidence documents are entered into the content-addressable evidence base, and corrected entries are added to the append-only event ledger and updated. Version relationships, cross-source consistency score Recalculate and write back based on the corrected event clusters; When performing version updates: partial recalculation is performed in the same way. In the next step, the commitment root is updated using the path recalculation method. The new commitment layer evidence package and the sampling disclosure layer evidence package are written into the reference of the previous commitment root and the reference of the previous disclosure entry. The path recalculation uses a backtracking algorithm to recalculate and schedule the work queue that can be reused in step two.

[0112] When used, the issue closed-loop checklist maps the verification feedback to event fingerprint values. Corrected entries are fixed in the append-only ledger and traceable to the fingerprint of evidence references; partial recalculation and commitment root path recalculation limit the scope of updates and maintain... The verification party can review the updated version under the same criteria; the version chain of the commitment layer evidence package and disclosure items retains historical disclosures, and new disclosures can be clearly identified as subsequent versions and the source of differences can be traced.

[0113] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0114] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0115] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0116] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0117] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A carbon footprint end-to-end data collection and reliable accounting system based on artificial intelligence, characterized by: include, The process involves collecting, cleaning, and normalizing raw records to a single time window, associating them with business affiliations, generating calculable events, calculating event fingerprint values, and writing them into an append-only event ledger; writing evidence files into a content-addressed evidence library to form evidence citations; establishing constraint sets, calculating cross-source consistency scores, and writing back calculable events. Generate a combined snapshot number, calculate a risk score based on cross-source consistency score, data quality and emission contribution under constraints, generate a verification sampling plan, a minimum evidence list template and a problem closure list, and aggregate key results summary based on the combined snapshot number; Based on calculable events, evidence citations, combined snapshot numbers, and key result summaries, a commitment structure is constructed to generate a full commitment root and form a commitment layer evidence package; for events that hit the verification sampling plan, a sampling disclosure layer evidence package is generated, which includes member proof, minimal evidence object or desensitized fragment, and recalculated formula.

2. The carbon footprint end-to-end data collection and reliable accounting system according to claim 1, characterized in that: The original records are normalized according to the field sequence of the event mode version. The unit mapping, occurrence time window opening and closing intervals are performed, and the numerical precision is fixed and null value placeholders are handled to form a normalized event string. The event fingerprint value is generated by concatenating the normalized event string with the domain separation label and then performing hash mapping.

3. The carbon footprint end-to-end data collection and reliable accounting system according to claim 2, characterized in that: Before being written into the content-addressable evidence database, the evidence files undergo byte normalization and archiving encoding processing according to the evidence type, generating evidence fingerprints and using the evidence fingerprints as object keys; the append-only event ledger records the ordered sequence of evidence fingerprints in the entries corresponding to the event fingerprint values ​​as evidence references.

4. The carbon footprint end-to-end data collection and reliable accounting system according to claim 3, characterized in that: The constraint set consists of constraints related to energy consumption and operating conditions, output and cycle time, billing cycle and measurement, logistics mileage and load, computing power usage and energy consumption model. The first step is to calculate the normalized residuals of the constraint terms and generate residual summaries. The residual summaries and cross-source consistency scores are then written back to the accountable events.

5. The carbon footprint end-to-end data collection and reliable accounting system according to claim 4, characterized in that: The combined snapshot number is formed by assembling rule entries, factor entries, algorithm entries, and parameter entries. Each entry includes a version number and scope of application. The combined snapshot number is written into the accounting task record, and a unique mapping table between the combined snapshot number and the entry is established for recalculation retrieval.

6. The carbon footprint end-to-end data collection and reliable accounting system according to claim 5, characterized in that: The risk score is obtained by fusing the data quality score, anomaly score, contribution score, change score, and cross-source consistency score; the data quality score is generated by mapping data quality labels to evidence citation completeness, and the contribution score is generated by normalizing the event emission estimates under the combined snapshot number.

7. The carbon footprint end-to-end data collection and trusted accounting system according to claim 6, characterized in that: The verification sampling plan includes a mandatory verification set and a sample verification set. Calculated events whose risk scores meet the mandatory verification threshold and cross-source consistency scores meet the consistency threshold are included in the mandatory verification set. The remaining calculated events are then stratified by activity type and their contribution scores are combined to generate the sample verification set.

8. The carbon footprint end-to-end data collection and reliable accounting system according to claim 7, characterized in that: The minimum evidence list template associates required evidence types with alternative evidence types according to activity type, and verifies the completeness of evidence citations by the evidence type corresponding to the evidence fingerprint in the evidence citation; the problem closed-loop list generates gap identification items, supplementary evidence request items and recalculation verification items by indexing the event fingerprint value.

9. The carbon footprint end-to-end data collection and reliable accounting system according to claim 8, characterized in that: The commitment structure includes a commitment tree. The leaf summary of the commitment tree is calculated by combining the snapshot number, the event fingerprint value, and the ordered sequence of evidence fingerprints in the evidence citation. The commitment root is calculated from bottom to top according to the leaf summary and a fixed splicing direction rule. The commitment root and the splicing direction rule version are written into the commitment layer evidence package.

10. The carbon footprint end-to-end data collection and trusted accounting system according to claim 9, characterized in that: The sampling disclosure layer evidence package includes the corresponding member proof, the desensitized fragment of the minimum evidence object, and the recalculated formula for each hit event, and associates the event fingerprint value with the problem closed-loop list. When a request for supplementary evidence is generated from the issue closure list, a correction entry is added to the supplementary event ledger and the evidence references are updated. The commitment layer evidence package and the sampling disclosure layer evidence package that establish a reference relationship with the previous version are regenerated.

Images