A privacy protection-based environmental factor deduction carbon emission transaction method and system

Abstract

The present application relates to a kind of environmental factor deduction carbon emission trading methods based on privacy protection, belong to carbon emission data management technical field.It includes: based on the observation data of each accounting subject calculation and determine the effective result of day, threshold signature and update global model;Calculate the hash of voucher anchor point of day certificate;Single deduction amount, total deduction amount and final emission are obtained using global model calculation, and generate remote report and corresponding specification hash, output three binding hashes;Build the zero-knowledge proof circuit that can be aggregated, jointly verify deduction equation, membership, uniqueness constraint and binding consistency in a proof;In single transaction, verify threshold signature, the specification hash corresponding to remote report, three binding hashes and zero-knowledge proof complete transaction.The present application method realizes the full process automation of environmental factor deduction carbon emission trading based on privacy protection, can be trusted and can be traced, realizes end-to-end verifiable, low-cost high-throughput carbon emission settlement.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of carbon emission data management, and particularly relates to an environmental factor deduction carbon emission transaction method and system based on privacy protection. BACKGROUND

[0002] With the promotion of global climate change policy, carbon emission trading as a market-based emission reduction mechanism has become an important means for enterprises to control greenhouse gas emissions. The typical approach is to complete emission accounting, environmental factor deduction and transaction settlement based on activity data (such as electricity consumption, fuel consumption), factors / baselines (emission factors / carbon intensity of the region or period), and registration system credentials (certificates / quantities and their status). However, in the engineering landing and large-scale collaborative scenarios, the existing technology generally faces many technical problems:

[0003] 1. Due to the significant multi-source heterogeneous characteristics of carbon emission related data, the power grid meter reading, enterprise self-testing and third party authentication data have great differences in time granularity, measurement unit, collection caliber and missing mode, and are often accompanied by abnormalities and mutations caused by holidays, maintenance, peak load, sensor drift, etc. The existing methods mostly use simple average or manual screening in the processing of multi-source observation, which is difficult to effectively suppress the influence of extreme values and noise, leading to baseline fluctuation and accounting deviation. At the same time, there is a lack of unified and verifiable "daily observation fingerprints" at the data level, and once a dispute occurs, the cost of subsequent verification and tracing is high;

[0004] 2. The lack of privacy protection and trusted computing capability restricts cross-subject data collaboration: activity data, capacity indicators and operation parameters are highly sensitive, and traditional centralized aggregation and unified modeling have leakage risks and compliance pressures. Even if some privacy technologies are used, there are still problems of insufficient precision, robustness and external verifiability, which makes it difficult to meet the requirements of "provable correctness" of the transaction counterpart and the audit party. In addition, the robustness of joint modeling in the face of abnormalities or malicious updates also weakens the public credibility of subsequent settlement;

[0005] 3. The environmental factors and regional carbon intensity have obvious time and space dynamic characteristics, and actual settlement needs timely, compliant and verifiable deduction calculation of dynamic factors. Traditional systems usually rely on static tables or manual review, which is difficult to reflect structural changes and generate trusted dynamic emission reduction results in a timely manner. For certificate factors, it also involves complex constraints such as time window, regional / gird synchronous area matching, proportion upper limit and one-time cancellation. The existing implementation mostly stays at the business rule level, lacking machine verifiable and automated execution technology expression. At the same time, different jurisdictions / scenarios (such as compliant emission trading and voluntary / ESG accounting) have differences in whether to include certain factors, and the existing system lacks a switchable strategy mechanism, making it difficult to reuse the same technology stack across scenarios;

[0006] 4. On-chain verification and settlement mechanisms are still imperfect. Many platforms regard "on-chain evidence storage" as a reliable guarantee, but they lack integrated and one-time technical proof of the correctness of deduction equations, membership relationships (source and validity of vouchers), uniqueness (preventing duplication), and abnormal data processing. Verification of multiple transactions or transactions over multiple days brings high on-chain costs and throughput bottlenecks. Voucher status management relies on business system marking and lacks on-chain one-way state machines and anti-replay mechanisms. Signature endorsements are mostly single-point signatures, which make it difficult to reflect the impact of data governance on settlement weights. At the same time, the calculation inputs, model versions, and output results lack synchronous binding and verifiable reports, which makes it difficult for post-event replay and independent auditing.

[0007] In summary, existing technologies suffer from numerous problems, including accounting bias, difficulty in tracing, difficulty in balancing privacy protection and verification auditing, difficulty in achieving compliant dynamic deductions, and imperfect on-chain verification and settlement mechanisms. Summary of the Invention

[0008] Based on the above analysis, this invention aims to disclose a privacy-preserving method and system for carbon emission trading by deducting environmental factors. Under the premise of protecting corporate privacy and compliance, it introduces robust anomaly removal and robust aggregation for multi-source data, combines scalable joint modeling and trusted execution computation, and completes one-time on-chain verification and atomic settlement with a verifiable proof mechanism and smart contracts, thereby systematically improving the accuracy, credibility and automation level of carbon emission trading.

[0009] On the one hand, this invention provides a privacy-preserving method for trading carbon emissions by deducting environmental factors, specifically including the following steps:

[0010] The global model is updated based on the observation data of each accounting entity, and the valid results and threshold signatures for the day are calculated and determined. The valid results for the day include the regional baseline carbon intensity and the hash of the observation anchor point for the day.

[0011] Calculate the anchor hash of the voucher for the day;

[0012] The global model is used to calculate the single-certificate deduction, the total certificate deduction and final emissions of each accounting entity based on the regional baseline carbon intensity, energy structure of each accounting entity, scenario vector and certificate snapshot, and generates remote reports and corresponding canonical hashes for each accounting entity, and outputs the corresponding three-item binding hashes.

[0013] Based on the regional baseline carbon intensity of the day, the anchor hash of the voucher of the day, the remote report, and the three binding hashes, an aggregateable zero-knowledge proof circuit is constructed to jointly verify the deduction equation, membership, uniqueness constraint and binding consistency within a single proof;

[0014] The threshold signature, the canonical hash corresponding to the remote report, the three binding hashes and the zero-knowledge proof are verified in sequence within a single transaction, and the single transaction is completed after the verification is passed.

[0015] Further, the accounting subjects include power grids and enterprises; the global model includes a server and clients corresponding to the accounting subjects; and updating the global model based on observation data of the accounting subjects comprises:

[0016] The client corresponding to each accounting subject locally updates parameters based on a feature sequence of a specified date;

[0017] Each client uploads the parameters processed by the mask to the server;

[0018] The server side performs robust federated aggregation on the updated parameters of each client by taking the median value of each dimension to obtain the global model.

[0019] Further, the calculation and determination of the daily effective result and the threshold signature based on the observation data of the accounting subjects comprise:

[0020] Each accounting subject generates an aggregated index value based on daily observation data;

[0021] The daily regional baseline carbon intensity and the trusted observation digest are calculated based on the aggregated index values of the accounting subjects;

[0022] The daily observation anchor hash is generated based on the daily regional baseline carbon intensity and the trusted observation digest;

[0023] Based on the cumulative weighted consensus weight of each accounting subject and the set threshold, in the Byzantine fault tolerance consensus, if the cumulative weighted consensus weight of each accounting subject is higher than the set threshold, the daily regional baseline carbon intensity and the daily observation anchor hash are confirmed as the daily effective result, and the threshold signature is obtained based on the daily observation anchor hash; otherwise, it is determined as invalid, and the rollback and retry process is automatically triggered.

[0024] Further, the cumulative weighted consensus weight of each accounting subject is calculated by the following steps:

[0025] Using the global model, the subject abnormality rate of each accounting subject is estimated based on the local reconstruction error and the update bias of each accounting subject;

[0026] The reputation corresponding to each accounting subject is calculated based on the multi-day observation data and the subject abnormality rate of each accounting subject;

[0027] The cumulative weighted consensus weight of each accounting subject is calculated based on the reputation and the staking ratio of each accounting subject.

[0028] Further, the calculating the single certificate deduction amount, the total certificate deduction amount of each accounting subject and the final emission amount based on the regional baseline carbon intensity of the day, the energy structure of each accounting subject, the context vector and the certificate snapshot using the global model comprises:

[0029] The dynamic emission reduction coefficient is obtained based on the regional baseline carbon intensity of the day, the energy structure of each accounting subject and the context vector using the global model;

[0030] The original emission amount of each accounting subject is calculated based on the regional baseline carbon intensity of the day, the energy structure of each accounting subject and the context vector;

[0031] The single certificate deduction amount and the total certificate deduction amount of each accounting subject are calculated based on the dynamic emission reduction coefficient and the certificate snapshot;

[0032] The final emission amount is determined based on the original emission amount of each accounting subject and the total certificate deduction amount.

[0033] Further, the three binding hashes are determined based on the following formula:

[0034] ;

[0035] ;

[0036] ;

[0037] wherein, is an anti-collision hash; , , are input hash, output hash and model hash respectively; is an observation anchor hash; is a certificate anchor hash; is additional context; is the original emission amount of each accounting subject; is the total certificate deduction amount of each accounting subject; is the final emission amount of each accounting subject; , , , are model identifier, model version, parameters of the global model of the dynamic emission reduction coefficient and the date of federal aggregation respectively; is a function for processing input data format.

[0038] Further, the zero-knowledge proof circuit that can be aggregated is constructed based on the regional baseline carbon intensity of the day, the certificate anchor hash of the day, the remote report and the three binding hashes, and the deduction equation, the membership relationship, the uniqueness constraint and the binding consistency are jointly verified in one proof, comprising:

[0039] Build an aggregatable zero-knowledge proof circuit with the following constraints in a proof:

[0040] , and , , ;

[0041] The certificate used for membership belongs to the snapshot of the day;

[0042] The identity corresponding to each certificate does not appear in the used set, satisfying the uniqueness constraint;

[0043] The binding consistency satisfies the consistency of the remote report, the three binding hashes, and the registration value.

[0044] Further, in a single transaction, the threshold signature, the corresponding standard hash of the remote report, the three binding hashes, and the zero-knowledge proof are verified in sequence, including:

[0045] In a single transaction, the threshold signature, the corresponding standard hash of the remote report, and the three binding hashes are verified in sequence. If the verification is passed, the dynamic certificate corresponding to the single certificate deduction is updated from unused to verified, and if the verification fails, the entire transaction is rolled back;

[0046] Based on the strategy mapping, output the policy parameter set for settlement and verification;

[0047] When it is determined based on the strategy mapping that a certain factor cannot be included, the single certificate deduction amount corresponding to the factor is recorded as 0.

[0048] Further, the calculation of the daily regional baseline carbon intensity and the trusted observation summary based on the daily aggregated indicator values of each accounting subject includes:

[0049] Each accounting subject determines the robust scale of each indicator observation data based on the daily indicator observation data;

[0050] Each accounting subject performs anomaly processing on each indicator observation data based on the robust scale of each indicator observation data;

[0051] Each accounting subject aggregates each indicator observation data after anomaly processing to obtain a daily robust aggregation value of each indicator;

[0052] Based on the robust aggregation values of each indicator of each accounting subject, the daily regional baseline carbon intensity and the trusted observation summary are obtained.

[0053] On the other hand, the present application also provides an environmental factor deduction carbon emission trading system based on privacy protection, comprising:

[0054] a data governance and federal consensus module, configured to calculate and determine a daily valid result, a threshold signature and update a global model based on observation data of each accounting subject; wherein the daily valid result comprises a daily regional baseline carbon intensity and a daily observation anchor hash;

[0055] a certificate snapshot and verification module, configured to calculate a daily certificate anchor hash;

[0056] a trusted execution calculation module, configured to calculate a single deduction amount, a total deduction amount of certificates of each accounting subject and a final emission amount of each accounting subject based on the daily regional baseline carbon intensity, an energy structure of each accounting subject, a context vector and a certificate snapshot by using the global model, and generate a remote report of each accounting subject and a corresponding standard hash, and output a corresponding three-item binding hash;

[0057] a proof generation module, configured to construct an aggregatable zero-knowledge proof circuit based on the daily regional baseline carbon intensity, the daily certificate anchor hash, the remote report and the three-item binding hash, and jointly verify a deduction equation, membership, uniqueness constraint and binding consistency in one proof;

[0058] a settlement and evidence storage module, configured to sequentially verify a threshold signature, a standard hash corresponding to the remote report, the three-item binding hash and a zero-knowledge proof in a single transaction, and complete the single transaction if the verification is passed.

[0059] The present application can at least achieve one of the following beneficial effects:

[0060] By introducing robust anomaly removal and robust aggregation for multi-source data under the premise of protecting enterprise privacy and compliance, combining scalable joint modeling and trusted execution calculation, and completing one-time on-chain verification and atomic settlement through a verifiable proof mechanism and a smart contract, the accuracy, credibility and automation level of carbon emission trading are systematically improved.

[0061] By daily scale alignment, Hampel anomaly cleaning and geometric median (Weiszfeld) robust aggregation, extreme values and multi-source deviations are robustly suppressed; an observation anchor hash is calculated for a key field of a trusted observation digest and a daily regional baseline carbon intensity , and the daily valid result is confirmed under a weighted Byzantine fault tolerance consensus, thereby significantly improving the accuracy and credibility of regional baseline carbon intensity and emission accounting.

[0062] Through the local training of each accounting subject and the submission of the parameter update after the mask, the server performs robust federated aggregation in a trusted execution environment (TEE) to obtain the same global model, which on one hand produces an abnormal rate to govern data quality and affect consensus weight, and on the other hand participates in deduction calculation by outputting a dynamic emission reduction coefficient γ, so that the plaintext is not out of the domain, the same model is isomorphic, and the privacy-friendly dynamic deduction is consistent.

[0063] The strategy mapping P (jurisdiction, scene, date) is used to programmably control whether to count a specific environmental factor / certificate and the upper limit thereof; in the ETS scene (compliant emission trading), the range two or specific certificate can be automatically set to zero, and in the ESG scene (voluntary environment, society and governance), the certificate is counted according to rules, so that the same certificate pipeline remains consistent and verifiable under different policy standards.

[0064] By adopting the three-way binding of the observation anchor point hash and the certificate anchor point hash dual-domain anchor point and the TEE, a traceable evidence chain is formed, strong binding of input-model-output and cross-domain anti-replacement are realized, and the third party can independently review and trace without obtaining the plaintext.

[0065] Through single certificate aggregation zero-knowledge proof and on-chain verification, the cost is reduced and the throughput is improved. In the zero-knowledge proof, the deduction equation, members / matching, uniqueness (non-reuse), and TEE binding consistency are jointly verified, and 、 、 、 、 、 are used as public inputs; after the aggregation of multiple proofs and multiple days of proof, the on-chain realizes one-time verification of a single transaction, and compared with the verification of each transaction, the on-chain verification complexity can be converged from O (n) to approximately O (1), which significantly reduces the cost and improves the throughput.

[0066] By verifying the threshold signature, the specification hash corresponding to the remote report, the three binding hashes and the zero-knowledge proof in sequence, the mechanism can prevent the use of one certificate multiple times / replay, and if any step fails, the whole process is rolled back, so that the state consistency and fund safety are guaranteed.

[0067] In the present application, the above technical solutions can be combined with each other to realize more preferred combination schemes. Other features and advantages of the present application will be described in the subsequent specification, and some advantages will become apparent from the specification or by implementing the present application. The purpose and other advantages of the present application can be achieved and obtained by the contents specifically pointed out in the specification and the drawings. BRIEF DESCRIPTION OF DRAWINGS

[0068] The accompanying drawings are included to provide a further understanding of the embodiments and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings:

[0069] Figure 1 A flow chart of the method of the present application. DETAILED DESCRIPTION

[0070] Preferred embodiments of the present application will be described in detail below with reference to the drawings, in which the same reference numerals will be used throughout the different drawings, and wherein:

[0071] Embodiments of the method

[0072] One embodiment of the present application discloses a privacy protection-based environmental factor deduction carbon emission trading method, which specifically comprises steps S1-S5.

[0073] Step S1, updating a global model based on observation data of each accounting subject, and calculating and determining a daily effective result and a threshold signature; wherein the daily effective result comprises a daily regional baseline carbon intensity and a daily observation anchor point hash.

[0074] The accounting subjects include power grids and enterprises in the region that carry out carbon trading, i.e., units with independent metering / settlement accounts.

[0075] The observation data includes power consumption, fuel consumption, etc.

[0076] The global model is used to process and analyze data from each accounting subject, including extracting data features, identifying data anomalies, aggregating data, and predicting dynamic emission reduction coefficients. The global model updates the model and parameters through federated learning, while ensuring the privacy and security of the data. The global model includes a server and a client corresponding to each accounting subject.

[0077] Specifically, updating the global model based on the observation data of each accounting subject comprises S1-11-S1-13.

[0078] S1-11, the client corresponding to each accounting subject updates parameters based on the feature sequence of the specified date locally; specifically, the client corresponding to each accounting subject updates parameters based on the feature sequence of the specified date locally using an LSTM autoencoder to reconstruct and learn the feature sequence after cleaning for a plurality of days, preferably for 60 days.

[0079] S1-12, the client of each accounting subject uploads the masked parameters to the server to ensure that the local plaintext privacy is not leaked; wherein the masking process is used to hide sensitive information during data transmission or storage to prevent unauthorized access and leakage.

[0080] S1-13, the server side performs robust federated aggregation on the updated parameters of each client corresponding to each accounting subject by taking the median of each dimension to obtain a global model LSTM-AE.

[0081] Further, the daily effective result and the threshold signature are calculated and determined based on the observation data of each accounting subject, including S1-01-S1-04.

[0082] S1-01, each accounting subject generates daily aggregation index values based on daily observation data.

[0083] Specifically, the observation data of the power grid and each enterprise (such as power consumption, fuel consumption, etc.) and the authority published factors / baseline need to be first preprocessed, including abnormality rejection, time and unit alignment, and missing value processing.

[0084] The authority published factors / baseline include carbon emission factors, source consumption baseline, industry benchmark, environmental quality standard, and other standard, parameter or baseline data published by government agencies, international organizations, industry associations or other recognized authority.

[0085] Further, Hampel abnormality rejection is performed on the daily observation data, including:

[0086] A 7-day window is constructed with a natural day d as the center For the observation data of each index point, let

[0087] ;

[0088] Wherein, is the robust scale of the current index; is the median absolute deviation of the current index; is the Gaussian consistency coefficient, and the preferred value is 1.4826; is the window median.

[0089] The threshold coefficient is preferably 3, that is, when is judged as abnormal, and is replaced.

[0090] Time and unit alignment includes ensuring that all observation data are consistent in time and unified in unit.

[0091] For observation data missing processing, exemplary interpolation method, mean filling, using model to predict missing value or deleting records containing missing value can be used

[0092] After completing the preprocessing step, the daily observation data of each index is geometrically robustly aggregated to generate daily aggregation index values of each index, and Weiszfeld iteration is used: ​

[0093] ;

[0094] wherein t represents the iteration number; represents the i-th data point of the current calculated indicator; represents the estimated value of the geometric median calculated in the t+1 iteration; is a robust norm, is a small positive number to prevent the denominator from being zero; until converges, the daily aggregated indicator value of the current calculated indicator is obtained .

[0095] S1-02, based on the daily aggregated indicator value of each accounting subject, a daily regional baseline carbon intensity and a credible observation summary are calculated.

[0096] Specifically, the daily aggregated indicator values of all indicators of each accounting subject in the summary region are aggregated to obtain a regional robust aggregation value, and exemplary methods for aggregation include weighted average method or multivariate robust regression method.

[0097] Further, based on the regional robust aggregation value and the adjustment coefficient, the daily regional baseline carbon intensity is calculated.

[0098] Further, based on the daily regional baseline carbon intensity and the number of accounting subjects, the daily aggregated indicator value of each indicator, the aggregation method, the time range, the abnormal value processing rule, etc., a credible observation summary is generated.

[0099] S1-03, based on the daily regional baseline carbon intensity and the credible observation summary, a daily observation anchor point hash is generated.

[0100] Specifically, based on the key fields of the credible observation summary and the daily regional baseline carbon intensity, the fields are encoded in a predetermined order and taken anti-collision hash to generate an observation anchor point hash .

[0101] S1-04, based on the cumulative weighted consensus weight of each accounting subject and the set threshold, in the Byzantine fault tolerance consensus, if the cumulative weighted consensus weight of each accounting subject is higher than the set threshold, the daily regional baseline carbon intensity and the daily observation anchor point hash are confirmed as the daily effective result, and a threshold signature is obtained based on the daily observation anchor point hash determined as the daily effective result; otherwise, it is determined as invalid, and the rollback and retry process is automatically triggered.

[0102] Further, in S1-04, the cumulative weighted consensus weight of each accounting subject is calculated by the following steps:

[0103] The global model is used to estimate the subject abnormality rate of each accounting subject based on the local reconstruction error and update deviation of each accounting subject, and the subject abnormality rate of each accounting subject is represented as where d represents the day, and j is the serial number of the accounting subject;

[0104] Based on the multi-day observation data and the subject abnormality rate of each accounting subject, the corresponding credit of each accounting subject is calculated, ; wherein, is the observation value of subject j on natural day t; and is the central value and the robust scale of the day; is the update coefficient;

[0105] Based on the credit and the pledge ratio of each accounting subject, the weighted consensus weight of each accounting subject is calculated:

[0106] ;

[0107] wherein, is the weighted consensus weight of accounting subject j; is the mixing coefficient, which is a configurable parameter with a value between 0 and 1, used to allocate weight between credit and pledge, the greater the alpha, the more emphasis on credit, the smaller the more emphasis on pledge, the specific value is adjusted according to the jurisdiction, scene and risk preference; The denominator represents the sum of the weighted consensus weights of all accounting subjects.

[0108] Further, based on the weighted consensus weight of each accounting subject, the cumulative weighted consensus weight is obtained, and under the Byzantine fault tolerance consensus, when the cumulative weight of the set is greater than the set threshold (the preferred value is 2 / 3), the daily regional baseline carbon intensity and the daily observation anchor hash is the daily effective result, and the message containing the daily observation anchor hash is threshold signed by the committee node ; Otherwise, it is determined that the daily result is invalid, no threshold signature is generated, no proof and on-chain settlement link is entered, the on-chain state remains unchanged, and the rollback and retry process is automatically triggered.

[0109] In this embodiment, the daily scale alignment-Hampel anomaly cleaning-geometric median (Weiszfeld) robust aggregation is used to robustly suppress extreme values and multi-source deviations; the observation anchor hash and confirm the daily effective result under the weighted Byzantine fault-tolerant consensus, thereby significantly improving the accuracy and reliability of regional baseline carbon intensity and emission accounting. Through local training of the sub-encoder by each accounting subject and submission of the parameter update after masking, the server performs robust federated aggregation in a trusted execution environment (TEE) to obtain the same global model, which on one hand produces an abnormal rate to govern data quality and affect consensus weight, and on the other hand outputs a dynamic emission reduction coefficient γ to participate in deduction calculation, thereby realizing privacy-friendly dynamic deduction in plaintext without domain, same model isomorphism, and consistent caliber.

[0110] Step S2, calculating a daily credential anchor point hash.

[0111] Specifically, a subset of environmental attribute credentials (including credential snapshots) used for settlement is obtained from a third-party registration system, the fields include type, quota, ownership, validity period / region and current state, consistency check is performed on the subset and the key field is encoded and hashed to obtain the credential anchor point hash . At the same time, a strategy mapping P (jurisdiction, scene, date) is introduced as a compliance switch: when it is determined that a certain factor cannot be counted in the target scene, the corresponding deduction amount is automatically set to zero, but the related anchor point is still retained to support auditing and proof.

[0112] In this embodiment, the strategy mapping P (jurisdiction, scene, date) is used to programmably control whether to count a specific environmental factor / credential and its upper limit; for subsequent automatic zeroing of scope two or specific credentials in the ETS scene (compliant emission trading), and counting according to rules in the ESG scene (voluntary environment, society and governance), the same evidence pipeline is kept consistent and verifiable under different policy caliber.

[0113] Step S3, using the global model to calculate the single certificate deduction amount, the total credential deduction amount of each accounting subject and the final emission amount based on the daily regional baseline carbon intensity, the energy structure of each accounting subject, the context vector and the credential snapshot, and generating the remote report of each accounting subject and the corresponding standard hash, and outputting the corresponding three binding hashes. Specifically, steps S31-S36 are included.

[0114] S31, using the global model to obtain a dynamic emission reduction coefficient based on the daily regional baseline carbon intensity, the energy structure of each accounting subject and the context vector.

[0115] Specifically, the daily regional baseline carbon intensity, the energy structure of each accounting subject, the context vector and the credential snapshot are taken as the global model input, and the output dynamic emission reduction coefficient is obtained , wherein is the daily global model encoder latent representation, is Sigmoid, and respectively, are weight and bias.

[0116] S32, the original emission amount is calculated based on the regional baseline carbon intensity of the day, the energy structure of each accounting subject, and the context vector, wherein the context vector represents a vector containing all context information related to carbon emission trading, including, for example, power consumption curve, production shift, proportion of self-provided renewable energy, etc., and the regional baseline carbon intensity of the day includes carbon intensity and corresponding time period label, etc., and the calculation formula is:

[0117] ;

[0118] is the original emission amount of each accounting subject; is the kth activity amount, which is obtained based on the energy structure and the context vector; K represents a set of activity categories; is the kth activity amount corresponding factor or baseline, wherein the baseline of the power category is the regional baseline carbon intensity of the day, and the baseline of the direct fuel / process category preferably adopts a national or industry standard factor; is an indicator function, which is used to determine whether a certain activity should be included in the calculation of the total emission amount according to a specific policy or rule, when , it means that the activity meets the policy requirements and can be included in the emission amount, and when , it is not allowed to be included.

[0119] S33, the certificate deduction amount and the total certificate deduction amount of each accounting subject are calculated based on the dynamic emission reduction coefficient and the certificate snapshot. The specific representation is as follows:

[0120] ;

[0121] .

[0122] wherein, is the total certificate deduction amount of each accounting subject; is the set of certificates of the day; is the dynamic emission reduction coefficient; is the equivalent emission reduction amount of certificate j in day d, which is the certificate deduction amount if the policy allows , and if the policy does not allow , the emission reduction amount of the certificate is not included in the deduction; is the regional baseline carbon intensity of the day; is the energy quota of certificate j available for cancellation in day d; are respectively the time window and the regional mapping determination function; means that j appears in the nullifier set only if.

[0123] S34, determine the final emission amount of each accounting subject based on the original emission amount and the total deduction amount of the certificate.

[0124] Specifically, the final emission amount of each accounting subject is determined based on the original emission amount and the total deduction amount of the certificate. The non-negative constraint needs to be met:

[0125] .

[0126] S35, the server generates a remote report in the trusted environment TEE and the corresponding specification hash .

[0127] S36, the server outputs three binding hashes to ensure verifiability:

[0128] ;

[0129] ;

[0130] ;

[0131] wherein, is an anti-collision hash; , , are input hash, output hash, and model hash, respectively; is an observation anchor hash; is a certificate anchor hash; is additional context; is the original emission amount of each accounting subject; is the total deduction amount of the certificate of each accounting subject; is the final emission amount of each accounting subject; , , , are model identifier, model version, parameters of the global model of dynamic emission reduction coefficient, and date of federal aggregation, respectively; is a function for processing input data format.

[0132] In this embodiment, by adopting the dual-domain anchor points of observation anchor hash and certificate anchor hash and the three-way binding of TEE, a traceable evidence chain is formed, strong binding of input-model-output and cross-domain anti-replacement are achieved, and independent review and traceability of the third party are facilitated without obtaining the plaintext.

[0133] Step S4, based on the regional baseline carbon intensity of the day, the certificate anchor hash of the day, the remote report, and the three binding hashes, build an aggregable zero-knowledge proof circuit to jointly verify the deduction equation, membership, uniqueness constraint, and binding consistency in one proof.

[0134] Specifically, the aggregatable zero-knowledge proof circuit is constructed with the daily regional baseline carbon intensity, the daily certificate anchor hash, the remote report, and the three binding hashes as public inputs, and the following constraints are jointly verified in one proof:

[0135] , and , , ;

[0136] The certificate used for membership belongs to the daily snapshot (the certificate anchor hash constraint), and satisfies , ;

[0137] The identity corresponding to each certificate does not appear in the used set, satisfying the uniqueness constraint;

[0138] The binding consistency satisfies The three binding hashes , , are consistent with the registration value, indicating that the data has not been tampered with since registration.

[0139] Further, in the present application, in order to balance the efficiency and verifiability of batch settlement, the zero-knowledge proof of the subject s on the single record of the certificate j in the natural day d is recorded as The set of multiple records in the natural day d is , wherein i represents a certificate record, and then when the set B is jointly settled, the aggregated proof can be constructed, which is semantically equivalent to "each in the set has passed the verification according to the above constraints, and the batch consistency constraint of the set is established".

[0140] In this embodiment, through single certificate aggregated zero-knowledge proof and on-chain verification, the cost is reduced and the throughput is improved. Among them, the deduction equation, membership / matching, uniqueness (non-repeated use), TEE binding consistency are jointly verified in one aggregatable zero-knowledge proof, and , , , , , , etc. are used as public inputs; after the aggregation of multiple and multi-day proofs, on-chain one-time verification of single transaction is realized, which can converge the on-chain verification complexity from O(n) to approximately O(1) compared with the verification of each transaction, significantly reducing the cost and improving the throughput.

[0141] Step S5, sequentially verifying the threshold signature, the standard hash corresponding to the remote report, the three binding hashes and the zero-knowledge proof in a single transaction, and completing the single transaction if the verification is passed.

[0142] The threshold signature, the standard hash corresponding to the remote report and the three binding hashes are sequentially verified in a single transaction, and if the verification is passed, the dynamic voucher corresponding to the single certificate deduction is updated from unused to already cancelled, and if the verification fails, the whole transaction is rolled back.

[0143] Based on the strategy mapping, the policy parameter set for settlement and verification is outputted.

[0144] When it is determined based on the strategy mapping that a certain factor cannot be counted, the single certificate deduction amount corresponding to the factor is recorded as 0.

[0145] In this embodiment, by sequentially verifying the threshold signature, the standard hash corresponding to the remote report, the three binding hashes and the zero-knowledge proof, the mechanism is prevented from using the same certificate multiple times / replaying, and the state consistency and fund safety are guaranteed.

[0146] System embodiment

[0147] Another specific embodiment of the present application discloses an environmental factor deduction carbon emission transaction system based on privacy protection, which specifically comprises a data governance and federated learning module, a voucher snapshot and verification module, a trusted execution calculation module, a proof generation module and a settlement and evidence storage module.

[0148] Specifically, the data governance and federated consensus module is used to calculate and determine the daily effective result, the threshold signature and update the global model based on the observation data of each accounting subject; wherein the daily effective result includes the daily regional baseline carbon intensity and the daily observation anchor hash.

[0149] The voucher snapshot and verification module is used to calculate the daily voucher anchor hash.

[0150] The trusted execution calculation module is used to calculate the single certificate deduction amount, the total voucher deduction amount and the final emission amount of each accounting subject based on the daily regional baseline carbon intensity, the energy structure of each accounting subject, the context vector and the voucher snapshot by using the global model, and generate the remote report of each accounting subject and the corresponding standard hash, and output the corresponding three binding hashes.

[0151] The proof generation module is based on the daily regional baseline carbon intensity, the daily voucher anchor hash, the remote report, the three binding hashes to construct an aggregatable zero-knowledge proof circuit, and jointly verify the deduction equation, the membership relationship, the uniqueness constraint and the binding consistency in a proof.

[0152] The settlement and storage module is configured to sequentially verify a threshold signature, a standard hash corresponding to the remote report, the three binding hashes and the zero-knowledge proof in a single transaction, and complete the single transaction if the verification is passed.

[0153] The embodiment discloses an environmental factor deduction carbon emission trading system based on privacy protection, which is modularized and split into "data governance-federation and consensus-certificate snapshot-trustworthy calculation-proof generation-chain settlement" modules. The on-chain smart contract automatically completes threshold signature, TEE authentication and zero-knowledge proof verification, and updates dNFT (dynamic carbon asset digital certificate) and state hash. Without manual intervention, the system can support multi-source, high-frequency and cross-regional carbon trading scenarios, and has good scalability and engineering usability.

[0154] It should be noted that the above embodiments are based on the same inventive concept, and the parts not repeated can be mutually referred to.

[0155] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A privacy-preserving environmental factor deduction carbon emission trading method, characterized in that, Comprising the following steps: updating the global model based on observation data of each accounting subject, calculating and determining the daily effective result, threshold signature; wherein the daily effective result includes the daily regional baseline carbon intensity and the daily observation anchor hash; calculating the daily certificate anchor hash; The global model is used to calculate the single certificate deduction amount, the total certificate deduction amount of each accounting subject, and the final emission amount based on the regional baseline carbon intensity on the day, the energy structure of each accounting subject, the context vector, and the certificate snapshot, and to generate the remote report of each accounting subject and the corresponding standard hash, and to output the corresponding three binding hashes; the three binding hashes are determined based on the following formula: ; ; ; wherein, is an anti-collision hash; , , are input hash, output hash, and model hash respectively; is an observation anchor hash; is a certificate anchor hash; is additional context; is the original emission amount of each accounting subject; is the total certificate deduction amount of each accounting subject; is the final emission amount of each accounting subject; , , , are the model identifier, model version, parameters of the global model of the dynamic emission reduction coefficient, and the date of federal aggregation respectively; is a function for processing input data format; Based on the regional baseline carbon intensity of the day, the certificate anchor hash of the day, the remote report, and the three binding hashes, a zero-knowledge proof circuit is constructed, and the deduction equation, membership, uniqueness constraint and binding consistency are jointly verified in a proof, including: constructing a zero-knowledge proof circuit with the regional baseline carbon intensity of the day, the certificate anchor hash of the day, the remote report, and the three binding hashes as common inputs, and jointly verifying the following constraints in a proof: , and , , ; the certificate used for membership belongs to the snapshot of the day; the identity corresponding to each certificate does not appear in the used set, which satisfies the uniqueness constraint; the binding consistency satisfies that the remote report, the three binding hashes and the registration value are consistent; verifying the threshold signature, the standard hash corresponding to the remote report, the three binding hashes and the zero-knowledge proof in turn within a single transaction, and completing the single transaction if the verification is passed.

2. The carbon emissions trading method according to claim 1, wherein, The accounting subjects include power grids and enterprises; the global model includes servers and clients corresponding to each accounting subject; updating the global model based on the observation data of each accounting subject comprises: The client corresponding to each accounting subject updates the parameters locally based on the feature sequence of the specified date; Each client uploads the masked parameters to the server; The server side performs robust federated aggregation on the updated parameters of each client by taking the median value of each dimension to obtain the global model.

3. The carbon emissions trading method of claim 2, wherein, The calculation and determination of the daily effective result and the threshold signature based on the observation data of each accounting subject comprises: Each accounting subject generates daily aggregate index values based on daily observation data; Based on the daily aggregate index values of each accounting subject, the daily regional baseline carbon intensity and the trusted observation digest are calculated; Based on the daily regional baseline carbon intensity and the trusted observation digest, the daily observation anchor hash is generated; Based on the cumulative weighted consensus weight of each accounting subject and the set threshold, under the Byzantine fault tolerance consensus, if the cumulative weighted consensus weight of each accounting subject is higher than the set threshold, the daily regional baseline carbon intensity and the daily observation anchor hash are confirmed as the daily effective result, and the threshold signature is obtained based on the daily observation anchor hash; otherwise, it is determined to be invalid, and the rollback and retry process is automatically triggered.

4. The carbon emissions trading method according to claim 3, wherein, The cumulative weighted consensus weight of each accounting subject is calculated by the following steps: Using the global model, the subject anomaly rate of each accounting subject is estimated based on the local reconstruction error and update bias of each accounting subject; Based on the multi-day observation data and the subject anomaly rate of each accounting subject, the reputation corresponding to each accounting subject is calculated; Based on the reputation and the staking ratio of each accounting subject, the cumulative weighted consensus weight of each accounting subject is calculated.

5. The carbon emissions trading method of claim 2, wherein, The use of the global model to calculate the single certificate deduction amount, the total certificate deduction amount of each accounting subject, and the final emission amount based on the daily regional baseline carbon intensity, the energy structure of each accounting subject, the context vector, and the certificate snapshot comprises: Using the global model, a dynamic emission reduction coefficient is obtained based on the daily regional baseline carbon intensity, the energy structure of each accounting subject, and the context vector; Based on the daily regional baseline carbon intensity, the energy structure of each accounting subject, and the context vector, the original emission amount of each accounting subject is calculated; Based on the dynamic emission reduction coefficient and the certificate snapshot, the single certificate deduction amount and the total certificate deduction amount of each accounting subject are calculated; Based on the original emission amount of each accounting subject and the total certificate deduction amount, the final emission amount is determined.

6. The carbon emissions trading method of claim 5, wherein, Verifying the threshold signature, the standard hash corresponding to the remote report, the three binding hashes and the zero-knowledge proof in turn within a single transaction comprises: The threshold signature, the corresponding standard hash of the remote report, and the three binding hashes are verified in sequence within a single transaction, if the verification is passed, the dynamic certificate corresponding to the single certificate deduction is updated from unused to already cancelled, if the verification fails, the whole transaction is rolled back; The policy parameter set for settlement and verification is output based on the strategy mapping; When it is determined based on the strategy mapping that a certain factor cannot be counted, the single certificate deduction amount corresponding to the factor is recorded as 0.

7. The carbon emissions trading method of claim 3, wherein, The daily regional baseline carbon intensity and the trusted observation summary are calculated based on the daily aggregated index value of each accounting subject, including: Each accounting subject determines the robust scale of each index observation data based on the daily index observation data; Each accounting subject performs anomaly processing on each index observation data based on the robust scale of each index observation data; Each accounting subject aggregates each index observation data after anomaly processing to obtain a daily robust aggregation value of each index; The daily regional baseline carbon intensity and the trusted observation summary are obtained based on the robust aggregation value of each index of each accounting subject.

8. A privacy-preserving environmental factor deduction carbon emission trading system, characterized in that, It includes: A data management and federal consensus module for calculating and determining daily effective results, threshold signatures, and updating global models based on observation data of each accounting subject; wherein the daily effective results include daily regional baseline carbon intensity and daily observation anchor hash; A certificate snapshot and verification module for calculating daily certificate anchor hash; A trusted execution computing module is configured to calculate, based on the regional baseline carbon intensity, the energy structure of each accounting subject, the context vector, and the certificate snapshot of the day, a single certificate deduction amount, a total certificate deduction amount of each accounting subject, and a final emission amount of each accounting subject, generate a remote report of each accounting subject and a corresponding standard hash, and output a corresponding three-item binding hash; the three-item binding hash is determined based on the following formula: ; ; ; wherein, is an anti-collision hash; , , are input hash, output hash, and model hash, respectively; is an observation anchor hash; is a certificate anchor hash; is additional context; is the original emission amount of each accounting subject; is the total certificate deduction amount of each accounting subject; is the final emission amount of each accounting subject; , , , are the model identifier, the model version, the parameters of the global model of the dynamic emission reduction coefficient, and the date of federal aggregation, respectively; is a function for processing input data format; The proof generation module, based on the regional baseline carbon intensity of the day, the certificate anchor point hash of the day, the remote report, the three binding hashes, constructs a zero-knowledge proof circuit that can be aggregated, and jointly verifies the deduction equation, membership, uniqueness constraint and binding consistency in a proof, including: constructing a zero-knowledge proof circuit that can be aggregated with the regional baseline carbon intensity of the day, the certificate anchor point hash of the day, the remote report, and the three binding hashes as common input, and jointly verifying the following constraints in a proof: , and , , ; the certificate used for membership belongs to the snapshot of the day; the identity corresponding to each certificate does not appear in the used set, which satisfies the uniqueness constraint; the binding consistency satisfies that the remote report, the three binding hashes and the registration value are consistent; A settlement and evidence module for verifying threshold signatures, the corresponding standard hash of the remote report, the three binding hashes and zero-knowledge proof in sequence within a single transaction, and completing the single transaction if the verification is passed.

Images