A carbon asset transaction risk prediction method and system
By constructing an order state DAG and using BFT Finality confirmation and Merkle Proof receipts to filter deliverable asset subgraphs, the problem of identifying state differences in carbon asset trading risk prediction in cross-chain transactions is solved, achieving more accurate risk prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHONGYAN GREEN FINANCE RES INST (NANJING) CO LTD
- Filing Date
- 2026-05-14
- Publication Date
- 2026-07-31
AI Technical Summary
Existing methods for predicting carbon asset trading risks fail to effectively identify state differences in different chain environments during cross-chain transactions, leading to discrepancies between risk prediction results and actual delivery status. In particular, in a multi-chain collaborative environment, assets that have not yet met the delivery conditions may be included in the scope of deliverable judgment.
By extracting the State Commitment direction from the cross-chain contract event stream of the target transaction order, an order state DAG is formed. Combined with the BFT Finality confirmation prefix and Merkle Proof receipt, a deliverable asset subgraph is selected to generate the target Temporal Graph state difference, which is then input into a pre-trained risk prediction TGN model for risk prediction.
This improves the reliability of predicting the risk of delivery failure in carbon asset trading, avoids risk judgment based solely on one side of asset records, and ensures that the prediction results are consistent with the actual delivery conditions.
Smart Images
Figure CN122492346A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of green finance technology, specifically to a method and system for predicting risks in carbon asset trading. Background Technology
[0002] As the green finance system continues to advance, carbon assets are gradually shifting from a single registration and management object to digital assets that are transferable, verifiable, and settleable. While improving trading efficiency, carbon asset trading systems also need to reliably manage asset status, transaction performance, and delivery risks, especially in a multi-chain collaborative environment where there may be a mismatch between on-chain records and actual delivery conditions.
[0003] Existing methods for predicting carbon asset trading risks typically focus on the asset records or account balances readable by the target platform, and use this to determine the ability to fulfill transaction obligations. These methods do not adequately consider the order of status submission and the finality of confirmation in different blockchain environments during cross-chain transactions. This can easily lead to assets that have not yet met the full delivery conditions being included in the scope of deliverable assessment, resulting in a discrepancy between the risk prediction results and the actual delivery status.
[0004] Therefore, how to improve the ability to identify cross-chain state differences in carbon asset trading risk prediction in the context of green finance, and avoid making risk judgments based solely on one-sided asset records, has become a technical problem that carbon asset digital trading systems need to solve. Summary of the Invention
[0005] The purpose of this invention is to provide a method and system for predicting carbon asset trading risks, in order to solve the problems mentioned in the background art.
[0006] To achieve the above objectives, the present invention provides the following technical solution:
[0007] In a first aspect, the present invention provides a method for predicting carbon asset trading risks, characterized in that it includes:
[0008] Extract the State Commitment acceptance direction from the cross-chain contract event stream of the target transaction order; project the State Commitment acceptance direction onto the happens-before partial order topology to form the order state DAG;
[0009] The BFT Finality confirmation prefix of the chain environment to which the state node in the order state DAG belongs is used as the truncation boundary; the fulfillment occupancy path is back-embedded into the causal predecessor path of the target chain mapping node to obtain the unfinished state subgraph;
[0010] The source chain reachability of the target chain mapping node is verified based on the Merkle Proof receipt; the target chain mapping node is filtered by combining the node coverage of the unfinished state subgraph to form a deliverable asset subgraph.
[0011] In the DAG of order status, the deliverable asset subgraph and the unfinished state subgraph are compared to generate the target Temporal Graph state difference; the target Temporal Graph state difference is input into the pre-trained risk prediction TGN model to obtain the carbon asset transaction delivery failure risk prediction result.
[0012] Secondly, this invention provides a carbon asset trading risk prediction system, implemented based on the method described above, comprising:
[0013] The topology module is used to extract the State Commitment acceptance direction from the cross-chain contract event stream of the target transaction order; and to project the State Commitment acceptance direction onto the happens-before partial order topology to form the order state DAG.
[0014] The delimiting module is used to use the BFT Finality confirmation prefix of the chain environment to which the state node in the order state DAG belongs as the truncation boundary; and to embed the fulfillment occupancy path into the causal predecessor path of the target chain mapping node in reverse to obtain the unfinished state subgraph.
[0015] The verification module is used to verify the source chain reachability of the target chain mapping nodes based on the Merkle Proof receipt; and to filter the target chain mapping nodes by combining the node coverage of the unfinished state subgraph to form a deliverable asset subgraph.
[0016] The prediction module compares the deliverable asset subgraph and the unfinished state subgraph in the order state DAG to generate the target Temporal Graph state difference; the target Temporal Graph state difference is input into the pre-trained risk prediction TGN model to obtain the carbon asset transaction delivery failure risk prediction result.
[0017] The technical effects and advantages provided by the present invention in the above technical solution are as follows:
[0018] This invention reconstructs the cross-chain contract event flow corresponding to the target transaction order into an order state DAG, incorporating the state continuity relationship between source chain locking, cross-chain verification, and target chain mapping into the risk prediction basis. Furthermore, it extracts the unfinalized state subgraph using BFTFinality confirmation prefixes, preventing mapped assets whose target chain has been generated but whose causal predecessors have not yet been finalized from being prematurely included in the deliverable judgment. Simultaneously, this invention combines Merkle Proof verification of the source chain reachability of the target chain mapping nodes, further filtering out mapped assets lacking source chain proof or still within the pending finalization range, thereby forming a deliverable asset subgraph consistent with the actual delivery conditions. Therefore, this invention can use cross-chain state differences rather than one-sided balance records as risk prediction input, improving the reliability of carbon asset transaction delivery failure risk prediction in green finance scenarios. Attached Figure Description
[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.
[0020] Figure 1 This is a flowchart of a carbon asset trading risk prediction method according to the present invention;
[0021] Figure 2 This is a framework diagram of a carbon asset trading risk prediction system according to the present invention. Detailed Implementation
[0022] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided to make the description of this application more complete and comprehensive, and to fully convey the concept of the exemplary embodiments to those skilled in the art. The drawings are merely illustrative illustrations of this application and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.
[0023] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more exemplary embodiments. Numerous specific details are provided in the following description to give a full understanding of the exemplary embodiments disclosed in this application. However, those skilled in the art will recognize that the technical solutions disclosed in this application can be practiced with one or more specific details omitted, or other methods, components, steps, etc., can be employed. In other instances, well-known structures, methods, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of the disclosure of this application.
[0024] Example 1
[0025] like Figure 1 As shown in the figure, this embodiment discloses a method for predicting carbon asset trading risks, including:
[0026] S101: Extract the State Commitment acceptance direction from the cross-chain contract event stream of the target transaction order; project the State Commitment acceptance direction onto the happens-before partial order topology to form the order state DAG;
[0027] S102: Use the BFT Finality confirmation prefix of the chain environment to which the state node in the order state DAG belongs as the truncation boundary; embed the fulfillment occupancy path in reverse into the causal predecessor path of the target chain mapping node to obtain the unfinished state subgraph.
[0028] S103: Verify the source chain reachability of the target chain mapping node based on the Merkle Proof receipt; filter the target chain mapping node by combining the node coverage of the unfinished state subgraph to form a deliverable asset subgraph;
[0029] S104: Compare the deliverable asset subgraph and the unfinished state subgraph in the order state DAG to generate the target Temporal Graph state difference; input the target Temporal Graph state difference into the pre-trained risk prediction TGN model to obtain the carbon asset transaction delivery failure risk prediction result.
[0030] The target transaction order is the order data for carbon asset delivery to be executed. The target transaction order includes an order identifier, an asset identifier used by the order, the delivery quantity, the order account field, and the fulfillment time field. The asset identifier used by the order points to the target chain-mapped asset to be used by the target transaction order. The target chain-mapped asset is a carbon asset mapping record generated by the target chain based on the cross-chain status submission results.
[0031] The cross-chain contract event stream is a collection of on-chain contract events generated during the cross-chain settlement process of a target transaction order. The cross-chain contract event stream originates from the source chain locking contract event log, the cross-chain verification contract event log, and the target chain mapping contract event log. The source chain locking contract event log records the source chain carbon asset locking status. The cross-chain verification contract event log records the cross-chain proof verification status. The target chain mapping contract event log records the target chain mapped asset generation status. These event logs are aggregated according to the asset identifier occupied by the target transaction order, forming the cross-chain contract event stream corresponding to the target transaction order.
[0032] Contract events in a cross-chain contract event stream include an on-chain position field, a State Commitment field, and a preceding State Commitment field. The on-chain position field indicates the commit position of the contract event within its respective chain environment, denoted as: ;
[0033] in, Indicates the first The on-chain location field of a contract event. Indicates chain identifier, Indicates the block height. Represents the transaction hash. Indicates the event sequence number.
[0034] The State Commitment field represents the state commitment formed when the current contract event is written onto the chain, denoted as... The preceding State Commitment field is used to represent the previous state commitment undertaken by the current contract event, denoted as... When the contract event is the starting event of the cross-chain settlement process corresponding to the target transaction order, the preceding StateCommitment field is recorded as the creation marker. .
[0035] The direction of acceptance of a State Commitment is used to indicate the direction of acceptance between two state commitments. If the... The preceding State Commitment field of a contract event With the The State Commitment field of a contract event If they are consistent, then the first The contract event is the first The preceding event of each contract event. The corresponding State Commitment direction is denoted as: ;
[0036] in, Indicates by point to The direction of State Commitment.
[0037] It should be noted that the State Commitment acceptance direction is based on the reference relationship between the state commitment fields. Through this acceptance direction, the target chain mapped asset can be traced back to the corresponding cross-chain verification state and source chain locking state. This process ensures that the target chain mapped asset is no longer merely a balance record on the target chain participating in subsequent risk prediction, but rather participates in subsequent processing as a state node with a cross-chain state source.
[0038] The happens-before partial order topology is used to represent the happen-before relationship between cross-chain contract events. This partial order topology includes partial order nodes and partial order edges. A partial order node is the target contract event node whose on-chain position field, State Commitment field, and preceding State Commitment field are written to. A partial order edge is a directed edge connecting two partial order nodes. The direction of the partial order edge is determined by the reference relationship in the State Commitment field and the order of the on-chain position fields.
[0039] Partial order nodes are denoted as Partially ordered nodes This corresponds to a target contract event. The partial order edge is denoted as... .when ,and Prior to in the execution order of the chain environment or cross-chain protocol At that time, a partial order node is formed. Pointing to partial order nodes Partial order edges: ;
[0040] The happens-before partial order topology is denoted as: ;
[0041] in, This represents the happens-before partial order topology. Represents a partially ordered set of nodes. This represents the set of partially ordered edges.
[0042] For two contract events that only have an on-chain positional relationship but no State Commitment field reference relationship, no partial order edge is generated. For two contract events that have a State Commitment field reference relationship but whose on-chain position fields do not satisfy the happen-before relationship, this reference relationship is not added to the partial order edge set. This dual constraint ensures that the happens-before partial order topology has both a state commitment basis and an on-chain commit order basis, avoiding the inclusion of target chain events without state origins into the order state DAG.
[0043] Specifically, the extraction process of the State Commitment receiving direction includes:
[0044] Retrieve the target contract event from the cross-chain contract event stream corresponding to the target transaction order; match the target contract event with the order asset identifier of the target transaction order; extract the on-chain position field, StateCommitment field, and preceding State Commitment field from the target contract event to obtain the state commitment field group.
[0045] When performing this step, the target transaction order's asset identifier is used as the retrieval basis to search for the target contract event in the cross-chain contract event stream. The target contract event is a contract event that has a matching relationship with the asset identifier of the order. The matching relationship includes source chain carbon asset identifier matching, target chain mapped asset identifier matching, cross-chain state commitment reference matching, and target transaction order binding field matching.
[0046] Target contract events include source chain locking events, cross-chain verification events, and target chain mapping events. Source chain locking events indicate that the source chain carbon asset has entered a locked state. Cross-chain verification events indicate that the cross-chain proof has been processed by the verification contract. Target chain mapping events indicate that the target chain mapped asset has been generated. All types of target contract events participate in the subsequent direction extraction in the form of a state commitment field group.
[0047] After extracting the on-chain position field, State Commitment field, and preceding State Commitment field from each target contract event, a state commitment field group is obtained. This state commitment field group is denoted as: ;
[0048] in, Indicates the first The state commitment field group for each target contract event.
[0049] Use the preceding State Commitment field in the State Commitment field group as an inverted index; connect the State Commitment field in the State Commitment field group with the preceding State Commitment field to form the State Commitment successor direction.
[0050] When performing this step, the preceding State Commitment field in the State Commitment field group is used as a reverse index to retrieve a State Commitment field from the State Commitment field group set that matches the preceding State Commitment field. If it exists... Then connect and and form a group point to The direction of State Commitment.
[0051] Each State Commitment receiving direction in the set includes a predecessor state commitment and a successor state commitment. When projecting a State Commitment receiving direction onto the happens-before partial order topology, the on-chain position field of the corresponding target contract event is synchronously invoked. If the corresponding on-chain position field satisfies Then the State Commitment's receiving direction enters the happens-before partial order topology. (Symbol) This indicates that the on-chain position field conforms to the precedence relationship in the execution order of the chain environment or cross-chain protocol.
[0052] Through the above processing, a valid basis for cross-chain state transfer can only be formed between target contract events if both a state commitment reference relationship and an on-chain prior occurrence relationship are simultaneously satisfied. This technical feature is used to exclude event data that only forms asset records on the target chain but lacks a cross-chain state transfer basis.
[0053] Specifically, the formation process of the order status DAG includes:
[0054] The partial order node corresponding to the asset identifier of the target transaction order is used as the merging end; the partial order nodes with the State Commitment acceptance direction are merged in reverse along the partial order edge to obtain the cross-chain state commit path.
[0055] During this step, the partial-order node corresponding to the asset identifier occupied by the order is located in the happens-before partial-order topology. This partial-order node corresponds to the target chain mapping event or the target chain mapping asset generation status. This partial-order node serves as the merging end. The merging end is used to indicate the aggregation position of the cross-chain state submission path corresponding to the target transaction order on the target chain mapping side.
[0056] Starting from the merging end, backtrack along the reverse direction of the partial order edges. The reverse direction of the partial order edges is the direction from the successor partial order node to the predecessor partial order node. Backtracking along this direction, the partial order nodes corresponding to cross-chain verification events, the partial order nodes corresponding to source chain locking events, and other partial order nodes that have a successor relationship with the above state commitments are successively gathered.
[0057] During the backtracking process, only partially ordered nodes with a State Commitment acceptance direction are aggregated. If a partially ordered node only has an on-chain positional relationship with the merging end but lacks a State Commitment acceptance direction, then that partially ordered node will not enter the cross-chain state commit path. If a partially ordered node has a State Commitment acceptance direction but fails the on-chain position verification in the happens-before partially ordered topology, then that partially ordered node will not enter the cross-chain state commit path.
[0058] The cross-chain state commit path represents the state commit chain of the carbon asset corresponding to the target transaction order, from locking on the source chain, cross-chain verification, to mapping on the target chain. This path can be a single chain or a directed acyclic structure with multiple predecessor branches. Through this path, the target chain mapped asset is expressed as a state commit result with both source chain state origin and cross-chain verification origin.
[0059] Convert the partial order nodes in the cross-chain state submission path into state nodes; convert the partial order edges in the cross-chain state submission path into state receiving edges; write the target transaction order binding field into the state node to form an order state DAG.
[0060] During this step, the partial-order nodes in the cross-chain state commit path are converted into state nodes. State nodes are used to carry cross-chain state information at the target transaction order level. A state node includes at least the following fields: on-chain position, StateCommitment, preceding State Commitment, the chain environment to which the state node belongs, and the target transaction order binding field. The chain environment to which the state node belongs is used to distinguish between the source chain environment, the chain environment where the cross-chain verification contract resides, and the target chain environment.
[0061] Transform the partial order edges in the cross-chain state commit path into state acceptance edges. State acceptance edges represent the state commitment acceptance relationship between two state nodes. The direction of the state acceptance edge is consistent with the direction of the State Commitment acceptance. If the state node... State Commitment field State Node Preceding State Commitment fields References are found in the state node. With state nodes A state-transfer edge is formed between them.
[0062] Write the target transaction order binding field to the state node. The target transaction order binding field includes at least the order identifier and the asset identifier used by the order. The target transaction order binding field is used to bind the cross-chain state submission path to the target transaction order, avoiding the mixing of state nodes between different orders or assets mapped to different target chains.
[0063] Order status DAG is denoted as: ;
[0064] in, This represents the DAG of order status. Represents a set of state nodes. This represents the set of edges that support the state.
[0065] The order status DAG provides a unified graph structure foundation for the generation of subsequent unfinished state subgraphs, source chain reachability verification, and risk prediction input generation.
[0066] Understandably, the target chain-mapped asset enters subsequent processing as the target chain-mapped node in the order state DAG. This approach preserves the state continuity between source chain locking, cross-chain verification, and target chain mapping, providing a traceable data foundation for identifying gaps in cross-chain final confirmation.
[0067] When executing S102, the order state DAG is used as the basis for processing. The order state DAG represents the cross-chain state succession relationship corresponding to the target transaction order, but does not directly indicate whether each state node has entered final confirmation. Since state nodes may be distributed in different chain environments, it is necessary to determine the BFT Finality confirmation prefix based on the chain environment to which the state node belongs.
[0068] The target chain mapping node is the state node in the order state DAG corresponding to the target chain mapping asset generation event. The fulfillment occupancy path is the occupancy path of the target transaction order on the target chain mapping asset. The fulfillment occupancy path includes at least the asset identifier and occupancy order of the order. The fulfillment occupancy path is used to represent the target chain mapping asset that the target transaction order actually depends on.
[0069] Embedding the fulfillment occupancy path in reverse into the causal precursor path of the target chain mapping node means starting from the target chain mapping node, tracing back along the reverse direction of the state-bearing edges in the order state DAG, and writing the occupancy order of the target transaction order into the causal precursor path obtained through backtracking. Thus, the fulfillment occupancy relationship of the target transaction order and the cross-chain state causal relationship are placed in the same graph structure.
[0070] The BFT Finality confirmation prefix includes a commit position sequence and a consensus commit certificate status. The commit position sequence consists of block positions arranged sequentially in the corresponding chain environment. The consensus commit certificate status is used to indicate that the block position has completed the BFT consensus commit. The final commit position of the commit position sequence serves as the truncation boundary of the causal precursor path.
[0071] The BFT Finality confirmation prefix is determined based on the chain environment to which the state node belongs. If the state node belongs to the source chain environment, the BFT Finality confirmation prefix corresponding to the source chain environment is used. If the state node belongs to the target chain environment, the BFT Finality confirmation prefix corresponding to the target chain environment is used. If the state node belongs to the chain environment where the cross-chain verification contract resides, the BFT Finality confirmation prefix corresponding to that chain environment is used.
[0072] The commit position sequence is read from the consensus commit record, block header extension field, or on-chain final confirmation record of the corresponding chain environment. The consensus commit certificate status is determined by the consensus commit certificate, block commit signature set, or final confirmation contract event of the corresponding chain environment. All of the above data comes from on-chain or consensus layer records of the corresponding chain environment and is used to provide verifiable final confirmation evidence.
[0073] The submission position sequence is denoted as: ;
[0074] in, Represents the chain environment The sequence of submission positions, This indicates the final commit position in the commit position sequence. The consensus commit certificate status is used to characterize whether the corresponding commit position has completed the BFT consensus commit. If the on-chain position field of the state node is located... If the on-chain position field of a state node is not located within the consensus commit certificate state, then that state node is marked as finalized. If a node is in a state where a consensus submission certificate is not in effect at the corresponding submission location, then that node is marked as pending finality.
[0075] The on-chain position field, block position, and commit position each have different meanings. The on-chain position field represents the contract event-level position, used to locate a specific contract event. The block position represents the block-level position within the chain environment, used to describe the order of blocks on the chain. The commit position is the block-level confirmation position in the BFT Finality confirmation prefix, used to characterize the block position where BFT consensus has been committed. This distinction is used to avoid confusion between event-level positions and block-level confirmation positions.
[0076] It should be understood that the BFT Finality confirmation prefix is determined before the generation process of the unfinalized state subgraph. The unfinalized state subgraph needs to be extracted based on whether the state nodes have entered the finality confirmation stage. If the BFT Finality confirmation prefix is not determined first, the state nodes in the order occupying the predecessor chain cannot generate the finality attribution marker.
[0077] Specifically, the process of generating the unfinished state subgraph includes:
[0078] Parse the fulfillment occupancy path from the target transaction order; bind the order occupancy asset identifier pointed to by the fulfillment occupancy path to the target chain mapping node in the order status DAG to obtain the order mapping anchor point.
[0079] During this step, the fulfillment occupancy path is parsed from the target transaction order. The fulfillment occupancy path includes at least the order-occupied asset identifier and the occupancy order. The order-occupied asset identifier points to the target chain-mapped asset that the target transaction order intends to use. The occupancy order indicates the order in which the target transaction order occupies multiple target chain-mapped assets.
[0080] The order asset identifier pointed to by the fulfillment path is bound to the target chain mapping node in the order status DAG. If the order asset identifier matches the mapped asset field in the target chain mapping node, a binding relationship is established. If multiple target chain mapping nodes have the same mapped asset field, the corresponding target chain mapping node is determined according to the target transaction order binding field in the status node.
[0081] After binding, an order mapping anchor is obtained. The order mapping anchor is the anchor node of the target chain mapping node under the dimension of the target transaction order. The order mapping anchor is used as the starting point for causal precursor path backtracking.
[0082] Using the order mapping anchor point as the backtracking starting point; extract the causal predecessor path along the incoming edge direction of the order state DAG; write the occupancy order of the fulfillment occupancy path into the causal predecessor path to obtain the order occupancy predecessor chain.
[0083] When performing this step, the order mapping anchor point is used as the backtracking starting point. The causal predecessor path is extracted along the incoming edge direction of the order state DAG. The incoming edge direction is the direction from the target chain mapping node to the predecessor state node. The causal predecessor path includes the order mapping anchor point, the predecessor state node that has a state inheritance relationship with the order mapping anchor point, and the state inheritance edge connecting the above state nodes.
[0084] The occupancy order of the fulfillment occupancy path is written into the causal predecessor path. After the occupancy order is written, the causal predecessor path simultaneously contains the cross-chain state succession relationship and the order fulfillment occupancy relationship, forming the order occupancy predecessor chain.
[0085] If an order mapping anchor has multiple predecessor branches, the predecessor branch with a state continuity relationship to the asset identifier occupied by the order is retained. If a predecessor branch cannot return to the order mapping anchor through a state continuity edge, then that predecessor branch will not enter the order-occupying predecessor chain. This filtering focuses the identification of unfinished states on the target chain mapped assets actually occupied by the target transaction order, reducing the interference of irrelevant cross-chain events on risk prediction.
[0086] Based on the commit position sequence and consensus commit certificate status in the BFT Finality confirmation prefix, a final ownership marker for the state node in the order occupies the preceding chain is generated; the residual branch representing the pending final state and connecting the order mapping anchor point is retained to obtain the unfinalized state subgraph.
[0087] During this step, the chain environment and on-chain position fields of each state node in the predecessor chain of the order are read. The corresponding BFT Finality confirmation prefix is invoked based on the chain environment of the state node. The on-chain position field of the state node is matched with the commit position sequence, and combined with the consensus commit certificate status, a final attribution marker is generated.
[0088] The finality attribution flag is used to characterize the finality state of a state node. If the on-chain position field of a state node is within the commit position sequence, and the corresponding commit position has completed the BFT consensus commit, then the finality attribution flag indicates a finalized state. If the on-chain position field of a state node is not within the commit position sequence, or the corresponding commit position has not completed the BFT consensus commit, then the finality attribution flag indicates a pending finality state.
[0089] Retain the residual branches that represent the pending final state and connect to the order mapping anchor point. The residual branches include the pending final state node, the state-carrying edge between the pending final state node and the order mapping anchor point, and the related state nodes located on the connection path. If there is no state-carrying path between a pending final state node and the order mapping anchor point, then the pending final state node will not enter the unfinalized state subgraph.
[0090] The unfinished state subgraph consists of residual branches. The unfinished state subgraph is used to represent the portion of the target chain mapped assets occupied by the target transaction order that still has unfinished causal predecessors.
[0091] Through S102, even after the target chain mapping asset has been written to the target chain, it still needs to accept the BFT Finality confirmation prefix constraint of its respective chain environment. This process can identify cross-chain finality confirmation gaps before risk prediction, reducing the possibility that the finalized mapping asset is prematurely included in the deliverable asset set.
[0092] When executing S103, the order state DAG, the unfinished state subgraph, and the Merkle Proof receipt are used as the processing basis. The order state DAG is used to provide the target chain mapping node, the source chain commit node, and the causal predecessor path between them. The unfinished state subgraph is used to provide the node coverage of the nodes to be finalized. The Merkle Proof receipt is used to provide membership proof data for the target chain mapping node to trace back to the source chain state root.
[0093] The target chain mapping node is the state node in the order state DAG corresponding to the target chain mapping asset generation event. The source chain commit node is the state node in the order state DAG corresponding to the source chain locking event or the source chain state commit event. Source chain reachability is used to characterize whether the target chain mapping node can reach the source chain commit node along the causal predecessor path in the order state DAG, and whether the proof verification can be completed through Merkle Proof receipt.
[0094] It's important to note that the existence of target chain mapping nodes in the order state DAG only indicates that the target chain has formed a mapped asset state; it does not equate to the target chain mapped asset being eligible for delivery. Target chain mapping nodes must also meet the source chain reachability verification conditions and be located outside the node coverage area of the unfinalized state subgraph. Source chain reachability verification confirms that the target chain mapped asset has a verifiable source chain state origin. Node coverage filtering excludes target chain mapped assets still within the pending finalized causal range. Both together constitute the node selection criteria for the deliverable asset subgraph.
[0095] The Merkle Proof receipt includes a membership proof path, a leaf commitment, and a source chain state root; the membership proof path is returned by the target chain verification contract for the source chain state root; the leaf commitment has a hash inheritance relationship with the State Commitment; and the source chain state root serves as the root hash boundary for root hash consistency verification.
[0096] Merkle Proof receipts are proof receipt data generated during the cross-chain verification process. This proof receipt data originates from cross-chain verification contract events or target chain verification contract events within the cross-chain contract event stream. The member proof path represents the hash path from the leaf commitment to the source chain state root. The leaf commitment is used to inherit the StateCommitment field from the source chain commit node. The source chain state root serves as the boundary value for root hash consistency verification.
[0097] The membership proof path is denoted as: ;
[0098] in, Indicates the membership proof path, This represents a path node in the membership proof path. A leaf commitment is denoted as... The source chain state root is denoted as If the State Commitment field of the source chain commit node is denoted as The hash relationship between the leaf commitment and the State Commitment field is then represented as follows: ;
[0099] in, This represents a hash function.
[0100] The root hash consistency check is used to verify whether the leaf commitment can be computed to the source chain state root via a membership proof path. During the check, the leaf commitment is used as the basis for the verification. As the initial hash value, hashes are merged according to the order of path nodes in the member proof path to obtain the computed state root. If the following conditions are met: ;
[0101] If the above relationship is not satisfied, the root hash consistency check result is passed.
[0102] Specifically, the process of forming the deliverable asset sub-map includes:
[0103] In the order status DAG, the Merkle Proof receipt is written into the proof field of the target chain mapping node to form a proof-associated node.
[0104] During this step, the target chain mapping node is located in the order status DAG. The target chain mapping node is determined by the order-occupied asset identifier, the target chain mapping asset identifier, and the target transaction order binding fields. After locating the target chain mapping node, the MerkleProof receipt corresponding to that target chain mapping node in the cross-chain contract event stream is written into the proof field of the target chain mapping node.
[0105] The proof field carries the member proof path, leaf commitment, and source chain state root. The target chain mapping node after the proof field is written forms a proof-associated node. The proof-associated node simultaneously retains the target chain mapping state and the cross-chain proof receipt, enabling subsequent source chain reachability verification to be completed within the order state DAG.
[0106] When a target chain mapping node corresponds to multiple Merkle Proof receipts, the corresponding Merkle Proof receipt is determined based on the on-chain position field and the State Commitment continuation direction in the cross-chain contract event stream. Specifically, a Merkle Proof receipt written into the proof field must meet the following conditions: the preceding State Commitment field of the contract event corresponding to the Merkle Proof receipt has a continuation relationship with the preceding State Commitment field of the target chain mapping node, and the on-chain position field of the contract event corresponding to the Merkle Proof receipt precedes the on-chain position field of the target chain mapping node. Merkle Proof receipts that cannot establish the above state continuation relationship with the target chain mapping node are not written into the proof field of that target chain mapping node.
[0107] Retrieve the source chain commit node along the causal predecessor path of the proof-related node; perform hash verification between the StateCommitment field of the source chain commit node and the leaf commitment in the Merkle Proof receipt to obtain the candidate reachable path of the source chain.
[0108] When performing this step, starting from the proof-related node, a causal precursor path is retrieved along the incoming edge direction of the order state DAG. The causal precursor path includes the proof-related node, the cross-chain verification node, the source chain commit node, and the state-connecting edges between the above state nodes. The source chain commit node is a state node in the causal precursor path whose chain environment is the source chain environment and which corresponds to the source chain lock event or the source chain state commit event.
[0109] Extract the State Commitment field from the source chain's commit node, denoted as... Read the leaf promises from the Merkle Proof receipt and denote them as... Subsequently, regarding and Perform hash concatenation verification. Hash concatenation verification is represented as follows: ;
[0110] When the above relationship holds, a hash connection exists between the source chain commit node and the Merkle Proof receipt. In this case, the causal predecessor path between the proof-associated node and the source chain commit node is determined as a candidate reachable path for the source chain. When the above relationship does not hold, the causal predecessor path is not considered a candidate reachable path for the source chain.
[0111] The candidate reachable path of the source chain includes the source chain submission node, the proof-related node, and the state-connecting edge between them. This path is used for subsequent root hash consistency verification. By performing hash connection verification first and then root hash consistency verification, the proof object can be limited to the target chain mapping nodes with source chain state connection relationships, reducing the number of irrelevant mapping nodes entering the deliverable asset screening process.
[0112] Based on the member proof path and the source chain state root, root hash consistency verification is performed on the candidate reachable paths of the source chain; the target chain mapping nodes in the order state DAG that are outside the coverage of the unfinalized state subgraph are formed into a finalized candidate node set; the target chain mapping nodes that pass the root hash consistency verification and belong to the finalized candidate node set are determined as deliverable nodes, forming a deliverable asset subgraph.
[0113] During this step, the leaf commitments, membership proof paths, and source chain state roots corresponding to the candidate reachable paths of the source chain are read. Using the leaf commitments as the initial hash values, a layer-by-layer hash merging process is performed according to the order of the path nodes in the membership proof paths to obtain the calculated state root. The source chain state root is denoted as The root hash consistency check is represented as: ;
[0114] like If so, the root hash consistency check result is passed. If the root hash consistency check fails, the result will be "failed".
[0115] Next, the node coverage of the unfinished state subgraph is read. The node coverage is the set of all state nodes in the unfinished state subgraph, denoted as . The set of target chain mapping nodes in the order status DAG is denoted as . The set of candidate nodes for the final game is denoted as... , is represented as: ;
[0116] in, This represents the target chain mapping node. The above expression indicates that the final candidate node set consists of the target chain mapping nodes in the order state DAG that are outside the coverage area of the nodes in the non-final state subgraph.
[0117] Target chain mapping nodes that pass the root hash consistency check and belong to the final candidate node set are determined as deliverable nodes. , is represented as: ;
[0118] in, Indicates the target chain mapping node The root hash consistency check result, This indicates that the root hash consistency check result is passed.
[0119] The deliverable asset subgraph consists of deliverable nodes, candidate reachable paths from the source chain corresponding to each deliverable node, and state-connecting edges between these nodes. The deliverable asset subgraph is denoted as: ;
[0120] in, This represents a sub-graph of deliverable assets. Represents the set of deliverable nodes. This represents the set of state-connecting edges associated with a deliverable node.
[0121] Through S103, the source chain reachability verification result and the node coverage of the unfinalized state subgraph jointly participate in the target chain mapping node selection. This process ensures that the deliverable asset subgraph simultaneously satisfies both the source chain proof validity and the finalized state exclusion condition. Therefore, target chain mapped assets that are unreachable from the source chain, as well as target chain mapped assets still within the unfinalized state range, will not be included in the deliverable asset subgraph.
[0122] During execution of S104, the deliverable asset subgraph and the incomplete state subgraph are compared using the order state DAG as a unified reference. The deliverable asset subgraph represents the target chain mapping nodes in the target transaction order that satisfy the source chain reachability verification and the incomplete exclusion condition. The incomplete state subgraph represents the residual branches in the target transaction order that still have causal predecessors awaiting completion. The structural differences between the two in the order state DAG are used to form the target Temporal Graph state difference.
[0123] The target Temporal Graph state difference is used to characterize the temporal graph state differences of the target transaction order. The target Temporal Graph state difference includes target node state difference, target edge state difference, and target time state difference. The target node state difference represents the difference in the attribution of the target chain-mapped node in the deliverable asset subgraph and the unfinished state subgraph. The target edge state difference represents the difference in the retention of state-bearing edges in the two subgraphs. The target time state difference represents the relative relationship between the on-chain position of a state node and the fulfillment time field.
[0124] The objective temporal graph state difference is denoted as: ;
[0125] in, Represents the objective Temporal Graph state difference. This represents the state difference of the target node. Represents the target edge state difference. This represents the target time state difference.
[0126] The generation process of the target Temporal Graph state difference is as follows: The attribution results of the target chain-mapped nodes in the deliverable asset subgraph and the unfinished state subgraph are written into the target node state difference. The retention results of the state-bearing edges in the deliverable asset subgraph and the unfinished state subgraph are written into the target edge state difference. The relative order between the on-chain position field of the state node and the fulfillment time field of the target transaction order is written into the target time state difference.
[0127] Furthermore, the target node state difference, target edge state difference, and target time state difference are aligned according to the state node identifier, state-connecting edge identifier, and on-chain position field to form the target Temporal Graph state difference matrix. This target Temporal Graph state difference matrix serves as the input data for the pre-trained risk prediction TGN model. Thus, the model input directly corresponds to the nodes, edges, and time fields in the order state DAG, avoiding the abstraction of graph structure differences into a single numerical input.
[0128] A pre-trained risk prediction TGN model is used to receive the target Temporal Graph state difference matrix and output the prediction result of carbon asset trading delivery failure risk. The TGN model is a Temporal Graph Network model. This model is used to process graph-structured data with event time.
[0129] The pre-trained risk prediction TGN model includes an event encoding module, a time encoding module, a node memory module, a message aggregation module, a graph update module, and a risk output module. The event encoding module receives the target temporal graph state difference matrix and encodes the target node state difference, target edge state difference, and target time state difference into event feature vectors. The time encoding module encodes the time interval between the on-chain position field and the fulfillment time field of a state node into a time feature vector. The node memory module stores the historical state representation of the same target chain mapped node in different cross-chain contract events. The message aggregation module aggregates the event feature vector, time feature vector, and node memory vector according to the state-bearing edge to obtain a state-bearing message vector. The graph update module updates the temporal graph representation of the target chain mapped node, the source chain commit node, and the cross-chain verification node based on the state-bearing message vector. The risk output module maps the temporal graph representation corresponding to the target chain mapped node to a settlement failure probability.
[0130] The target Temporal Graph state difference matrix includes a node difference numerator matrix, an edge difference numerator matrix, and a time difference numerator matrix. The rows of the node difference numerator matrix correspond to state nodes in the order state DAG, and the columns correspond to the ownership state of the state node in the deliverable asset subgraph and the unfinished state subgraph. The rows of the edge difference numerator matrix correspond to state-receiving edges in the order state DAG, and the columns correspond to the retained state of the state-receiving edges in the deliverable asset subgraph and the unfinished state subgraph. The rows of the time difference numerator matrix correspond to state nodes, and the columns correspond to the relative time characteristics between the on-chain position field and the fulfillment time field of the state node. The event encoding module aligns the three sub-matrices according to the state node identifier and the state-receiving edge identifier, generating the event sequence corresponding to the target Temporal Graph state difference matrix.
[0131] Specifically, the training process of the risk prediction TGN model includes:
[0132] Obtain the historical cross-chain contract event stream and historical delivery failure tags corresponding to the historical target transaction orders; generate a historical order state DAG, a historical unfinished state subgraph, and a historical deliverable asset subgraph based on the historical cross-chain contract event stream.
[0133] When performing this step, the historical cross-chain contract event stream corresponding to the historical target transaction order is obtained. The historical cross-chain contract event stream has the same field structure as the cross-chain contract event stream of the current target transaction order. The historical cross-chain contract event stream includes the historical on-chain position field, the historical State Commitment field, the historical preceding State Commitment field, and the historical Merkle Proof receipt.
[0134] The historical settlement failure tag is used to characterize the settlement result of historical target transaction orders. The historical settlement failure tag is determined by the settlement completion record, target chain mapped asset rollback record, source chain lock revocation record, or cross-chain proof verification failure record of the historical target transaction order. If a historical target transaction order is settled within the period corresponding to the fulfillment time field, and there is no target chain mapped asset rollback, source chain lock revocation, or cross-chain proof verification failure, then the historical settlement failure tag indicates successful settlement. If a historical target transaction order is not settled within the period corresponding to the fulfillment time field, or if the above-mentioned settlement anomaly records occur, then the historical settlement failure tag indicates settlement failure.
[0135] Based on historical cross-chain contract event flows, a historical order state DAG, a historical incomplete state subgraph, and a historical deliverable asset subgraph are generated. The generation rules for the historical order state DAG are consistent with those for the order state DAG. The generation rules for the historical incomplete state subgraph are consistent with those for the incomplete state subgraph. The generation rules for the historical deliverable asset subgraph are consistent with those for the deliverable asset subgraph. Therefore, historical samples and current prediction samples have a consistent data structure, which can be used for supervised training.
[0136] By comparing the historical deliverable asset subgraph with the historical unfinished state subgraph in the historical order state DAG, a historical Temporal Graph state difference is generated.
[0137] When performing this step, the historical order status DAG is used as a reference to compare the historical deliverable asset subgraph and the historical unfinished status subgraph. The comparison includes differences in node affiliation of historical target chain mapping nodes, differences in retention of historical state-bearing edges, residual status of historical fulfillment-occupied paths, and relative positions of historical event times.
[0138] The historical temporal graph state differences are denoted as: ;
[0139] in, Represents the differences between historical temporal graph states. Represents the state difference of historical nodes. Represents the historical edge state difference. This represents the difference in state at historical time.
[0140] Historical node state differences are used to characterize the differences in the attribution of historical target chain mapped nodes in the historical deliverable asset subgraph and the historical unfinished state subgraph. Historical edge state differences are used to characterize the differences in the retention of historical state-bearing edges in the two subgraphs. Historical time state differences are used to characterize the temporal relationship between the on-chain position of historical state nodes and the historical fulfillment time field.
[0141] By differentiating historical temporal graph states, training samples can preserve graph states in historical target transaction orders, such as those that are mapped but awaiting finalization, those that are mapped and ready for settlement, and those that are reachable from the source chain but whose finalization is incomplete. This differential data can reflect the structural sources of settlement failure risk.
[0142] The historical Temporal Graph state differences are used as training inputs; historical delivery failure labels are used as supervised outputs; the initial risk prediction TGN model is trained to obtain the trained risk prediction TGN model.
[0143] In this step, historical temporal graph state differences are used as training input, and historical delivery failure labels are used as supervision output. After receiving the historical temporal graph state differences, the initial risk prediction TGN model updates its parameters and learns the mapping between the temporal graph state differences and the delivery failure labels.
[0144] During training, the initial risk prediction TGN model generates prediction outputs based on the historical temporal graph state differences. The training loss is calculated based on the prediction outputs and historical settlement failure labels. The training loss is represented by the cross-entropy loss function. ;
[0145] in, This indicates the number of historical target transaction orders. Indicates the first Historical delivery failure tags for historical target trading orders. This represents the predicted probability of delivery failure output by the initial risk prediction TGN model.
[0146] When training the pre-trained risk prediction TGN model, the node memory dimension, event encoding dimension, time encoding dimension, and message aggregation layer number of the initial risk prediction TGN model are determined based on the number of historical target transaction orders, the number of state nodes in the historical order state DAG, and the number of state connecting edges. Training batches are obtained by dividing historical target transaction orders. Each training batch includes historical temporal graph state differences and historical delivery failure labels. After training is completed, a pre-trained risk prediction TGN model is obtained for risk prediction of the current target transaction order.
[0147] In the current prediction phase, the deliverable asset subgraph and the unfinished state subgraph are compared in the order state DAG to generate the target Temporal Graph state difference. Subsequently, the target Temporal Graph state difference is input into a pre-trained risk prediction TGN model to obtain the carbon asset transaction delivery failure risk prediction result.
[0148] The prediction result for the risk of delivery failure in carbon asset transactions is the probability of delivery failure. The probability of delivery failure is used to represent the probability that the target transaction order will fail to be delivered within the period corresponding to the performance time field.
[0149] The above process allows risk prediction to no longer rely on the target chain's mapped asset balance or the target chain's mapped asset record itself, but rather on the target TemporalGraph state difference between the deliverable asset subgraph and the unfinished state subgraph. This approach can incorporate cross-chain final confirmation gaps, source chain reachability, and order fulfillment occupancy relationships into the risk prediction input, which helps to reduce the underestimation of the risk of delivery failure.
[0150] Example 2
[0151] like Figure 2 As shown in the example, the parts not detailed in this embodiment are as shown in Example 1. This embodiment discloses a carbon asset trading risk prediction system, including:
[0152] Topology module 201 is used to extract the State Commitment acceptance direction from the cross-chain contract event stream of the target transaction order; and project the State Commitment acceptance direction onto the happens-before partial order topology to form the order state DAG.
[0153] Bounding module 202 is used to use the BFT Finality confirmation prefix of the chain environment to which the state node in the order state DAG belongs as the truncation boundary; and to embed the performance occupancy path in reverse into the causal predecessor path of the target chain mapping node to obtain the unfinished state subgraph.
[0154] Verification module 203 is used to verify the source chain reachability of the target chain mapping node based on the Merkle Proof receipt; and to filter the target chain mapping node by combining the node coverage of the unfinished state subgraph to form a deliverable asset subgraph.
[0155] The prediction module 204 is used to compare the deliverable asset subgraph and the unfinished state subgraph in the order state DAG to generate the target Temporal Graph state difference; the target Temporal Graph state difference is input into the pre-trained risk prediction TGN model to obtain the carbon asset transaction delivery failure risk prediction result.
[0156] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.
Claims
1. A method for predicting risks in carbon asset trading, characterized in that, include: Extract the State Commitment acceptance direction from the cross-chain contract event stream of the target transaction order; project the State Commitment acceptance direction onto the happens-before partial order topology to form the order state DAG; The BFT Finality confirmation prefix of the chain environment to which the state node in the order state DAG belongs is used as the truncation boundary; the fulfillment occupancy path is back-embedded into the causal predecessor path of the target chain mapping node to obtain the unfinished state subgraph; The source chain reachability of the target chain mapping node is verified based on the Merkle Proof receipt. By combining the node coverage of the unfinished state subgraph, the target chain mapping nodes are filtered to form a deliverable asset subgraph; In the DAG of order status, the deliverable asset subgraph and the unfinished state subgraph are compared to generate the target Temporal Graph state difference; the target Temporal Graph state difference is input into the pre-trained risk prediction TGN model to obtain the carbon asset transaction delivery failure risk prediction result.
2. The method according to claim 1, characterized in that, The extraction process for the State Commitment receiving direction includes: Retrieve the target contract event from the cross-chain contract event stream of the target transaction order; match the target contract event with the order asset identifier of the target transaction order; extract the on-chain position field, State Commitment field, and preceding State Commitment field from the target contract event to obtain the state commitment field group; Use the preceding State Commitment field in the State Commitment field group as an inverted index; connect the State Commitment field in the State Commitment field group with the preceding State Commitment field to form the State Commitment successor direction.
3. The method according to claim 2, characterized in that, The happens-before partial order topology includes partial order nodes and partial order edges; a partial order node is a target contract event node for which the on-chain position field, State Commitment field, and preceding State Commitment field are written; a partial order edge is a directed edge connecting two partial order nodes; the direction of the partial order edge is determined by the reference relationship of the State Commitment field and the order relationship of the on-chain position field.
4. The method according to claim 3, characterized in that, The formation process of the order status DAG includes: The partial order node corresponding to the asset identifier of the target transaction order is used as the merging end; the partial order nodes with the State Commitment acceptance direction are merged in reverse along the partial order edge to obtain the cross-chain state commit path; Convert the partial order nodes in the cross-chain state submission path into state nodes; convert the partial order edges in the cross-chain state submission path into state receiving edges; write the target transaction order binding field into the state node to form an order state DAG.
5. The method according to claim 1, characterized in that, The BFT Finality confirmation prefix includes a commit position sequence and a consensus commit certificate status; the commit position sequence consists of block positions arranged consecutively in the commit order in the corresponding chain environment; the consensus commit certificate status is used to indicate that the block position has completed the BFT consensus commit. The final commit position of the commit position sequence serves as the truncation boundary of the causal precursor path.
6. The method according to claim 5, characterized in that, The process of generating the unfinished state subgraph includes: Parse the fulfillment occupancy path from the target transaction order; bind the order occupancy asset identifier pointed to by the fulfillment occupancy path to the target chain mapping node in the order status DAG to obtain the order mapping anchor point; Using the order mapping anchor point as the backtracking starting point; extract the causal predecessor path along the incoming edge direction of the order state DAG; write the occupancy order of the fulfillment occupancy path into the causal predecessor path to obtain the order occupancy predecessor chain; Based on the commit position sequence and consensus commit certificate status in the BFT Finality confirmation prefix, a final ownership marker for the state node in the order occupies the preceding chain is generated; the residual branch representing the pending final state and connecting the order mapping anchor point is retained to obtain the unfinalized state subgraph.
7. The method according to claim 1, characterized in that, The Merkle Proof receipt includes a membership proof path, a leaf commitment, and the source chain state root; the membership proof path is returned by the target chain verification contract for the source chain state root; the leaf commitment and the State Commitment have a hash inheritance relationship; the source chain state root serves as the root hash boundary for root hash consistency verification.
8. The method according to claim 7, characterized in that, The process of forming the deliverable asset submap includes: In the order status DAG, the Merkle Proof receipt is written into the proof field of the target chain mapping node to form a proof-associated node; Retrieve the source chain commit node along the causal predecessor path of the proof-related node; perform hash verification between the StateCommitment field of the source chain commit node and the leaf commitment in the Merkle Proof receipt to obtain the candidate reachable path of the source chain; Based on the member proof path and the source chain state root, root hash consistency verification is performed on the candidate reachable paths of the source chain; the target chain mapping nodes in the order state DAG that are outside the coverage of the unfinalized state subgraph are formed into a finalized candidate node set; the target chain mapping nodes that pass the root hash consistency verification and belong to the finalized candidate node set are determined as deliverable nodes, forming a deliverable asset subgraph.
9. The method according to claim 1, characterized in that, The training process of the risk prediction TGN model includes: Obtain the historical cross-chain contract event stream and historical delivery failure tags corresponding to historical target transaction orders; generate a historical order state DAG, a historical unfinished state subgraph, and a historical deliverable asset subgraph based on the historical cross-chain contract event stream; In the historical order status DAG, compare the historical deliverable asset subgraph with the historical unfinished state subgraph to generate the historical Temporal Graph state difference; The historical Temporal Graph state differences are used as training inputs; historical delivery failure labels are used as supervised outputs; the initial risk prediction TGN model is trained to obtain the trained risk prediction TGN model.
10. A carbon asset trading risk prediction system, implemented based on the method of any one of claims 1-9, characterized in that, include: The topology module is used to extract the State Commitment acceptance direction from the cross-chain contract event stream of the target transaction order; and to project the State Commitment acceptance direction onto the happens-before partial order topology to form the order state DAG. The delimiting module is used to use the BFT Finality confirmation prefix of the chain environment to which the state node in the order state DAG belongs as the truncation boundary; and to embed the fulfillment occupancy path into the causal predecessor path of the target chain mapping node in reverse to obtain the unfinished state subgraph. The verification module is used to verify the source chain reachability of the target chain mapping node based on the Merkle Proof receipt; By combining the node coverage of the unfinished state subgraph, the target chain mapping nodes are filtered to form a deliverable asset subgraph; The prediction module compares the deliverable asset subgraph and the unfinished state subgraph in the order state DAG to generate the target Temporal Graph state difference; the target Temporal Graph state difference is input into the pre-trained risk prediction TGN model to obtain the carbon asset transaction delivery failure risk prediction result.