Engineering financial cost automatic accounting method based on fund chain tracing

Abstract

The present application relates to the technical field of engineering cost tracing accounting, and discloses an automatic engineering financial cost accounting method based on fund chain tracing. The method comprises receiving project-related original financial voucher flow, generating a standardized fund transaction sequence through structured deconstruction and semantic standardization processing. According to a preset rule atlas, the sequence is associated and aggregated into a multi-level fund circulation network with hierarchical relationships, and the edge attributes of the network include circulation amount and cost subject direction. On the network, deep mining and propagation tracing are performed from the end cost consumption node to the source fund allocation node, and a complete cost influence link set is identified. Based on the link set, dynamic cost allocation and re-accounting are performed, a tracing accounting report for each cost subject is generated, and the contribution of different fund sources to the final cost is quantified. The method improves the cost accounting from static subject collection to dynamic tracing along the actual path, and reveals the complete fund context of cost formation.

Description

Technical Field

[0001] This invention relates to the field of engineering cost traceability and accounting technology, specifically to an automatic accounting method for engineering financial costs based on capital chain traceability. Background Technology

[0002] Current financial cost accounting for engineering projects generally relies on traditional accounting subject systems and pre-defined allocation rules. Existing technology typically allocates expenses directly to the corresponding cost accounts based on voucher summaries and account information, or distributes them among multiple cost objects according to a fixed ratio. This method treats financial transactions as independent records, and the accounting process focuses more on the summarization and allocation of results.

[0003] Existing technical solutions have shortcomings, primarily in the opaque nature of cost composition and the lack of traceability. Accounting-based cost aggregation only reflects the final allocation of expenses, failing to reveal the specific flow of funds within the project and intermediate consumption stages. Furthermore, static allocation ratios are completely detached from the actual movement of funds, making it impossible for the final cost data to answer which sources of funds and through which paths specific costs are incurred. This leads to vague cost driver analysis and a lack of accurate financial process data to support management decisions.

[0004] The field of engineering cost accounting urgently needs to address the transition from "outcome accounting" to "process traceability." A method is needed to transform discrete financial documents into structured fund flow graphs, enabling reverse path tracing from any cost consumption point to the funding source point on this graph. This would provide transparency into the cost formation process and accurately quantify the contribution of different funding sources to the final cost. Summary of the Invention

[0005] The purpose of this invention is to provide an automatic accounting method for engineering financial costs based on capital chain tracing, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this invention provides an automatic accounting method for engineering financial costs based on capital chain tracing, the method comprising:

[0007] Receive the stream of original financial documents associated with the target project;

[0008] The original financial document flow is subjected to structured deconstruction and semantic standardization processing to generate a standardized financial transaction sequence;

[0009] Based on the preset cost aggregation rule graph, the standardized fund transaction sequence is processed by transaction chain association and aggregation to construct a multi-level fund flow network with hierarchical relationship. In the multi-level fund flow network, the nodes represent the fund status, the edges represent fund transactions, and the edge attributes include the flow amount and cost account.

[0010] In-depth cost driver mining and propagation tracing are performed on the multi-level capital flow network to identify a complete set of influence links from the end cost consumption node to the source capital allocation node. Each link in the set of influence links represents the source and flow path of a cost structure.

[0011] Based on the set of influence links, dynamic cost allocation and recalculation are performed on the multi-level fund flow network to generate a traceability accounting report for each cost item. The traceability accounting report quantifies the contribution of different funding sources to the final cost value.

[0012] Preferably, the step of performing structured deconstruction and semantic standardization processing on the original financial document flow to generate a standardized financial transaction sequence includes the following steps:

[0013] The original financial voucher stream contains multi-dimensional voucher records arranged in chronological order. Each voucher record is associated with a description of fund flow, an amount, a payment recipient identifier, and a cost account classification code.

[0014] The structured deconstruction and semantic standardization process includes: separating structured data fields and unstructured text fields from each voucher record; performing domain terminology recognition and synonym substitution operations on the unstructured text fields, mapping them to the standardized expressions in a preset engineering financial standard terminology library; and recombining the mapped standardized expressions with the structured data fields to form the standardized financial transaction sequence.

[0015] The original financial voucher stream is subjected to batch preprocessing, which includes voucher format verification, key field integrity verification, and duplicate record removal.

[0016] A dedicated named entity recognition model based on the engineering finance field is invoked to extract entities from unstructured text fields in batch preprocessed voucher records. The target types of the entity extraction include project stage entities, specific work content entities, material and equipment entities, and partner entities.

[0017] The extracted entities are matched and aligned with the preset engineering financial standard terminology library. The matching and alignment process includes: calculating the semantic similarity between the extracted entities and the candidate standard terms in the terminology library; replacing the extracted entities with semantic similarity exceeding a set threshold with the corresponding standard terms; and storing the extracted entities as new term candidates in the pending review queue as semantic similarity candidates.

[0018] All fields that have completed terminology standardization are reorganized and serialized according to the preset "transaction-object-amount-account" quadruple template to generate the standardized financial transaction sequence.

[0019] Preferably, the step of calling a dedicated named entity recognition model based on the engineering finance field to extract entities from unstructured text fields in batch preprocessed voucher records includes the following steps:

[0020] Load the dedicated named entity recognition model, which adopts a sequence labeling architecture based on a deep bidirectional attention mechanism and is pre-trained using massive amounts of engineering contract texts, settlement reports, and audit opinion texts;

[0021] The unstructured text field is segmented and tagged with parts of speech to generate an initial word sequence and a part-of-speech tag sequence;

[0022] The initial word sequence is concatenated with the part-of-speech tag sequence and then input into the dedicated named entity recognition model. The model outputs the probability distribution of each word belonging to a preset entity type.

[0023] The probability distribution is processed using the Viterbi decoding algorithm to obtain the globally optimal entity type label sequence, thereby completing the entity extraction.

[0024] Preferably, the step of performing transaction chain association and aggregation processing on the standardized fund transaction sequence based on a preset cost aggregation rule graph to construct a multi-level fund flow network with hierarchical relationships includes the following steps:

[0025] The initiator, recipient, transaction type, associated cost item, and amount of the fund transaction are extracted from the standardized fund transaction sequence.

[0026] Using the initiator and receiver as candidate nodes and financial transactions as candidate edges, the system determines whether candidate edges should be established based on the preset cost aggregation rule graph. The cost aggregation rule graph defines the allowed flow relationships between different transaction types and different cost items.

[0027] For edges that are allowed to be established, they are added to the network and assigned the properties parsed from the standardized financial transaction sequence;

[0028] Aggregate analysis is performed on all edges that have been added to the network. Multiple edges that describe the same flow of funds but belong to different times or different vouchers are aggregated and merged according to their associated cost accounts to form aggregated edges.

[0029] Based on the actual flow of funds, starting from the top-level fund pool node, the aggregated edges and nodes are linked layer by layer downwards to form the multi-level fund flow network in a tree or graph structure.

[0030] Preferably, the step of performing in-depth cost driver mining and propagation tracing processing on the multi-level capital flow network to identify the complete set of influence links from the end cost consumption node to the source capital allocation node includes the following steps:

[0031] In the multi-level capital flow network, all nodes directly marked as specific cost consumption types are marked as end nodes;

[0032] Starting from each end node, perform a reverse backtracking traversal along the opposite direction of the capital inflow, that is, along the direction of the edge.

[0033] During the reverse backtracking process, each node and edge visited is recorded, and the amount ratio or sharing coefficient of the edges visited is accumulated. The amount ratio or sharing coefficient is obtained from the edge attributes of the multi-level capital flow network.

[0034] When traversing to a node that has no predecessor node or whose predecessor node is the top-level fund pool, stop tracing the path and record the path from the end node to this stopped node as a complete influence link.

[0035] Collect all paths traced from all end nodes to form the set of influencing links.

[0036] Preferably, the step of performing dynamic cost allocation and recalculation processing on the multi-level fund transfer network based on the set of influence links to generate a traceability accounting report for each cost item includes the following steps:

[0037] Select a cost item to be analyzed, and filter out all impact links whose end nodes belong to the cost item from the set of impact links;

[0038] For each selected influencing link, starting from the cost amount of the end node of the influencing link, the cost amount is reversibly allocated to the source node level by level according to the cost ratio or allocation coefficient of each side recorded on the link, so as to obtain the theoretical contribution amount of the source node to the end cost.

[0039] Summarize the reverse allocation calculation results of all the impact links belonging to the cost item to obtain the total contribution amount distribution of each funding source node to the cost item;

[0040] The total contribution amount distribution is compared with and normalized to the actual total cost of the cost item to generate the contribution percentage.

[0041] The contribution percentage, source node information, and circulation path information are integrated to form the traceability accounting report for the cost item.

[0042] Preferably, for each selected influencing link, starting from the cost amount of the end node of the influencing link, the cost amount is reversibly allocated back to the source node according to the cost ratio or allocation coefficient recorded on each side of the link, to obtain the theoretical contribution amount of the source node to the end cost, including the following steps:

[0043] Obtain the actual cost consumption amount corresponding to the affected end node of the link, and use it as the initial value for reverse allocation calculation;

[0044] Starting from the end node of the affected link, move along the link towards the source node. At each step, process the edge between the current node and its direct predecessor node.

[0045] Read the amount ratio or sharing coefficient recorded by the currently processed edge;

[0046] Multiply the cost amount to be allocated at the current node by the amount ratio or allocation coefficient to obtain the cost amount to be allocated to its direct predecessor node.

[0047] The calculated cost amount is assigned to the direct predecessor node as a new cost amount that the predecessor node needs to continue to allocate forward.

[0048] Repeat the steps until all edges in the link have been processed and the source node is reached;

[0049] Record the cost amount that the source node is ultimately allocated to, which is the theoretical contribution of the source node to the end cost through this influence link;

[0050] For the same impact chain, ensure that during the cost allocation process, the sum of the costs flowing from the end node to the source node is consistent with the initial cost incurred by the end node.

[0051] Preferably, the step of comparing and normalizing the total contribution amount distribution with the actual total cost of the cost item to generate the contribution percentage includes the following steps:

[0052] Calculate the sum of the contribution amounts of all source nodes in the total contribution amount distribution;

[0053] Divide the contribution amount of each source node by the sum of the contribution amounts to obtain the initial contribution degree of the source node;

[0054] Obtain the actual total cost value of the aforementioned cost item in the financial system;

[0055] Multiply the initial contribution by the actual total cost to obtain the actual cost amount allocated to each source node;

[0056] Based on the actual allocated cost amount, the contribution percentage of each source node to the actual total cost is recalculated.

[0057] Preferably, the application of the Viterbi decoding algorithm to process the probability distribution to obtain the globally optimal entity type label sequence, thereby completing the entity extraction, includes the following steps:

[0058] Construct a state transition probability matrix, which describes the constraint rules and possibilities for mutual transformation between different entity type labels;

[0059] The probability distribution of each word output by the dedicated named entity recognition model belonging to a preset entity type is used as the observation probability.

[0060] Initialize a path probability matrix and a path backtracking matrix to record the optimal path probability and its source for each tag at each position during the decoding process;

[0061] Starting from the first word of the text sequence, the local optimal path probability of each entity type label of each word is calculated sequentially. The local optimal path probability is jointly determined by the path probability of the previous word, the state transition probability, and the observation probability of the current word.

[0062] During the calculation process, the label of the previous word corresponding to each optimal probability is recorded through the path backtracking matrix;

[0063] After processing the last word of the text sequence, starting from the label with the highest path probability, reverse tracing is performed based on the path backtracking matrix to determine the entity type label corresponding to each word in turn.

[0064] The label sequence obtained by reverse tracing is output as the globally optimal entity type label sequence.

[0065] Preferably, when traversing to a node without a predecessor node or whose predecessor node is the top-level fund pool, stopping the path tracing and recording the path from the end node to this stopped node as a complete influence link includes the following steps:

[0066] During the reverse backtracking traversal, a path stack is maintained for the current traversal path to record the nodes and edges visited in sequence.

[0067] When a node is accessed, all incoming edges pointing to that node are queried in the multi-level fund transfer network. The source nodes corresponding to these incoming edges are the predecessor nodes of that node.

[0068] If the query result is empty, it means that the current node has no predecessor node, and the current node is determined to be the source of the path;

[0069] If the query result is not empty, but all predecessor nodes are marked as top-level funding pool nodes, it is determined that the current path has been traced back to the top-level funding source.

[0070] In both of the above cases, stop backtracking and treat the sequence of nodes and edges from the end node to the current stopping node recorded in the current path stack as a complete candidate influence link;

[0071] The candidate impact links are validated to check whether there is a valid fund transfer relationship between adjacent nodes on the link, and invalid paths with logical contradictions are eliminated.

[0072] The verified candidate impact links are formally recorded into the impact link set.

[0073] Compared with the prior art, the beneficial effects of the present invention are:

[0074] By aggregating standardized financial transactions into a multi-level fund flow network based on a pre-defined rule graph, and assigning cost account pointing attributes to the edges in the network, dynamic and explicit binding of fund flows and cost accounting objects in the topological structure is achieved. This transforms unstructured transaction data into a network model with clear hierarchy and relationships, ensuring that the direction and purpose of each fund flow are marked and structured during the process. This provides a directly calculable and traversable underlying data framework for cost traceability, overcoming the limitations of traditional methods that rely on loose associations between fund flows and cost objects and post-event manual interpretation.

[0075] This fund flow network performs in-depth mining and propagation tracing from the end cost consumption node to the source fund allocation node, identifying a complete set of cost impact links. This transforms the basis for cost allocation from a preset static ratio to a dynamically mined set of actual fund flow paths. Costs can be reversed and allocated along multiple potentially coexisting fund flow links, thereby accurately characterizing and quantifying the specific contribution of each fund flow path to the end cost value. This achieves a fundamental change in cost drivers from "rough allocation estimation" to "precise path tracing," generating a traceability accounting report that reflects the details of cost composition and the contribution of fund sources. Attached Figure Description

[0076] Figure 1 This is a schematic diagram illustrating the working principle of the automatic accounting method for engineering financial costs based on capital chain tracing as described in this invention.

[0077] Figure 2A flowchart for entity extraction for a dedicated named entity recognition model;

[0078] Figure 3 A flowchart for constructing a multi-level fund transfer network;

[0079] Figure 4 A bar chart comparing the sources of funds for allocating project financial costs;

[0080] Figure 5 A dual-axis analysis chart is provided to support the source of project funding. Detailed Implementation

[0081] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0082] Please see Figure 1 This invention provides an automatic accounting method for engineering financial costs based on capital chain tracing. The method includes: receiving a stream of original financial documents associated with a target engineering project, the stream containing multi-dimensional document records of capital flow events arranged chronologically; performing structured deconstruction and semantic standardization on the original financial document stream to generate a standardized sequence of capital transactions, which describes the core elements of capital transactions in a standardized format; performing transaction chain association and aggregation processing on the standardized sequence of capital transactions according to a preset cost aggregation rule graph to construct a hierarchical multi-level capital flow network, where nodes represent capital status, edges represent capital transactions, and edge attributes include the amount of flow and the cost account indicated; and performing in-depth cost driver mining and propagation tracing processing on the multi-level capital flow network to identify a complete set of influence links from the end cost consumption node to the source capital allocation node, where each link represents the source and flow path of a cost component. Based on the set of influence links, dynamic cost allocation and recalculation are performed on the multi-level fund flow network to generate a traceability accounting report for each cost item. This report quantifies the contribution of different funding sources to the final cost value.

[0083] Example 1: In practical implementation, the automatic accounting method for engineering financial costs based on capital chain tracing receives the original financial voucher stream associated with the target engineering project. The original financial voucher stream contains multi-dimensional voucher records arranged in chronological order. Each voucher record is associated with a description of capital flow, an amount, a payment recipient identifier, and a cost account classification code. In practical implementation, an example of the original financial voucher stream could include a bank payment voucher stating, "Payment to XX Construction Company for the fifth phase of the project progress payment, involving the reinforcement binding and concrete pouring work of the third floor of the main building, amounting to 500,000 yuan, recorded under the account 'Construction Engineering - Main Structure Engineering'". The goal of performing structured deconstruction and semantic standardization processing on the original financial voucher stream is to generate a standardized sequence of financial transactions. In practical implementation, the processing begins with batch preprocessing of the original financial voucher stream, which includes voucher format verification, key field integrity verification, and duplicate record removal. In practice, the voucher format verification confirms whether each voucher has the required fields, while the key field integrity verification checks whether fields such as cost account code and counterparty name have empty values; duplicate record removal is based on the combination of voucher number, amount, and timestamp to determine and remove identical records.

[0084] In some embodiments, a dedicated named entity recognition model based on the engineering finance domain is invoked to extract entities from unstructured text fields in batch-preprocessed voucher records. The target types for entity extraction include project phase entities, specific work content entities, material and equipment entities, and partner entities. Optionally, the extracted entities are matched and aligned with a pre-defined engineering finance standard terminology library. The matching and alignment process includes calculating the semantic similarity between the extracted entities and candidate standardized terms in the terminology library. Semantic similarity can be calculated using a cosine similarity method based on word vectors, expressed by the formula:

[0085]

[0086] in: Indicates entity extraction With candidate normative terms in the terminology database Semantic similarity between them It is a physical entity Word vector representation, Candidate standard terms Word vector representation, Represents the vector dot product. It represents the magnitude of the vector.

[0087] In some embodiments, all fields that have undergone terminology standardization are reorganized and serialized according to a preset "transaction-object-amount-account" quadruple template to generate a standardized financial transaction sequence. In a specific implementation, for the original record "Payment to Building Materials Company A for the purchase of 200 cubic meters of C30 commercial concrete for the pouring of the foundation slab of Building #1, amount 100,000 yuan, account code 510101", after the aforementioned processing, the structured data fields "Building Materials Company A", "100,000 yuan", and "510101" are retained, and the unstructured text, after entity extraction and terminology standardization, yields the standardized expressions "purchase", "C30 commercial concrete", and "foundation slab pouring".

[0088] Example 2: See Figure 2 In its implementation, the automatic accounting method for engineering financial costs based on capital chain tracing calls a dedicated named entity recognition model in the engineering finance field to extract entities from unstructured text fields in batch-preprocessed voucher records. Specifically, the processed objects are unstructured text fields such as "Payment to Installation Company B for completing the fire pipe laying and pressure testing work on the 10th floor of Building #1". The call process begins by loading the dedicated named entity recognition model, which employs a sequence labeling architecture based on a deep bidirectional attention mechanism and is pre-trained using massive amounts of engineering contract text, settlement reports, and audit opinion text. This pre-training process enables the dedicated named entity recognition model to understand the contextual semantic relationships of professional terms such as "pressure testing," "fire pipe," and "10th floor" within the engineering finance field.

[0089] The Viterbi decoding algorithm is applied to process the probability distribution to obtain the globally optimal entity type label sequence, thus completing entity extraction. The process of applying the Viterbi decoding algorithm includes constructing a state transition probability matrix. This matrix describes the constraints and probabilities of transitions between different entity type labels. For example, the state transition probability matrix can define that the label "specific job content entity" is more likely to be followed by the label "materials and equipment entity," rather than directly by another "specific job content entity" label. In specific implementation, a path probability matrix and a path backtracking matrix are initialized to record the optimal path probability and its source for each label at each position during the decoding process. Starting from the first word of the text sequence, the locally optimal path probability for each entity type label of each word is calculated sequentially. The locally optimal path probability is jointly determined by the path probability of the previous word, the state transition probability, and the observation probability of the current word. The calculation of the locally optimal path probability can be expressed as:

[0090]

[0091] in: Indicates at time step Label The probability of the local optimal path, This represents the labels of all possible previous time steps. Take the maximum value. It is the previous time step tag The probability of the optimal path. From the label Move to label The state transition probability, In the label The current word was observed below. The observed probability. During the calculation process, the label of the previous word corresponding to each optimal probability is recorded through the path backtracking matrix. For example, when processing the word "laying", if its optimal path probability is transferred from the label of "materials and equipment entities" of the previous word "pipeline", then this association is recorded in the path backtracking matrix.

[0092] It is understandable that after processing the last word of the text sequence, the system starts by backtracking from the label with the highest path probability, using the path backtracking matrix to determine the entity type label corresponding to each word. The label sequence obtained from the backtracking is then output as the globally optimal entity type label sequence, thus completing entity extraction. This process identifies "B Installation Company" as a partner entity, "10th Floor" as an engineering project phase entity, "Fire Pipeline" as a material and equipment entity, and "Laying" and "Pressure Testing" as specific work content entities. This provides structured semantic information for subsequent terminology standardization and transaction sequence generation.

[0093] Example 3: See Figure 3In practice, standardized fund transaction sequences are linked and aggregated based on a pre-defined cost aggregation rule graph to construct a multi-level fund flow network with hierarchical relationships. The input to the process is a standardized fund transaction sequence generated through semantic standardization. The initiator, recipient, transaction type, associated cost item, and amount of each fund transaction are extracted from this sequence. In the example above, the extraction result is: Initiator "Project Total Fund Pool", Recipient "Building Materials Company A", Transaction Type "Procurement Payment", Associated Cost Item "Construction Engineering - Main Structure Engineering - Concrete Material Costs", and Amount "100,000". Using the initiator and recipient as candidate nodes and fund transactions as candidate edges, the pre-defined cost aggregation rule graph determines whether candidate edges should be established. This graph defines the allowed flow relationships between different transaction types and different cost items. In practice, the pre-defined cost aggregation rule graph may include a rule that "procurement payment" transactions can flow from the "project total fund pool" node to the material supplier node and be associated with specific material cost items. For edges that are allowed to be established, they are added to the network and given attributes parsed from the standardized fund transaction sequence. The attributes of the edges must at least include the amount of the transaction and the cost item they point to.

[0094] In some embodiments, aggregation analysis is performed on all edges joined to the network. Multiple edges describing the same fund flow but belonging to different times or different vouchers are aggregated and merged according to their associated cost accounts to form an aggregated edge. In a specific implementation, assume there are three independent edges, all describing "procurement payment" transactions flowing from the "Project Total Fund Pool" to "Building Materials Company A," and all associated with the cost account "Construction Engineering - Main Structure Engineering - Concrete Material Costs," with amounts of 100,000 yuan, 80,000 yuan, and 120,000 yuan respectively. Aggregation analysis merges these three edges into one aggregated edge, whose amount is the sum of the three amounts. The amount of the aggregated edge... It can be calculated in the following ways:

[0095]

[0096] in: This represents the total amount of the aggregated edges. This represents the number of edges to be aggregated. Representing the The original amount of the edge to be aggregated. Based on the actual flow of funds, starting from the top-level fund pool node, the aggregated edges and nodes are linked layer by layer downwards to form a tree-like or graph-like multi-level fund flow network. In specific implementation, the top-level fund pool node "Project Total Fund Pool" is connected to the "Building Materials Company A" node through an aggregated edge. The "Building Materials Company A" node may further connect to a node representing a specific construction activity, "Concrete Pouring of Building #1 Foundation Slab," through another edge describing the payment, thus forming a fund flow hierarchy of "Project Total Fund Pool" -> "Building Materials Company A" -> "Concrete Pouring of Building #1 Foundation Slab."

[0097] It is understandable that in-depth cost driver mining and propagation tracing are performed on a multi-level capital flow network to identify the complete set of influence links from the end cost consumption node to the source capital allocation node. In the multi-level capital flow network, all nodes directly marked as specific cost consumption types are marked as end nodes. In specific implementation, a node marked as "labor cost for fire pipe installation on the 10th floor of Building #1" is identified as an end node. Starting from each end node, a reverse traversal is performed along the direction of capital inflow, that is, along the direction of the edge. During the reverse traversal, each node and edge traversed is recorded, and the amount ratio or allocation coefficient of the edges traversed is accumulated. The amount ratio or allocation coefficient is obtained from the edge attributes of the multi-level capital flow network.

[0098] Optionally, when traversing to a node without a predecessor node or whose predecessor node is a top-level funding pool, the path tracing stops, and the path from the end node to this stopping node is recorded as a complete influence link. During the reverse backtracking traversal, a path stack is maintained for the current traversal path to record the nodes and edges traversed in sequence. When a node is visited, all incoming edges pointing to the current node in the multi-level funding flow network are queried. The source nodes corresponding to these incoming edges are the predecessor nodes of the current node. If the query result is empty, it indicates that the current node has no predecessor node, and the current node is determined to be the source of the path. If the query result is not empty, but all predecessor nodes are marked as top-level funding pool nodes, the current path is determined to have traced back to the top-level funding source. In both of the above cases, the backtracking stops, and the sequence of nodes and edges recorded in the current path stack from the end node to the current stopping node is recorded as a complete candidate influence link. The candidate impact links are validated to check for valid fund transfer relationships between adjacent nodes. Invalid paths with logical contradictions are eliminated; for example, it is verified whether the cost item recorded on an edge matches the cost type of its connected downstream node. The validated candidate impact links are then formally recorded in the impact link set.

[0099] Example 4: Dynamic cost allocation and recalculation of a multi-level fund transfer network based on an impact link set, generating a traceability accounting report for each cost item. A cost item to be analyzed is selected, and all impact links belonging to that cost item at the end node are filtered from the impact link set. For each selected impact link, starting from the cost amount of the end node, the cost amount is recalculated backwards to the source node according to the amount ratio or allocation coefficient recorded on each side of the link, obtaining the theoretical contribution amount of the source node to the end cost. The recalculation results of all impact links belonging to that cost item are summarized to obtain the total contribution amount distribution of each fund source node to that cost item. The total contribution amount distribution is compared and normalized with the actual total cost of that cost item to generate a contribution percentage. The contribution percentage, source node information, and transfer path information are integrated to form a traceability accounting report for that cost item.

[0100] In practice, dynamic cost allocation and recalculation are performed on the multi-level capital flow network based on the influence link set, generating a traceability accounting report for each cost item. A cost item to be analyzed is selected, and all influence links with end nodes belonging to this cost item are filtered from the influence link set. The cost item to be analyzed can be "Construction Engineering - Main Structure Engineering - Concrete Material Costs". The influence links with end nodes all marked as "Construction Engineering - Main Structure Engineering - Concrete Material Costs" selected from the influence link set may contain three links, denoted as link α, link β, and link γ.

[0101] In some embodiments, for each selected influencing link, starting from the cost amount of the end node of the influencing link, the cost amount is regressively allocated back to the source node according to the amount ratio or allocation coefficient recorded on each side of the link, to obtain the theoretical contribution amount of the source node to the end cost. In a specific implementation, assuming that the actual cost of the end node "Concrete material cost of Building #1 foundation slab" of link α is 150,000 yuan, and link α records two sides from the end node to the source node, the allocation coefficient of the first side is 1.0, and the amount ratio of the second side is 0.7. The regressive allocation calculation first traces the full amount of 150,000 yuan back to the relay node connected by the first side, and then, according to the amount ratio of the second side of 0.7, traces 105,000 yuan of the 150,000 yuan back to the source node "Special loan for main structure", to obtain the theoretical contribution amount of the source node "Special loan for main structure" to the end cost through link α as 105,000 yuan. Similarly, the same reverse amortization calculation process is performed on links β and γ.

[0102] Optionally, the reverse allocation calculation results of all impact links belonging to cost items are summarized to obtain the total contribution amount distribution of each funding source node to the cost item. In specific implementation, it is assumed that after calculation, three impact links trace back to three funding source nodes: "Special Loan for Main Structure", "Project Capital", and "Transfer of Equipment Leasing Revenue". After summing the theoretical contribution amounts of all links, the total theoretical contribution amount of "Special Loan for Main Structure" is 255,000 yuan, the total theoretical contribution amount of "Project Capital" is 120,000 yuan, and the total theoretical contribution amount of "Transfer of Equipment Leasing Revenue" is 25,000 yuan. These data constitute the total contribution amount distribution of the funding source nodes to "Construction Engineering - Main Structure Engineering - Concrete Material Costs".

[0103] The total contribution amount distribution is compared and normalized with the actual total cost of the cost item to generate a contribution percentage. In specific implementation, the actual total cost of "Construction Engineering - Main Structure Engineering - Concrete Material Costs" in the financial system is 410,000 yuan. The sum of the theoretical contribution amounts of all source nodes in the total contribution amount distribution is calculated to be 400,000 yuan. The initial contribution of each source node is obtained by dividing its theoretical contribution amount by the sum of its theoretical contribution amounts. The initial contribution is multiplied by the actual total cost to obtain the actual cost amount allocated to each source node. Based on the actual allocated cost amount, the proportion of each source node's contribution to the actual total cost is recalculated to obtain the contribution percentage. The contribution percentage can be expressed as follows:

[0104]

[0105] in, Representing the The percentage contribution of each funding source node. Representing the The actual cost amount allocated to each funding source node. This represents the actual total cost of the cost item. See Table 1 for the traceability accounting data for the cost item "Construction Engineering - Main Structural Engineering - Concrete Material Costs".

[0106] Table 1: Analysis of Funding Sources for Cost Item "Construction Engineering - Main Structure Engineering - Concrete Material Costs"

[0107] Funding source node Theoretical contribution amount (RMB) Preliminary contribution Actual allocated cost (RMB) Contribution percentage (ρ) Special loan for main structure 255,000 63.75% 261,375 63.75% Project capital 120,000 30.00% 123,000 30.00% Equipment leasing revenue transferred 25,000 6.25% 25,625 6.25% total 400,000 100.00% 410,000 100.00%

[0108] This approach integrates contribution percentages, source node information, and flow path information to form a traceability accounting report for each cost item. The traceability accounting report is presented in a structured document format, containing the name of the analyzed cost item, the actual total cost, detailed information on each funding source node, and a description of the path through which each contributing fund flows. The traceability accounting report quantifies the contribution of different funding sources to the final cost figure.

[0109] See Figure 4 This is a bar chart comparing the funding sources for the allocation of project financial costs. The "actual allocated cost amount" for each funding source is slightly higher than the "theoretical contribution amount," consistent with the actual scenario in cost accounting where "theoretical values ​​and actual values ​​have slight discrepancies." The amount of the main structure special loan (approximately 250,000-260,000 RMB) is significantly higher than the other two sources, making it the primary source of funding for this cost item. The main structure special loan accounts for over 60%, project capital accounts for approximately 30%, and equipment leasing revenue accounts for less than 10%, reflecting the concentration of funding sources. This type of chart is used in the project financial cost retrospective accounting phase. By comparing the theoretical contribution with the actual allocated amount, it verifies the rationality of the cost allocation logic and clarifies the contribution magnitude of different funding sources to the target cost item. It is one of the core visual materials for generating cost retrospective reports.

[0110] Example 5: For each selected influencing link, starting from the cost amount of the end node of the influencing link, the cost amount is calculated in reverse from the end node to the source node according to the cost ratio or allocation coefficient recorded on each edge of the link, to obtain the theoretical contribution amount of the source node to the end cost. The actual cost consumption amount corresponding to the end node of the influencing link is obtained as the initial value for reverse allocation calculation. Starting from the end node of the influencing link, move along the link towards the source node, processing the edge between the current node and its direct predecessor node at each step. Read the cost ratio or allocation coefficient recorded on the currently processed edge. Multiply the cost amount to be allocated by the current node by the cost ratio or allocation coefficient to obtain the cost amount to be allocated to its direct predecessor node. Assign the calculated cost amount to the direct predecessor node as the new cost amount that the predecessor node needs to continue to allocate forward. Repeat the steps until all edges in the link are processed and the source node is reached. Record the cost amount that the source node is finally allocated, which is the theoretical contribution amount of the source node to the end cost through this influencing link. For the same impact chain, ensure that the sum of the costs flowing from the end node to the source node during the allocation process remains consistent with the initial cost consumption of the end node. Compare and normalize the total contribution amount distribution with the actual total cost of the cost item to generate a contribution percentage, including calculating the sum of the contribution amounts of all source nodes in the total contribution amount distribution. Divide the contribution amount of each source node by the sum of the contribution amounts to obtain the initial contribution of the source node. Obtain the actual total cost value of the cost item in the financial system. Multiply the initial contribution by the actual total cost value to obtain the actual cost amount allocated to each source node. Recalculate the proportion of each source node's contribution to the actual total cost based on the actual allocated cost amount to obtain the contribution percentage.

[0111] In practice, for each selected impact link, starting from the cost amount of the end node of the impact link, the cost amount is recalculated backwards towards the source node based on the proportion or allocation coefficient of each edge recorded on the link, thus obtaining the theoretical contribution amount of the source node to the end cost. In practice, assuming a selected impact link is "Construction Site C Area Safety Fence Maintenance Fee" <- "Miscellaneous Project Reserve Fund Pool" <- "Project Total Fund Pool", the actual cost consumption amount corresponding to the end node "Construction Site C Area Safety Fence Maintenance Fee" is 8,000 yuan. The actual cost consumption amount of 8,000 yuan corresponding to the end node of the impact link is obtained as the initial value for the reverse allocation calculation. Starting from the end node "Construction Site C Area Safety Fence Maintenance Fee", the process moves along the link towards the source node, processing the edge between the current node and its direct predecessor node at each step. In practice, the first step is to process the edge between the end node "Construction Site C Area Safety Fence Maintenance Fee" and its direct predecessor node "Miscellaneous Project Reserve Fund Pool".

[0112] In some embodiments, the amount ratio or allocation coefficient recorded on the currently processed edge is read. In a specific implementation, the allocation coefficient recorded in the edge attribute connecting the "Construction Site C Zone Safety Fence Maintenance Fee" node and the "Miscellaneous Projects Reserve Fund Pool" node is 0.8, indicating that 80% of the cost of the end node can be traced back to the "Miscellaneous Projects Reserve Fund Pool" node. The cost amount to be allocated at the current node is multiplied by the amount ratio or allocation coefficient to obtain the cost amount to be allocated to its direct predecessor node. In a specific implementation, the calculation process is 8,000 yuan multiplied by 0.8, resulting in 6,400 yuan, which is the cost amount to be allocated to the direct predecessor node "Miscellaneous Projects Reserve Fund Pool". The calculated cost amount is assigned to the direct predecessor node "Miscellaneous Projects Reserve Fund Pool" as a new cost amount that the "Miscellaneous Projects Reserve Fund Pool" node needs to continue to allocate forward. The steps are repeated to process the edge between the "Miscellaneous Projects Reserve Fund Pool" node and its direct predecessor node "Project Total Fund Pool".

[0113] Optionally, read the amount ratio or allocation coefficient recorded on the edge connecting the "Miscellaneous Project Reserve Fund Pool" and the "Project Total Fund Pool," assuming it is 1.0. Multiply the current cost amount to be allocated in the "Miscellaneous Project Reserve Fund Pool" node, 6,400 yuan, by 1.0 to obtain 6,400 yuan. This 6,400 yuan is the cost amount to be allocated to the source node, "Project Total Fund Pool." Assign the calculated cost amount of 6,400 yuan to the "Project Total Fund Pool" node. Repeat the steps until all edges in the link are processed and the source node, "Project Total Fund Pool," is reached. Record the final cost amount of 6,400 yuan allocated to the source node, which is the theoretical contribution amount of the source node, "Project Total Fund Pool," to the end cost, "Construction Site C Area Safety Fence Maintenance Fee," through this influence link. For the same influence link, ensure that during the allocation process, the sum of the cost amounts of each node flowing from the end node to the source node remains consistent with the initial cost consumption amount of the end node.

[0114] It is understandable that the total contribution amount distribution is compared and normalized with the actual total cost of the cost item to generate the contribution percentage. In specific implementation, the cost item "Construction Safety Measures Fee" is selected, with an actual total cost of 50,000 yuan. The reverse allocation calculation results of all impact links belonging to "Construction Safety Measures Fee" are summarized to obtain the total contribution amount distribution, where the sum of the contribution amounts of the funding source node "Project Total Fund Pool" is 30,000 yuan, and the sum of the contribution amounts of the funding source node "Safety Production Special Fund" is 15,000 yuan. The sum of the contribution amounts of all source nodes in the total contribution amount distribution is calculated to be 45,000 yuan. Dividing the contribution amount of each source node by the sum of the contribution amounts yields the preliminary contribution degree of the source node.

[0115] In some embodiments, the actual total cost value of the cost item "Construction Safety Measures Fee" in the financial system is obtained as 50,000 yuan. The initial contribution is multiplied by the actual total cost value to obtain the actual cost amount allocated to each source node. The actual cost amount allocated to the funding source node "Project Total Fund Pool" is calculated as 0.6667 * 50,000 = 33,335 yuan, and the actual cost amount allocated to the funding source node "Safety Production Special Fund" is calculated as 0.3333 * 50,000 = 16,665 yuan. Based on the actual allocated cost amount, the contribution of each source node to the actual total cost is recalculated to obtain the contribution percentage. The contribution percentage can be expressed as:

[0116]

[0117] in: Represents the source node of funds The percentage of contribution, Represents the source node of funds The actual cost amount allocated, This represents the actual total cost figure for each cost item.

[0118] See Figure 5 This is a dual-axis chart analyzing the contribution of project funding sources. The contribution amount and contribution percentage show a strong positive correlation: the total project funding pool contributes the highest amount (approximately 330,000 yuan), corresponding to the highest contribution percentage (approximately 70%). As the amount of funding from each source decreases, the contribution percentage also decreases, reflecting the logic that "the larger the amount, the higher the contribution to costs." The first two funding sources account for over 80% of the total contribution, representing the core funding source for costs; subsequent sources contribute a lower percentage, serving as supplementary funding. This type of dual-axis chart is used in the contribution analysis phase of project financial cost accounting, simultaneously displaying "absolute amount" and "relative percentage," helping to intuitively understand the actual contribution level and structural proportion of different funding sources. It is a key visualization tool for demonstrating the rationality of funding sources in cost traceability reports.

[0119] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0120] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for automatically calculating engineering financial costs based on capital chain tracing, characterized in that, The method includes the following steps: Receive the stream of original financial documents associated with the target project; The original financial document flow is subjected to structured deconstruction and semantic standardization processing to generate a standardized financial transaction sequence; Based on the preset cost aggregation rule graph, the standardized fund transaction sequence is processed by transaction chain association and aggregation to construct a multi-level fund flow network with hierarchical relationship. In the multi-level fund flow network, the nodes represent the fund status, the edges represent fund transactions, and the edge attributes include the flow amount and cost account. In-depth cost driver mining and propagation tracing are performed on the multi-level capital flow network to identify a complete set of influence links from the end cost consumption node to the source capital allocation node. Each link in the set of influence links represents the source and flow path of a cost structure. Based on the set of influence links, dynamic cost allocation and recalculation are performed on the multi-level fund flow network to generate a traceability accounting report for each cost item. The traceability accounting report quantifies the contribution of different funding sources to the final cost value.

2. The automatic accounting method for engineering financial costs based on capital chain tracing according to claim 1, characterized in that, The process of performing structured deconstruction and semantic standardization on the original financial document flow to generate a standardized financial transaction sequence includes the following steps: The original financial voucher stream contains multi-dimensional voucher records arranged in chronological order. Each voucher record is associated with a description of fund flow, an amount, a payment recipient identifier, and a cost account classification code. The structured deconstruction and semantic standardization process includes: separating structured data fields and unstructured text fields from each voucher record; performing domain terminology recognition and synonym substitution operations on the unstructured text fields, mapping them to the standardized expressions in a preset engineering financial standard terminology library; and recombining the mapped standardized expressions with the structured data fields to form the standardized financial transaction sequence. The original financial voucher stream is subjected to batch preprocessing, which includes voucher format verification, key field integrity verification, and duplicate record removal. A dedicated named entity recognition model based on the engineering finance field is invoked to extract entities from unstructured text fields in batch preprocessed voucher records. The target types of the entity extraction include project stage entities, specific work content entities, material and equipment entities, and partner entities. The extracted entities are matched and aligned with the preset engineering financial standard terminology library. The matching and alignment process includes: calculating the semantic similarity between the extracted entities and the candidate standard terms in the terminology library; replacing the extracted entities with semantic similarity exceeding a set threshold with the corresponding standard terms; and storing the extracted entities as new term candidates in the pending review queue as semantic similarity candidates. All fields that have completed terminology standardization are reorganized and serialized according to the preset "transaction-object-amount-subject" quadruple template to generate the standardized financial transaction sequence.

3. The automatic accounting method for engineering financial costs based on capital chain tracing according to claim 2, characterized in that, The process of calling a specialized named entity recognition model based on the engineering finance field to extract entities from unstructured text fields in batch preprocessed voucher records includes the following steps: Load the dedicated named entity recognition model, which adopts a sequence labeling architecture based on a deep bidirectional attention mechanism and is pre-trained using massive amounts of engineering contract texts, settlement reports, and audit opinion texts; The unstructured text field is segmented and tagged with parts of speech to generate an initial word sequence and a part-of-speech tag sequence; The initial word sequence is concatenated with the part-of-speech tag sequence and then input into the dedicated named entity recognition model. The model outputs the probability distribution of each word belonging to a preset entity type. The probability distribution is processed using the Viterbi decoding algorithm to obtain the globally optimal entity type label sequence, thereby completing the entity extraction.

4. The automatic accounting method for engineering financial costs based on capital chain tracing according to claim 1, characterized in that, The process of performing transaction chain association and aggregation on the standardized fund transaction sequence based on a preset cost aggregation rule graph to construct a multi-level fund flow network with hierarchical relationships includes the following steps: The initiator, recipient, transaction type, associated cost item, and amount of the fund transaction are extracted from the standardized fund transaction sequence. Using the initiator and receiver as candidate nodes and financial transactions as candidate edges, the system determines whether candidate edges should be established based on the preset cost aggregation rule graph. The cost aggregation rule graph defines the allowed flow relationships between different transaction types and different cost items. For edges that are allowed to be established, they are added to the network and assigned the properties parsed from the standardized financial transaction sequence; Aggregate analysis is performed on all edges that have been added to the network. Multiple edges that describe the same flow of funds but belong to different times or different vouchers are aggregated and merged according to their associated cost accounts to form aggregated edges. Based on the actual flow of funds, starting from the top-level fund pool node, the aggregated edges and nodes are linked layer by layer downwards to form the multi-level fund flow network in a tree or graph structure.

5. The automatic accounting method for engineering financial costs based on capital chain tracing according to claim 4, characterized in that, The process of performing in-depth cost driver mining and propagation tracing on the multi-level capital flow network to identify the complete set of influence links from the end cost consumption node to the source capital allocation node includes the following steps: In the multi-level capital flow network, all nodes directly marked as specific cost consumption types are marked as end nodes; Starting from each end node, perform a reverse backtracking traversal along the opposite direction of the capital inflow, that is, along the direction of the edge. During the reverse backtracking process, each node and edge visited is recorded, and the amount ratio or sharing coefficient of the edges visited is accumulated. The amount ratio or sharing coefficient is obtained from the edge attributes of the multi-level capital flow network. When traversing to a node that has no predecessor node or whose predecessor node is the top-level fund pool, stop tracing the path and record the path from the end node to this stopped node as a complete influence link. Collect all paths traced from all end nodes to form the set of influencing links.

6. The automatic accounting method for engineering financial costs based on capital chain tracing according to claim 5, characterized in that, The process of dynamically allocating and recalculating costs for the multi-level fund transfer network based on the set of influence links, and generating a traceability accounting report for each cost item, includes the following steps: Select a cost item to be analyzed, and filter out all impact links whose end nodes belong to the cost item from the set of impact links; For each selected influencing link, starting from the cost amount of the end node of the influencing link, the cost amount is reversibly allocated to the source node level by level according to the cost ratio or allocation coefficient of each side recorded on the link, so as to obtain the theoretical contribution amount of the source node to the end cost. Summarize the reverse allocation calculation results of all the impact links belonging to the cost item to obtain the total contribution amount distribution of each funding source node to the cost item; The total contribution amount distribution is compared with and normalized to the actual total cost of the cost item to generate the contribution percentage. The contribution percentage, source node information, and circulation path information are integrated to form the traceability accounting report for the cost item.

7. The automatic accounting method for engineering financial costs based on capital chain tracing according to claim 6, characterized in that, For each selected influencing link, starting from the cost amount of the end node of the influencing link, the cost amount is reversibly allocated back to the source node according to the cost ratio or allocation coefficient recorded on each side of the link, to obtain the theoretical contribution amount of the source node to the end cost. This includes the following steps: Obtain the actual cost consumption amount corresponding to the affected end node of the link, and use it as the initial value for reverse allocation calculation; Starting from the end node of the affected link, move along the link towards the source node. At each step, process the edge between the current node and its direct predecessor node. Read the amount ratio or sharing coefficient recorded by the currently processed edge; Multiply the cost amount to be allocated at the current node by the amount ratio or allocation coefficient to obtain the cost amount to be allocated to its direct predecessor node. The calculated cost amount is assigned to the direct predecessor node as a new cost amount that the predecessor node needs to continue to allocate forward. Repeat the steps until all edges in the link have been processed and the source node is reached; Record the cost amount that the source node is ultimately allocated to, which is the theoretical contribution of the source node to the end cost through this influence link; For the same impact chain, ensure that during the cost allocation process, the sum of the costs flowing from the end node to the source node is consistent with the initial cost incurred by the end node.

8. The automatic accounting method for engineering financial costs based on capital chain tracing according to claim 6, characterized in that, The step of comparing and normalizing the total contribution amount distribution with the actual total cost of the cost item to generate the contribution percentage includes the following steps: Calculate the sum of the contribution amounts of all source nodes in the total contribution amount distribution; Divide the contribution amount of each source node by the sum of the contribution amounts to obtain the initial contribution degree of the source node; Obtain the actual total cost value of the aforementioned cost item in the financial system; Multiply the initial contribution by the actual total cost to obtain the actual cost amount allocated to each source node; Based on the actual allocated cost amount, the contribution percentage of each source node to the actual total cost is recalculated.

9. The automatic accounting method for engineering financial costs based on capital chain tracing according to claim 3, characterized in that, The application of the Viterbi decoding algorithm to process the probability distribution yields the globally optimal entity type label sequence, thereby completing the entity extraction. This includes the following steps: Construct a state transition probability matrix, which describes the constraint rules and possibilities for mutual transformation between different entity type labels; The probability distribution of each word output by the dedicated named entity recognition model belonging to a preset entity type is used as the observation probability. Initialize a path probability matrix and a path backtracking matrix to record the optimal path probability and its source for each tag at each position during the decoding process; Starting from the first word of the text sequence, the local optimal path probability of each entity type label of each word is calculated sequentially. The local optimal path probability is jointly determined by the path probability of the previous word, the state transition probability, and the observation probability of the current word. During the calculation process, the label of the previous word corresponding to each optimal probability is recorded through the path backtracking matrix; After processing the last word of the text sequence, starting from the label with the highest path probability, reverse tracing is performed based on the path backtracking matrix to determine the entity type label corresponding to each word in turn. The label sequence obtained by reverse tracing is output as the globally optimal entity type label sequence.

10. The automatic accounting method for engineering financial costs based on capital chain tracing according to claim 5, characterized in that, When traversing to a node that has no predecessor node or whose predecessor node is the top-level fund pool, the path tracing stops, and the path from the end node to this stopped node is recorded as a complete influence link, including the following steps: During the reverse backtracking traversal, a path stack is maintained for the current traversal path to record the nodes and edges visited in sequence. When a node is accessed, all incoming edges pointing to that node are queried in the multi-level fund transfer network. The source nodes corresponding to these incoming edges are the predecessor nodes of that node. If the query result is empty, it means that the current node has no predecessor node, and the current node is determined to be the source of the path; If the query result is not empty, but all predecessor nodes are marked as top-level funding pool nodes, it is determined that the current path has been traced back to the top-level funding source. In both of the above cases, stop backtracking and treat the sequence of nodes and edges from the end node to the current stopping node recorded in the current path stack as a complete candidate influence link; The candidate impact links are validated to check whether there is a valid fund transfer relationship between adjacent nodes on the link, and invalid paths with logical contradictions are eliminated. The verified candidate impact links are formally recorded into the impact link set.

Images