A scientific research project fund whole track dynamic supervision and risk early warning system

Abstract

The present application relates to the technical field of computer data processing and information supervision, and discloses a scientific research project fund full-track dynamic supervision and risk early warning system, which comprises a state machine modeling module, a heterogeneous graph construction module, a metric parameter reconstruction module, a persistent homology evolution module and a closed-loop feedback control module.The state machine modeling module outputs the deterministic business state of scientific research entities based on clock synchronization and deterministic automata; the heterogeneous graph construction module extracts the physical features of fund flow; the metric parameter reconstruction module generates dynamic penalty coefficients and compresses the topological simplex metric parameters according to the compliance situation; the persistent homology evolution module constructs a simple complex sequence and extracts persistent barcode features to generate a one-dimensional topological ring; the closed-loop feedback control module reconstructs the entity flow loop path according to the abnormal topological ring and outputs the early warning result, and intercepts low-priority conflict requests by using the highest responsibility priority parameter, and forcibly updates the illegal entity approval state to the blocking and locking state.

Description

Technical Field

[0001] This invention relates to the field of computer data processing and information supervision technology, specifically to a dynamic monitoring and risk warning system for the entire trajectory of scientific research project funds. Background Technology

[0002] Supervision of research project funds is an important means of ensuring the compliant use of research funds. In actual supervision, relevant business data is usually scattered across multiple independent systems such as science and technology integrated service platforms and external financial institution networks. These cross-platform systems generally have the problem of asynchronous system clocks when exchanging data, which leads to time sequence deviations when summarizing government approval logs and financial transaction records. When facing concurrent operations of multiple systems, this asynchronous data phenomenon makes it impossible for the supervision system to accurately deduce the deterministic business status of research entity nodes under the absolute time of fund transactions, which is prone to misjudgment of status and makes it difficult to achieve accurate traceability of rights and responsibilities.

[0003] In terms of risk feature identification, existing regulatory technologies mostly rely on linear rule matching or conventional flow path queries. While this approach has a certain monitoring capability for direct fund violations, it is difficult to deal with hidden fund violations in scientific research projects that involve multiple nodes and multiple levels of nesting. Because conventional path query methods cannot transform business violations into structural evolution features in cyberspace, it is difficult to effectively extract multi-level fund flow loops in complex fund network structures, resulting in low accuracy in identifying complex risks.

[0004] Existing regulatory early warning mechanisms typically separate risk identification from business execution. After discovering clues of abnormal fund flows, the system often only provides early warning information to management personnel, lacking an automated closed-loop control chain that communicates with the source business state machine. Because it cannot directly trigger underlying state interception after identifying a violation risk, the system cannot block or lock the approval status of the entities involved in the violation in the first instance. As a result, it cannot effectively prevent the further transfer of illegal funds and subsequent unauthorized business approvals, resulting in regulatory lag. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a dynamic monitoring and risk warning system for the entire trajectory of scientific research project funds. It solves the problems of misjudgment of status caused by time sequence deviation of cross-platform interactive data in the existing scientific research fund monitoring process, the inability of conventional path query methods to accurately identify multi-level hidden illegal fund flow loops, and the inability to achieve automated blocking control due to the separation between risk warning and business execution links.

[0006] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a dynamic monitoring and risk warning system for the entire trajectory of scientific research project funding, comprising: a state machine modeling module, used to construct a responsibility state machine based on business process data and responsibility list data, and establish a state query mapping function to output the business approval status of scientific research entity nodes; a heterogeneous graph construction module, used to integrate government interaction data and fund flow data to construct a dynamic attribute graph containing a set of nodes and a set of edges, and extract the physical feature vector of the fund flow edges in the edge set; a metric parameter reconstruction module, used to calculate the basic interaction distance between associated nodes, combine the business approval status and responsibility list data to generate a state conflict indication function and a dynamic penalty coefficient, and calculate the basic interaction distance and the dynamic penalty coefficient to output the topological simplex metric parameter; a continuous homogeneous evolution module, used to construct a simplex sequence by introducing a filtering threshold in the topological space based on the topological simplex metric parameter, extract the birth threshold and death threshold of the one-dimensional topological loop, and generate continuous barcode features; and a closed-loop feedback control module, used to compare the continuous barcode features to filter abnormal topological loops, parse the entity flow closed-loop path mapped to the dynamic attribute graph to output the warning result, and input the forced intervention event to the state machine modeling module to update the state transition function.

[0007] The specific operational logic and evolutionary processing steps for the aforementioned dynamic monitoring and risk warning system for the entire funding trajectory of scientific research projects are as follows: The state machine modeling module uses a network time protocol server as the reference time source to perform clock synchronization calibration, correct system time differences, merge approval logs, data upload records, and transaction details, and sort them in absolute time order to construct a global time-series event queue. Based on the responsibility list data, the module constructs a deterministic finite automaton model containing state transition functions. The state query mapping function retrieves historical triggering events prior to the timestamp parameter from the global time-series event queue, outputting the deterministic business state of the research entity node at absolute time.

[0008] The heterogeneous graph construction module materializes business data through node and edge sets. The physical feature vector includes the proportion of the transaction amount corresponding to the fund flow edge to the total allocation amount in the upstream node's account, the time span from the starting node receiving funds to transferring funds to the target node, and the business action identifier. The metric parameter reconstruction module performs a negative exponential operation based on the time span feature and the time decay constant to obtain the basic distance in the time dimension. It then obtains the basic distance in the monetary dimension based on the absolute difference between the proportion feature and the numerical value. Finally, it adds the basic distance in the time dimension and the basic distance in the monetary dimension to generate the basic interaction distance.

[0009] When the metric parameter reconstruction module determines that a fund flow edge violates compliance principles, the control state conflict indicator function outputs a value of one. This value is then accumulated with the violation penalty weight to generate a dynamic penalty coefficient. The metric parameter reconstruction module divides the basic interaction distance by the dynamic penalty coefficient to obtain the topological simplex metric parameter. When a fund flow behavior violates the rights and responsibilities state machine constraints, causing the dynamic penalty coefficient to exceed one value, the topological simplex metric parameter between the corresponding associated nodes undergoes numerical compression according to the division operation rules.

[0010] The continuous homogeneous evolution module maps the node set to the topological space, and monotonically increases the filtering threshold by a preset step size. A one-dimensional simplex is generated when the topological simplex metric parameter is less than or equal to the current filtering threshold. A higher-order simplex is generated when there is a complete subgraph consisting of three or more nodes in the topological space. As the filtering threshold monotonically increases, the topological simplex metric parameter that has undergone numerical compression satisfies the judgment condition and becomes connected and closed at a smaller filtering threshold. The continuous homogeneous evolution module extracts topological structures with a one-dimensional Betti number greater than zero to construct one-dimensional topological loops. The filtering threshold when the one-dimensional topological loop forms a hole structure is recorded as the birth threshold, and the filtering threshold when a two-dimensional simplex covers the interior of the one-dimensional topological loop is recorded as the death threshold. The difference between the death threshold and the birth threshold is taken as the topological duration.

[0011] The closed-loop feedback control module sets birth and duration thresholds, and extracts one-dimensional topological loops with birth thresholds less than the birth threshold and topological durations greater than the duration thresholds as abnormal topological loops. The module then maps these abnormal topological loops back to the dynamic attribute graph to reconstruct the entity flow closed-loop path, traversing the flow path to extract the identifiers of research entity nodes associated with the illegal fund flow edges, and generating risk warning results. The state machine modeling module receives the forced intervention event from the reverse input and configures a global highest priority parameter. The state transition function intercepts low-priority conflict requests based on the highest priority parameter, forcibly updating the business approval status of the illegally associated research entity nodes to a blocked / locked state.

[0012] This invention provides a dynamic monitoring and risk early warning system for the entire trajectory of scientific research project funding. It has the following beneficial effects: 1. This invention eliminates timing deviations and misjudgments of state during cross-platform data interaction by constructing a global time-series event queue and a responsibility state machine. The system uses a network time protocol server to perform unified clock synchronization calibration on government approval and financial transaction data. Combined with a state query mapping function and a deterministic finite automaton model, it can accurately deduce the deterministic business state of scientific research entity nodes in absolute time, solving the problems of data asynchrony and difficulty in tracing responsibility states caused by concurrent operations of multiple business systems.

[0013] 2. Based on dynamic penalty coefficients and continuous coherent evolution calculations, this invention achieves accurate extraction of the closed-loop flow characteristics of concealed illegal funds. The metric parameter reconstruction module generates dynamic penalty coefficients based on business compliance verification results, and numerically compresses the basic interaction distance between related nodes, so that the network structure containing illegal behavior is connected and closed in advance during the evolution of the topology space. By calculating the Betty number and extracting the birth threshold and topology duration of the one-dimensional topological ring, the system directly captures multi-level illegal fund flow closed loops that are difficult to detect by conventional path query methods, thereby improving the accuracy of complex risk identification.

[0014] 3. Based on the reverse control of the highest priority mandatory intervention event and the state transition function, this invention constructs an automated closed-loop control chain from risk identification to business blocking. After parsing the flow entity corresponding to the abnormal topology loop, the closed-loop feedback control module encapsulates the warning signal as a mandatory intervention event and inputs it into the state machine modeling module. The state transition function intercepts low-priority requests based on the global highest priority parameter, directly updating the business approval status of the illegal associated entity to the blocking and locking status, effectively preventing further allocation of illegal funds and unauthorized business approval. Attached Figure Description

[0015] Figure 1 This is a system architecture diagram of the present invention. Detailed Implementation

[0016] The technical solutions in 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.

[0017] Please see Figure 1 This invention provides a dynamic monitoring and risk warning system for the entire trajectory of scientific research project funds. The system may include: The state machine modeling module receives business process data and responsibility list data from the science and technology integrated service platform. It maps the business process data to state nodes and the responsibility list data to state constraints to construct a responsibility state machine. The responsibility state machine is configured with a finite set of states, a set of triggering events, and state transition functions. Based on the responsibility state machine, a state query mapping function with timestamps is constructed. The state query mapping function outputs the business approval status of the scientific research entity node at each time point.

[0018] The heterogeneous graph construction module integrates government interaction data and external financial fund flow data to construct a dynamic attribute graph containing entity attributes and time-series actions. The dynamic attribute graph contains a set of nodes consisting of project teams, suppliers, and management personnel entities, as well as a set of edges consisting of fund allocation actions and business approval actions. It also extracts the physical feature vector of each fund flow edge in the edge set from the external financial fund flow data. The physical feature vector contains the transaction amount ratio feature and the fund turnover time span feature.

[0019] The metric parameter reconstruction module calculates the heterogeneous fusion metric between nodes in the dynamic attribute graph. Specifically, it calculates the basic interaction distance between related nodes based on the physical feature vector, and calls the state machine modeling module to obtain the business approval status when related nodes have fund transfers. Based on the business approval status and the rules of the list of rights and responsibilities, it calculates the state conflict indication function and generates a dynamic penalty coefficient. Then, it performs nonlinear weighting operations on the basic interaction distance and the dynamic penalty coefficient to output the topological simplex metric parameters.

[0020] The continuous homology evolution module performs continuous homology topological evolution calculations based on the topological simplex metric parameters. In the topological space composed of dynamic attribute graph nodes, an increasing filtering threshold is introduced to construct a simplex sequence, and the homology group and Betti number features of the simplex sequence are calculated to extract the one-dimensional topological ring feature set in the network. The feature parameters of the one-dimensional topological ring feature set as the filtering threshold evolves are recorded. The feature parameters include the birth threshold and the death threshold, and continuous barcode features are generated accordingly.

[0021] The closed-loop feedback control module performs topology feature parsing and reverse control of the responsibility state machine. Specifically, it compares the continuous barcode features with preset threshold judgment conditions, filters the abnormal topology loop set, parses the abnormal topology loop set to the entity flow path of the original dynamic attribute graph, extracts the state conflict indication function corresponding to the entity flow path, outputs the warning result containing business process violation information and fund flow closed-loop information, and finally inputs the abnormal topology loop signal as a forced intervention event into the state machine modeling module to update the state transition function of the corresponding scientific research entity node and perform business permission blocking operation.

[0022] Based on the overall architecture of the aforementioned embodiments, the specific operational logic of the state machine modeling module is as follows.

[0023] The state machine modeling module connects to the science and technology integrated service platform and the financial institution network. It extracts government approval logs, scientific research data upload records, and transaction details from the science and technology integrated service platform and the financial institution, respectively. It reads the timestamp fields from the government approval logs, scientific research data upload records, and transaction details, and performs clock synchronization calibration using a network time protocol server as the reference time source. It calculates and corrects the system time difference between different server nodes, and then merges the calibrated government approval logs, scientific research data upload records, and transaction details, sorts them according to absolute time order, and finally constructs a global time-series event queue.

[0024] Next, a deterministic finite automaton model is constructed based on the preset list of rights and responsibilities. The deterministic finite automaton model defines a finite set of business states. Triggering event set and state transition function , where the set of finite business states This includes the node states throughout the lifecycle of a research project. These node states are mapped to a state configuration table in the database, triggering a set of events. It includes form actions and approval operation instructions submitted by system users, and the aforementioned form actions and approval operation instructions are mapped to the event mapping table in the database.

[0025] State transition function Used to configure the association constraints between the state configuration table and the event mapping table, limiting the set of business states. The specific trigger events allowed to be received under a given business state, when the set of business states is limited. The current state receives a set of triggering events. When a valid operation instruction is executed, the state transition function is used. Update the status field of the corresponding scientific research entity node and perform a hierarchical jump in business status.

[0026] Then, a status query mapping function is established. Status query mapping function Receive scientific research entity nodes Identifier parameters and timestamp parameters Retrieve timestamp parameters from the global time-series event queue. Previous scientific research entity nodes All historical triggering events that occurred, based on the state transition function. Proceed sequentially to output scientific research entity nodes. In absolute time The current state of the business is one of certainty.

[0027] Configure a state concurrency control window during state transition. When the same research entity node is received within the state concurrency control window... When multiple state change requests are received, the priority parameters corresponding to the multiple state change requests are extracted from the event mapping table. By comparing the priority parameters, the state transition function is determined. State transitions are performed only based on the highest priority triggering event, while low-priority conflicting requests are discarded, thus maintaining the research entity nodes. The uniqueness of business status is ultimately recorded in the status query mapping function. All output results are written to the status log database, which records the research entity nodes. The business state evolution trajectory provides constraint boundary conditions for subsequent measurement operations and supports the extraction of business attribute parameters of the corresponding scientific research entity node at any time when a fund transaction occurs.

[0028] Based on the global time-series event queue and business states output by the aforementioned state machine modeling module, the heterogeneous graph construction module executes the entity definition of the dynamic attribute graph, specifically within the set observation time window. The built-in dynamic property graph includes entity attributes and temporal actions. Dynamic attribute graph In Represents a set of nodes, a set of nodes Includes a set of project research group nodes External supplier node set and management personnel node set The entity that constitutes Represents the set of edges, edge set Includes the relationships between various entities within the observation time window Directed edges for fund transfers and directed edges for business approvals occurring within the organization, for the set of edges. Any fund transfer edge The heterogeneous graph construction module extracts the capital flow edge. physical feature vector Physical feature vector In Indicates the flow of funds The corresponding transaction amount represents a percentage of the total allocation amount in the upstream node's account. Indicates the starting node Funds received to the starting node To the target node The time interval for transferring out the corresponding funds. Indicates the side accompanying the flow of funds The business action identifier declared at the time of occurrence.

[0029] The metric parameter reconstruction module reads the physical feature vector. Calculate the starting node With the target node Basic interaction distance between Basic interaction distance The calculation formula is: In the calculation formula The weight balancing parameters configured for system initialization. The time decay constant is the time decay constant when the starting node is... With the target node The time span of capital turnover between Approaching zero and the proportion of transaction amount A temporary trend, basic interaction distance Output the minimum value, the base interaction distance. The minimum value represents the starting node. With the target node There is a close connection in the capital flow chain.

[0030] The metric parameter reconstruction module calls the state query mapping function to obtain the starting node. With the target node The business approval status at the moment of fund interaction, combined with business action identifiers. Calculate the state conflict indication function State conflict indicator function Used to determine the flow of funds Does it violate item # in the list of powers and responsibilities? According to the compliance principle, when the fund transfer behavior meets the business approval status, a status conflict indicator function is used. The output value is zero. This is a state conflict indicator function that responds to unauthorized operations or irregular fund disbursements outside the approval period. The output value is one, and the metric parameter reconstruction module then uses the state conflict indication function. Generate dynamic penalty coefficient Dynamic penalty coefficient The calculation formula is: In the calculation formula This represents the total number of compliance principles in the list of powers and responsibilities. For the first The corresponding penalty weights for violations of each compliance principle For the flow of funds The absolute time of occurrence and The starting node is respectively With the target node In absolute time The business approval status.

[0031] The metric parameter reconstruction module combines basic interaction distance With dynamic penalty coefficient Perform nonlinear weighting operations to generate topological simplex metric parameters for subsequent topology calculations. Topological simplex metric parameters The calculation formula is: According to the aforementioned calculation formula, when the starting node With the target node The violation of the rights and responsibilities state machine constraints in the fund interactions between them leads to a dynamic penalty coefficient. When the value is greater than one, the topological simplex metric parameter Numerical compression occurs when the system transforms the violation status of business flow into distance reduction in mathematical metric space, thereby enabling the early connection of violation entity nodes when constructing simple complex sequences.

[0032] The aforementioned heterogeneous graph construction module and metric parameter reconstruction module output the topological simplex metric parameters. Based on this, the continuous coherence evolution module reads the dynamic attribute graph. Node set and node set Topological simplex metric parameters between any two nodes , set of nodes Mapping to the metric space to construct a basic point cloud set, and setting filtering threshold parameters. Let the filter threshold parameter Starting from zero, a monotonically increasing operation is performed with a preset step size to characterize the connectivity of the topological simplex at different distance scales.

[0033] In the filter threshold parameter During the monotonically increasing process, the continuous homogeneous evolution module constructs a Vitris-Lipps complex sequence based on the basic point cloud set, specifically determining the topological simplex metric parameters between any two nodes in the metric space. Is it less than or equal to the current filter threshold parameter? When the topological simplex metric parameter Less than or equal to the filter threshold parameter At the starting node With the target node A one-dimensional simplex is generated between them, when there exists a simplex in the metric space consisting of... When nodes form a complete subgraph that is connected pairwise, the generated dimension is... The higher-order simplex, here For natural numbers greater than or equal to three, the filtering threshold parameter varies. As the value of increases, lower-order simplexes connect and gradually form higher-order simplexes, thus the output varies with the filter threshold parameter. Evolved Vitry-Lips complex sequence.

[0034] The continuous homology evolution module performs homology group algebra operations on the Vitry-Lips complex sequence to calculate the Vitry-Lips complex sequence at various filtering threshold parameters. The Betty number features include zero-dimensional Betty numbers and one-dimensional Betty numbers. One-dimensional Betty number The number of independent one-dimensional topological rings in the characterization metric space is used to extract the one-dimensional Betty number using the continuous homology evolution module. A topology with values ​​greater than zero is used to construct a one-dimensional topological ring feature set in the network.

[0035] For each one-dimensional topological ring in the feature set of one-dimensional topological rings The continuous homology evolution module extracts one-dimensional topological rings. Lifecycle parameters during the evolutionary process record a one-dimensional topological cycle. The filtering threshold parameter corresponding to the formation of a one-dimensional porous structure through interconnection. As a birth threshold Record a one-dimensional topological ring The corresponding filtering threshold parameter when the interior is covered by a two-dimensional simplex As a death threshold Extract birth threshold With the death threshold Calculate the topology duration, where the topology duration is the death threshold. With birth threshold The difference, the continuous homogeneous evolution module is based on the birth threshold of all one-dimensional topological rings in the feature set of one-dimensional topological rings. With the death threshold A persistent barcode feature is generated, which is used to characterize the stability of the closed-loop transaction structure in the time-series funding network. The persistent barcode feature is then output to the closed-loop feedback control module to perform subsequent risk analysis.

[0036] Based on the continuous barcode features output by the aforementioned continuous coherence evolution module, the closed-loop feedback control module reads each one-dimensional topological loop in the continuous barcode features. Birth threshold With the death threshold Set birth determination threshold Duration determination threshold Calculate a one-dimensional topological ring Topological duration Topological duration The calculation formula is: Based on the computational logic of numerical compression of topological simplex metric parameters in the metric parameter reconstruction module, a one-dimensional topological loop containing violations will prematurely close under a very small filtering threshold parameter. The closed-loop feedback control module then extracts the birth threshold accordingly. Less than the birth determination threshold And topological duration Greater than the duration threshold One-dimensional topological rings are used as anomalous topological rings, and all the selected anomalous topological rings are used to construct an anomalous topological ring set.

[0037] For any anomalous topological loop in the set of anomalous topological loops, the closed-loop feedback control module reverse-maps the topological simplex contained in the anomalous topological loop to the dynamic property graph. In the process, the node sequence and edge sequence constituting the abnormal topological loop are extracted to reconstruct the corresponding entity flow closed-loop path. The closed-loop feedback control module traverses each capital flow edge in the entity flow closed-loop path. Call the metric parameter reconstruction module to obtain the capital flow edge Corresponding state conflict indication function According to the state conflict indication function Filter out edges with a value of 1 indicating irregular fund transfers, and extract the identifiers of the research entity nodes associated with these edges, as well as the violation and compliance principle serial numbers. It integrates the identity data of scientific research entities and the records of violations of rights and responsibilities rules in the closed-loop path of entity circulation, and generates risk warning results that include information on violations of business processes and information on the closed loop of fund circulation.

[0038] The closed-loop feedback control module encapsulates the abnormal topological loop signal in the risk warning result into a mandatory intervention event, and inputs the mandatory intervention event back into the global time-series event queue of the state machine modeling module. The state machine modeling module receives the mandatory intervention event, extracts the scientific research entity node identifier associated with the illegal fund transfer edge, configures the mandatory intervention event to have the highest global priority parameter, and sets the state transition function. Based on the highest priority parameter of global authority and responsibility, low-priority conflict requests are intercepted, and the business approval status of the corresponding scientific research entity node is forcibly updated to the blocked and locked status. Subsequent approval and fund transfer instructions initiated by the non-compliant scientific research entity node in the science and technology integrated service platform and financial institution network are rejected, thereby realizing closed-loop control of the authority and responsibility status of scientific research projects and dynamic self-locking of business timing constraints.

[0039] 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 dynamic monitoring and risk early warning system for the entire trajectory of scientific research project funding, characterized in that, include: The state machine modeling module is used to construct a responsibility state machine based on business process data and responsibility list data, and to establish a state query mapping function to output the business approval status of scientific research entity nodes. The heterogeneous graph construction module is used to integrate government interaction data and fund flow data to construct a dynamic attribute graph containing a set of nodes and a set of edges, and to extract the physical feature vectors of the fund flow edges in the set of edges; The metric parameter reconstruction module is used to calculate the basic interaction distance between associated nodes, combine the business approval status and the list of rights and responsibilities data to generate a status conflict indication function and a dynamic penalty coefficient, and perform calculations on the basic interaction distance and the dynamic penalty coefficient to output the topological simplex metric parameters. The continuous homology evolution module is used to construct a simplex sequence by introducing a filtering threshold in the topological space based on the topological simplex metric parameters, extract the birth threshold and death threshold of the one-dimensional topological loop, and generate continuous barcode features. The closed-loop feedback control module is used to compare the abnormal topological loop of the continuous barcode feature screening, parse the entity flow closed-loop path mapped to the dynamic attribute graph to output the early warning result, and input the forced intervention event to the state machine modeling module to update the state transition function.

2. The research project funding dynamic monitoring and risk early warning system according to claim 1, characterized in that, The state machine modeling module is specifically used for: Extract the timestamp field from the approval log, data upload records, and transaction details; Perform clock synchronization calibration using a Network Time Protocol (NTP) server as the reference time source to correct system time differences; The approval log, the data upload record, and the transaction details after calibrating the system time difference are merged and sorted in absolute time order to construct a global time-series event queue.

3. The research project funding dynamic monitoring and risk early warning system according to claim 2, characterized in that, The state machine modeling module constructs a deterministic finite automaton model containing the state transition function based on the list of responsibilities data; The state query mapping function retrieves all historical triggering events of the scientific research entity node before the timestamp parameter in the global time-series event queue, and outputs a deterministic business state based on the state transition function; When multiple state change requests are received from the same research entity node, the state transition function performs a state transition based on the triggering event corresponding to the highest priority parameter, discarding low-priority conflict requests.

4. The research project funding dynamic monitoring and risk early warning system according to claim 1, characterized in that, The node set includes entities consisting of project teams, external suppliers, and management personnel; The physical feature vector includes: the proportion of the transaction amount corresponding to the fund transfer edge to the total allocation amount of the upstream node's account, the time span from when the starting node receives the funds to when the starting node transfers the corresponding funds to the target node, and the business action identifier reported at the moment when the fund transfer edge occurs.

5. The research project funding dynamic monitoring and risk early warning system according to claim 4, characterized in that, The specific steps for the metric parameter reconstruction module to calculate the basic interaction distance are as follows: After multiplying the time span feature by the time decay constant, perform a negative exponential operation, subtract the result from the value, and multiply by the weight balance parameter to obtain the basic distance in the time dimension. Calculate the absolute difference between the proportional feature and the value one, and multiply the absolute difference by the complementary value obtained by subtracting the weight balance parameter from the value one to obtain the basic distance of the monetary dimension. The basic interaction distance is obtained by adding the basic distance in the time dimension to the basic distance in the monetary dimension.

6. The research project funding dynamic monitoring and risk early warning system according to claim 4, characterized in that, The metric parameter reconstruction module is specifically used for: Determine whether the fund transfer violates the compliance principles in the list of rights and responsibilities data. When the fund transfer behavior conforms to the business approval status, the status conflict indicator function outputs a value of zero; when a violation occurs, it outputs a value of one. The dynamic penalty coefficient is generated by multiplying the state conflict indication function by the violation penalty weight, summing the product results under all compliance principles, and then adding the result to the value. Divide the basic interaction distance by the dynamic penalty coefficient to obtain the topological simplex metric parameter.

7. The research project funding dynamic monitoring and risk early warning system according to claim 1, characterized in that, The continuous coherence evolution module is specifically used for: The node set is mapped to the topological space to construct a basic point cloud set, and the filtering threshold is monotonically increased from zero with a preset step size. Determine whether the topological simplex metric parameter between any two nodes is less than or equal to the current filtering threshold. If the condition is met, generate a one-dimensional simplex between the starting node and the target node.

8. The research project funding dynamic monitoring and risk early warning system according to claim 7, characterized in that, When there exists a complete subgraph in the topological space consisting of three or more nodes connected in pairs, a higher-order simplex with a dimension of the total number of nodes minus one is generated, and the simplex sequence that evolves with the filtering threshold is output. Calculate the Betti number features of the simple complex sequence, extract the topological structures with a one-dimensional Betti number greater than zero, and construct the one-dimensional topological ring; The filtering threshold corresponding to the formation of a hole structure by connecting the one-dimensional topological ring is recorded as the birth threshold, and the filtering threshold corresponding to the formation of a hole structure by connecting the one-dimensional topological ring is recorded as the death threshold.

9. A dynamic monitoring and risk warning system for the entire trajectory of scientific research project funds according to claim 8, characterized in that, The closed-loop feedback control module is specifically used for: The one-dimensional topological loops where the birth threshold is less than the birth determination threshold and the difference between the death threshold and the birth threshold is greater than the duration determination threshold are extracted as the abnormal topological loops; The abnormal topological loop is reverse-mapped onto the dynamic attribute graph, and the node sequence and edge sequence constituting the abnormal topological loop are extracted to reconstruct the entity flow closed-loop path.

10. A dynamic monitoring and risk warning system for the entire trajectory of scientific research project funds according to claim 9, characterized in that, The closed-loop feedback control module traverses the fund flow edges in the closed-loop path of the entity flow, extracts the scientific research entity node identifier and the violation compliance principle sequence number associated with the violation fund flow edge according to the state conflict indication function, and generates the warning result containing business process violation information and fund flow closed-loop information. The state machine modeling module receives the forced intervention event as a reverse input, configures the forced intervention event to have the highest global priority parameter, and forcibly updates the business approval status of the scientific research entity node associated with the illegal fund transfer edge to the blocked and locked state through the state transition function.

Images