Industry decision support method based on coupling of big data and knowledge base

Abstract

This invention relates to the field of decision support technology and discloses an industry decision support method based on the coupling of big data and knowledge base. The method includes: collecting streaming business big data and mapping it to a time-series knowledge graph to determine active nodes; generating a business association deviation vector based on the deviation between business flow frequency and business dependency weights; obtaining cash asset reserves and available credit balances and adding them to determine the total liquidity buffer, and constructing a risk mitigation coefficient based on the average default penalty ratio; adjusting the magnitude of the business association deviation vector using the reciprocal of the risk mitigation coefficient to obtain the topology evolution increment; generating predicted association edges with time decay attributes based on the topology evolution increment, and locking the first-order adjacency edges of the active nodes to achieve local topology dynamic synchronization. This invention reconstructs the data processing mechanism through business logic, accurately captures the asynchronous state of static rules and dynamic data logic, and effectively distinguishes between normal business fluctuations and substantial risk shocks.

Description

Technical Field

[0001] This invention belongs to the field of decision support technology, and in particular relates to an industry decision support method based on the coupling of big data and knowledge base. Background Technology

[0002] Currently, the supply chain finance decision support field generally adopts a data processing architecture that couples industry knowledge bases with streaming business big data. It uses static rules as prior logical constraints and extracts real-time transaction features to input into predictive models to calculate the probability of default risk of business entities. Under the existing technical framework, there is a significant temporal asynchrony contradiction between the low-frequency updated knowledge system and the high-frequency fluctuating streaming data. When dealing with locally clustered abnormal transaction signals, conventional solutions implicitly presuppose the technical assumption of frictionless risk transmission, abstracting business entities into theoretical calculation nodes that do not have risk absorption capabilities. However, in reality, core enterprises or suppliers in the supply chain generally have risk tolerance constituted by working capital reserves and grace periods of commercial contracts. Existing processing systems strip away the logical damping characteristics of this financial dimension, and when faced with sudden changes in local data flow, the graph is prone to oversensitive global topological oscillations.

[0003] Besides the bias in the modeling logic regarding the abstraction of business entities, existing algorithmic risk identification and transmission mechanisms are inadequate. For example, Chinese invention patent application CN120088071A discloses a big data mining method and system for supply chain finance, which uses a pre-trained temporal graph convolutional network to aggregate multimodal features and generates a risk transmission topology graph based on the fluctuation of transaction intensity weights. This technical solution is data-driven at the mechanism level, and the risk identification logic relies on the transient mathematical fluctuations of feature components. It does not establish a regulatory mechanism to characterize the financial resilience of business entities. When faced with local data aggregation caused by non-risk factors such as payment period adjustments or settlement delays, the system lacks logical damping, and the graph topology structure oscillates violently in response to short-term noise, causing massive compliance false alarms. It is difficult to balance prediction accuracy and system processing efficiency.

[0004] Therefore, the technical problem to be solved by this invention is how to construct a local map reconstruction mechanism driven by streaming data and with commercially rigid hysteresis characteristics, so as to accurately identify substantive risks while resisting transient noise interference. Summary of the Invention

[0005] This invention provides an industry decision support method based on the coupling of big data and knowledge base, comprising the following steps:

[0006] Step S1: Acquire streaming business big data in real time and map the streaming business big data to a time-series knowledge graph to determine the active nodes in the time-series knowledge graph;

[0007] Step S2: Calculate the frequency of business flow generated by streaming business big data in the local neighborhood of the activated node, and perform a difference calculation between the frequency of business flow and the preset business dependency weight to generate a business association deviation vector.

[0008] Step S3: Obtain the cash asset reserve data and available credit balance of the business entity corresponding to the activated node, add the cash asset reserve data and available credit balance to determine the total liquidity buffer, and construct a risk resistance coefficient that represents the credit risk hedging ability in combination with the average default penalty ratio.

[0009] Step S4: Compare the magnitude of the business association deviation vector with the preset graph topology steady-state threshold. When the magnitude exceeds the preset graph topology steady-state threshold, use the reciprocal of the risk resistance coefficient as an adjustment ratio to reduce the magnitude of the business association deviation vector by weight, and obtain the topology evolution increment that characterizes the fluctuation intensity of the graph topology structure.

[0010] Step S5: Based on the topology evolution increment, predictive association edges with time decay attributes are generated in the local association topology domain of the activated node, and the first-order adjacent edges of the activated node are locked to prevent the weight update signal from spreading outward through the local neighborhood of the activated node, thereby realizing the dynamic synchronization of the local topology structure of the time-series knowledge graph and outputting the decision support results.

[0011] Preferably, step S3 further includes the following sub-steps: Step S31, read the cash and cash equivalents reserve value and the unused bank credit line of the commercial entity through the data interface, and add the cash and cash equivalents reserve value and the bank credit line to calculate the total liquidity buffer representing the short-term liquidity level; Step S32, extract the average default penalty ratio of the commercial entity's historical contracts as a weighting factor, and convert the total liquidity buffer into a risk resistance coefficient through a preset mapping function, wherein the risk resistance coefficient is positively correlated with both the total liquidity buffer and the average default penalty ratio of historical contracts.

[0012] Preferably, step S2 specifically includes: extracting transient components representing transaction activity from streaming business big data and determining the projection direction of the transient components in the temporal knowledge graph topology space; using the projection direction to perform directional weighting on the degree of deviation, generating a business association deviation vector with spatial orientation and amplitude attributes.

[0013] Preferably, step S4 further includes the following sub-steps: step S41, calculating the excess deviation value of the business association deviation vector above the preset graph topology steady-state threshold; step S42, multiplying the excess deviation value with the reciprocal of the risk resistance coefficient to obtain the mapped adjustment amount; step S43, determining the mapped adjustment amount as the topology evolution increment to limit the impact intensity of local data mutations on the graph topology structure.

[0014] Preferably, locking the first-order adjacent edges of the active node to prevent the weight update signal from spreading outward through the local neighborhood of the active node includes: configuring a weight update lock on the first-order adjacent edges of the active node; and restricting the transmission of topology evolution increments to the non-active regions of the temporal knowledge graph through the local associated topology domain during dynamic synchronization, thereby achieving local logical isolation of risk impacts.

[0015] Preferably, after dynamic synchronization, the method further includes: monitoring the weight evolution trend of the predicted associated edges; when the weight disappearance rate is lower than the preset decay benchmark and the duration exceeds the observation threshold, modifying the attribute of the predicted associated edges to a persistent topological edge and storing it in the static industry rule base.

[0016] Preferably, the output decision support results include: calculating the graph Laplacian operator eigenvalues ​​of the synchronized local topology; assessing the change in the default probability of business entities based on the drift of the eigenvalues; and outputting a risk penetration prediction report.

[0017] Preferably, the activated nodes include at least core enterprise nodes, supplier nodes, logistics nodes, and capital flow nodes; the preset business dependency weights are determined based on the historical settlement cycle and order coupling degree between nodes.

[0018] Preferably, the decision support results also include: when multiple industry rules are identified to conflict, the dynamically synchronized local topology is invoked to perform a logical consistency comparison, and the highest priority risk handling strategy instruction is determined and output.

[0019] Compared to existing technologies, the industry decision support method based on the coupling of big data and knowledge base in this invention has the following advantages:

[0020] 1. In industry decision support methods, based on the relation stress vector generated by streaming business big data in the local neighborhood of the activated node in the time-series knowledge graph, and combined with the comparison results of the magnitude of the relation stress vector with the preset graph topology steady-state threshold, the mechanism can accurately capture the logical asynchronous state between the static industry rule base and the dynamic transient data stream. This mechanism provides the decision support system with transient feedback reflecting the changes in the business environment by quantifying the deviation between the actual business flow frequency and the preset business dependency weight, thus alleviating the prediction blind spot problem caused by the traditional batch processing update strategy.

[0021] 2. By extracting the short-term available funds and credit balance of the business entities corresponding to the activated nodes, the total liquidity buffer is determined, and a business rigidity coefficient is constructed by combining the average default penalty ratio. This introduces logical damping that conforms to objective business laws into the topological evolution of the knowledge graph. This method of deeply coupling financial indicators with the stress buffer pool accumulation mechanism enables the system to effectively distinguish between normal cash flow fluctuations and substantial risk shocks, avoid overly sensitive topological oscillations caused by the aggregation of local data flows, and reduce the compliance false alarm rate of the decision support system for short-term fluctuations in the supply chain.

[0022] 3. By utilizing the generated fictitious association edges with time decay properties, and in conjunction with blocking the backpropagation path of the edge weights in the first-order Markov blanket graph of the activated node, dynamic in-situ reconstruction of the local topology of the knowledge graph is achieved. The time decay factor adjusted based on the credit rating mean ensures that the temporarily generated business associations gradually weaken over time. This ensures that the prediction model can use dynamic prior knowledge to deal with undefined business behaviors, while preventing transient abnormal data from causing permanent logical pollution to the underlying industry common sense base. Attached Figure Description

[0023] Figure 1 This is the core steps and execution flowchart of the industry decision support method of this invention;

[0024] Figure 2 This is a logical diagram of data flow and topology update within the decision-making system of this invention. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0026] It should be noted that all directional and positional terms used in this invention, such as: up, down, left, right, front, back, vertical, horizontal, inner, outer, top, bottom, transverse, longitudinal, center, etc., are only used to explain the relative positional relationship and connection between components in a specific state (as shown in the accompanying drawings). They are only for the convenience of describing this invention and do not require that this invention be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention. In addition, the descriptions of "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated.

[0027] In the description of this invention, unless otherwise explicitly specified and limited, the terms installation, connection, and linking should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections; they can refer to direct connections or indirect connections through an intermediate medium; they can refer to the internal connection of two components. For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances.

[0028] In the description of this specification, references to the terms "an embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example, and the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0029] An industry decision support method based on the coupling of big data and knowledge base includes the following steps:

[0030] Step S1: Acquire streaming business big data in real time and map the streaming business big data to a time-series knowledge graph to determine the active nodes in the time-series knowledge graph;

[0031] Step S2: Calculate the frequency of business flow generated by streaming business big data in the local neighborhood of the activated node, and perform a difference calculation between the frequency of business flow and the preset business dependency weight to generate a business association deviation vector.

[0032] Step S3: Obtain the cash asset reserve data and available credit balance of the business entity corresponding to the activated node, add the cash asset reserve data and available credit balance to determine the total liquidity buffer, and construct a risk resistance coefficient that represents the credit risk hedging ability in combination with the average default penalty ratio.

[0033] Step S4: Compare the magnitude of the business association deviation vector with the preset graph topology steady-state threshold. When the magnitude exceeds the preset graph topology steady-state threshold, use the reciprocal of the risk resistance coefficient as an adjustment ratio to reduce the magnitude of the business association deviation vector by weight, and obtain the topology evolution increment that characterizes the fluctuation intensity of the graph topology structure.

[0034] Step S5: Based on the topology evolution increment, predictive association edges with time decay attributes are generated in the local association topology domain of the activated node, and the first-order adjacent edges of the activated node are locked to prevent the weight update signal from spreading outward through the local neighborhood of the activated node, thereby realizing the dynamic synchronization of the local topology structure of the time-series knowledge graph and outputting the decision support results.

[0035] Preferably, step S3 further includes the following sub-steps: Step S31, read the cash and cash equivalents reserve value and the unused bank credit line of the commercial entity through the data interface, and add the cash and cash equivalents reserve value and the bank credit line to calculate the total liquidity buffer representing the short-term liquidity level; Step S32, extract the average default penalty ratio of the commercial entity's historical contracts as a weighting factor, and convert the total liquidity buffer into a risk resistance coefficient through a preset mapping function, wherein the risk resistance coefficient is positively correlated with both the total liquidity buffer and the average default penalty ratio of historical contracts.

[0036] Preferably, step S2 specifically includes: extracting transient components representing transaction activity from streaming business big data and determining the projection direction of the transient components in the temporal knowledge graph topology space; using the projection direction to perform directional weighting on the degree of deviation, generating a business association deviation vector with spatial orientation and amplitude attributes.

[0037] Preferably, step S4 further includes the following sub-steps: step S41, calculating the excess deviation value of the business association deviation vector above the preset graph topology steady-state threshold; step S42, multiplying the excess deviation value with the reciprocal of the risk resistance coefficient to obtain the mapped adjustment amount; step S43, determining the mapped adjustment amount as the topology evolution increment to limit the impact intensity of local data mutations on the graph topology structure.

[0038] Preferably, in step S5, the weights of the predicted associated edges with time decay properties are... It conforms to the following quantification rules: ,in, To predict the initial weight values ​​of associated edges, The time decay factor is adjusted for the mean of the activated node's credit rating. To predict the duration of associated edges after they are generated.

[0039] Preferably, locking the first-order adjacent edges of the active node to prevent the weight update signal from spreading outward through the local neighborhood of the active node includes: configuring a weight update lock on the first-order adjacent edges of the active node; and restricting the transmission of topology evolution increments to the non-active regions of the temporal knowledge graph through the local associated topology domain during dynamic synchronization, thereby achieving local logical isolation of risk impacts.

[0040] Preferably, after dynamic synchronization, the method further includes: monitoring the weight evolution trend of the predicted associated edges; when the weight disappearance rate is lower than the preset decay benchmark and the duration exceeds the observation threshold, modifying the attribute of the predicted associated edges to a persistent topological edge and storing it in the static industry rule base.

[0041] Preferably, the output decision support results include: calculating the graph Laplacian operator eigenvalues ​​of the synchronized local topology; assessing the change in the default probability of business entities based on the drift of the eigenvalues; and outputting a risk penetration prediction report.

[0042] Preferably, the activated nodes include at least core enterprise nodes, supplier nodes, logistics nodes, and capital flow nodes; the preset business dependency weights are determined based on the historical settlement cycle and order coupling degree between nodes.

[0043] Preferably, the decision support results also include: when multiple industry rules are identified to conflict, the dynamically synchronized local topology is invoked to perform a logical consistency comparison, and the highest priority risk handling strategy instruction is determined and output.

[0044] Example 1: The application environment of the method of the present invention is a risk monitoring system for a large-scale electronics manufacturing enterprise and its suppliers. It collects streaming commercial big data containing transaction records and contract change information, maps this data to a time-series knowledge graph containing enterprise entity nodes and supply agreement edges, identifies active nodes, calculates the business flow frequency of the active node within a preset time window, and performs a difference operation between the business flow frequency and a preset business dependency weight to generate a business association deviation vector. Specifically, the system extracts a one-dimensional unit topology vector from the active node to the associated node based on the adjacency matrix of the time-series knowledge graph. The scalar difference calculated between the actual business flow frequency and the preset business dependency weight is used as a scalar multiplier and multiplied by the unit topology vector. The topological vector is multiplied by scalar. If the scalar difference is positive, it indicates an overflow of business volume, and the original unit topological vector is kept pointing unchanged. If the scalar difference is negative, it indicates a shrinkage of business volume, and the unit topological vector is reversed. This transforms the scalar difference without direction into a business association deviation vector with source-sink node pointing characteristics in the graph topological space. The cash asset reserves of the business entity and the unused bank credit line are read, and the cash asset reserves and bank credit line are added to determine the total liquidity buffer. The average default penalty ratio between nodes is extracted as a weighting factor, and the total liquidity buffer and the average default penalty ratio are positively correlated and transformed through a mapping function to construct a risk resistance coefficient.

[0045] Under extreme commercial liquidation conditions, the cash assets and available bank credit at a specific activated node are simultaneously depleted, and the total liquidity buffer parameter approaches zero. Based on the overflow prevention boundary control principle of computer arithmetic logic units, the system incorporates an extreme value bottom-line protection mechanism at the interface between the risk mitigation unit and the topology reconstruction unit to truncate the code link that triggers the underlying division-by-zero exception. Solve for the weight adjustment parameters of the actual delivered lower level, and calculate the variables within the computational structure. The output is a dimensionless adjustment ratio for use by the topology evolution module; the input items are... Characterized by the initial risk mitigation coefficient passed from the previous level numerical mapper, a constant. The smallest positive real number is set as the constraint floating-point register to calculate the baseline to ensure that the denominator is always greater than zero; the magnitude of the business association deviation vector is compared with the preset graph topology steady-state threshold. When the magnitude exceeds the preset graph topology steady-state threshold, the reciprocal of the risk resistance coefficient is used as an adjustment ratio to reduce the magnitude of the business association deviation vector, determine the topology evolution increment, and based on the topology evolution increment, predictive association edges are generated in the local association topology domain of the activated node. The weight of the predictive association edges is... Satisfying the formula: ,in, To predict the real-time weights of associated edges, To predict the initial weight values ​​of associated edges, The time decay factor is adjusted for the mean of the activated node's credit rating. To predict the duration of associated edges after they are generated.

[0046] The system locks the first-order adjacent edges of the active nodes to restrict the propagation of topology evolution increments to inactive regions of the temporal knowledge graph, synchronizing the local topology of the temporal knowledge graph. During local topology synchronization, the system continuously monitors the temporal decay status of predicted associated edges. When the vanishing rate of their weights is lower than a preset decay benchmark and the duration exceeds an observation threshold, the system triggers the ontology alignment module. This module extracts the industry classification codes and business transaction semantics of the entities at both ends of the predicted edge and performs triple similarity retrieval in the static industry rule base. When it is identified that the predicted edge and the existing static rules do not have ontology logical mutual exclusion, the system directly overwrites the attributes of the predicted edge as a persistent topology edge for long-term storage. If a logical conflict is identified, the system forces a conflict comparison. The actual flow frequency of the link over the past six months is used to overwrite the original static rules of failure with high-frequency real data as the truth value, so as to realize the smooth closed loop of the graph evolution logic. The graph Laplacian operator eigenvalues ​​of the local topology after synchronization are calculated. The change value of the default probability of the business entity is evaluated according to the drift of the eigenvalues. Based on the sampling theorem and the basic principles of graph representation learning in the field of signal processing, converting discrete event streams into equally spaced continuous signals and establishing an entity coordinate system is the physical premise for performing subsequent frequency domain extraction functions. The system sets a constant width sliding observation window, continuously segments discrete transfer and performance logs in streaming commercial big data along the time axis, accumulates and calculates the frequency of interactive behaviors within a single time window, and outputs a continuous activity time series signal with a fixed sampling rate.

[0047] In this physical transformation process, the system abstracts discrete order transaction timestamps into a Dirac function sequence with time pulse characteristics. Through discrete convolution integration with a sliding observation window function of constant width, the discontinuous random point process is reconstructed into an equally spaced band-limited signal satisfying the Shannon sampling theorem. Through this reconstruction mechanism, the transaction flow, which originally lacked continuous fluctuation in the time domain, is transformed into an amplitude sequence containing definite frequency components. This supports the subsequent legitimate invocation of high-pass algorithms such as Chebyshev digital filters at the underlying mathematical logic level to filter out low-frequency baseline fluctuations reflecting regular payment periods, avoiding filtering failures due to mismatches in the mathematical properties of the original signal. The system then initiates a random walk node mapping model, extracts the connection density distribution matrix of each activated node and its nearest neighbors in the time-series knowledge graph, calculates the continuous real coordinate components of all nodes in the multidimensional Euclidean space, and establishes a graph topology space with geometric distance measures and direction vector dimensions. In this model operation mechanism, the system sets a mapping objective function... To maximize the co-occurrence probability of actual transaction-related nodes within the local neighborhood of the graph topology, the core walk transition probability is dynamically determined by the scalar product of the magnitude of the business association deviation vector and the preset business dependency weights. This guides the random walk path to preferentially tilt towards nodes with strong financial connections. Based on the aforementioned transition probability, the system performs multiple biased samplings of fixed length to generate node transaction sequences. Using a word vectorization embedding algorithm, the system performs dimensionality reduction on the transaction sequence through a single hidden layer neural network, mapping the discrete entity network into continuous real-valued coordinate components in each dimension. To determine the projection direction of the business association deviation vector in the temporal knowledge graph topology space, the system uses a high-pass filtering algorithm to extract the transient components representing transaction activity from streaming business big data. By calculating the business traffic distribution gradient of the transient components on each adjacent edge of the activated node, the resultant force direction of the business traffic distribution gradient is determined as the projection direction, and a direction weighting operation is applied. This allows the finally generated business association deviation vector to quantitatively represent the diffusion trend of risk signals in the graph topology dimension.

[0048] Example 2: The verification environment for the method of this invention uses a simulation system containing 500 enterprise entity nodes and 2000 supply agreement edges. The data source is an electronic manufacturing industry supply relationship dataset generated based on the Monte Carlo algorithm. When the system acquires streaming commercial big data, Gaussian white noise with a signal-to-noise ratio of 20dB and power frequency interference with a frequency of 50Hz are superimposed on the original signal. A preset time window is used. The setting is based on the transient rate of change of business flow frequency. When the rate of change exceeds a preset threshold, the system reduces... To cover high-frequency fluctuations, the test was conducted over a 24-hour period, with a preset graph topological steady-state threshold. The method for determining the threshold is to extract three times the standard deviation of the magnitude of the business correlation deviation vector within the historical stable period. It should be noted that the threshold quantization procedure based on three times the standard deviation is only specific to the current embodiment of the electronic manufacturing industry scenario where the order frequency follows an approximately Gaussian stationary distribution. To resolve the issue of threshold failure caused by cross-industry feature offset, the system has a threshold equation adaptive routing module nested at the bottom layer: when a strong trend time series caused by exchange rates is identified in the cross-border trade scenario, the system automatically cuts off the standard deviation statistical caliber and switches to the method of solving the second derivative of the business correlation deviation vector to capture the mutation rate; while in the mixed leasing scenario with huge data noise, the system calls up the second moment measure as the energy bias parameter to perform the judgment, ensuring that the physical measurement method of the steady-state threshold always dynamically matches the statistical feature essence of the current input big data.

[0049] The validation groups were determined to be an experimental group, control group 1, and control group 2. The experimental group adopted the complete logic of the method of this invention, control group 1 adopted a fixed-cycle batch processing mode, and control group 2 removed the risk mitigation coefficient adjustment step when calculating the topology evolution increment. The experiment set gradients of 10%, 30%, 50%, and 80% for the fluctuation intensity of business flow frequency. When the fluctuation intensity was between 10% and 30%, the prediction accuracy deviation of the three groups was within 5%. When the fluctuation intensity increased to 50%, the accuracy of control group 1 decreased to 65.2% due to the lag in static knowledge base updates. Control group 2, lacking a damping adjustment mechanism based on risk mitigation coefficients, experienced a false alarm rate of 28.5%. The experimental group, by adding cash asset reserves to unused bank credit lines to obtain the total liquidity buffer, maintained a prediction accuracy of 94.8% with a false alarm rate below 3.2% under 80% volatility conditions. The system acquired streaming business big data with added noise, containing random pulses with amplitudes ranging from 0.15 to 0.30. Control group 2's calculated business correlation deviation vector magnitude frequently exceeded the preset graph topological steady-state threshold. Discontinuous graph reconstruction signals are generated. The experimental group extracts the cash reserves of 5 million yuan and the credit balance of 2 million yuan from the activated nodes and calculates the total liquidity buffer. A risk mitigation coefficient is determined based on an average default penalty ratio of 15%. The system uses the reciprocal of this coefficient to process the magnitude of the business correlation deviation vector. This mechanism eliminates random fluctuations caused by 20dB noise, ensuring that the topology evolution increment only responds to substantive order defaults. When the risk mitigation coefficient is in the range of 0.8 to 1.2, the default risk identification delay is less than 150ms; the weights of the predicted associated edges are also analyzed. Follow the time decay rule ,in, To predict the real-time weights of associated edges, These are the initial weight values. The time decay factor is adjusted for the mean of the credit rating. To predict the duration of associated edges after their generation, experimental data show that the weights of the predicted associated edges change over time within 48 hours. The increase in risk leads to an exponential decrease, reflecting the convergence of business risk over time. The degree of drift of the graph Laplace operator eigenvalues ​​of the synchronized local topology is used to characterize the change in the default probability of the business entity.

[0050] Example 3: The method of this invention is applied to a risk penetration prediction system for cross-border electronic trade settlement. The system accesses a static industry knowledge base containing 500,000 supplier entities. The computing platform has a floating-point operation capability of 10 TFLOPS per second. The system uses the average monthly settlement amount over the past 12 months to determine the preset business dependency weights. The calculation formula is as follows: ,in, As the initial weights, This represents the average traffic volume per individual node. The system determines the total business volume of the industry; it synchronously determines the process judgment quantification procedure, and calculates the second derivative of the business correlation deviation vector during historical stable periods and extracts the fluctuation amplitude corresponding to the 95% confidence interval, thereby determining the preset graph topology steady-state threshold. The system acquires asynchronous streaming commercial big data, uses a semantic recognition unit to extract taxpayer identification numbers and bank account numbers from transaction records, converts them into unique hash indexes, and compares them with existing nodes in the time-series knowledge graph. When a hash index match is successful, the system marks the corresponding entity as an active node, then extracts the active node's cash asset reserve data of 20 million yuan and available credit balance of 5 million yuan, calculates the total liquidity buffer, sets the search depth of the local related topology to second-order adjacency edges, retrieves the average default penalty ratio of 12% from historical contracts to determine the risk resistance coefficient, deploys weight update locks on the first-order adjacency edges of the active node to restrict the propagation of topology evolution increments to inactive subgraphs, and calculates the Laplace matrix of the reconstructed local subgraph. ,in, For Laplace matrix, For degree matrix, Given an adjacency matrix containing predicted associated edge weights, the system extracts the Laplace matrix. The eigenvalue sequence is analyzed, and its Euclidean distance relative to the previous time step is calculated to determine the eigenvalue drift. When the eigenvalue drift When the risk exceeds the preset change threshold for three consecutive time windows, the system issues a risk penetration instruction and determines the change value of the default probability.

[0051] Based on the principle of logistic regression probability transformation, the system introduces a nonlinear mapping operator to force the dimensionless geometric anomalies of the graph structure to converge to the standardized probability statistical interval. The system extracts the eigenvalue drift parameters of the computation processor and substitutes them into the transformation formula. Execute the solution command, parameters The final output is a scalar representing the change in the probability of default, with values ​​bounded between 0 and 1. The operational variables are... The multiplier factor replaces the dimensionless eigenvalue drift input read by the preceding module. The calibration factor is a positive real-valued amplification factor obtained by fitting and calibrating an array of anonymized contract default records from the past five years in the industry, with a constant term. The system is configured with a fixed bias value generated by referencing the current macroeconomic credit benchmark interest rate. With 30% of companies in the supply chain simultaneously experiencing order changes, the system response time is 85ms. The accuracy rate of its penetrating prediction of default risk from second-tier and higher suppliers is 96.1%. A linear transformation function is applied when constructing the risk resilience coefficient. The total liquidity buffer is mapped to the average default penalty ratio to a physical damping value, where As a risk resistance coefficient, and This is a sensitivity constant preset based on the risk tolerance of a specific industry. For the total amount of liquidity buffer, To determine the average default penalty ratio, the system transforms the financial resilience of business entities into logical operators for constraint graph topological fluctuations through this parameterized correlation representation. This ensures that the incremental topological evolution only reflects substantial shocks exceeding the entity's capacity to withstand them. During the parameter calibration phase of the system's actual deployment, the computing kernel extracts the industry's supply chain default blacklist database and corresponding financial statements from the past sixty months as supervised learning samples. The system uses the actual oscillation amplitude of the graph topology corresponding to the time node when the enterprise declares default as the regression target, inputting the total liquidity buffer and average default penalty ratio for the same period. It then uses the least squares method for multiple linear regression iteration to fit a baseline sensitivity coefficient. Based on the principle of blocking extreme long-tail risks, the system also incorporates physical boundary hard-cutoff constraints to reduce the impact of funding factors. The value boundary is forcibly locked in the range (0.05, 0.15), thus limiting the default penalty factor. Locking the liquidity parameters within the (1.2, 2.5) range prevents them from excessively masking the actual default evolution signals due to overfitting.

[0052] Example 4: In the deployment scenario of a risk monitoring system in the new energy equipment leasing industry, the system reads historical transaction records of 1000 leasing entities, divides them into 10 credit tier groups, extracts the cash asset reserves and unused bank credit lines of each group, calculates the average total liquidity buffer for each group, extracts the average default penalty ratio for the corresponding contracts, determines the distribution range of the risk resistance coefficient through historical data fitting, and then calculates the distribution range according to the formula. Determine the parameter mapping table corresponding to the credit rating of different industries, where, As a risk resistance coefficient, For the total amount of liquidity buffer, To calculate the average default penalty ratio, the parameter mapping table is finally injected into the risk resistance coefficient calculation unit, providing an initial physical damping reference for streaming commercial big data.

[0053] When the system is applied to a mixed business environment with different settlement cycles, it collects 72 hours of simulated streaming business big data under no-load operation, calculates the background noise amplitude sequence of the business correlation deviation vector under the absence of substantial risk signal interference, and extracts the second moment of this sequence as the preset graph topological steady-state threshold. The initial bias term is used to obtain the mean corporate credit rating of each node and map it to a time decay factor. The step value determines the weight of the predicted associated edges. The physical meaning of the decay slope under different credit ratings is determined. A full topology observer with first-order adjacent edges is deployed on all active nodes of the temporal knowledge graph. The physical trigger point of the weight update lock is determined by comparing the difference between the feature value drift and the initial noise bias term, and a system readiness report is output.

[0054] Example 5: In the deployment scenario of a risk monitoring system in the automotive parts supply industry, the system faces the challenge of pre-defined business dependency weights in a static industry knowledge base. For missing business scenarios, an initial dataset containing 5,000 sample contracts was obtained through a pre-calibration procedure. Semantic recognition operators were used to extract purchase amounts and delivery cycles from the contract text, and the business scale proportion of supplier entity nodes was calculated. To address identity ambiguity issues in the mapping stage of streaming commercial big data, a hash matching similarity threshold was set. With a value of 0.95, the characters of the enterprise name and the geographical coordinates of the registered location are extracted and concatenated under the condition that the taxpayer identification number is incomplete. A composite hash index is generated and compared with the existing nodes in the time-series knowledge graph to determine the unique mapping of the active node.

[0055] The system initiates a 168-hour background noise acquisition procedure to determine the preset topological steady-state threshold. The quantitative benchmark continuously records normal transaction data streams without substantial risk signals during the collection period. It calculates the variance of the business-related deviation vector and extracts the volatility corresponding to the 99% confidence interval as the benchmark. The initial setting value is used to verify the triggering mechanism of the weight update lock. A locking signal is deployed on the first-order adjacent edge of the active node, and the eigenvalue drift is recorded. Below for 12 consecutive sampling periods When the local associated topological domain reaches 50%, it is determined that the evolution state of the local associated topological domain has entered the steady state range, and a system ready state report is output.

[0056] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit of this application and the scope of protection of this invention, and all of these forms are within the protection scope of this application.

Claims

1. An industry decision support method based on the coupling of big data and knowledge base, characterized in that, Includes the following steps: Step S1: Acquire streaming business big data in real time and map the streaming business big data to a time-series knowledge graph to determine the active nodes in the time-series knowledge graph; Step S2: Calculate the frequency of business flow generated by streaming business big data in the local neighborhood of the activated node, and perform a difference calculation between the frequency of business flow and the preset business dependency weight to generate a business association deviation vector. Step S3: Obtain the cash asset reserve data and available credit balance of the business entity corresponding to the activated node, add the cash asset reserve data and available credit balance to determine the total liquidity buffer, and construct a risk resistance coefficient that represents the credit risk hedging ability in combination with the average default penalty ratio. Step S4: Compare the magnitude of the business association deviation vector with the preset graph topology steady-state threshold. When the magnitude exceeds the preset graph topology steady-state threshold, use the reciprocal of the risk resistance coefficient as an adjustment ratio to reduce the magnitude of the business association deviation vector by weight, and obtain the topology evolution increment that characterizes the fluctuation intensity of the graph topology structure. Step S5: Based on the topology evolution increment, predictive association edges with time decay attributes are generated in the local association topology domain of the activated node, and the first-order adjacent edges of the activated node are locked to prevent the weight update signal from spreading outward through the local neighborhood of the activated node, thereby realizing the dynamic synchronization of the local topology structure of the time-series knowledge graph and outputting the decision support results.

2. The industry decision support method based on the coupling of big data and knowledge base as described in claim 1, characterized in that, Step S3 further includes the following sub-steps: Step S31, read the cash and cash equivalents reserves and unused bank credit lines of the commercial entity through the data interface, and add the cash and cash equivalents reserves to the bank credit lines to calculate the total liquidity buffer representing the short-term liquidity level; Step S32, extract the average default penalty ratio of the commercial entity's historical contracts as a weighting factor, and convert the total liquidity buffer into a risk resistance coefficient through a preset mapping function, wherein the risk resistance coefficient is positively correlated with both the total liquidity buffer and the average default penalty ratio of historical contracts.

3. The industry decision support method based on the coupling of big data and knowledge base according to claim 1, characterized in that, Step S2 specifically includes: extracting transient components representing transaction activity from streaming business big data and determining the projection direction of the transient components in the temporal knowledge graph topology space; using the projection direction to perform directional weighting on the degree of deviation, generating a business association deviation vector with spatial orientation and amplitude attributes.

4. The industry decision support method based on the coupling of big data and knowledge base according to claim 1, characterized in that, Step S4 further includes the following sub-steps: Step S41, calculate the excess deviation value of the business association deviation vector above the preset graph topology steady state threshold; Step S42, multiply the excess deviation value by the reciprocal of the risk resistance coefficient to obtain the mapped adjustment amount. Step S43: The mapped adjustment amount is determined as the topology evolution increment to limit the impact of local data mutations on the graph topology.

5. The industry decision support method based on the coupling of big data and knowledge base according to claim 1, characterized in that, Locking the first-order adjacent edges of the active node to prevent weight update signals from spreading outward through the local neighborhood of the active node includes: configuring weight update locks on the first-order adjacent edges of the active node; and restricting the transmission of topology evolution increments to non-active regions of the temporal knowledge graph through the local associated topology domain during dynamic synchronization.

6. The industry decision support method based on the coupling of big data and knowledge base according to claim 1, characterized in that, Following dynamic synchronization, the process also includes: monitoring the evolution trend of the weights of predicted associated edges; when the weight disappearance rate is lower than the preset decay benchmark and the duration exceeds the observation threshold, modifying the attributes of the predicted associated edges to persistent topological edges and storing them in the static industry rule base.

7. The industry decision support method based on the coupling of big data and knowledge base according to claim 1, characterized in that, The output decision support results include: calculating the graph Laplacian operator eigenvalues ​​of the synchronized local topology; assessing the change in the default probability of business entities based on the drift of the eigenvalues; and outputting a risk penetration prediction report.

8. The industry decision support method based on the coupling of big data and knowledge base according to claim 1, characterized in that, Activated nodes include at least core enterprise nodes, supplier nodes, logistics nodes, and cash flow nodes; preset business dependency weights are determined based on the historical settlement cycle and order coupling degree between nodes.

9. The industry decision support method based on the coupling of big data and knowledge base according to claim 1, characterized in that, The decision support results also include: when multiple industry rules are identified to conflict, the dynamically synchronized local topology is invoked to perform a logical consistency comparison, determine the highest priority risk handling strategy instruction, and output it.

Images