A risk assessment and task scheduling method and system based on multi-source business data association network and adaptive prediction window
By constructing a network of business object associations and dynamically adjusting the prediction time window, the problems of prediction lag and redundant calculation in computer business processing systems under rapid changes in multi-source heterogeneous data are solved, achieving high efficiency and accuracy in task scheduling.
Patent Information
- Application Number
- CN202611610004.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-19
- Publication Date
- 2026-11-17
AI Technical Summary
Existing computer business processing systems struggle to adjust the prediction calculation frequency in a timely manner when faced with rapidly changing multi-source heterogeneous data, leading to delayed prediction results or duplicate calculations, and making it difficult to accurately identify the scope of risk propagation and task scheduling priorities.
A network of business object relationships is constructed, and a closed-loop data processing process is formed by dynamically adjusting the prediction time window, risk propagation analysis, and task priority calculation, combined with task execution feedback to update parameters.
It improves task scheduling response efficiency, reduces redundant calculations, promptly identifies the scope of risk propagation and optimizes task scheduling, and achieves timeliness and accuracy of prediction results.
Smart Images

Figure CN123117729A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of computer data processing, graph structure computation, and task scheduling. Specifically, it relates to a method and system for performing data verification, format standardization, time alignment, and business object identification mapping on heterogeneous data from multiple business systems to construct a business object association network; adaptively adjusting the prediction calculation time window based on dynamic risk characteristics; determining affected business objects and risk propagation paths based on the predicted risk status and the business object association network; limiting the set of target tasks that need to be recalculated based on the affected business objects; and then performing task priority updates, resource matching, and task scheduling in conjunction with the schedulable resource status. Finally, it relates to updating risk assessment parameters, prediction window adjustment parameters, and association edge parameters based on task execution feedback. Background Technology
[0002] Existing computer business processing systems typically perform state assessments and predictive calculations at fixed time intervals. When input data changes rapidly within a short period, the fixed prediction time window makes it difficult to increase the prediction calculation frequency in a timely manner, easily causing the prediction results to lag behind the current state. When input data remains stable for a long period, continuing to perform predictive calculations at the same frequency on all business objects at a fixed interval will result in redundant calculations and unnecessary overhead on processor, memory, and task scheduling resources.
[0003] On the other hand, existing systems typically perform object state assessment, correlation impact analysis, and task scheduling separately. For multiple business objects with data references, business dependencies, resource sharing, historical anomaly co-occurrence, or task dependencies, calculating the state of each object individually makes it difficult to directly limit the scope of related objects to which state changes may propagate. This results in subsequent task scheduling still needing to recalculate and sort all incomplete tasks.
[0004] Therefore, in computer systems with continuously updated multi-source heterogeneous data and dynamically changing task queues, the following technical problems need to be solved: adaptively adjusting the prediction calculation time window based on changes in input state; calculating the range of affected objects using the correlation between business objects, and forming a subset of target tasks that need to be recalculated to reduce redundant calculations of unaffected tasks; and updating subsequent calculation parameters based on the prediction error between task execution results and actual state, so that prediction frequency, correlation propagation calculation, and task scheduling can form a closed-loop data processing process between continuous analysis time windows. Summary of the Invention
[0005] In view of this, the purpose of this invention is to provide a risk assessment and task scheduling method and system based on a multi-source business data association network and an adaptive prediction time window. This method is used to dynamically change the prediction calculation frequency according to the degree of state change when multi-source heterogeneous data is continuously updated, limit the set of target tasks that need to be recalculated according to the calculation results of the association network, and perform incremental task scheduling in combination with the status of schedulable resources. At the same time, it uses task execution feedback to update risk assessment parameters, prediction window adjustment parameters, and association edge parameters, so as to reduce the repeated calculations caused by fixed prediction cycles and full task rearrangement, and improve the task scheduling response efficiency after dynamic state changes.
[0006] To achieve the above objectives, the present invention provides the following technical solution:
[0007] A risk assessment and task scheduling method based on multi-source business data association network and adaptive prediction window
[0008] The steps include: S1, acquiring and associating multi-source business data, acquiring multi-source business data corresponding to the business object to be managed;
[0009] The multi-source business data includes at least: basic business object data; historical business result data; real-time business status data; abnormal event data; executed task data; and schedulable resource data.
[0010] The business objects can be suppliers, materials, business orders, business projects, assets, or other business management objects.
[0011] Based on the business object identifier, business time, and business relationship, multi-source business data from different business systems are associated to form a dynamic business data set corresponding to the business object.
[0012] Based on this, the system constructs a business object association network according to the business dependencies, data associations, and temporal associations between business objects. Specifically, the system establishes association edges based on at least one of the following: upstream and downstream business process relationships, data reference relationships, resource sharing relationships, historical anomaly co-occurrence relationships, and task dependencies between business objects; it sets association strength parameters, association direction parameters, historical association parameters, and time decay parameters for the association edges; and it determines candidate risk propagation paths between business objects based on the association strength parameters, association direction parameters, historical association parameters, and time decay parameters.
[0013] Heterogeneous data from multiple business systems, including procurement, warehousing, quality, assets, finance, and personnel, including business object attributes, historical results, real-time status, abnormal events, task execution, and resource status.
[0014] Outliers are removed through data validation, data from different sources is converted to a unified format through format standardization, data with different time granularities is mapped to the same time coordinate system through time alignment, and the same business object scattered across different systems is merged through object identifier mapping. Then, association edges are established based on business dependencies, data reference relationships, and time association relationships, and association strength parameters, association direction parameters, historical association parameters, and time decay parameters are set for each association edge.
[0015] A network of business objects with business objects as nodes and relationships as edges, and a dynamic set of business data for each business object.
[0016] This network is used for subsequent risk propagation analysis—when a business object experiences a risk change, the system traces the risk propagation path along the associated edges in this network.
[0017] By constructing a network of relationships between business objects, data that was originally scattered across multiple isolated systems is integrated into a unified data structure that reflects the dependencies between business objects, providing a computable graph structure foundation for risk propagation analysis.
[0018] Step S2: Dynamic risk feature extraction and current risk status assessment;
[0019] For each business object in the business object association network, the following risk characteristics are extracted within the current analysis time window:
[0020] (1) Current business status characteristics: including current business indicator values, indicator deviation degree, and importance of business objects;
[0021] (2) Historical trend characteristics: including the direction, magnitude and number of consecutive deteriorations of business indicators within a continuous historical time window;
[0022] (3) Characteristics of abnormal events: including the number of abnormal events, the density of abnormal events, the severity, and the duration;
[0023] (4) Task execution characteristics: including the number of timed-out tasks, the number of failed tasks, and the backlog of unfinished tasks.
[0024] The system calculates the current dynamic risk value for each business object based on the extracted features. In one implementation, the current dynamic risk value is represented as:
[0025] ;
[0026] in, This represents the current dynamic risk value. Basic risk value; This is a historical trend risk correction value; This is a correction value for the risk of abnormal events; Adjustment value for task execution risk; For the corresponding weight parameters; This is the normalization function.
[0027] Base risk value Determined based on the degree of deviation of business indicators in the current business status characteristics. Historical trend risk correction value. The risk correction value for abnormal events is determined based on the direction and magnitude of changes in business indicators within multiple consecutive historical analysis time windows—positive values are used when the direction of change is deteriorating, and negative values are used when the direction of change is improving. The risk adjustment value for task execution is determined based on the severity, frequency, and duration of the abnormal event. Determined based on the status of timed-out tasks, failed tasks, and backlog of incomplete tasks.
[0028] The system compares the current dynamic risk value with a preset risk level range to determine the current risk level of the business object. The system also determines the risk change trend based on the difference between the current dynamic risk value and one or more historical dynamic risk values; risk change trends include risk increase, risk stabilization, and risk decrease.
[0029] A dynamic set of business data for each business object in the business object association network, including current business status, historical business results, abnormal event records, and task execution records.
[0030] Within the current analysis time window, calculate the current business status characteristics, historical trend characteristics, anomaly event characteristics, and task execution characteristics. Input these characteristics into the dynamic risk value calculation formula. The resulting quantified dynamic risk value is compared with a preset risk level range to determine the risk level.
[0031] The current dynamic risk value, current risk level, and current risk change trend of each business object.
[0032] The current dynamic risk value and risk change trend are used to drive the three subsequent key processes: (1) Determining the adaptive prediction time window (step S3), the faster the risk changes, the shorter the prediction window; (2) Generating the predicted risk state (step S3), the current risk value is used as the initial state for prediction; (3) Calculating the dynamic priority of the task (step S5), tasks with high risk and an upward trend are given higher priority.
[0033] By introducing dynamic adjustments across three dimensions—historical trend correction, real-time anomaly correction, and task execution correction—the risk value can be continuously updated as business status changes.
[0034] Step S3: Determining the adaptive prediction time window and predicting future risk status
[0035] The system determines the rate of risk change based on the dynamic risk value changes of each business object within multiple consecutive historical analysis time windows. The rate of risk change is calculated by averaging the differences in risk values between adjacent time windows. The system also determines the anomaly density based on the ratio of the number of anomalies within the current analysis time window to the window duration. And determine the rate of change of task backlog based on the change of the number of unfinished tasks over time. .
[0036] In one implementation, the prediction window adjustment factor is expressed as:
[0037] ;
[0038] in, Adjustment factor for the forecast window; Adjustment amount based on the base; For the speed of risk change; Density of abnormal events; The rate of change in task backlog; and These are the corresponding weight parameters.
[0039] The prediction time window is represented as follows:
[0040] ;
[0041] in, For adaptive prediction time window; Based on the forecast time window; and These are the lower and upper limits of the prediction time window, respectively; This indicates that the value is restricted to a specified range.
[0042] When the rate of risk change Abnormal event density or the rate of change in task backlog When it rises, Increase the prediction time window Shortening the timeframe allows for more timely risk prediction; when all the above indicators are below the corresponding preset thresholds... Reduce the prediction time window Extend the timeframe to reduce unnecessary computational and scheduling overhead.
[0043] To ensure that the predicted risk status yields repeatable calculation results according to the processing flow of this invention, this embodiment preferentially employs a trend extrapolation method based on the current dynamic risk value, the rate of risk change, and the current prediction time window to calculate the future risk status. The system first calculates the rate of risk change based on the dynamic risk values of multiple consecutive historical analysis time windows. Then, it determines a prediction window adjustment factor based on the rate of risk change, the density of abnormal events, and the rate of change of task backlog. The system then uses this adjustment factor to adjust the basic prediction time window to obtain the current prediction time window.
[0044] When using trend extrapolation for risk prediction, the current dynamic risk value is used as the starting value for prediction, the rate of risk change is used as the input for the direction and magnitude of risk change, and the current prediction time window is used as the prediction time range. The predicted risk value within the current prediction time window is calculated. Then, the predicted risk value is compared with a preset risk level range to determine the predicted risk level, and the prediction occurrence time is determined based on the time when the predicted risk value reaches the corresponding risk level threshold.
[0045] The rate of risk change is calculated by the difference in dynamic risk values between adjacent historical analysis time windows. The density of abnormal events is calculated by the number of abnormal events in the current analysis time window and the corresponding window duration. The rate of change in task backlog is calculated by the change in the number of unfinished tasks in adjacent analysis time windows. The basic prediction time window, prediction window adjustment parameters, and risk level range are determined by historical operating data, verification data, or pre-set business operating parameters of the actual application scenario, and are consistent with the calculation process described in the specification.
[0046] When the rate of risk change, the density of abnormal events, or the rate of change in task backlog changes, the system recalculates the prediction window adjustment factor and obtains a new current prediction time window. This current prediction time window is used for the next risk prediction calculation, rather than just as a condition for determining the risk level. This allows the update frequency of the prediction calculation to be dynamically adjusted according to changes in the risk status.
[0047] Risk propagation analysis uses the predicted risk status as input. For a given source business object, the system calculates the risk propagation impact value according to the association edges in the business object's association network, and determines the predicted risk impact on the target business object based on association strength, propagation direction, propagation attenuation parameters, and propagation path length. When the risk propagation impact value of the target business object reaches a preset propagation threshold, the target business object is identified as the risk propagation impact object, and the corresponding risk propagation path is recorded.
[0048] After task scheduling is completed, the system continues to acquire task execution status and task processing results, and compares the predicted risk status with the subsequent actual risk status to obtain the prediction error. Based on whether the task is completed, whether it has timed out, whether there is an execution anomaly, and the prediction error, the risk assessment parameters, prediction window adjustment parameters, and related edge parameters are updated; the updated parameters enter the next analysis time window, so that risk assessment, risk prediction, risk propagation analysis, and task scheduling form a continuous data processing closed loop.
[0049] Dynamic risk value sequences, abnormal event records, and task backlog data for each business object within multiple consecutive historical analysis time windows.
[0050] First, calculate the rate of change of risk. (Average of the difference in risk values between adjacent windows), anomaly event density (Ratio of the number of exceptions to the window duration) and the rate of change in task backlog (The rate of change of the number of unfinished tasks over time). Then, substitute these three indicators into the formula for the forecast window adjustment factor. Then through An adaptive forecasting time window is calculated. Finally, within this window, the future risk status is predicted based on the current risk value and historical trends.
[0051] Adaptive determination of the prediction time window and various business objects in The predicted risk status within the range (predicted risk value, predicted risk level, predicted trend, predicted occurrence time).
[0052] The length of the prediction time window determines the update frequency of risk prediction—the shorter the window, the more frequently the system performs risk prediction and task scheduling assessment; the predicted risk status is used to calculate the impact value of subsequent risk propagation and to predict the generation of risk events.
[0053] By dynamically linking the prediction window with the rate of risk change, the density of abnormal events, and the rate of change in task backlog, the prediction cycle is automatically shortened when the risk changes rapidly to ensure timely early warning, and the prediction cycle is automatically extended when the risk is stable to reduce unnecessary computational overhead.
[0054] Step S4: Calculation of risk propagation path and identification of affected objects
[0055] The system identifies the source business object whose risk status has changed based on the business object association network. The source business object can be a business object whose current risk level has increased or whose predicted risk level has increased.
[0056] For candidate risk propagation paths between the source business object and the target business object, the system calculates the path risk propagation impact value based on the predicted risk value of the source business object and the association strength parameters, association direction parameters, historical association parameters, and time decay parameters of each associated edge in the path.
[0057] In one implementation, the impact value of path risk propagation is expressed as:
[0058] ;
[0059] in, The path risk propagation impact value of the source business object s along path p to the target business object j; The predicted risk value for the source business object; Let e be the association strength parameter for the associated edge e; For the associated direction parameter; Historical correlation parameters; κ represents the time interval corresponding to the correlation; κ is the time decay coefficient.
[0060] When multiple propagation paths exist for the same target business object, the system uses a probabilistic union approach to obtain the aggregated risk propagation impact value of the target business object:
[0061] ;
[0062] in, The aggregated risk propagation impact value of the target business object j; The set of valid paths to reach the target business object j.
[0063] When the risk propagation impact value of the target business object When the preset propagation threshold is reached, the system identifies the target business object as the object affected by risk propagation and determines the corresponding candidate risk propagation path as the effective risk propagation path.
[0064] Business object association network (including nodes and parameterized association edges), predicted risk value of source business object .
[0065] For each candidate path p from the source business object s to each target business object j, the association strength parameter of each associated edge on the path is... Associative direction parameters Historical correlation parameters and time decay term Multiply by the predicted risk value of the source business object. The impact value of path risk propagation was obtained. For multiple effective propagation paths reaching the same target business object, the risk propagation impact value of each path is considered. Aggregate the data to obtain the aggregated risk propagation impact value of the target business object.
[0066] Aggregate risk propagation impact value for each target business object The set of objects affected by risk transmission and the set of effective risk transmission paths.
[0067] The objects and scope of risk propagation are used to predict the generation of risk events and calculate the dynamic priority of tasks.
[0068] By calculating the risk propagation impact value along the associated edges of the business object association network, the risk scoring of a single business object is extended to multi-object risk propagation analysis, enabling the system to identify the direction, path, and scope of risk propagation from the source business object to associated business objects.
[0069] Step S5: Predicting Risk Event Generation and Calculating Dynamic Task Priorities
[0070] The system generates a set of predicted risk events based on the predicted risk status, risk propagation direction, risk propagation path, and affected objects of each business object. A predicted risk event must include at least the predicted risk source, predicted risk status, risk change trend, predicted occurrence time, risk impact scope, risk propagation path, and related business objects.
[0071] The system correlates predicted risk events with the current set of incomplete tasks. This correlation matching is based on the business object identifier, the business object association path, and the task type. If a predicted risk event corresponds to an incomplete task, that incomplete task is identified as the target task; otherwise, candidate preventative tasks are generated based on the predicted risk event.
[0072] In one implementation, the dynamic priority score for tasks is represented as:
[0073] ;
[0074] in, The dynamic priority score for the k-th target task; Risk level factor; Risk change trend factor; Risk transmission influencing factors; The remaining processing time factor for the task; to These are the corresponding weight parameters.
[0075] The system categorizes tasks into Level 1, Level 2, Level 3, and routine tasks based on dynamic priority scoring. For tasks whose predicted risk level increases significantly, whose risk trend is upward, whose impact is expanding, or whose predicted occurrence time is earlier than the task's expected completion time, the system raises their dynamic priority.
[0076] The predicted risk status of each business entity, the set of objects affected by risk propagation, and the set of currently unfinished tasks.
[0077] First, predicted risk events are generated based on the predicted risk status and risk propagation results. Then, these predicted risk events are matched with currently incomplete tasks—if a corresponding task exists, it is identified as a target task; otherwise, a preventative task is generated. Finally, a dynamic priority score is calculated for each target task. .
[0078] The set of predicted risk events, the set of target tasks associated with the predicted risk events, and the dynamic priority score and priority level of each target task.
[0079] Dynamic priority scoring is used for calculating resource matching scores and determining task refactoring trigger conditions.
[0080] By updating the priority of only the target tasks associated with the predicted risk events (instead of reordering all tasks), the computational scope of task priority adjustment is narrowed from all incomplete tasks to a subset of target tasks related to the risk events, which significantly reduces the scheduling computational overhead.
[0081] Step S6: Resource matching scoring and dynamic task scheduling
[0082] The system acquires resource status data for currently schedulable resources. This resource status data includes at least resource capability tags, historical task completion rates, historical task average completion times, current task load, current availability status, and response constraints.
[0083] In one implementation, the resource matching score is represented as:
[0084] ;
[0085] in, The resource matching score is given to the i-th schedulable resource; For resource capability matching degree; For the effectiveness of historical mission execution; This indicates the current availability of resources. Current task load; Cost of task response constraints; to These are the corresponding weight parameters.
[0086] The system matches and scores schedulable resources according to the dynamic priority of tasks from high to low, and schedules tasks to the resource with the highest matching score that meets preset scheduling conditions. Preset scheduling conditions include that the resource's capability tag meets the task requirements, its current availability status is available, and the current task load is below the upper limit.
[0087] When a scheduled resource fails to receive a task within a preset response time or the task execution status meets the timeout condition, the system recalculates the resource matching score of the remaining schedulable resources and reschedules the incomplete task to other resources that meet the conditions.
[0088] This invention employs a multi-condition jointly triggered task refactoring mechanism. The task refactoring triggering conditions include at least one of the following:
[0089] (1) The predicted risk level will be upgraded by a level relative to the current risk level;
[0090] (2) The number of objects affected by risk transmission increases relative to the current number of objects affected and reaches a preset threshold;
[0091] (3) The predicted occurrence time of the risk event is earlier than the expected completion time of the associated target task;
[0092] (4) The predicted resource demand for the predicted risk event exceeds the current available resource capacity;
[0093] (5) The prediction error between the predicted risk status and the actual risk status exceeds the preset error threshold.
[0094] When at least one of the above conditions is met, the system performs task refactoring on the target task. Task refactoring operations include at least one of the following: adjusting task priority, adjusting task execution order, adjusting task execution resources, splitting tasks, merging tasks, and generating preventative tasks.
[0095] Scheduled resource status data (capability tags, historical completion rate, current load, availability status, response constraints), and dynamic priority of target tasks.
[0096] Calculate a resource matching score for each schedulable resource. Tasks are scheduled to the resource with the highest matching score and that meets preset scheduling conditions, based on their dynamic priority from highest to lowest. Simultaneously, five task refactoring trigger conditions are continuously monitored; task refactoring is triggered when any one of these conditions is met.
[0097] Each target task is allocated schedulable resources and a task scheduling scheme; when a task refactoring is triggered, the refactored task priority, execution order, resource allocation, and task splitting / merging scheme are output.
[0098] The results of task scheduling and refactoring enter the execution phase (step S7), and the system continuously acquires the task execution status and results.
[0099] Resource matching is performed by comprehensively considering resource capabilities, historical execution results, current load, availability, and response constraints, prioritizing the allocation of high-risk, high-urgency tasks to the most suitable resources. A multi-condition joint triggering mechanism avoids scheduling jitter caused by single threshold triggers.
[0100] Step S7: Task execution feedback and rolling update of risk parameters
[0101] After a task is generated or restructured, the system continuously acquires the task execution status and task processing results. The task execution status includes at least the following: pending, received, executing, completed, timed out, execution failed, and execution exception.
[0102] The system correlates the task execution results with the subsequent actual risk status and calculates the prediction error based on the difference between the predicted risk status and the actual risk status.
[0103] When the prediction error between the predicted risk state and the actual risk state reaches the preset error condition, the system updates the risk assessment parameters, prediction window adjustment parameters, and association edge parameters in the business object association network according to the prediction error.
[0104] Among them, the risk assessment parameter is used to calculate the current dynamic risk value in the next analysis time window; the prediction window adjustment parameter is used to redetermine the prediction window adjustment factor based on the rate of risk change, the density of abnormal events, and the rate of change of task backlog; and the correlation edge parameter is used to calculate the risk propagation impact value in the next analysis time window.
[0105] For the implementation method of updating risk assessment parameters using parameter optimization, the system constructs a loss function based on the predicted risk state and the actual risk state, and iteratively updates the risk assessment parameters according to the loss function. The parameter optimization method can be implemented using gradient descent, and the updated parameters are used for risk assessment in the next analysis time window.
[0106] For the prediction window adjustment parameters, the system adjusts the prediction window adjustment factor according to the correspondence between the prediction error and the subsequent actual risk changes, so that the prediction time window can be readjusted when the prediction error is large. For the association edge parameters, the system adjusts the association strength parameter, historical association parameter, or time decay parameter according to the actual risk change relationship between the associated business objects and the task execution results.
[0107] The updated risk assessment parameters, prediction window adjustment parameters, and associated edge parameters are entered into the next analysis time window and used sequentially for risk assessment, prediction time window determination, risk prediction, risk propagation analysis, task dynamic priority calculation, and task scheduling, thus forming a continuous update process based on execution feedback.
[0108] When the task is completed and the corresponding abnormal event is eliminated, the system reduces the abnormal event risk correction value. and task execution risk adjustment value When a task times out, fails to execute, or generates a new abnormal event, the system increases the abnormal event risk correction value or the task execution risk correction value, and re-executes the task priority calculation and resource scheduling.
[0109] The system calculates the prediction error based on the difference between the predicted risk state and the actual risk state. The system then updates the risk assessment parameters, prediction window adjustment parameters, associated edge parameters, and propagation threshold parameters based on this prediction error.
[0110] In one implementation, the risk assessment parameters are represented as a parameter vector w, and the system updates the parameters according to the loss function L(w):
[0111] ;
[0112] in, This is the current parameter vector; The updated parameter vector; L( is the learning rate; L( It is constructed based on the difference between the predicted risk status and the actual risk status.
[0113] Through the above feedback updates, the system can re-execute risk assessment and task scheduling based on the updated multi-source business data and risk assessment parameters in the next analysis time window, forming a continuous closed loop.
[0114] The status data (received, in progress, completed, timed out, execution error) and processing results during task execution, as well as the predicted risk status and the actual risk status.
[0115] Update the dynamic business data set based on task execution status—reduce the impact when the task is completed and the anomaly is resolved. and Increase efficiency when a task times out or fails. Then calculate the prediction error, by... ;
[0116] Updated dynamic business data set, updated risk assessment parameters (weight parameters) Prediction window adjustment parameters (Associated edge parameters, etc.)
[0117] The updated data and parameters are used in steps S1 to S6 of the next analysis time window, enabling the system to perform risk assessment and task scheduling based on the latest business data and execution feedback in each cycle.
[0118] By feeding back task execution results and prediction errors to risk assessment parameters, the system gradually optimizes risk prediction in each cycle, achieving a continuous closed loop of "data acquisition - dynamic analysis - risk assessment - risk prediction - propagation analysis - task scheduling - execution feedback - parameter update".
[0119] Compared with the prior art, the present invention has at least the following beneficial effects:
[0120] This invention determines a prediction window adjustment factor based on the rate of risk change, the density of abnormal events, and the rate of change of task backlog, and adjusts the prediction time window according to the prediction window adjustment factor. This allows the update frequency of prediction calculations to be dynamically adjusted according to changes in the business risk status, thereby reducing the lag in risk status updates caused by a fixed prediction time window and reducing unnecessary repeated prediction calculations when the risk status is relatively stable.
[0121] This invention inputs the predicted risk status into the business object association network, and determines the objects affected by risk propagation and the risk propagation path based on the association relationship, association strength, propagation direction, historical association parameters, time decay parameters and propagation path between the source business object and the target business object. This enables the predicted risk status to propagate along the association relationship between business objects, providing a calculation basis for subsequent target task selection.
[0122] This invention filters target tasks based on the business object association between the risk propagation impact object and the unfinished tasks, and performs dynamic priority calculation and resource rematch only for the target tasks, thereby avoiding a full recalculation of all unfinished tasks when the risk status changes, and reducing the repeated scheduling of unrelated tasks.
[0123] This invention further acquires the task execution status and task processing results, and updates the risk assessment parameters, prediction window adjustment parameters, and association edge parameters in the business object association network based on the prediction error between the predicted risk status and the actual risk status. This allows the updated parameters to enter the next analysis time window, thereby forming a continuous processing process of data acquisition, risk assessment, risk prediction, risk propagation, task scheduling, execution feedback, and parameter updating. Attached Figure Description
[0124] Figure 1 This is a schematic diagram of the overall process of the risk assessment and task scheduling method based on multi-source business data association network and adaptive prediction window of the present invention;
[0125] Figure 2 This is a schematic diagram of the multi-source business data dynamic analysis structure of the present invention;
[0126] Figure 3 This is a schematic diagram of the closed-loop dynamic risk assessment and task scheduling of the present invention;
[0127] Figure 4 This is a schematic diagram of the system structure of the present invention. Detailed Implementation
[0128] The present invention will be further described below with reference to the accompanying drawings and embodiments. It should be understood that the following embodiments are only for explaining the present invention and are not intended to limit the scope of protection of the present invention.
[0129] Overall implementation process:
[0130] like Figure 1 As shown, the method provided in this embodiment includes steps S1 to S7.
[0131] S1: Acquisition of Multi-Source Business Data and Construction of Business Object Association Network
[0132] The system acquires multi-source business data through database connections, application programming interfaces (APIs), message queues, file imports, mobile terminal synchronization, or device data collection. This multi-source business data includes at least:
[0133] Business object attribute data, such as supplier level, material category, order type, project stage, asset status, and personnel skill tags;
[0134] Historical business results data, such as historical pass rate, historical on-time delivery rate, historical anomaly rate, and historical task completion rate;
[0135] Real-time business status data, such as the current quantity to be processed, current execution progress, current inventory status, and current device status;
[0136] Abnormal event data, such as quality anomalies, delivery delays, asset failures, process blockages, and threshold violations;
[0137] Task execution data, such as task creation time, receipt time, start time, completion time, execution status, and processing result;
[0138] Scheduled resource status data, such as personnel capability tags, equipment capability tags, current task load, historical completion rate, available time period, and response constraints.
[0139] The system performs data validation, missing value handling, duplicate data merging, outlier marking, format standardization, timestamp alignment, and business object identifier mapping on multi-source business data.
[0140] During the data association phase, the system establishes association edges between business objects based on at least one of the following: upstream and downstream relationships in business processes, data reference relationships, resource sharing relationships, historical anomaly co-occurrence relationships, and task dependency relationships. The association edges are configured with the following parameters: association strength parameter. Associative direction parameters Historical correlation parameters and time decay parameters .
[0141] S2: Dynamic risk feature extraction and current risk status assessment;
[0142] Within the current analysis time window, the system extracts risk characteristics for each business object in the business object association network: current business status characteristics, historical change trend characteristics, abnormal event characteristics, and task execution characteristics.
[0143] The current dynamic risk value is expressed as:
[0144] ;
[0145] The system compares the current dynamic risk value with a preset risk level range to determine the current risk level of the business object. The system also determines the risk change trend based on the difference between the current dynamic risk value and historical dynamic risk values.
[0146] S3: Adaptive prediction time window determination and future risk status prediction;
[0147] S3-1 Input, Calculation, Output and Parameter Definition of the Prediction Model
[0148] (1) Input data
[0149] The input data for the prediction model includes at least:
[0150] The dynamic risk value sequence of the business object within the most recent n consecutive historical analysis time windows: Current dynamic risk value Speed of risk change Calculated by the average of the differences in risk values between adjacent windows; anomaly event density. Task backlog change rate Adaptive prediction time window Importance of business objects It is determined based on the business object attribute data, such as supplier level, material criticality, asset importance, etc.
[0151] The prediction model can be implemented using any one or a combination of the following methods.
[0152] Method 1: Trend Extrapolation Model
[0153] ;
[0154] in, To predict risk values; This represents the current dynamic risk value. For the speed of risk change; For adaptive prediction time window; This is the trend decay coefficient, used to represent the degree to which the rate of change of risk decays over time. ∈(0,1).
[0155] Method 2: Weighted Moving Average Trend Model
[0156] ;
[0157] in, Weighted by historical risk values, satisfying Furthermore, the closer to the current time, the greater the weight, for example... .
[0158] Method 3: Time Series Model
[0159] Using ARIMA, exponential smoothing, or regression models, with historical risk value sequences, outlier density, and task backlog change rates as input features, the output is a predicted risk value.
[0160] Method 4: Classification Model
[0161] Using logistic regression, decision tree, or gradient boosting models, the current risk characteristics are taken as input, and the probability distribution of each risk level is output. The level with the highest probability is taken as the predicted risk level.
[0162] The prediction model outputs:
[0163] Predicted risk value Predicted risk level The risk level is determined by comparing the predicted risk value with the preset risk level range; the trend of predicted risk changes is also considered. This includes risk escalation, risk stabilization, and risk depreciation; and the predicted timing of occurrence. See S3-2 for details.
[0164] The predicted risk value Used for calculating the risk propagation impact value in S4, serving as the input for the predicted risk value of the source business object; the predicted risk level and predicted time of occurrence Used in S5 for predicting the generation of risk events and calculating dynamic task priorities; the predicted risk change trend Used to determine whether to increase task priority.
[0165] By outputting quantified future risk states through predictive models, the system can generate early warnings and schedule tasks in advance before risks actually occur, thus achieving proactive risk management.
[0166] This embodiment preferably uses a trend extrapolation method to predict the future risk status of the business object, so that the predicted risk value can be determined based on the current dynamic risk value, the rate of risk change, and the current prediction time window.
[0167] The input data for the prediction model includes the dynamic risk value sequence of the business object within a continuous historical analysis time window, the current dynamic risk value, the rate of risk change, the density of abnormal events, the rate of change of task backlog, and the current prediction time window.
[0168] The rate of risk change is determined based on the difference in dynamic risk values between adjacent historical analysis time windows; the density of abnormal events is determined based on the number of abnormal events associated with business objects within the current analysis time window and the corresponding window duration; the rate of change in task backlog is determined based on the change in the number of unfinished tasks in adjacent analysis time windows; and the current prediction time window is jointly determined by the basic prediction time window and the prediction window adjustment factor.
[0169] Within the current forecast time window, the current dynamic risk value is used as the starting value for forecasting, the rate of risk change is used as the direction and magnitude of risk value change, and the rate of risk change is adjusted according to a preset trend decay coefficient to obtain the forecast risk value for future time.
[0170] The trend extrapolation model is expressed as follows:
[0171] Predicted risk value = Current dynamic risk value + Risk change rate × Prediction time × Trend decay coefficient.
[0172] The trend decay coefficient represents the degree to which the rate of risk change decreases as the forecast period increases. This trend decay coefficient is determined based on historical risk change data or pre-set operating parameters.
[0173] The system determines the predicted risk level by comparing the predicted risk value with the preset risk level range; it determines the trend of predicted risk changes based on the relationship between the current predicted risk value and the historical risk value; and it determines the predicted occurrence time based on the time required for the predicted risk value to reach the target risk level threshold.
[0174] The predicted risk value serves as the input for risk propagation analysis, while the predicted risk level, predicted risk change trend, and predicted occurrence time serve as the input for predicting the generation of risk events and the dynamic priority calculation of target tasks.
[0175] In other implementations, the predicted risk status can also be obtained by using weighted moving average, time series analysis or classification prediction methods according to the specific business scenario. However, all of the above different prediction methods use the dynamic risk characteristics formed by multi-source business data and the current prediction time window as prediction inputs, and input the prediction results into subsequent risk propagation analysis and task scheduling processing.
[0176] S3-2 Input, Calculation, Output and Parameter Definition for Predicting Occurrence Time
[0177] (1) Input data
[0178] Current dynamic risk value Speed of risk change Target risk level threshold For example, high-risk threshold =60, significant risk threshold =80; Adaptive prediction time window Current time .
[0179] (2) Calculation process
[0180] Predicted time of occurrence Defined as: at the current rate of risk change If the value remains unchanged, the dynamic risk value will change from the current value. Reaching the target risk level threshold Time required
[0181] ;
[0182] Among them, when ≥ When the target risk level is reached or above, the predicted occurrence time is 0; when < and When >0, the time to reach the threshold is calculated by linear extrapolation; when < and When the value is ≤0, it indicates that the risk has not increased and the target threshold will not be reached within the prediction window. The predicted occurrence time is taken as the upper limit of the prediction window. .
[0183] The predicted time of occurrence is:
[0184] ;
[0185] (3) Output results
[0186] Predicted time of occurrence The unit is hours or days; the predicted time of occurrence. .
[0187] The predicted occurrence time Task remaining processing time factor used for dynamic priority calculation of tasks in S5 When the predicted occurrence time is earlier than the expected completion time of the task, the task priority is increased; at the same time, the predicted occurrence time is used to determine the task refactoring trigger condition in S6 - if the predicted occurrence time of the risk event is earlier than the expected completion time of the associated target task, then task refactoring is triggered.
[0188] By calculating and predicting the occurrence time, the system can quantify the urgency of the risk as a scheduling basis, avoiding scheduling based solely on risk level while ignoring the time dimension, thus improving the foresight and timeliness of task scheduling.
[0189] The system determines the rate of risk change based on the dynamic risk value changes within multiple consecutive historical analysis time windows. The rate of risk change is calculated by averaging the differences in risk values between adjacent time windows. The system also determines the density of anomalous events. and the rate of change of task backlog .
[0190] The forecast window adjustment factor is expressed as:
[0191] ;
[0192] The prediction time window is represented as follows:
[0193] ;
[0194] When the rate of risk change Abnormal event density or the rate of change in task backlog When it rises, Increase the prediction time window Shorten; when all of the above indicators are below the corresponding preset thresholds. Reduce the prediction time window extend.
[0195] Within the forecast time window, the system forecasts the risk status of each business object, including the forecast risk value, forecast risk level, forecast trend, and forecast occurrence time.
[0196] S4: Risk propagation path calculation and impact target identification;
[0197] The system determines the source business object based on the network associated with the business object. The impact value of path risk propagation is expressed as:
[0198] ;
[0199] When multiple propagation paths exist for the same target business object, the system aggregates them using a probabilistic union approach:
[0200] ;
[0201] When the risk propagation impact value of the target business object When the preset propagation threshold is reached, the system identifies the target business object as the object affected by the risk propagation.
[0202] S5: Predicting risk event generation and calculating dynamic task priorities;
[0203] The system generates a set of predicted risk events based on the predicted risk status of each business object and the objects affected by the risk propagation.
[0204] Task dynamic priority scoring is represented as follows:
[0205] ;
[0206] The system categorizes tasks into first-priority tasks, second-priority tasks, third-priority tasks, and routine tasks based on dynamic priority scoring.
[0207] S6: Resource matching scoring and dynamic task scheduling;
[0208] Resource matching score is expressed as follows:
[0209] ;
[0210] The system schedules tasks to the resources with the highest resource matching score and that meet the preset scheduling conditions, according to the dynamic priority of the tasks from high to low.
[0211] The triggering conditions for task refactoring include at least one of the following: (1) the predicted risk level is upgraded across levels; (2) the number of objects affected by risk propagation increases and reaches a preset quantity threshold; (3) the predicted occurrence time is earlier than the expected completion time of the task; (4) the predicted resource demand exceeds the available resource capacity; (5) the prediction error exceeds a preset error threshold.
[0212] S7: Task execution feedback and rolling updates of risk parameters
[0213] The system updates the dynamic business data set based on the task execution status.
[0214] The system updates the risk assessment parameters based on the prediction error:
[0215] .
[0216] Example 1: Detailed calculation process for adaptive adjustment of the prediction window;
[0217] This embodiment details the specific calculation process of the adaptive prediction time window in step S3.
[0218] Suppose a supplier business object exists in a company's quality management system, and the system maintains a dynamic risk value sequence for its past 5 analysis time windows (each window is 1 hour long): .
[0219] Step 1: Calculate the rate of change of risk .
[0220] The system calculates the risk value difference between adjacent windows:
[0221] ;
[0222] ;
[0223] ;
[0224] ;
[0225] The rate of change of risk is the average of the above differences:
[0226] ;
[0227] Step 2: Calculate the density of abnormal events .
[0228] The system counts 13 abnormal events related to this supplier within the current analysis time window (1 hour). Therefore:
[0229] =13 / 1=13;
[0230] Step 3: Calculate the rate of change of task backlog .
[0231] The system counts the number of incomplete tasks associated with this supplier. There were 8 tasks at the end of the previous time window, and 17 tasks at the end of the current time window.
[0232] ;
[0233] Step 4: Calculate the prediction time window. To illustrate the calculation process, the following parameters are only examples of formula calculations and do not represent actual business test results. Let the risk change rate threshold be set. =2, Abnormal event density threshold =5, Threshold for the rate of change of task backlog =2, corresponding to the scale parameter =5、 =10、 =10, weight parameter = = =0.2.
[0234] In scenarios where risks change rapidly =6.5 =13、 =9, then = (6.5 - 2) / 5 = 0.9, = (13 - 5) / 10 = 0.8, = (9 - 2) / 10 = 0.7. Therefore... =0.2×0.9+0.2×0.8+0.2×0.7=0.48.
[0235] Set the basic forecast time window =24 hours =4 hours =48 hours, then = clip(24 / (1+0.48), 4, 48) ≈ 16.22 hours. Therefore, when the dynamic indicator exceeds the corresponding threshold, the prediction time window is shortened from 24 hours to approximately 16.22 hours.
[0236] Step 5: Predict future risk conditions.
[0237] Let the current dynamic risk value be... =68, Risk Change Rate = 6.5 minutes / hour, trend decay parameter =0.12 / hour, then the predicted risk value at the end of the prediction time window is =clip(68+6.5×16.22×exp(-0.12×16.22),0,100)≈83.06. If the target risk level threshold... =80, then the system searches for the time when it first reaches 80 points within the time frame [0, 16.22] hours, and obtains... Approximately 2.49 hours. The above figures are for illustrative purposes only.
[0238] In the comparison scenario, if =0.5 =2、 = -1, then = -0.3, = -0.3, = -0.3, =0.2×0.3+0.2×0.3+0.2×0.3=0.18. At this time... =clip(24×(1+0.18),4,48)=28.32 hours. Therefore, when none of the three dynamic indicators exceed the corresponding thresholds, the prediction time window is extended from 24 hours to 28.32 hours, thereby reducing the frequency of repeated prediction calculations in a steady state.
[0239] Within the current prediction time window, the current dynamic risk value is used as the starting value for prediction, and the rate of risk change is used as the direction and magnitude of risk value change. The rate of risk change is adjusted according to a preset trend decay coefficient to obtain the predicted risk values corresponding to different prediction times. The trend decay coefficient characterizes the degree of influence of the current risk change trend on the future risk state as the prediction time increases. Its value range and determination method can be preset based on the prediction error within the historical analysis window or updated using historical data. To ensure the comparability of calculation results between different business objects, the trend decay coefficient is determined according to the same calculation rules under the same risk assessment model and parameter configuration.
[0240] The trend extrapolation model is expressed as follows:
[0241] Predicted risk value = Current dynamic risk value + Risk change rate × Prediction time × Trend decay coefficient.
[0242] Wherein, the prediction time is the length of time between the current analysis time and the time to be predicted; when calculating the predicted risk value at the end of the prediction time window, the prediction time is equal to the current prediction time window; when determining the prediction occurrence time that reaches the preset risk level threshold, the time required for the predicted risk value to reach the corresponding risk level threshold is calculated in reverse according to the trend extrapolation model, and this time is determined as the prediction occurrence time.
[0243] Example 2: Complete application for IQC quality management;
[0244] Taking the incoming material inspection quality management scenario of an enterprise as an example, the business objects include suppliers, materials, purchase orders, inspection tasks, and inspection personnel.
[0245] The system acquires data from procurement management, warehouse management, quality management, and mobile inspection terminals. It establishes a network of business objects linking suppliers, materials, purchase orders, inspection tasks, and inspection personnel.
[0246] The current analysis time window is set to 30 days. The historical pass rate of a certain supplier (supplier number S-10086) has continuously decreased from 98.5% to 91.2%. Several new quality anomaly events have been added, and there is one overdue unprocessed task. System calculations:
[0247] Base risk value =40 (based on the current pass rate deviation);
[0248] Historical trend risk correction value =12 (The pass rate has declined for four consecutive weeks, decreasing by about 1.8 percentage points each week);
[0249] Abnormal event risk correction value =18 (6 new abnormal records were added within 3 days);
[0250] Task execution risk adjustment value =8 (1 task timed out by 72 hours);
[0251] Current dynamic risk value:
[0252] ;
[0253] ;
[0254] If a risk level of 60-79 is preset to be high, then the supplier is identified as high-risk, and the risk trend is rapidly increasing (the risk level in the previous time window was 55).
[0255] The current dynamic risk value is 78 points, the preset minimum threshold for a major risk level is 80 points, and the risk change rate is 5.2 points / hour. The system first calculates the prediction window adjustment factor based on the risk change rate, abnormal event density, and task backlog change rate, and obtains the current prediction time window. For the predicted risk value at the end of the current prediction time window, the corresponding predicted risk value is calculated using a trend extrapolation model; for the predicted occurrence time, a reverse calculation is performed based on the time required for the predicted risk value to reach 80 points. Specifically, when the trend decay coefficient is... And when the predicted risk score reaches 80 points, then:
[0256] 80 = 78 + 5.2 × predicted occurrence time × .
[0257] Therefore, the predicted occurrence time is based on the determined trend decay coefficient. Calculated. If the system determines based on historical prediction errors. If the value is a preset value, then substitute this preset value into the above formula to obtain the corresponding predicted occurrence time. Thus, the predicted occurrence time and the predicted time window are used as two time parameters with different meanings. The predicted time window is used to determine the time range for the prediction calculation, and the predicted occurrence time is used to determine the expected time when the risk level reaches the target threshold.
[0258] The system detected the rate of change in risk. =5.2, Abnormal Event Density =8 and the rate of change in task backlog =6, calculate the prediction window adjustment factor ( ):
[0259] ;
[0260] Hour;
[0261] The system predicts that this supplier's risk level may escalate from high to significant within the next 16.48 hours (predicted risk score of 85), and calculates the risk propagation impact along the "supplier—material—purchase order" path. Assume this supplier is associated with 5 materials, among which material M-001 has a strong association... =0.9, direction of association =1.0, historical correlation parameters =0.85, time interval =2 hours, κ=0.1, then:
[0262] ;
[0263] The risk propagation impact value of 53.3 exceeds the preset propagation threshold of 40, and the system identifies material M-001 as the object affected by the risk propagation.
[0264] The system generates "Supplier Quality Anomaly Review Task" and "Related Material Tightened Inspection Task", and calculates the dynamic priority of the tasks. ):
[0265] ;
[0266] ;
[0267] The system marks the task as a first-priority task (P>70).
[0268] The system retrieves data from three executable resources A, B, and C (as shown in the table below):
[0269]
[0270] use calculate:
[0271] Resource A: =0.35×95+0.25×90+0.25×70-0.10×60-0.05×20=66.25. Resource B: =0.35×85+0.25×95+0.25×90-0.10×35-0.05×10=71.00. Resource C: =0.35×90+0.25×80+0.25×60-0.10×30-0.05×30=62.00. Therefore, under the above example parameters, resource B has the highest resource matching score, and the system will schedule the target task to resource B.
[0272] If the number of anomalies decreases after the task is completed, timeout tasks are eliminated, and new test results are normal, the system recalculates the risk value:
[0273] ;
[0274] The risk level has been downgraded from high risk to medium risk. The system has reduced the priority of subsequent tasks and released some resources.
[0275] Example 3: Task Restructuring Triggering and Solution Selection;
[0276] This embodiment details the process of judging task reconstruction triggered by multiple conditions, the process of generating candidate task reconstruction schemes, and the process of calculating and selecting scheme costs.
[0277] (A) Determining the triggering conditions for task refactoring;
[0278] Suppose there is a "Supplier Quality Anomaly Review Task" in the system, and its current status is as follows:
[0279] Current risk level: Medium risk (risk score 55).
[0280] Predicted risk level: Major risk (predicted risk score: 88 points);
[0281] Number of objects affected by the current spread: 2 items;
[0282] Predicted number of objects affected by the spread: 7 materials;
[0283] Estimated completion time for the task: 48 hours later;
[0284] Predicted risk occurrence time: 24 hours later;
[0285] Current available resources: 3 inspection personnel;
[0286] Forecasted resource requirements: 8 inspection personnel;
[0287] Current prediction error: The risk levels are the same, but the risk value deviation is 22 points (predicted 88 vs. actual 66).
[0288] The system checks each of the five triggering conditions one by one:
[0289] Condition 1 (Skip-level upgrade): The current risk level is medium risk, and the predicted risk level is major risk. The risk level change path is "medium risk → high risk → major risk," skipping one risk level (high risk), which constitutes a skip-level upgrade. Condition 1 is satisfied.
[0290] Condition 2 (Increased number of affected objects): Currently affecting 2 materials, predicted to affect 7 materials, the increase is 5. A preset threshold of 3 is set; if 5 > 3, Condition 2 is satisfied.
[0291] Condition 3 (Predicted time is earlier than completion time): The predicted time of risk occurrence is 24 hours later, and the estimated time of task completion is 48 hours later. Since 24 < 48, condition 3 is satisfied.
[0292] Condition 4 (Resource Demand Exceeds Limit): The predicted resource demand is 8 inspectors, and the current available number is 3. Since 8 > 3, Condition 4 is satisfied.
[0293] Condition 5 (Prediction Error Exceeds Limit): The risk levels are the same, but the predicted risk value deviation is 22 points. A preset error threshold of 15 points (absolute deviation of risk value) is set; 22 > 15, so condition 5 is satisfied.
[0294] If five out of the five conditions are met (the number of conditions met is ≥ 1), the system determines that task refactoring is triggered.
[0295] (B) Generation of candidate task reconstruction schemes;
[0296] Based on the characteristics of the target task and the current resource status, the system generates three candidate task refactoring schemes:
[0297] Option 1 (Priority Upgrade + Resource Addition): Upgrade the task priority from medium priority to first priority, allocate an additional 5 inspection personnel to support it, and split the task so that 3 people can perform the inspection of different material categories in parallel.
[0298] Option 2 (Task Splitting + Parallel Execution): The original "Supplier Quality Anomaly Review Task" is split into 5 sub-tasks, each corresponding to an independent review of 5 material categories. Each sub-task is assigned to 1 inspection personnel, for a total of 5 people to execute in parallel.
[0299] Option 3 (Preventive Task Generation + Resource Reservation): Retain the original tasks and generate 3 additional preventive tasks (material pre-inspection, supplier on-site audit, and alternative supplier evaluation), and reserve 2 inspection personnel as emergency reserves.
[0300] (C) Cost calculation of the proposed solution;
[0301] The system uses the following cost calculation formula:
[0302] ;
[0303] in, The overall cost of the restructuring scheme for the m-th task (the lower the value, the better); The expected remaining risks after the implementation of the plan; The total delay in task processing caused by the solution; This represents the total amount of resources used by the plan; To determine the resource load balancing level after the implementation of the solution; =0.40、 =0.20、 =0.15、 =0.25 is the corresponding weight parameter.
[0304] The costs of each option are calculated as follows:
[0305] Option 1: Expecting Residual Risk =15 (High-risk events have been handled but residual risks remain), mission delayed =4 hours (resource allocation time), resource usage =8 people, load balancing =75 (Normalized standard deviation of resource load; higher values indicate better load balance):
[0306] ;
[0307] ;
[0308] Option 2: Expecting Residual Risk =20 (Task splitting may result in insufficient coordination of some subtasks), task delay =2 hours (can be started quickly), resource consumption =5 people, load balancing =85 (tasks evenly distributed):
[0309] ;
[0310] ;
[0311] Option 3: Expecting Residual Risk =10 (Comprehensive coverage of preventative tasks), task delay =8 hours (time spent generating and approving new tasks), resource consumption =6 people, load balancing =65 (New tasks have caused some resource overload):
[0312] ;
[0313] ;
[0314] Of the three schemes, with the preset comprehensive cost function and corresponding weight parameters remaining unchanged, the comprehensive cost of Scheme 2 is -12.1, the comprehensive cost of Scheme 1 is -10.75, and the comprehensive cost of Scheme 3 is -9.75. Therefore, Scheme 2 has the lowest comprehensive cost. The system selects Scheme 2 as the current task reconstruction scheme according to the preset comprehensive cost comparison rules, that is, splitting the original task into multiple sub-tasks and executing them in parallel. The above scheme selection results are used to illustrate the task reconstruction decision-making process under the current input data, resource status, and preset weight parameters, and do not limit the present invention to selecting the same task reconstruction scheme under other parameter configurations or other business scenarios.
[0315] The system selects a candidate task reconstruction scheme with a lower overall cost based on a preset comprehensive cost comparison rule, and generates corresponding sub-tasks, task execution order, and resource allocation results according to the selected scheme. During task execution, the system continuously collects the reception status, execution status, completion status, and processing results of each sub-task, and feeds back the actual status to the next analysis time window.
[0316] The risk value, latency, resource consumption, load balancing degree, and overall cost of the above candidate solutions are only used to illustrate the solution comparison calculation process. In actual implementation, each parameter should be obtained from real operating data, simulation data, or verified system configurations, and is not intended to limit the invention based on the example values in this embodiment.
[0317] Through a multi-condition joint triggering mechanism, the system can initiate task reconstruction when there are significant changes in the predicted risk level, the scope of the spread of the impact, the predicted occurrence time, the resource requirements, or the prediction error. By using a preset comprehensive cost function to uniformly compare candidate reconstruction schemes, the system can select the task reconstruction scheme with the lower comprehensive cost under the current input data and current weight parameters.
[0318] Example 4: Feedback Parameter Update;
[0319] This embodiment details the specific calculation process for updating the feedback parameters in step S7.
[0320] Suppose the system generates the following data in a certain round of risk prediction:
[0321] Predicted risk value =85 (predicted as a major risk);
[0322] Actual risk value =66 (actually high risk);
[0323] Prediction window adjustment factor =0.456;
[0324] Current risk assessment parameter vector ;
[0325] Step 1: Calculate the prediction error.
[0326] The system uses mean squared error as the loss function:
[0327] ;
[0328] Step 2: Calculate the gradient of the loss function with respect to each parameter.
[0329] For linear weighted models The gradient is:
[0330] ;
[0331] ;
[0332] ;
[0333] ;
[0334] The current risk characteristic value is ,but:
[0335] ;
[0336] ;
[0337] ;
[0338] ;
[0339] Step 3: Update parameters (gradient descent).
[0340] Set the learning rate η = 0.0001:
[0341] ;
[0342] ;
[0343] ;
[0344] ;
[0345] The updated parameter vector is .
[0346] Step 4: Synchronously update the prediction window adjustment parameters.
[0347] The system also updates the prediction window and adjusts parameters based on the prediction error. If the predicted risk value is too high (predicted value > actual value), it indicates that the prediction window may be too long, and the base prediction window needs to be shortened appropriately or the weights adjusted.
[0348] ;
[0349] set up =0.005, current =6.5:
[0350] ;
[0351] Step 5: Update the associated edge parameters.
[0352] The system also updates the parameters of associated edges using the task execution results. If the task corresponding to a certain associated edge executes successfully and eliminates the risk, the historical associated parameters of that edge are strengthened. If execution fails, then the following measures will be taken:
[0353] ;
[0354] set up =0.05, =0.85, task executed successfully (success=1, failure=0):
[0355] ;
[0356] Step 6: Feedback parameters take effect.
[0357] Updated parameters ;
[0358] This was used for risk calculation in the next analysis time window. Assuming the next window... =45、 =15、 =20、 =10, then the updated risk value is:
[0359] ;
[0360] ;
[0361] If the old parameters w=[1.0,0.8,0.9,0.7] are used, the risk value is:
[0362] ;
[0363] The updated parameters reduced the tendency to overestimate risk values (82 → 71.85), improving the accuracy of risk prediction.
[0364] System-level technical effects: This embodiment demonstrates the gradient descent parameter update process based on prediction error. This is achieved by feeding back the difference between the actual risk state and the predicted risk state after task execution to the risk assessment parameters (…). ), Prediction window adjustment parameters ( ) and associated edge parameters ( In each cycle, the system continuously optimizes the accuracy of risk assessment and prediction, achieving a self-correcting closed loop of "prediction → execution → feedback → update".
[0365] This embodiment demonstrates a parameter update process based on prediction error. The system uses the difference between the actual risk state and the predicted risk state after task execution as feedback information, and updates the risk assessment parameters, prediction window adjustment parameters, and related edge parameters in the business object association network based on this feedback information. In this calculation example, the updated parameters adjust the predicted value corresponding to the current sample towards the actual risk value. This result illustrates that prediction error can be fed back to the update process of risk assessment parameters, prediction window adjustment parameters, and related edge parameters. Whether it can improve the overall prediction accuracy should be determined by statistically comparing the prediction errors before and after the update based on independent validation data or continuously running data. Therefore, this embodiment does not directly limit the improvement of overall prediction accuracy based on the calculation results of a single sample.
[0366] Taking the parameters in this embodiment as an example, the risk value for the next analysis time window recalculated using the updated parameters is 71.85, while the risk value calculated using the parameters before the update is 82. This demonstrates that parameter updates change the risk calculation result for the next analysis time window. This value is used to illustrate the parameter update process and its calculation results, and is not directly used as proof of improved prediction accuracy. For a quantitative evaluation of prediction accuracy, prediction error, mean absolute error, or other predetermined evaluation indicators should be used to compare the results before and after the parameter update under the same dataset and operating conditions.
[0367] like Figure 4 As shown, the system of the present invention adopts a layered architecture, including a data access layer, a correlation analysis layer, a risk decision layer, a task scheduling layer, and an execution feedback layer.
[0368] The data access layer includes a multi-source data acquisition module, an interface adaptation module, and a data verification module, which are used to acquire multi-source business data from business databases, interface services, message queues, mobile terminals, and file import channels.
[0369] The association analysis layer includes a data association module, an association network construction module, and a dynamic analysis module, which are used to complete data cleaning, standardization, time alignment, object identifier mapping, association edge construction, and risk feature extraction.
[0370] The risk decision-making layer includes a risk assessment module, a forecast window determination module, a risk prediction module, and a risk propagation analysis module, used to calculate the current dynamic risk value. Determine the risk level and risk change trend, and adaptively determine the forecast time window. Predict future risk conditions and identify risk propagation paths. .
[0371] The task scheduling layer includes a task priority generation module, a resource matching module, a dynamic scheduling module, and a task refactoring module, which are used to generate predicted risk events, determine target tasks, and calculate dynamic priorities. Computational resource matching score Execute task scheduling and task refactoring.
[0372] The execution feedback layer includes an execution feedback module, a prediction error calculation module, and a parameter update module. These modules are used to obtain task execution status and processing results, calculate prediction errors, and update risk assessment parameters. .
[0373] In summary, this solution can be applied to enterprise quality management systems, supply chain management systems, procurement management systems, operations management systems, asset management systems, and other enterprise business management scenarios. By combining adaptive prediction window adjustment, risk propagation analysis, task priority generation, resource matching, and execution feedback loop, this invention can improve the timeliness of risk warning and the efficiency of computing resource utilization, demonstrating clear industrial applicability.
[0374] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention; any equivalent substitutions or conventional modifications that can be made by those skilled in the art without departing from the technical concept of the present invention should be determined in accordance with the scope of protection defined by the claims.
Claims
1. A risk assessment and task scheduling method based on multi-source business data correlation network and adaptive prediction window, characterized in that, include: The system acquires multi-source business data from multiple business objects, performs data verification, format standardization, time alignment, and business object identification mapping on the multi-source business data, and constructs a business object association network based on at least one of the following: business dependency relationship, data association relationship, historical risk co-occurrence relationship, resource sharing relationship, and task dependency relationship. Within the current analysis time window, the system calculates the current dynamic risk value of each business object based on the current business status, historical change trend, abnormal events, and task execution status, and calculates the risk change rate, abnormal event density, and task backlog change rate. Based on the risk change rate, abnormal event density, and task backlog change rate, the system determines the prediction window adjustment amount, shortens or extends the basic prediction time window based on the prediction window adjustment amount, obtains the current prediction time window, and obtains the predicted risk status within the current prediction time window based on the current dynamic risk value and risk change rate. The predicted risk status is input into the business object association network. The risk propagation impact value is calculated based on the association edge parameters and candidate propagation paths to determine the risk propagation impact objects and risk propagation paths. A target task set is generated based on the association relationship between the risk propagation impact objects and business objects with incomplete tasks. Dynamic priority and resource matching score of schedulable resources are calculated only for the target task set. A task scheduling scheme is generated and executed based on the calculation results. Obtain the task execution status, task processing results, and prediction error between the predicted risk status and the actual risk status. Update the risk assessment parameters, prediction window adjustment parameters, and associated edge parameters based on the prediction error, and use the updated parameters for the next analysis time window.
2. The risk assessment and task scheduling method based on multi-source service data correlation network and adaptive prediction window according to claim 1, characterized in that, The multi-source business data includes basic business data, business process data, business result data, real-time status data, abnormal event data, historical risk data, task execution data, and resource status data. The multi-source business data is associated according to the business object identifier and timestamp, and a unified data input is formed through data quality verification, outlier handling, missing value handling, and data standardization. 3.The risk assessment and task scheduling method based on multi-source service data correlation network and adaptive prediction window according to claim 1, wherein, The business object association network includes multiple business object nodes and association edges connecting the business object nodes. Each association edge is configured with an association strength parameter. The association strength parameter is determined based on at least one of the following: business dependency relationship, data association relationship, historical risk co-occurrence relationship, resource sharing relationship, and task dependency relationship between business objects. 4.The risk assessment and task scheduling method based on multi-source service data correlation network and adaptive prediction window according to claim 1, wherein, The adjustment amount of the prediction window is determined based on the degree of deviation of the risk change rate, abnormal event density, and task backlog change rate from their respective preset thresholds. When at least one of the risk change rate, abnormal event density, and task backlog change rate exceeds the corresponding preset threshold, the basic prediction time window is shortened according to the corresponding degree of deviation. When none of the risk change rate, abnormal event density, and task backlog change rate exceed the corresponding preset threshold, the basic prediction time window is extended according to the corresponding degree of deviation. The adjusted current prediction time window is not less than a preset lower limit and not greater than a preset upper limit.
5. The risk assessment and task scheduling method based on multi-source service data correlation network and adaptive prediction window according to claim 1, characterized in that, The risk propagation impact value is calculated based on the predicted risk value of the source business object, the correlation strength parameter between the source business object and the target business object, the propagation attenuation parameter, and the propagation path length. The degree of change in the predicted risk of the target business object is determined based on the risk propagation impact value.
6. The risk assessment and task scheduling method based on multi-source service data correlation network and adaptive prediction window according to claim 1, characterized in that, The dynamic priority of the target task is determined by combining at least two of the following: predicted risk level, predicted risk change trend, risk propagation impact range, predicted occurrence time, and target task processing time limit, and the target tasks are sorted according to the dynamic priority.
7. The risk assessment and task scheduling method based on multi-source business data correlation network and adaptive prediction window according to claim 1, characterized in that, The resource matching score is calculated jointly based on the resource capacity of schedulable resources, historical task execution results, current availability, current task load, and task response constraints. Schedulable resources that meet preset scheduling conditions are selected based on the resource type, resource quantity, and task response time requirements of the target task. 8.The risk assessment and task scheduling method based on multi-source business data correlation network and adaptive prediction window of claim 1, wherein, When the predicted risk level changes exceed a preset level threshold, the scope of risk propagation exceeds a preset range threshold, the predicted occurrence time enters a preset emergency time range, or the prediction error between the predicted risk and the actual risk exceeds a preset error threshold, task reconfiguration is triggered; the task reconfiguration includes at least one of the following: task priority adjustment, task execution order adjustment, resource reallocation, task splitting, task merging, and preventive task generation. 9.The risk assessment and task scheduling method based on multi-source business data correlation network and adaptive prediction window of claim 1, wherein, The prediction error is calculated based on the predicted risk status and the actual risk status obtained during task execution. The risk assessment parameters, prediction window adjustment parameters, and associated edge parameters are updated based on the prediction error so that the updated parameters can be used for risk assessment, prediction window determination, risk propagation calculation, and task scheduling in the next analysis cycle. 10.A risk assessment and task scheduling system based on multi-source business data correlation network and adaptive prediction window, characterized in that, include: The multi-source data processing module is used to acquire multi-source business data from multiple business objects and perform data validation, format standardization, time alignment, and business object identifier mapping. The network construction module is used to construct a network of business objects based on at least one of the following: business dependency, data dependency, historical risk co-occurrence, resource sharing, and task dependency, and to set the parameters of the associated edges. The dynamic risk assessment module is used to calculate the current dynamic risk value of each business object within the current analysis time window based on the current business status, historical change trends, abnormal events, and task execution status, and to obtain the risk change rate, abnormal event density, and task backlog change rate. The prediction window and risk prediction module are used to determine the adjustment amount of the prediction window based on the rate of change of risk, the density of abnormal events and the rate of change of task backlog, adjust the basic prediction time window and obtain the predicted risk status. The risk propagation analysis module is used to input the predicted risk status into the business object association network, calculate the risk propagation impact value based on the association edge parameters and candidate propagation paths, and determine the risk propagation impact objects and risk propagation paths; The task screening and scheduling module is used to generate a set of target tasks based on the business object association between the objects affected by risk propagation and the unfinished tasks, and to calculate the dynamic priority and resource matching score of schedulable resources only for the set of target tasks and to execute task scheduling. The feedback update module is used to update the risk assessment parameters, prediction window adjustment parameters, and associated edge parameters based on the task execution status, task processing results, and the prediction error between the predicted risk status and the actual risk status, and then use the updated parameters for the next analysis time window.
