An artificial intelligence-based workflow intelligent management method and system

Abstract

The application discloses a kind of based on artificial intelligence's work flow intelligent management method and system, it is related to work flow intelligent management technical field, including: using incremental learning model to analyze work flow historical data, construct the probability distribution model of task completion time, and calculate the risk coefficient of task based on probability distribution model;According to risk coefficient, combine expert rule base dynamically adjusts the buffer time and resource reservation strategy of task, when risk coefficient exceeds preset threshold, trigger early warning mechanism;Classify processing is carried out to early warning task by incremental decision tree, automatically generate risk response scheme, and feedback processing result to expert rule base and carry out rule update.The application can accurately predict task delay and resource conflict risk, realizes the self-adaptive optimization of system, and provides a more intelligent, adaptive work flow management solution for enterprise.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of workflow management, in particular to an intelligent workflow management method and system based on artificial intelligence. BACKGROUND

[0002] With the deepening of enterprise digital transformation, workflow management system (WfMS) has become a key tool to improve organizational operational efficiency. Traditional workflow management mainly relies on static process definition and fixed resource allocation strategy, coordinating task execution through predefined business rules and scheduling algorithms. In recent years, with the rapid development of artificial intelligence technology, especially the wide application of machine learning and deep learning in process mining and predictive analysis, intelligent workflow management has gradually become a research hotspot. Existing technologies have made significant progress in workflow modeling, task scheduling optimization, resource allocation, etc., such as automatic discovery of workflow model based on process mining, task scheduling optimization based on heuristic algorithm, resource collaborative allocation based on multi-agent system, etc.

[0003] However, the existing workflow management methods still have many limitations when facing complex and variable business environment. First, traditional static threshold and fixed rules are difficult to adapt to dynamic changes in business needs, resulting in insufficient prediction accuracy of system for task delay risk and resource conflict. Second, existing risk assessment models often use a single assessment index, lacking comprehensive consideration of multi-dimensional risk factors, and cannot accurately reflect the actual risk situation in the task execution process. Third, existing technologies generally adopt passive response strategy in resource scheduling, lack of active early warning and dynamic adjustment mechanism, resulting in significant performance decline of the system under high load or emergency state. Finally, existing methods have obvious shortcomings in knowledge accumulation and experience inheritance, making it difficult to realize continuous optimization and improvement of system performance.

[0004] The existing workflow management technology mainly has the technical problems of low task risk prediction accuracy, low resource scheduling efficiency, poor system adaptability, etc. The present application effectively solves the above technical problems by introducing an incremental learning model, a dynamic risk assessment mechanism and an intelligent early warning response, providing a more intelligent and adaptive workflow management solution for enterprises. SUMMARY

[0005] In view of the above problems, the present application is proposed.

[0006] Therefore, the present application provides an intelligent workflow management method and system based on artificial intelligence, which can solve the problems mentioned in the background art.

[0007] To solve the above technical problems, the application provides the following technical scheme: a work flow intelligent management method based on artificial intelligence, comprising: using an incremental learning model to analyze work flow historical data, constructing a probability distribution model of task completion time, and calculating a risk coefficient of a task based on the probability distribution model; the risk coefficient comprises a task delay probability and a resource conflict probability; according to the risk coefficient, combining an expert rule base to dynamically adjust a buffer time of the task and a resource reservation strategy, and triggering an early warning mechanism when the risk coefficient exceeds a preset threshold; the early warning mechanism sets different early warning levels based on the criticality of the task; classifying and processing the early warning task through an incremental decision tree, automatically generating a risk response scheme, and feeding back the processing result to the expert rule base for rule updating.

[0008] As a preferred scheme of the work flow intelligent management method based on artificial intelligence, the process of calculating the risk coefficient comprises the following steps: using an incremental learning model to extract features of task start and end time, resource occupation information and task dependency relationship in the work flow historical data to obtain a feature vector; constructing a probability distribution model of task completion time by using the feature extraction result; calculating a task delay probability based on the probability distribution model; calculating a resource conflict probability between tasks; and calculating a risk coefficient of the task by FMEA method based on the delay probability and the resource conflict probability.

[0009] As a preferred scheme of the work flow intelligent management method based on artificial intelligence, the probability distribution model is represented by the following formula,

[0010]

[0011] wherein μ t is the average completion time of the historical task; is the variance of the completion time; and P(t) is the output of the probability distribution model.

[0012] The delay probability is represented by the following formula: for a specified completion time T deadline , the delay probability is:

[0013]

[0014] Through standardization transformation, the following is obtained:

[0015]

[0016] wherein Φ(x) is the cumulative distribution function of the standard normal distribution; and P d is the delay probability.

[0017] The calculation of the resource conflict probability comprises the following steps:

[0018] Resource-time occupancy set O of a construction task task :

[0019] O task = (r m , t s ) | r m ∈ R set , t s ∈ [T start , T end ];

[0020] wherein r m is the mth required resource; R set is the resource set required by the task; t s is the time point; T start is the task start time; T end is the task estimated end time.

[0021] Calculate the resource conflict probability P based on the resource-time intersection c :

[0022]

[0023] wherein O i is the resource-time occupancy set of task i; O j is the resource-time occupancy set of task j; |O i ∩O j | represents the intersection cardinality of the occupancy set; |O i | represents the cardinality of the occupancy set; Γ is the task pair set satisfying the following conditions: there is an overlap in the execution time period of task i and task j; there is an intersection in the resource requirement set of task i and task j.

[0024] The risk coefficient is represented by the following formula:

[0025] R = O risk × S risk × D risk ;

[0026] wherein O risk is the risk occurrence rate: O risk = max(P d , P c ) × 10;

[0027] S risk is the risk severity: S risk = -ln(1-P d ) × 5;

[0028] D risk is the risk detection difficulty: Drisk = (1 + P c ) 2 × 5.

[0029] As a preferred scheme of the artificial intelligence-based workflow intelligent management method, the expert rule base contains a risk coefficient threshold matrix M threshold and a task criticality matrix M critical ; the task level includes high-level tasks, medium-level tasks, and low-level tasks, and the running state includes normal state, bottleneck state, and emergency state.

[0030] As a preferred scheme of the artificial intelligence-based workflow intelligent management method, the expression of the risk coefficient threshold matrix M threshold is:

[0031]

[0032] The expression of the task criticality matrix M critical is:

[0033]

[0034] wherein R tij is the risk threshold of the i-level task in the j-state; K ij is the criticality coefficient of the i-level task in the j-state; i∈[1, 3], indicating the task level, i=1: low-level task, i=2: medium-level task, i=3: high-level task; j∈[1, 3], indicating the running state, j=1: normal state, j=2: bottleneck state, j=3: emergency state; R t21 is the risk threshold of the medium-level task in the normal state, serving as a medium-level threshold for primary warning determination; R t31 is the risk threshold of the high-level task in the normal state, serving as a high-level threshold for secondary and tertiary warning determination; K 21 is the criticality coefficient of the medium-level task in the normal state, serving as a medium-level threshold for primary warning determination; K 31 is the criticality coefficient of the high-level task in the normal state, serving as a high-level threshold for secondary and tertiary warning determination.

[0035] As a preferred scheme of the artificial intelligence-based workflow intelligent management method, according to the risk coefficient, the buffer time and resource reservation strategy of the task are dynamically adjusted in combination with the expert rule base, and the warning mechanism is triggered when the risk coefficient exceeds the preset threshold, including the following steps: establishing an expert rule base;

[0036] calculating the buffer time T buffer:

[0037] T buffer =T base ×(1+α t R task +β t K task );

[0038] Among them, T base R is the baseline buffer time. task K represents the risk factor for the current task. task α represents the criticality coefficient of the current task. t β is the risk weighting coefficient. t This is the keyness weighting coefficient.

[0039] Resource reservation ratio P for computational tasks reserve :

[0040]

[0041] Among them, P min Minimum resource reservation ratio; P max R represents the maximum resource reservation ratio. task K represents the risk factor for the current task. task R represents the criticality coefficient of the current task. threshold K represents the upper limit of the risk coefficient threshold. max This represents the upper limit of the criticality coefficient.

[0042] The risk coefficient and the risk coefficient threshold matrix M threshold Advanced threshold R in t31 and intermediate threshold R t21 The task criticality is compared with the task criticality matrix M. critical Advanced threshold K 31 and intermediate threshold K 21 Compare them.

[0043] Based on the comparison results, the warning levels are divided into Level 0, Level 1, Level 2, and Level 3; when the risk coefficient exceeds R... t31 And the criticality of the task exceeds K. 31 When the risk factor exceeds R, it is determined to be a Level 3 warning; when the risk factor exceeds R... t31 Or the criticality of the task exceeds K. 31 When the risk factor exceeds R, it is determined to be a Level 2 warning; when the risk factor exceeds R... t21 And the criticality of the task exceeds K. 21 When the warning is in effect, it is classified as a Level 1 warning; otherwise, it is classified as a Level 0 warning.

[0044] The pre-warning level is corresponded with a pre-warning response; wherein, the three-level pre-warning triggers three pre-warning responses of pre-warning notification, emergency resource pool starting and task re-planning, the two-level pre-warning triggers two pre-warning responses of pre-warning notification and emergency resource pool starting, the one-level pre-warning triggers one pre-warning response of pre-warning notification, and the zero-level pre-warning does not trigger pre-warning response.

[0045] The buffer time T buffer , the resource reservation ratio P reserve and the pre-warning response Action warn are cooperatively scheduled, and a scheduling scheme matrix is output.

[0046] As a preferred scheme of the work flow intelligent management method based on artificial intelligence, the pre-warning task is classified and processed through the incremental decision tree, a risk response scheme is automatically generated, and the processing result is fed back to the expert rule base for rule updating, including the following steps: a feature space of the incremental decision tree is constructed based on a risk coefficient, a task criticality, a buffer time, a resource reservation ratio, a pre-warning level and a pre-warning response, and an information gain is used as a node splitting criterion to construct an incremental decision tree model; processing effect evaluation indexes including a pre-warning accuracy, a response time and resource consumption are set, the decision tree is pruned and optimized when the processing effect of the sub-tree is less than or equal to the product of the processing effect of the parent node and a pruning threshold coefficient, and the decision tree leaf node path meeting a rule updating threshold is updated to the expert rule base.

[0047] To further solve the above technical problems, the present application provides the following technical scheme: a work flow intelligent management system based on artificial intelligence, comprising: an intelligent analysis module for analyzing work flow historical data using an incremental learning model, constructing a probability distribution model of task completion time, and calculating a risk coefficient of the task based on the probability distribution model; a dynamic adjustment module for dynamically adjusting the buffer time and resource reservation strategy of the task according to the risk coefficient in combination with an expert rule base, and triggering a pre-warning mechanism when the risk coefficient exceeds a preset threshold; and an intelligent response module for classifying and processing pre-warning tasks through an incremental decision tree, automatically generating a risk response scheme, and feeding back the processing result to the expert rule base for rule updating.

[0048] A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the work flow intelligent management method based on artificial intelligence when executing the computer program.

[0049] A computer readable storage medium having a computer program stored thereon, characterized in that the computer program is executed by a processor to implement the steps of the work flow intelligent management method based on artificial intelligence.

[0050] The application has the advantages that the application can accurately predict task delay and resource conflict risks by adopting an incremental learning model to construct a probability distribution model of task completion time and combining FMEA method to calculate risk coefficients; the system is adaptively optimized by dynamically adjusting task buffer time and resource reservation strategies through an expert rule base; the early warning tasks are classified and processed by an incremental decision tree, and the processing results are fed back to the expert rule base, forming a closed-loop continuous optimization mechanism. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0052] Figure 1 A schematic diagram of the overall process of the work flow intelligent management method based on artificial intelligence proposed by the present application;

[0053] Figure 2 A computer device diagram in the work flow intelligent management method based on artificial intelligence proposed by the present application. DETAILED DESCRIPTION

[0054] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings in the specification. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0055] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the scope of the present application. Therefore, the present application is not limited to the specific embodiments disclosed below.

[0056] Embodiment 1, refer to Figure 1 An embodiment of the present application provides a work flow intelligent management method based on artificial intelligence.

[0057] In the related art, first, the traditional static threshold and fixed rule are difficult to adapt to the dynamically changing business needs, resulting in insufficient prediction accuracy of the system for task delay risk and resource conflict. Second, the existing risk assessment model often uses a single evaluation index, lacks comprehensive consideration of multi-dimensional risk factors, and cannot accurately reflect the actual risk status in the task execution process. Third, the existing technology generally adopts a passive response strategy in resource scheduling, lacks active early warning and dynamic adjustment mechanism, resulting in significant performance degradation of the system under high load or emergency state. Finally, the existing method has obvious shortcomings in knowledge accumulation and experience inheritance, and it is difficult to realize the continuous optimization and improvement of system performance.

[0058] The present application provides a method for intelligently managing a workflow based on artificial intelligence, which can effectively solve the above-mentioned problems. Next, how to realize the method will be described in detail in combination with multiple embodiments.

[0059] Figure 1 A flowchart of the overall process of the method for intelligently managing a workflow based on artificial intelligence is shown, which includes:

[0060] S1: Analyze the workflow historical data using an incremental learning model, build a probability distribution model of task completion time, and calculate the risk coefficient of the task based on the probability distribution model.

[0061] Specifically, the risk coefficient includes the task delay probability and the resource conflict probability.

[0062] S1.1: Feature extraction is performed on the task start and end time T h , resource occupation information R t , and task dependency relationship D t in the workflow historical data using an incremental learning model to obtain a feature vector F = [T t , R t , D t ].

[0063] wherein the incremental learning model updates the feature weight based on the following function:

[0064] w k+1 = w k + η (y k - f(x k )) x k ;

[0065] wherein w k is the feature weight of the kth iteration; η is the learning rate; y k is the true value of the kth iteration; f(x k ) is the predicted value of the kth iteration; and x k is the input feature of the kth iteration.

[0066] It should be noted that by using the incremental learning model for data feature extraction, the online processing capability of historical data is realized. Compared with the traditional batch learning method, this step does not need to load all historical data at once, effectively reducing the system storage overhead, and achieving the effect of dynamic updating of feature extraction process.

[0067] S1.2: Construct the probability distribution model P(t) of task completion time using the feature extraction result:

[0068]

[0069] Wherein, μ t is the average completion time of historical tasks; is the variance of the completion time; t i is the completion time of the i-th historical task; N is the total number of historical tasks.

[0070] It should be noted that by constructing the task completion time probability model based on normal distribution, the discrete historical data is converted into continuous probability distribution function, which provides the necessary mathematical basis for the subsequent delay risk calculation.

[0071] S1.3: Calculate the task delay probability P d based on the probability distribution model P(t) of S1.2:

[0072] For the specified completion time T deadline , the delay probability is:

[0073]

[0074] Through standardization transformation, we get:

[0075]

[0076] Wherein, Φ(x) is the cumulative distribution function of standard normal distribution.

[0077] It should be noted that by establishing the calculation method of task delay probability and introducing standardization transformation to simplify the calculation process, the quantitative evaluation of task delay risk is realized. This step not only provides a quantitative index of delay risk, but also improves the calculation efficiency through standardization processing, and finally achieves the effect of accurate measurement of delay risk.

[0078] S1.4: Calculate the resource conflict probability P d between tasks:

[0079] First, construct the resource-time occupation set O task of the task:

[0080] O task= (r m , t s ) | r m ∈ R set , t s ∈ [T start , T end ];

[0081] wherein r m is the mth required resource; R set is the resource set required by the task; t s is the time point; T start is the task start time; T end is the task estimated end time.

[0082] Then the conflict probability P c is calculated based on the resource-time intersection:

[0083]

[0084] wherein O i is the resource-time occupancy set of task i; O j is the resource-time occupancy set of task j; |O i ∩O j | represents the intersection cardinality of the occupancy set; |O i | represents the cardinality of the occupancy set; Γ is a set of task pairs satisfying the following conditions:

[0085] The execution time period of task i and task j has overlap;

[0086] The resource requirement set of task i and task j has intersection;

[0087] It should be noted that by proposing a resource conflict evaluation scheme based on set operation, the unified processing of time dimension and resource dimension is realized, and the calculation of resource conflict probability is converted into set intersection operation, which breaks through the limitation of traditional method considering only single dimension, and finally achieves the accurate quantification effect of resource conflict.

[0088] S1.5: Based on the delay probability P d and the resource conflict probability P c , the risk coefficient R of the task is calculated by using the improved FMEA method:

[0089] R = O risk × S risk × D risk ;

[0090] wherein the calculation of the three risk elements is as follows:

[0091] Risk occurrence rate O risk : Orisk = max(P d , P c ) x 10;

[0092] Risk severity S risk : S risk = -ln(1-P d ) x 5;

[0093] Risk detection difficulty D risk : D risk = (1+P c ) 2 x 5.

[0094] Further, compared with the traditional FMEA method, the present application mainly makes the following improvements:

[0095] 1. The traditional FMEA method based on experience scoring is improved to a quantitative calculation method based on probability;

[0096] 2. The risk occurrence rate O risk is no longer using 1-10 discrete scoring, but using max(P d , P c ) and multiplying 10 for quantification;

[0097] 3. The risk severity S risk introduces the logarithmic function-ln(1-P d ) to establish the relationship with the delay probability, which reflects the nonlinear loss caused by the delay;

[0098] 4. The risk detection difficulty D risk adopts a quadratic function (1+P c ) 2 to depict the detection complexity growth caused by resource conflict.

[0099] It should be noted that by constructing an improved FMEA risk assessment scheme, a precise mapping from probability indicators to risk coefficients was achieved. In this step, the risk occurrence rate uses the maximum value of the delay probability and resource conflict probability as a benchmark, reflecting the weakest link theory in the "barrel effect." The risk severity introduces a logarithmic transformation, increasing rapidly when the delay probability approaches 1, reflecting the nonlinear loss caused by task delays. The risk detection difficulty is enhanced by a quadratic transformation to amplify the impact of resource conflicts, highlighting the complexity of resource scheduling in multi-task scenarios. This assessment method based on mathematical transformation not only maintains the universality of the FMEA framework but also enhances the discriminative power of risk indicators through nonlinear mapping, making the risk assessment results more consistent with empirical patterns in engineering practice. Ultimately, it provides a reliable quantitative basis for workflow risk management and resource scheduling decisions. In particular, by unifying delay risk and resource conflict risk under the FMEA framework, a comprehensive assessment of multi-dimensional risks is achieved, overcoming the problem of fragmented risk assessment indicators in traditional methods, and providing more comprehensive decision support for risk prevention and control in workflow systems.

[0100] S2: Based on the risk coefficient and combined with the expert rule base, dynamically adjust the task's buffer time and resource reservation strategy. When the risk coefficient exceeds the preset threshold, trigger the early warning mechanism.

[0101] Specifically, the early warning mechanism sets different early warning levels based on the criticality of the task.

[0102] S2.1: Establish an expert rule base.

[0103] The expert rule base contains a risk coefficient threshold matrix M for different task levels and different operating states. threshold and the task criticality matrix M critical The task level is divided into three levels: high, medium and low, and the operation status includes three states: normal, bottleneck and emergency.

[0104] Risk coefficient threshold matrix M threshold The expression is:

[0105]

[0106] Task criticality matrix M critical The expression is:

[0107]

[0108] Among them, R tij Represents the risk threshold of the i-th level task in the j-th state, in R tij In this context, 't' stands for 'threshold', used to distinguish it as a threshold type parameter; K ijKey degree coefficient of the ith level task in the jth state; i∈[1, 3] represents the task level, i = 1: low-level task, i = 2: medium-level task, i = 3: high-level task; j∈[1, 3] represents the running state, j = 1: normal state, j = 2: bottleneck state, j = 3: emergency state.

[0109] Risk coefficient threshold matrix M threshold The meaning of each element in the matrix is as follows:

[0110] R t11 : Risk threshold of low-level task in normal state;

[0111] R t12 : Risk threshold of low-level task in bottleneck state;

[0112] R t13 : Risk threshold of low-level task in emergency state;

[0113] R t21 : Risk threshold of medium-level task in normal state, used as a medium-level threshold for first-level early warning judgment;

[0114] R t22 : Risk threshold of medium-level task in bottleneck state;

[0115] R t23 : Risk threshold of medium-level task in emergency state;

[0116] R t31 : Risk threshold of high-level task in normal state, used as a high-level threshold for second-level and third-level early warning judgment;

[0117] R t32 : Risk threshold of high-level task in bottleneck state;

[0118] R t33 : Risk threshold of high-level task in emergency state.

[0119] Task key degree matrix M critical The meaning of each element in the matrix is as follows:

[0120] K 11 : Key degree coefficient of low-level task in normal state;

[0121] K 12 : Key degree coefficient of low-level task in bottleneck state;

[0122] K 13 : Key degree coefficient of low-level task in emergency state;

[0123] K 21: the criticality coefficient of the middle-level task in the normal state, which is used as a middle-level threshold for the first-level early warning judgment;

[0124] K 22 : the criticality coefficient of the middle-level task in the bottleneck state;

[0125] K 23 : the criticality coefficient of the middle-level task in the emergency state;

[0126] K 31 : the criticality coefficient of the high-level task in the normal state, which is used as a high-level threshold for the second-level and third-level early warning judgment;

[0127] K 32 : the criticality coefficient of the high-level task in the bottleneck state;

[0128] K 33 : the criticality coefficient of the high-level task in the emergency state.

[0129] Preferably, the design of the expert rule base adopts a double-matrix (risk coefficient threshold matrix and task criticality matrix) structure, which realizes the accurate definition of the threshold values of different task levels in different system states by taking the task level and the running state as the two dimensions of the matrix. This double-matrix structure design overcomes the limitations of the existing technology which often only uses a single threshold or simple classification, so that the present application can more accurately depict the task risk and criticality characteristics, providing more reliable decision-making basis for subsequent early warning judgment.

[0130] S2.2: Calculate the buffer time T of the task buffer :

[0131] T buffer = T base × (1 + α t R task + β t K task );

[0132] wherein, T base is the baseline buffer time; R task is the risk coefficient of the current task; K task is the criticality coefficient of the current task; α t is the risk weight coefficient, with a value range of [0, 1]; β t is the criticality weight coefficient, with a value range of [0, 1].

[0133] Preferably, the calculation of the buffer time introduces a risk weight coefficient and a criticality weight coefficient, and quantitatively weights the influence of the task risk coefficient and the criticality coefficient, thereby breaking through the problem of oversimplification or rigidity in the traditional method of setting the buffer time, realizing dynamic adaptive adjustment of the buffer time, and significantly improving the flexibility and accuracy of time scheduling.

[0134] S2.3: Calculate the resource reservation ratio P of the task reserve :

[0135]

[0136] wherein, P min is the minimum resource reservation ratio; P max is the maximum resource reservation ratio; R task is the risk coefficient of the current task; K task is the criticality coefficient of the current task; R threshold is the upper limit of the risk coefficient; K max is the upper limit of the criticality coefficient.

[0137] Preferably, the calculation of the resource reservation ratio adopts a nonlinear mapping scheme based on the risk coefficient and the criticality coefficient, establishes a functional relationship between resource reservation and task characteristics, thereby solving the problem of fixed proportion or empirical value in the resource reservation in the prior art, realizing the intelligentization and precision of resource reservation, and effectively improving the resource utilization efficiency.

[0138] S2.4: Compare the risk coefficient with the high-level threshold R t31 and the medium-level threshold R t21 in the risk coefficient threshold matrix, compare the task criticality with the high-level threshold K 31 and the medium-level threshold K 21 in the task criticality matrix, and divide the warning level into four levels of zero, one, two and three according to the comparison results, wherein when the risk coefficient exceeds R t31 and the task criticality exceeds K 31 , it is determined as a three-level warning, when the risk coefficient exceeds R t31 or the task criticality exceeds K 31 , it is determined as a two-level warning, when the risk coefficient exceeds R t21 and the task criticality exceeds K 21 , it is determined as a one-level warning, and the rest is determined as a zero-level warning.

[0139] Preferably, S2.4 adopts a "double-threshold double-index" judgment scheme, combines the use of high and low thresholds of risk coefficient and criticality, and constructs a four-level early warning system to overcome the limitations of existing early warning mechanisms that often rely on a single index or use simple threshold judgment, achieving more comprehensive and accurate risk early warning and significantly improving the system's risk prevention and control capabilities.

[0140] S2.5: Establish a corresponding relationship between the early warning level and the early warning response, wherein the third-level early warning triggers three early warning responses of early warning notification, emergency resource pool start, and task re-planning, the second-level early warning triggers two early warning responses of early warning notification and emergency resource pool start, the first-level early warning triggers one early warning response of early warning notification, and the zero-level early warning does not trigger an early warning response.

[0141] S2.6: Coordinate the scheduling of buffer time P buffer , resource reservation ratio P reserve , and early warning response Action warn :

[0142] Schedule=f(T buffer , P reserve , Action warn );

[0143] Wherein, f is a scheduling function, and the output is a scheduling scheme matrix:

[0144] Schedule=[Task id T adj R adj A exe ];

[0145] Wherein, Task id is the task identification number; T adj is the adjusted time arrangement; R adj is the adjusted resource allocation; and A exe is the type of early warning response executed.

[0146] It should be noted that the scheduling scheme matrix includes four dimensions of task identification number, adjusted time arrangement, adjusted resource allocation, and executed early warning response type, and the time dimension, resource dimension, and early warning dimension of the task are uniformly coordinated and dynamically adjusted through the scheduling scheme matrix.

[0147] Preferably, the coordinated scheduling adopts a multi-dimensional scheduling matrix structure containing task identification number, time arrangement, resource allocation, and early warning response type, and realizes the unified coordination of time dimension, resource dimension, and early warning dimension through the scheduling function. This design solves the problem of independent adjustment of each dimension in the prior art, realizes the systematicness and integrity of task adjustment, and significantly improves the scientificity and feasibility of the scheduling scheme.

[0148] S3: classifying and processing the early warning task through the incremental decision tree, automatically generating a risk response scheme, and feeding back the processing result to the expert rule base for rule updating.

[0149] S3.1: constructing a feature space of the incremental decision tree based on a risk coefficient, a task criticality, a buffer time, a resource reservation ratio, an early warning level and an early warning response, and using information gain as a node splitting criterion to construct an incremental decision tree model;

[0150] S3.2: setting processing effect evaluation indexes including early warning accuracy, response time and resource consumption, pruning and optimizing the decision tree when the processing effect of a sub-tree is less than or equal to the product of the processing effect of a parent node and a pruning threshold coefficient, and updating the decision tree leaf node path meeting a rule updating threshold to the expert rule base.

[0151] In summary, the present application can accurately predict task delay and resource conflict risks by using an incremental learning model to construct a probability distribution model of task completion time and combining the FMEA method to calculate the risk coefficient, dynamically adjusts the task buffer time and resource reservation strategy through the expert rule base to realize the adaptive optimization of the system, classifies and processes the early warning task through the incremental decision tree, and feeds back the processing result to the expert rule base to form a closed-loop continuous optimization mechanism.

[0152] Embodiment 2, which is an embodiment of the present application, provides an artificial intelligence-based workflow intelligent management system, comprising: an intelligent analysis module for analyzing workflow historical data using an incremental learning model, constructing a probability distribution model of task completion time, and calculating a risk coefficient of the task based on the probability distribution model; a dynamic adjustment module for dynamically adjusting a buffer time and a resource reservation strategy of the task according to the risk coefficient in combination with an expert rule base, and triggering an early warning mechanism when the risk coefficient exceeds a preset threshold; and an intelligent response module for classifying and processing the early warning task through the incremental decision tree, automatically generating a risk response scheme, and feeding back the processing result to the expert rule base for rule updating.

[0153] Embodiment 3, refer to Figure 2For one embodiment of the present application, different from the previous embodiment, the function, if realized in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application or the part of the technical solution that essentially contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in various embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0154] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a list of executable instructions for implementing logic functions, which can be specifically embodied in any computer readable medium for use by or in conjunction with an instruction execution system, apparatus or device, such as a computer-based system, a system including a processor or other system that can fetch and execute instructions from the instruction execution system, apparatus or device. For the purpose of the present specification, the "computer readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by or in conjunction with the instruction execution system, apparatus or device or in conjunction with these instruction execution systems, apparatus or devices.

[0155] More specific examples (a non-exhaustive list) of the computer readable medium include the following: an electrical connection having one or more wires (electrical devices), a portable computer diskette (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer readable medium can even be paper or other suitable medium on which the program can be printed, as the program can be electronically obtained, for example, by optical scanning of the paper or other medium, followed by editing, interpreting or otherwise processing, if necessary, in other suitable ways to be electronically obtained, and then stored in the computer memory.

[0156] It should be understood that various portions of the application can be implemented with hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented with software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following technologies, or a combination thereof, can be used: discrete logic circuitry having logic gates for implementing logic functions upon an application data signal, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like, as is well known in the art.

[0157] Example 4, which is an embodiment of the present application, provides an artificial intelligence-based intelligent management method for work flow. In order to verify the beneficial effects of the present application, economic benefit calculation and simulation experiments are used for scientific demonstration.

[0158] In order to verify the effectiveness of the present application, the present application selects 100 actual work flow tasks of an enterprise as test samples, and processes these tasks using the traditional work flow management method and the intelligent management method of the present application respectively. In the experiment, the present application focuses on key indicators such as task completion rate, resource utilization efficiency, risk early warning accuracy, etc. The traditional method uses fixed task planning and resource allocation strategy, while the present application method dynamically adjusts task parameters based on incremental learning model and performs intelligent early warning. The experiment lasts for 3 months, and tests are carried out under normal load, high load and emergency state respectively.

[0159] Table 1: Comparison data of different management methods on various indicators

[0160] Performance indicators Traditional method The present invention Lifting range Task completion rate 68.5％ 92.0％ 23.5％ Resource utilization efficiency 72.3％ 91.0％ 18.7％ System response time (ms) 850 551 35.2％ Risk early warning accuracy 60.8％ 92.3％ 31.5％ Risk processing timeliness 65.2％ 89.8％ 24.6％ Resource conflict rate 25.3％ 8.7％ 65.6％

[0161] As shown in Table 1, the comparison of six key performance indicators demonstrates that this invention achieves significant performance improvements over traditional methods in all aspects. The most notable improvement is in the resource conflict rate, which has decreased dramatically from 25.3% in the traditional method to 8.7%, a reduction of 65.6%. This fully verifies the effectiveness of the resource conflict prediction mechanism based on the probability distribution model proposed in this invention. The on-time task completion rate and risk warning accuracy have increased by 23.5% and 31.5% respectively, reaching high levels of 92.0% and 92.3%, indicating that the incremental learning model of this invention can accurately grasp task patterns, provide timely warnings, and adjust strategies. The system response time has decreased from 850ms to 551ms, an improvement of 35.2%, indicating that the decision tree model of this invention has good real-time performance. Resource utilization efficiency has increased to 91.0%, an improvement of 18.7% compared to the traditional method, thanks to the dynamic resource reservation strategy proposed in this invention. The 24.6% improvement in the timeliness of risk handling reflects the advantages of this invention in risk response. Overall, these data strongly demonstrate the comprehensive advantages of this invention in all aspects of intelligent workflow management.

[0162] Table 2 System performance data under different operating conditions

[0163]

[0164]

[0165] As shown in Table 2, the adaptive capability and stability of this invention are demonstrated in detail by comparing the system performance under three different operating states (normal load, high load, and emergency state). Under normal load, the performance improvement of this invention compared to traditional methods is relatively moderate, with improvements in warning accuracy, resource utilization, and task completion rate of 18.3%, 17.7%, and 19.5%, respectively. However, the advantages of this invention become more apparent as system pressure increases. Under high load, the improvement in all three indicators increases significantly, reaching 35.4%, 25.3%, and 28.4%, respectively. Especially under emergency state, this invention exhibits excellent performance stability, with a warning accuracy of up to 92.1%, an improvement of 40.7% compared to traditional methods, while resource utilization and task completion rate also improve by 30.9% and 32.3%, respectively. It is worth noting that this invention maintains relatively stable high performance under different pressure conditions (warning accuracy remains between 91% and 93%), while the performance of traditional methods decreases significantly with increasing pressure (warning accuracy drops from 75.2% to 51.4%). This comparison fully demonstrates that the management method based on incremental learning and dynamic adjustment proposed in this invention has excellent stress adaptability and system stability, and is particularly suitable for handling complex and ever-changing workflow scenarios.

[0166] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. An artificial intelligence-based intelligent management method of a workflow, characterized by, Comprise: The incremental learning model is used to analyze the workflow historical data, a probability distribution model of task completion time is constructed, and the risk coefficient of the task is calculated based on the probability distribution model; the risk coefficient includes the task delay probability and the resource conflict probability; According to the risk coefficient, the buffer time and resource reservation strategy of the task are dynamically adjusted combined with the expert rule base, and the early warning mechanism is triggered when the risk coefficient exceeds the preset threshold; the early warning mechanism sets different early warning levels based on the criticality of the task; The early warning task is classified and processed by the incremental decision tree, the risk response scheme is automatically generated, and the processing result is fed back to the expert rule base for rule updating; The process of calculating the risk coefficient includes the following steps, The incremental learning model is used to extract features from the task start and end time, resource occupation information and task dependency relationship in the workflow historical data to obtain a feature vector; A probability distribution model of task completion time is constructed using the feature extraction results; The task delay probability is calculated based on the probability distribution model; The resource conflict probability between tasks is calculated; The risk coefficient of the task is calculated by FMEA method based on the delay probability and the resource conflict probability; The probability distribution model is represented by the following formula, where μ t is the average completion time of historical tasks; is the variance of completion time; P(t) is the output of the probability distribution model; The delay probability is represented by the following formula: For a given completion time T deadline , the probability of delay is: Through standardization transformation, we get: wherein Φ(x) is the cumulative distribution function of the standard normal distribution; P d is the delay probability; The calculation of the resource conflict probability includes the following steps: A resource-time occupancy set O of a build task task : O task = (r m , t s )|r m ∈ R set , t s ∈ [T start , T end ]; wherein r m is the mth required resource; R set is the set of resources required for the task; t s is the time point; T start is the start time of the task; T end is the estimated end time of the task; Calculating resource conflict probability P based on resource-time intersection c : where O i is the resource-time occupancy set of task i; O j is the resource-time occupancy set of task i; |O i ∩O j | denotes the cardinality of the intersection of occupancy sets; |O i | denotes the cardinality of the occupancy set; Γ is the set of task pairs that satisfy the following conditions: The execution time period of task i and task j overlaps; The resource demand set of task i and task j has an intersection; The risk coefficient is represented by the following formula: R = O risk x S risk x D risk ; where O risk is the risk occurrence rate: O risk = max(P d , P c ) x 10; S risk For risk severity: S risk = -ln(l - P d ) x 5; D risk Risk detection difficulty: D risk = (1 + P c ) 2 x 5; The expert rule base comprises a risk coefficient threshold matrix M under different task levels and different running states threshold and a task criticality matrix M critical ; the task levels include high-level tasks, medium-level tasks and low-level tasks, and the running states include normal states, bottleneck states and emergency states; The risk coefficient threshold matrix M threshold The expression is: The mission criticality matrix M critical is expressed by the formula: wherein R tij is the risk threshold of the i-th level task in the j-th state; K ij is the criticality coefficient of the i-th level task in the j-th state; i∈[1,3], indicating the task level, i=1: low-level task, i=2: intermediate-level task, i=3: high-level task; j∈[1,3], indicating the running state, j=1: normal state, j=2: bottleneck state, j=3: emergency state; R t21 is the risk threshold of the intermediate-level task in the normal state, serving as an intermediate-level threshold for primary warning determination; R t31 is the risk threshold of the high-level task in the normal state, serving as a high-level threshold for secondary and tertiary warning determination; K 21 is the criticality coefficient of the intermediate-level task in the normal state, serving as an intermediate-level threshold for primary warning determination; K 31 is the criticality coefficient of the high-level task in the normal state, serving as a high-level threshold for secondary and tertiary warning determination; According to the risk coefficient, the buffer time and resource reservation strategy of the task are dynamically adjusted combined with the expert rule base, and the early warning mechanism is triggered when the risk coefficient exceeds the preset threshold, including the following steps: Establish an expert rule base; Buffer time T of the computing task buffer : T buffer = T base × (1 + α t R task + β t K task ); Wherein, T base is the reference buffer time; R task is the risk coefficient of the current task; K task is the criticality coefficient of the current task; α t is the risk weight coefficient; β t is the criticality weight coefficient; Resource reservation ratio P of a computing task reserve : wherein P min is the minimum resource reservation ratio; P max is the maximum resource reservation ratio; R task is the risk coefficient of the current task; K task is the criticality coefficient of the current task; R threshold is the upper limit of the risk coefficient; K max is the upper limit of the criticality coefficient; comparing the risk coefficient with high-level thresholds R threshold and medium-level thresholds R t31 in the risk coefficient threshold matrix M t21 and comparing the task criticality with high-level thresholds K critical and medium-level thresholds K 31 in the task criticality matrix M 21 ​ According to the comparison result, the early warning level is divided into zero level early warning, first level early warning, second level early warning, and third level early warning; when the risk coefficient exceeds R t31 , and the task criticality exceeds K 31 , it is determined as third level early warning; when the risk coefficient exceeds R t31 or the task criticality exceeds K 31 , it is determined as second level early warning; when the risk coefficient exceeds R t21 , and the task criticality exceeds K 21 , it is determined as first level early warning, and the remaining cases are determined as zero level early warning; The early warning level and the early warning response are correspondingly related; among them, the three-level early warning triggers three early warning responses of early warning notification, emergency resource pool start and task re-planning, the two-level early warning triggers two early warning responses of early warning notification and emergency resource pool start, the one-level early warning triggers one early warning response of early warning notification, and the zero-level early warning does not trigger the early warning response; The buffer time T buffer , the resource reservation ratio P reserve , and the early warning response Action warn are cooperatively scheduled, and a scheduling scheme matrix is output. 2.The artificial intelligence-based workflow intelligentization management method of claim 1, wherein: The early warning task is classified and processed by the incremental decision tree, the risk response scheme is automatically generated, and the processing result is fed back to the expert rule base for rule updating, including the following steps: Based on the risk coefficient, the criticality of the task, the buffer time, the resource reservation ratio, the early warning level and the early warning response, the feature space of the incremental decision tree is constructed, and the information gain is used as the node splitting criterion to construct the incremental decision tree model; Set the processing effect evaluation index including early warning accuracy, response time and resource consumption, when the processing effect of the subtree is less than or equal to the product of the processing effect of the parent node and the pruning threshold coefficient, the decision tree is pruned and optimized, and the decision tree leaf node path that meets the rule updating threshold is updated to the expert rule base.

3. A system employing the artificial intelligence-based workflow intelligent management method according to any one of claims 1 to 2, characterized in that, Comprise: An intelligent analysis module is configured to analyze workflow historical data using an incremental learning model, construct a probability distribution model of task completion time, and calculate a risk coefficient of the task based on the probability distribution model; The dynamic adjustment module is configured to dynamically adjust the buffer time and resource reservation strategy of the task according to the risk coefficient in combination with the expert rule base, and trigger a warning mechanism when the risk coefficient exceeds a preset threshold. The intelligent response module is configured to classify and process the warning task through an incremental decision tree, automatically generate a risk response scheme, and feed back the processing result to the expert rule base for rule updating.

4. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the artificial intelligence-based workflow intelligent management method in any one of claims 1 to 2.

5. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the artificial intelligence-based workflow intelligent management method in any one of claims 1 to 2.

Images