Collaborative Prediction Method and System for Completion Time of Distributed Business Processes

Abstract

The present invention provides a collaborative prediction method and system for the completion time of distributed business processes. The present invention proposes a selection criterion for the completion time prediction models of the main process and sub-processes, a multi-agent model for sub-processes, a multi-agent model for the main process, and a collaborative prediction method for the completion time of distributed business processes. The present invention innovatively proposes a collaborative prediction method for the completion time of distributed business processes based on multi-agent reinforcement learning technology.

Description

Technical Field

[0001] The present invention relates to the technical field of predictive business process monitoring. Specifically, it relates to a collaborative prediction method and system for the completion time of distributed business processes, and particularly to a collaborative prediction method for the completion time of distributed business processes based on multi-agent reinforcement learning. Background Art

[0002] Predictive business process monitoring refers to using historical business process data to predict the future state of the current business process instance, so as to achieve the monitoring of the business process. Predicting the completion time of the business process is an important task of predictive business process monitoring. With the increasingly prominent characteristic of cross-organizational operation of business processes, a new form of business process, namely distributed business processes, has emerged. A distributed business process corresponds to a main process and several sub-processes. The main process consists of a start activity, several key activities, and an end activity. The specific execution of the activity is responsible for the corresponding organizational unit. The main process only cares about when the activity is completed and its impact on the progress of the main process, rather than how the activity is executed. For the organizational unit responsible for executing a certain activity, the execution process of this activity is a complete sub-process. Currently, the research on business process completion time prediction methods is all for business processes within the same organizational unit, lacking a prediction method for the completion time of distributed business processes involving multiple organizational units.

[0003] In the patent document with the publication number CN115714741A, a routing decision method and system based on collaborative multi-agent reinforcement learning are disclosed. The method includes two parts: offline training and online decision-making. Offline training collects the public information in the inter-domain routing in the past period of time, including inter-domain link information and historical routing information, and then uses collaborative multi-agent reinforcement learning to train the offline intelligent routing decision model, and distributes the trained model to the control modules in each domain. During the online decision-making process, for the traffic demand requesting inter-domain routing, the decision model is used to generate the end-to-end inter-domain path for each flow, send the end-to-end path to the control modules in the inter-domain, and split it into multiple local paths in different domains. All local paths constitute an end-to-end inter-domain path.

[0004] Regarding the above related technologies, the inventors believe that the current research on business process completion time prediction methods is all for business processes within the same organizational unit, lacking a prediction method for the completion time of distributed business processes involving multiple organizational units. Therefore, a new technical solution needs to be proposed to improve the above technical problems. Summary of the Invention

[0005] Aiming at the deficiencies in the prior art, the purpose of the present invention is to provide a collaborative prediction method and system for the completion time of distributed business processes.

[0006] A collaborative prediction method for the completion time of a distributed business process provided by the present invention, the method comprising the following steps:

[0007] Step S1: Criteria for selecting the main process and sub-process completion time prediction models;

[0008] Step S2: The sub-process multi-agent model uses the sub-process instance-related information and prediction information as state information to decide whether to report the sub-process completion time prediction value output by the sub-process completion time prediction model to the corresponding main process multi-agent model;

[0009] Step S3: The main process multi-agent model uses the main process instance-related information and the prediction information uploaded by the corresponding sub-process as state information to decide whether to adopt the prediction information uploaded by the sub-process as the start time prediction value of the corresponding key activity, and iteratively calls the main process completion time prediction model to obtain a new main process completion time prediction value;

[0010] Step S4: There is a main process completion time prediction model at the main process end of the distributed business process completion time collaborative prediction method, and several main process multi-agents corresponding to different sub-processes; each of several sub-process ends has a sub-process completion time prediction model and a sub-process multi-agent.

[0011] Preferably, in the step S1:

[0012] The criteria for selecting the main process completion time prediction model, this model takes the prefixes of several activities of the process as input, outputs the start time prediction value of the next activity, iteratively makes predictions, and finally obtains the start time prediction value of the end activity, that is, the prediction value of the main process completion time; uses the mean absolute error between the main process completion time prediction value and the true value of the main process completion time as an index to measure the pros and cons of the model prediction performance;

[0013] The criteria for selecting the sub-process completion time prediction model, this model takes the prefixes of the activities that have occurred in the process as input, and directly outputs the start time prediction value of the sub-process end activity, that is, the prediction value of the sub-process completion time; uses the mean absolute error between the sub-process completion time prediction value and the true value of the sub-process completion time as an index to measure the pros and cons of the model prediction performance.

[0014] Preferably, in the step S2:

[0015] The state of the sub-process multi-agent:

[0016] s k =(σ (k) , σ′ (k) )

[0017] where k is the number of the sub-process, σ(k) is the sub - process prefix, including activity category, name, occurrence time, and a sequence of predicted values of the sub - process completion time given by the sub - process completion time prediction model; σ′ (k) is the main - process prediction prefix, including the predicted value of the sub - process completion time sent from the main - process side to the sub - process.

[0018] Preferably, in the step S3:

[0019] Main - process multi - agent state:

[0020] s k =(σ (k) , σ′ (k) )

[0021] where k is the number of the sub - process corresponding to the main - process multi - agent, σ (k) is the main - process prefix, including activity category, name, occurrence time, the predicted value of the sub - process completion time given by the main - process completion time prediction model, and a sequence of predicted values of the main - process completion time; σ ′(k) is the sub - process prediction prefix, including the predicted value of the sub - process completion time reported by the sub - process to the main - process.

[0022] Preferably, in the step S4:

[0023] When the distributed business process starts an activity, the main - process is started. The main - process side calls the main - process completion time prediction model to obtain the predicted values of the sub - process and main - process completion times, and sends the predicted value of the sub - process completion time to the sub - process side to start the execution of the sub - process.

[0024] The present invention also provides a collaborative prediction system for the completion time of a distributed business process. The system includes the following modules:

[0025] Module M1: The selection criterion for the main - process and sub - process completion time prediction models;

[0026] Module M2: The sub - process multi - agent model uses the sub - process instance - related information and prediction information as state information to decide whether to report the predicted value of the sub - process completion time output by the sub - process completion time prediction model to the corresponding main - process multi - agent model;

[0027] Module M3: The main - process multi - agent model uses the main - process instance - related information and the prediction information uploaded by the corresponding sub - process as state information to decide whether to adopt the prediction information uploaded by the sub - process as the predicted value of the start time of the corresponding key activity, and iteratively calls the main - process completion time prediction model to obtain a new predicted value of the main - process completion time;

[0028] Module M4: There is a main process completion time prediction model at the main process end of the distributed business process completion time collaborative prediction system, and several main process multi-agents corresponding to different sub-processes; each of several sub-process ends has a sub-process completion time prediction model and a sub-process multi-agent.

[0029] Preferably, in the module M1:

[0030] The selection criterion of the main process completion time prediction model: This model takes the prefixes of several activities of the process as input, outputs the predicted value of the start time of the next activity, and iteratively makes predictions to finally obtain the predicted value of the start time of the end activity, that is, the predicted value of the main process completion time; the mean absolute error between the predicted value of the main process completion time and the true value of the main process completion time is used as an index to measure the pros and cons of the model prediction performance.

[0031] The selection criterion of the sub-process completion time prediction model: This model takes the prefixes of the activities that have occurred in the process as input and directly outputs the predicted value of the start time of the end activity of the sub-process, that is, the predicted value of the sub-process completion time; the mean absolute error between the predicted value of the sub-process completion time and the true value of the sub-process completion time is used as an index to measure the pros and cons of the model prediction performance.

[0032] Preferably, in the module M2:

[0033] The state of the sub-process multi-agent:

[0034] s k =(σ (k) , σ' (k) )

[0035] where k is the number of the sub-process, σ (k) is the sub-process prefix, including activity category, name, occurrence time, and the sequence information of the predicted values of the sub-process completion time given by the sub-process completion time prediction model; σ' (k) is the main process prediction prefix, including the predicted value of the sub-process completion time sent from the main process end to the sub-process.

[0036] Preferably, in the module M3:

[0037] The state of the main process multi-agent:

[0038] s k =(σ (k) , σ' (k) )

[0039] where k is the number of the sub-process corresponding to the main process multi-agent, σ (k) is the main process prefix, including activity category, name, occurrence time, the predicted value of the sub-process completion time given by the main process completion time prediction model, and the sequence information of the predicted value of the main process completion time; σ'(k) It is a prefix for sub - process prediction, including the predicted value of the sub - process completion time reported by the sub - process to the main process.

[0040] Preferably, in the module M4:

[0041] When the distributed business process start activity begins, the main process is started. The main - process side calls the main - process completion - time prediction model to obtain the predicted values of the sub - process and the main - process completion times, and sends the predicted value of the sub - process completion time to the sub - process side to start the execution of the sub - process.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] For the brand - new business - process form of the distributed business process, the present invention improves the efficiency of collaborative prediction by reducing the number of calls to the business - process completion - time prediction models and the number of communications between the main - process side and the sub - process side, and improves the accuracy of collaborative prediction by providing more accurate predicted values of the business - process completion time. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] By reading the following detailed description of non - restrictive embodiments with reference to the accompanying drawings, other features, objects, and advantages of the present invention will become more apparent:

[0045] Figure 1 It is a schematic diagram of the main process and sub - processes of the distributed business process of the present invention;

[0046] Figure 2 It is a flowchart of the method for collaborative prediction of the distributed - business - process completion time of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] The present invention will be described in detail below with reference to specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several changes and improvements can still be made. These all belong to the protection scope of the present invention.

[0048] Example 1:

[0049] According to a method for collaborative prediction of the distributed - business - process completion time provided by the present invention, the method includes the following steps:

[0050] Step S1: Criteria for selecting the main process and sub-process completion time prediction models; Criteria for selecting the main process completion time prediction model, which takes the prefixes of several activities in the process as inputs, outputs the predicted value of the start time of the next activity, and iteratively makes predictions to finally obtain the predicted value of the start time of the end activity, that is, the predicted value of the main process completion time; The mean absolute error between the predicted value of the main process completion time and the true value of the main process completion time is used as an indicator to measure the prediction performance of the model; Criteria for selecting the sub-process completion time prediction model, which takes the prefixes of the activities that have occurred in the process as inputs and directly outputs the predicted value of the start time of the end activity of the sub-process, that is, the predicted value of the sub-process completion time; The mean absolute error between the predicted value of the sub-process completion time and the true value of the sub-process completion time is used as an indicator to measure the prediction performance of the model.

[0051] Step S2: The sub-process multi-agent model uses the sub-process instance-related information and prediction information as state information to decide whether to report the predicted value of the sub-process completion time output by the sub-process completion time prediction model to the corresponding main process multi-agent model; Sub-process multi-agent state:

[0052] s k =(σ (k) ,σ′ (k) )

[0053] where k is the number of the sub-process, σ (k) is the sub-process prefix, including the activity category, name, occurrence time, and the sequence information of the predicted values of the sub-process completion time given by the sub-process completion time prediction model; σ′ (k) is the main process prediction prefix, including the predicted value of the sub-process completion time sent from the main process end to the sub-process.

[0054] Step S3: The main process multi-agent model uses the main process instance-related information and the prediction information uploaded by the corresponding sub-process as state information to decide whether to adopt the prediction information uploaded by the sub-process as the predicted value of the start time of the corresponding key activity, and iteratively calls the main process completion time prediction model to obtain a new predicted value of the main process completion time; Main process multi-agent state:

[0055] s k =(σ (k) ,σ′ (k) )

[0056] where k is the number of the sub-process corresponding to the main process multi-agent, σ (k) is the main process prefix, including the activity category, name, occurrence time, the predicted value of the sub-process completion time given by the main process completion time prediction model, and the sequence information of the predicted values of the main process completion time; σ′ (k) is the sub-process prediction prefix, including the predicted value of the sub-process completion time reported by the sub-process to the main process.

[0057] Step S4: At the main process end of the distributed business process completion time collaborative prediction method, there is a main process completion time prediction model and several main process multi-agents corresponding to different sub-processes; at each of several sub-process ends, there is a sub-process completion time prediction model and a sub-process multi-agent; when the distributed business process starts, the main process is started. The main process end calls the main process completion time prediction model to obtain the predicted values of the sub-process and main process completion times, and sends the predicted value of the sub-process completion time to the sub-process end to start the execution of the sub-process.

[0058] The present invention also provides a collaborative prediction system for the distributed business process completion time. The collaborative prediction system for the distributed business process completion time can be implemented by executing the process steps of the collaborative prediction method for the distributed business process completion time. That is, those skilled in the art can understand the collaborative prediction method for the distributed business process completion time as the preferred implementation manner of the collaborative prediction system for the distributed business process completion time.

[0059] Example 2:

[0060] The present invention also provides a collaborative prediction system for the distributed business process completion time. The system includes the following modules:

[0061] Module M1: Selection criteria for the main process and sub-process completion time prediction models; selection criteria for the main process completion time prediction model, which takes several activity prefixes of the process as input, outputs the predicted value of the start time of the next activity, and iteratively makes predictions to finally obtain the predicted value of the start time of the end activity, that is, the predicted value of the main process completion time; uses the mean absolute error between the predicted value of the main process completion time and the true value of the main process completion time as an index to measure the performance of the model prediction; selection criteria for the sub-process completion time prediction model, which takes the activity prefixes that have occurred in the process as input and directly outputs the predicted value of the start time of the end activity of the sub-process, that is, the predicted value of the sub-process completion time; uses the mean absolute error between the predicted value of the sub-process completion time and the true value of the sub-process completion time as an index to measure the performance of the model prediction.

[0062] Module M2: The sub-process multi-agent model uses the sub-process instance related information and prediction information as state information to decide whether to report the predicted value of the sub-process completion time output by the sub-process completion time prediction model to the corresponding main process multi-agent model; sub-process multi-agent state:

[0063] s k =(σ (k) , σ' (k) )

[0064] where k is the number of the sub-process, σ(k) is the prefix of the subprocess, including information such as activity category, name, occurrence time, and the predicted subprocess completion time value sequence given by the subprocess completion time prediction model; σ′ (k) is the prefix of the main process prediction, including the predicted subprocess completion time value sent from the main process end to the subprocess.

[0065] Module M3: The main process multi-agent model uses the information related to the main process instance and the prediction information uploaded by the corresponding subprocess as state information, determines whether to adopt the prediction information uploaded by the subprocess as the predicted start time value of the corresponding key activity, and iteratively calls the main process completion time prediction model to obtain a new predicted main process completion time value; Main process multi-agent state:

[0066] s k =(σ (k) , σ′ (k) )

[0067] where k is the number of the subprocess corresponding to the main process multi-agent, σ (k) is the prefix of the main process, including activity category, name, occurrence time, the predicted subprocess completion time value given by the main process completion time prediction model, and the main process completion time prediction value sequence information; σ′ (k) is the prefix of the subprocess prediction, including the predicted subprocess completion time value reported by the subprocess to the main process.

[0068] Module M4: The main process end of the distributed business process completion time collaborative prediction system has a main process completion time prediction model and several main process multi-agents corresponding to different subprocesses; each of several subprocess ends has a subprocess completion time prediction model and a subprocess multi-agent; when the distributed business process starts an activity, the main process is started, the main process end calls the main process completion time prediction model to obtain the predicted values of the subprocess and the main process completion time, and sends the predicted subprocess completion time value to the subprocess end to start the execution of the subprocess.

[0069] Example 3:

[0070] A collaborative prediction method for the completion time of a distributed business process based on multi-agent reinforcement learning. For the new business process form of the distributed business process, multi-agent reinforcement learning technology is used to collaboratively predict the process completion time, including the selection criteria for the main process and subprocess completion time prediction models, the subprocess multi-agent model, the main process multi-agent model, and the collaborative prediction method for the completion time of the distributed business process.

[0071] The selection criteria for the main process and subprocess completion time prediction models require that the main process completion time prediction model iteratively predicts the start time of the next activity and finally obtains the predicted value of the main process completion time. The subprocess completion time prediction model is required to directly output the predicted value of the subprocess completion time.

[0072] For the subprocess multi-agent model, the state, action, and reward functions of the subprocess multi-agent and the state, action, and reward functions of the main process multi-agent are proposed.

[0073] The collaborative prediction method for the distributed business process completion time jointly uses the main process completion time prediction model, the subprocess completion time prediction model, the main process multi-agent, and the subprocess multi-agent to collaboratively complete the prediction of the distributed business process completion time.

[0074] In view of the deficiencies of the prior art, the present invention designs a collaborative prediction method for the distributed business process completion time based on multi-agent reinforcement learning. The present invention is realized through the following technical solutions:

[0075] A collaborative prediction method for the distributed business process completion time based on multi-agent reinforcement learning mainly includes: the selection criteria for the main process and subprocess completion time prediction models, the subprocess multi-agent model, the main process multi-agent model, and the collaborative prediction method for the distributed business process completion time.

[0076] Among them, it includes:

[0077] The selection criteria for the main process and subprocess completion time prediction models: The selection criteria for the main process completion time prediction model. This model takes the prefixes of several activities in the process as input, outputs the predicted value of the start time of the next activity, and iteratively makes predictions. Finally, the predicted value of the start time of the end activity, that is, the predicted value of the main process completion time, is obtained. The mean absolute error between the predicted value of the main process completion time and the true value of the main process completion time is used as an index to measure the quality of the model's prediction performance.

[0078] The selection criteria for the subprocess completion time prediction model. This model takes the prefixes of the activities that have occurred in the process as input and directly outputs the predicted value of the start time of the end activity of the subprocess, that is, the predicted value of the subprocess completion time. The mean absolute error between the predicted value of the subprocess completion time and the true value of the subprocess completion time is used as an index to measure the quality of the model's prediction performance.

[0079] The subprocess multi-agent model: The subprocess multi-agent uses the subprocess instance-related information and prediction information as state information to decide whether to report the predicted value of the subprocess completion time output by the subprocess completion time prediction model to the corresponding main process multi-agent model.

[0080] The state of the subprocess multi-agent:

[0081] sk = (σ (k) , σ' (k) )

[0082] where k is the number of the subprocess, and σ (k) is the subprocess prefix, including information such as the activity category, name, occurrence time, and the sequence of predicted subprocess completion times given by the subprocess completion time prediction model; σ' (k) is the main process prediction prefix, including the predicted subprocess completion time sent from the main process end to the subprocess.

[0083] Subprocess multi-agent actions:

[0084]

[0085] Subprocess multi-agent reward, r = R(s, a) is the reward obtained by taking action a in state s. The rewards obtained by taking different actions in state s are: j C is a pre-set magnification coefficient and should be set as a relatively large positive number.

[0086]

[0087] The prediction sequence refers to the sequence of pairs of the subprocess completion time prediction model call time and the subprocess completion time prediction value time.

[0088] The reporting sequence refers to the sequence of pairs of the subprocess completion time prediction model call time and the corresponding subprocess completion time prediction value time when the subprocess multi-agent decides to report to the main process.

[0089] The cumulative error (prediction sequence) is the integral over time of the absolute value of the difference between the predicted subprocess completion time value in the prediction sequence and the true subprocess completion time. The smaller the value, the closer the predicted subprocess completion time value in the prediction sequence is to the true value overall.

[0090] The cumulative error (reporting sequence) is the integral over time of the absolute value of the difference between the predicted subprocess completion time value in the reporting sequence and the true subprocess completion time. The smaller the value, the closer the predicted subprocess completion time value in the reporting sequence is to the true value overall.

[0091] The optimal reporting sequence refers to the sequence composed of the time pairs in the subsequence of the prediction sequence that can minimize the cumulative error (reporting sequence), which can be calculated by using the greedy algorithm or the method of enumerating all subsequences.

[0092]

[0093] ​The approximately optimal reporting sequence refers to the sequence composed of time pairs corresponding to the reports selected by a model that takes the current state as input and outputs 0 or 1 (0 indicates that the current does not belong to the optimal reporting sequence, and 1 indicates that the current belongs to the optimal reporting sequence). The training objective of this model is to output a reporting sequence close to the optimal reporting sequence.

[0094] Main process multi-agent model: The main process multi-agent model uses the information related to the main process instance and the prediction information uploaded by the corresponding sub-process as state information, decides whether to adopt the prediction information uploaded by the sub-process as the predicted start time of the corresponding key activity, and iteratively calls the main process completion time prediction model to obtain a new predicted value of the main process completion time.

[0095] Main process multi-agent state:

[0096] s k =(σ (k) , σ' (k) )

[0097] where k is the number of the sub-process corresponding to the main process multi-agent, σ (k) is the main process prefix, including information such as activity category, name, occurrence time, predicted sub-process completion time given by the main process completion time prediction model (corresponding to the predicted start time of the key activity), and the sequence of predicted main process completion time values; σ' (k) is the sub-process prediction prefix, including the predicted sub-process completion time value reported by the sub-process to the main process.

[0098] Main process multi-agent action:

[0099]

[0100] The reward of the main process multi-agent responsible for sub-process i, r = R(s, a) is the reward obtained by taking action a in state s. The rewards obtained by taking different actions in state s j are as follows:

[0101]

[0102] C is a pre-set magnification coefficient, which should be set as a relatively large positive number. The reporting sequence refers to the sequence of time pairs of the sub-process reporting time and the corresponding predicted main process completion time value.

[0103] The adoption sequence refers to the sequence of time pairs of the sub-process reporting time and the new predicted main process completion time value obtained by the main process multi-agent actually deciding to adopt the predicted sub-process completion time value reported by the sub-process multi-agent as the predicted start time of the corresponding key activity and iteratively predicting by calling the main process completion time prediction model after receiving the report from the sub-process multi-agent.

[0104] The cumulative error (reporting sequence) is the integral of the absolute value of the difference between the predicted value and the true value of the main process completion time of the reporting sequence with respect to time. The smaller the value, the closer the predicted value of the main process completion time of the reporting sequence is to the true value overall.

[0105] The cumulative error (adoption sequence) is the integral of the absolute value of the difference between the predicted value and the true value of the main process completion time of the adoption sequence with respect to time. The smaller the value, the closer the predicted value of the main process completion time of the adoption sequence is to the true value overall.

[0106] The optimal adoption sequence refers to the sequence composed of time pairs corresponding to the subsequence of the reporting sequence that can minimize the cumulative error (adoption sequence), which can be calculated by using the greedy algorithm or the method of enumerating all subsequences.

[0107] The approximate optimal adoption sequence refers to the sequence composed of time pairs selected by a model that takes the current state as input and outputs 0 or 1 (0 indicates that the current is not considered to belong to the optimal adoption sequence, and 1 indicates that the current is considered to belong to the optimal adoption sequence). The training objective of this model is to output an adoption sequence close to the optimal adoption sequence.

[0108] Distributed business process completion time collaborative prediction method: At the main process end of the distributed business process completion time collaborative prediction method, there is a main process completion time prediction model and several main process multi-agents corresponding to different sub-processes; at each of several sub-process ends, there is a sub-process completion time prediction model and a sub-process multi-agent.

[0109] When the distributed business process starts an activity, the main process is started. The main process end calls the main process completion time prediction model to obtain the predicted values of the completion time of sub-process 1 and the main process, and sends the predicted value of the completion time of sub-process 1 to the sub-process 1 end to start the execution of sub-process 1. When each activity of sub-process 1 starts, the sub-process completion time model is called to obtain the predicted value of the sub-process completion time, and the sub-process multi-agent decides whether to report. If reporting, the main process multi-agent corresponding to sub-process 1 decides whether to adopt the report and sends the latest predicted value of the sub-process completion time to the sub-process 1 end. After sub-process 1 ends, the main process end calls the main process completion time prediction model to obtain the predicted values of the completion time of sub-process 2 and the main process, sends them to the sub-process 2 end, and starts the execution of sub-process 2. Repeat this process until the execution of the last sub-process is completed, and the entire distributed business process execution is completed.

[0110] A collaborative prediction method for the completion time of distributed business processes based on multi-agent reinforcement learning mainly includes: the selection criteria for the main process and sub-process completion time prediction models, the sub-process multi-agent model, the main process multi-agent model, and the collaborative prediction method for the completion time of distributed business processes.

[0111] It includes:

[0112] The selection criteria for the main process and sub-process completion time prediction models: For a distributed business process consisting of one main process and n sub-processes, the present invention requires a main process completion time prediction model and n sub-process completion time prediction models corresponding to the n sub-processes respectively. The present invention does not require the use of specific main process and sub-process completion time prediction models. For existing various business process completion time prediction methods, as long as they meet the corresponding completion time prediction model selection criteria for the main process or sub-process, they can be used for predicting the process completion time of the main process or sub-process.

[0113] The selection criteria for the main process completion time prediction model. This model takes the prefixes of several activities of the process as input, outputs the predicted value of the start time of the next activity, and iteratively makes predictions. Finally, the predicted value of the start time of the end activity, that is, the predicted value of the main process completion time, is obtained. When the start activity of the process occurs, this model takes the start activity information of the process as the prefix input and outputs the predicted value of the start time of the key activity a1; this model takes the start activity information of the process and the predicted information of the key activity a1 as the prefix input and outputs the predicted value of the start time of the key activity a2, and so on, iteratively making predictions. Finally, the predicted value of the start time of the end activity, that is, the predicted value of the main process completion time, the predicted value of the completion time of the distributed business process, is obtained. When the key activity a i occurs, this model takes the real information of the start activity of the process, the key activities a1, a2,..., a i as the prefix input and obtains the predicted value of the start time of the key activity a i+1 . Iteratively making predictions, finally, the predicted value of the start time of the end activity of the process, that is, the predicted value of the main process completion time, is obtained. The start time of the start activity and the predicted value of the start time of the key activity a1 correspond to the start time and end time predicted values of sub-process 1; the predicted value of the start time of the key activity a i-1 and the predicted value of the start time of the key activity a i correspond to the start time and end time predicted values of sub-process ii. The predicted value of the start time of the key activity a i output by the main process completion time prediction model is, for sub-process i, the predicted value of the sub-process completion time given by the main process completion time prediction model. The mean absolute error between the predicted value of the main process completion time and the real value of the main process completion time is used as an index to measure the performance of the model prediction.

[0114] Sub - process completion time prediction model selection criteria. This model takes the activity prefix that has occurred in the process as input and directly outputs the predicted start time of the end activity of the sub - process, that is, the predicted value of the sub - process completion time. When the start activity of the sub - process occurs, this model takes the sub - process start activity information as the prefix input and directly outputs the start time of the end activity of the sub - process, that is, the predicted value of the sub - process completion time. When the activity e of the sub - process i occurs, this model takes the start activity of the sub - process, the activities e1, … i and related information as the prefix input and directly outputs the start time of the end activity of the sub - process, that is, the predicted value of the sub - process completion time. The mean absolute error between the predicted value of the sub - process completion time and the true value of the sub - process completion time is used as an indicator to measure the performance of the model prediction.

[0115] Sub - process multi - agent model: For a distributed business process composed of one main process and n sub - processes, the present invention requires n sub - process multi - agent models corresponding to the n sub - processes respectively. The sub - process multi - agent takes the sub - process instance - related information and prediction information as state information and decides whether to report the predicted value of the sub - process completion time output by the sub - process completion time prediction model to the main - process multi - agent model. This model is a deep double Q - network model based on priority experience replay.

[0116] Sub - process multi - agent state:

[0117] s k =(σ (k) , σ′ (k) )

[0118] where k is the number of the sub - process, σ (k) is the sub - process prefix, including information such as activity category, name, occurrence time, and the sequence of predicted values of the sub - process completion time given by the sub - process completion time prediction model; σ′ (k) is the main - process prediction prefix, including the predicted value of the sub - process completion time sent from the main - process side to the sub - process.

[0119] Sub - process multi - agent action:

[0120]

[0121] Reporting to the main process means that the multi-agent of the sub-process believes that the predicted sub-process completion time output by the sub-process completion time prediction model and the predicted sub-process completion time output by the main process completion time prediction model (the predicted start time of the corresponding key activity) have a large gap. The predicted sub-process completion time obtained from the sub-process completion time prediction model should be reported to the multi-agent of the main process, which decides whether to accept this value as the predicted start time of the corresponding key activity and conducts iterative prediction. Finally, a new predicted main process completion time is obtained, and the main process side issues the new predicted sub-process completion time to the sub-process side. Not reporting to the main process means not reporting the predicted sub-process completion time output by the sub-process completion time prediction model to the multi-agent of the main process and continuing to use the predicted sub-process completion time issued by the main process side.

[0122] The reward for the multi-agent of the sub-process, r = R(s,a), is the reward obtained by taking action a in state s. The rewards obtained by taking different actions in state s are as follows: j in state s

[0123]

[0124] C is a pre-set magnification coefficient, which should be set as a relatively large positive number, such as 10. The prediction sequence refers to the sequence of the call time of the sub-process completion time prediction model and the pair of the predicted sub-process completion time. For a sub-process with a length of N + 2, the sub-process conducts N + 1 times of sub-process completion time prediction at the start activity of the sub-process and N intermediate activities respectively. t0, t1, …, t n are the call times of the sub-process completion time prediction model respectively, and p0, p1, …, p n are the predicted sub-process completion times output by the sub-process completion time prediction model:

[0125] σ = {(t0, p0), (t1, p1), (t2, p2), …, (t n , p n )}

[0126] The reporting sequence refers to the sequence of the call time of the sub-process completion time prediction model and the corresponding pair of the predicted sub-process completion time when the multi-agent of the sub-process decides to report to the main process. Assume that there are k reports:

[0127] σ′ = {(t′1, p′10, (t′2, p′2), …, (t′ k , p′ k )}

[0128] The cumulative error (prediction sequence) is the integral of the absolute value of the difference between the predicted value of the subprocess completion time and the true value of the subprocess completion time with respect to time. The smaller the value, the closer the predicted value of the subprocess completion time in the prediction sequence is to the true value as a whole. Let \(t_0\) be the start time of the subprocess and \(t\) be the true value of the subprocess completion time. The specific calculation is as follows:

[0129]

[0130] The cumulative error (reporting sequence) is the integral of the absolute value of the difference between the predicted value of the subprocess completion time in the reporting sequence and the true value of the subprocess completion time with respect to time. The smaller the value, the closer the predicted value of the subprocess completion time in the reporting sequence is to the true value as a whole. The specific calculation is as follows:

[0131]

[0132] The optimal reporting sequence refers to the sequence composed of time pairs in the subsequence of the prediction sequence that can minimize the cumulative error (reporting sequence), which can be calculated by using the greedy algorithm or the method of enumerating all subsequences.

[0133] The approximate optimal reporting sequence refers to the sequence composed of time pairs corresponding to the reports selected by a model that takes the current state as input and outputs 0 or 1 (0 indicates that the current is considered not to belong to the optimal reporting sequence, and 1 indicates that the current is considered to belong to the optimal reporting sequence). The training objective of this model is to output a reporting sequence close to the optimal reporting sequence. Various models that meet the requirements can be used to obtain the approximate optimal reporting sequence, such as a binary classifier using a convolutional neural network.

[0134] For \(n\) subprocesses, the training set data of the \(n\) subprocesses are used to train the \(n\)-subprocess multi-agent model respectively.

[0135] Master process multi-agent model: For a distributed business process composed of one master process and \(n\) subprocesses, the present invention requires \(n\) master process multi-agent models corresponding to the \(n\)-subprocess multi-agent models respectively. Each master process multi-agent model takes the information related to the master process instance and the predicted information uploaded by the corresponding subprocess as state information, decides whether to adopt the predicted information uploaded by the subprocess as the predicted value of the start time of the corresponding key activity, and iteratively calls the master process completion time prediction model to obtain a new predicted value of the master process completion time. This model is a deep double Q-network model based on prioritized experience replay.

[0136] Master process multi-agent state:

[0137] s k =(σ (k) ,σ′ (k) )

[0138] Among them, k is the number of the sub-process corresponding to the multi-agent in the main process, and σ (k) is the prefix of the main process, including information such as activity category, name, occurrence time, predicted sub-process completion time value given by the main process completion time prediction model (corresponding to the predicted key activity start time value), and the sequence of predicted main process completion time values; σ' (k) is the prefix of the sub-process prediction, including the predicted sub-process completion time value reported by the sub-process to the main process.

[0139] Actions of the multi-agent in the main process:

[0140]

[0141] Adopting the sub-process report means that the multi-agent in the main process believes that the predicted sub-process completion time value (corresponding to the predicted key activity start time value) reported by the multi-agent in the sub-process is more reliable than the predicted key activity start time value output by the current main process completion time prediction model. Therefore, it decides to use the reported value as the new predicted key activity start time value, send the reported value to the sub-process side as the new predicted sub-process completion time value, and iteratively call the main process completion time prediction model to obtain a new predicted main process completion time value. Not adopting the sub-process report means that the multi-agent in the main process still believes that the predicted key activity start time value output by the current main process completion time prediction model is more reliable, still maintains the original predicted key activity start time value and the predicted main process completion time value, and sends the original predicted key activity start time value to the sub-process side as the predicted sub-process completion time value.

[0142] Responsible for the reward of the multi-agent in the main process for sub-process i, r = R(s, a) is the reward obtained by taking action a in state s. The rewards obtained by taking different actions in state s j are as follows:

[0143]

[0144] C is a pre-set magnification coefficient, which should be set as a relatively large positive number, such as 10. The reported sequence refers to the sequence of the sub-process reporting time and the corresponding predicted main process completion time value time pairs. Assume that sub-process i reports the predicted sub-process completion time value (the predicted start time of the key activity i in the main process) at times t1, t2,... t k respectively, and the predicted main process completion time values at the corresponding times are p1, p2,..., p k :

[0145] σ i = {(t1, p1), (t2, p2),..., (t k , p k )}

[0146] The adoption sequence refers to the sequence of time pairs of the reported time of the subprocess and the newly predicted main process completion time value obtained by iteratively predicting with the main process completion time prediction model after the multi-agent of the main process actually decides to adopt the predicted subprocess completion time value reported by the multi-agent of the subprocess as the predicted start time value of the corresponding key activity. For subprocess i, assuming that the multi-agent of the main process adopts the reports at times t′1, t′2, …, t′ j and the corresponding newly predicted main process completion time values are p′1, p′2, …, p′ j :

[0147] σ i ′ = {(t′1, p′1), (t′2, p′2), …, (t j ′, p j ′)}

[0148] The cumulative error (reporting sequence) is the integral of the absolute value of the difference between the predicted main process completion time value of the reporting sequence and the true main process completion time value over time. The smaller the value, the closer the predicted main process completion time value of the reporting sequence is to the true value overall. The start time of subprocess i (the start time of the start activity of the subprocess, the start time of key activity i - 1) is t0, and the predicted main process completion time value given by the main process completion time prediction model at t0 is p0, and the completion time (the start time of the end activity of the subprocess, the start time of key activity i) is t f , and the specific calculation is as follows:

[0149]

[0150] The cumulative error (adoption sequence) is the integral of the absolute value of the difference between the predicted main process completion time value of the adoption sequence and the true main process completion time value over time. The smaller the value, the closer the predicted main process completion time value of the adoption sequence is to the true value overall. The specific calculation is as follows:

[0151]

[0152] The optimal adoption sequence refers to the sequence composed of the time pairs of the reported time of the subprocess and the newly predicted main process completion time value corresponding to the subsequence of the reporting sequence that can minimize the cumulative error (adoption sequence), which can be calculated by the greedy algorithm or the method of enumerating all subsequences.

[0153] The approximately optimal adoption sequence refers to the sequence composed of time pairs selected by a model that takes the current state as input and outputs 0 or 1 (0 indicates that the current does not belong to the optimal adoption sequence, and 1 indicates that the current belongs to the optimal adoption sequence). The training objective of this model is to output an adoption sequence close to the optimal adoption sequence. Various models that meet the requirements can be used to obtain the approximately optimal adoption sequence, such as a binary classifier of a convolutional neural network.

[0154] For n subprocesses, the training set data generated by each of the n subprocesses is used to train the corresponding n main process multi-agent models.

[0155] Distributed business process completion time collaborative prediction method: For a distributed business process consisting of one main process and n subprocesses, when the training of a main process completion time prediction model and n subprocess completion time prediction models is completed, and the training of n subprocess multi-agents and n corresponding main process multi-agents is completed, the following method is used for the collaborative prediction of the distributed business process completion time.

[0156] When the start activity of the distributed business process starts, the main process starts. The main process completion time prediction model inputs the relevant information of the start activity of the main process and outputs the predicted value of the start time of key activity 1, which is the predicted value of the process completion time of subprocess 1. The main process completion time prediction model iteratively makes predictions to obtain the predicted value of the main process completion time. The main process multi-agent model 1 responsible for subprocess 1 is started at the main process end, and the predicted value of the process completion time of subprocess 1 is sent to the subprocess 1 end. Subprocess 1 is started. Its start activity is the start activity of the distributed business process and also the start activity of the main process. Its end activity is key activity 1. The subprocess multi-agent 1 at the subprocess 1 end is started.

[0157] When the start activity of subprocess 1 and other activities except the end activity start, the subprocess 1 end calls the subprocess completion time prediction model 1 to predict the completion time of subprocess 1, and calls the subprocess multi-agent 1 to compare with the latest predicted value of the subprocess 1 completion time sent by the main process end to decide whether to report the predicted value of the subprocess 1 completion time output by the subprocess completion time prediction model 1 to the main process end. After receiving the predicted value of the subprocess 1 completion time reported by subprocess 1, the main process multi-agent 1 at the main process end decides whether to adopt this value as the predicted value of the start time of the new key activity 1. If it decides to adopt, it sends this value to the subprocess 1 end and calls the main process completion time prediction model to iteratively predict to obtain the new predicted value of the main process completion time. If it decides not to adopt, it sends the original predicted value of the start time of key activity 1 to the subprocess 1 end.

[0158] When the end activity of Sub - process 1 starts, Sub - process 1 ends, and the Sub - process 1 side notifies the main - process side that Key Activity 1 starts. Key Activity 1 of the main process starts. The main - process completion - time prediction model inputs the relevant information of the main - process start activity and Key Activity 1, and outputs the predicted start time of Key Activity 2, which is the predicted value of the Sub - process 2 completion time. The main - process completion - time prediction model iteratively makes predictions to obtain a new predicted value of the main - process completion time. The main - process side starts the main - process multi - agent 2 responsible for Sub - process 2 and sends the predicted value of the Sub - process 2 completion time to the Sub - process 2 side. Sub - process 2 is started. Its start activity is Key Activity 1, and its end activity is Key Activity 2. The Sub - process 2 multi - agent 2 on the Sub - process 2 side is started.

[0159] Repeat this process until the end activity of the last sub - process, i.e., Sub - process n, starts. Sub - process n ends, and the Sub - process n side notifies the main - process side that the distributed business - process end activity starts. The main process ends, and the entire distributed business process ends.

[0160] Those skilled in the art can understand this embodiment as a more specific description of Embodiment 1 and Embodiment 2.

[0161] Those skilled in the art know that in addition to implementing the system and its various devices, modules, and units provided by the present invention in the form of pure computer - readable program code, the method steps can be logically programmed to enable the system and its various devices, modules, and units provided by the present invention to be implemented in the form of logic gates, switches, application - specific integrated circuits, programmable logic controllers, and embedded microcontrollers, etc., to achieve the same functions. Therefore, the system and its various devices, modules, and units provided by the present invention can be considered as a kind of hardware component, and the devices, modules, and units included therein for implementing various functions can also be regarded as the structures within the hardware component; the devices, modules, and units for implementing various functions can also be regarded as either software modules for implementing the method or structures within the hardware component.

[0162] The specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above - mentioned specific embodiments. Those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Without conflict, the embodiments of the present application and the features in the embodiments can be combined arbitrarily.

Claims

1. A collaborative prediction method for the completion time of distributed business processes, characterized in that, The method includes the following steps: Step S1: Criteria for selecting the main process and sub-process completion time prediction models; Step S2: The sub-process multi-agent model uses the sub-process instance related information and prediction information as state information to decide whether to report the sub-process completion time prediction value output by the sub-process completion time prediction model to the corresponding main process multi-agent model; Step S3: The main process multi-agent model uses the main process instance related information and the prediction information uploaded by the corresponding sub-process as state information to decide whether to adopt the prediction information uploaded by the sub-process as the start time prediction value of the corresponding key activity, and iteratively calls the main process completion time prediction model to obtain a new main process completion time prediction value; Step S4: There is a main process completion time prediction model and several main process multi-agents corresponding to different sub-processes at the main process end of the distributed business process completion time collaborative prediction method; each of several sub-process ends has a sub-process completion time prediction model and a sub-process multi-agent; In the said Step S1: Criteria for selecting the main process completion time prediction model. This model takes the prefixes of several activities of the process as input, outputs the start time prediction value of the next activity, and iteratively makes predictions to finally obtain the start time prediction value of the end activity, that is, the prediction value of the main process completion time; Use the mean absolute error between the main process completion time prediction value and the true value of the main process completion time as an indicator to measure the performance of the model prediction; Criteria for selecting the sub-process completion time prediction model. This model takes the prefixes of the activities that have occurred in the process as input and directly outputs the start time prediction value of the sub-process end activity, that is, the prediction value of the sub-process completion time; Use the mean absolute error between the sub-process completion time prediction value and the true value of the sub-process completion time as an indicator to measure the performance of the model prediction; In the said Step S2: Sub-process multi-agent state: s k =(σ (k) , σ' (k) ) where k is the number of the subprocess, and σ (k) is the subprocess prefix, including the activity category, name, occurrence time, and the sequence information of the predicted subprocess completion time given by the subprocess completion time prediction model; σ' (k) is the main process prediction prefix, including the predicted subprocess completion time sent from the main process end to the subprocess; In the said Step S3: Main process multi-agent state: s k =(σ (k) , σ′ (k) ) where k is the number of the sub-process corresponding to the multi-agent in the main process, and σ (k) is the prefix of the main process, including the activity category, name, occurrence time, the predicted sub-process completion time given by the main process completion time prediction model, and the sequence information of the main process completion time prediction value; σ′ (k) is the prefix of the sub-process prediction, including the predicted sub-process completion time reported by the sub-process to the main process; In the said Step S4: When the start activity of the distributed business process starts, the main process is started. The main process end calls the main process completion time prediction model to obtain the prediction values of the sub-process and main process completion times, and sends the sub-process completion time prediction value to the sub-process end to start the execution of the sub-process.

2. A collaborative prediction system for the completion time of distributed business processes, characterized in that, The system includes the following modules: Module M1: Criteria for selecting the main process and sub-process completion time prediction models; Module M2: The sub-process multi-agent model uses the sub-process instance related information and prediction information as state information to decide whether to report the sub-process completion time prediction value output by the sub-process completion time prediction model to the corresponding main process multi-agent model; Module M3: The main process multi-agent model uses the main process instance related information and the prediction information uploaded by the corresponding sub-process as state information to decide whether to adopt the prediction information uploaded by the sub-process as the start time prediction value of the corresponding key activity, and iteratively calls the main process completion time prediction model to obtain a new main process completion time prediction value; Module M4: There is a main process completion time prediction model and several main process multi - agents corresponding to different sub - processes at the main process end of the distributed business process completion time collaborative prediction system; each of several sub - process ends has a sub - process completion time prediction model and a sub - process multi - agent; In the said Module M1: The selection criterion of the main process completion time prediction model: This model takes the prefixes of several activities of the process as input, outputs the predicted value of the start time of the next activity, and iteratively makes predictions. Finally, the predicted value of the start time of the end activity, that is, the predicted value of the main process completion time, is obtained; The average absolute error between the predicted value of the main process completion time and the true value of the main process completion time is used as an index to measure the quality of the model's prediction performance; The selection criterion of the sub - process completion time prediction model: This model takes the prefixes of the activities that have occurred in the process as input and directly outputs the predicted value of the start time of the end activity of the sub - process, that is, the predicted value of the sub - process completion time; The average absolute error between the predicted value of the sub - process completion time and the true value of the sub - process completion time is used as an index to measure the quality of the model's prediction performance; In the said Module M2: The state of the sub - process multi - agent: s k =(σ (k) ,σ' (k) ) where k is the number of subprocesses, and σ (k) is the subprocess prefix, including the activity category, name, occurrence time, and the sequence information of the predicted subprocess completion time given by the subprocess completion time prediction model; σ' (k) is the main process prediction prefix, including the predicted subprocess completion time sent from the main process end to the subprocess; In the said Module M3: The state of the main process multi - agent: s k =(σ (k) ,σ' (k) ) Among them, k is the number of the sub-process corresponding to the main-process multi-agent, and σ (k) is the main-process prefix, including the activity category, name, occurrence time, the predicted sub-process completion time given by the main-process completion time prediction model, and the sequence information of the main-process completion time prediction value; σ' (k) is the sub-process prediction prefix, including the predicted sub-process completion time reported by the sub-process to the main process; In the said Module M4: When the start activity of the distributed business process starts, the main process is started. The main process end calls the main process completion time prediction model to obtain the predicted values of the sub - process and the main process completion time, and sends the predicted value of the sub - process completion time to the sub - process end to start the execution of the sub - process.

Images