A process prediction method and device based on historical model integration

By filtering and updating the prediction model and selecting the final prediction activities using the contribution set, the problem of degradation of prediction accuracy under dynamic changes in process state in the existing technology is solved, and accurate and timely process feature perception in dynamic scenarios is achieved, which improves the Predictive activity accuracy and efficiency.

CN118606718BActive Publication Date: 2024-11-01安徽思高智能科技有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411080753.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-11-01
Estimated Expiration
2044-08-08

AI Technical Summary

Technical Problem

When existing prediction process methods deal with dynamically changing process states, they are difficult to adapt, resulting in a decrease in prediction accuracy. It takes a lot of time and resources to train new models, and only consider the recent trajectory, lack of observations on historical information, which affects the prediction. Integrity.

Method used

By obtaining the set of process trajectories within the recent time interval, filtering and updating the prediction model, building a contribution set, and selecting the final prediction activities, we can achieve effective utilization of historical information and continuous optimization of the model.

Benefits of technology

In the constantly updated business scenarios, provide more accurate and timely process feature perception, reduce training costs, enhance the accuracy of prediction activities, and help achievement prediction and abnormal detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118606718B_ABST
    Figure CN118606718B_ABST
Patent Text Reader

Abstract

The present invention provides a process prediction method and device based on historical model integration, which relates to the field of process mining monitoring, including: obtaining a process trajectory set Σ of a business log L in a time period T∈[(n-1)t,nt] n ; Calculate and obtain the first prediction model set S in each prediction model in Σ n The prediction scores on the second prediction model set S' are obtained; through Σ n Train to obtain the current prediction model M n , M n Put S' into S' to obtain the third prediction model set S"; obtain the process trajectory σ being executed in T∈[nt,(n+1)t] k , through each prediction model in S' k Perform activity prediction and obtain the predicted activity set X n ; Construct the contribution set P through S” n , through P n From X n The present invention selects the final prediction activity from the prediction model by using the process trajectory set in the most recent time interval, thereby maintaining a relatively constant prediction model, which can more accurately monitor future prediction activities while avoiding the increase in training costs over time.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of process mining monitoring, and in particular to a process prediction method and device based on historical model integration. Background Art

[0002] Predictive process monitoring (PPM) is an important task in the field of process mining. Its main purpose is to use machine learning methods to build a predictive model from a set of historical process trajectories, and to help enterprises better understand and perceive the future state of process instances based on predictive information.

[0003] Predicting the next activity type of a process instance is the most important prediction task in PPM. Current technical methods mainly focus on building complex and precise prediction models to continuously improve the accuracy of predictions for future activities. However, considering that processes in actual business scenarios are highly dynamic, the construction of prediction models relies on constant business logs and cannot adapt to capture dynamically changing process states, affecting prediction accuracy. As the number of historical process trajectories increases, training a new prediction model based on all historical process information consumes more time and resources. Considering only the most recent process trajectories to build a prediction model will result in a lack of observation of historical process information in the prediction process, and the process characteristics within different time ranges cannot be fully considered. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a process prediction method and device based on historical model integration to solve the problem that the existing prediction process lacks observation of historical process information.

[0005] The present invention provides a process prediction method based on historical model integration, comprising the steps of:

[0006] S1: Get the process trajectory set Σ of the business log L in the time period T∈[(n-1)t,nt] n , where n is the iteration number and t is the time interval;

[0007] S2: Get the first prediction model set S, and calculate the prediction model in S in Σ n The prediction scores on the , eliminate the prediction models whose prediction scores are less than the preset value, and obtain the second prediction model set S';

[0008] S3: Through the process trajectory set Σ n Train to obtain the current prediction model M n , M n Put in S' to obtain the third prediction model set S";

[0009] S4: Get the process trajectory σ being executed in the time period T∈[nt, (n+1)t] k, through each prediction model in S' k Perform activity prediction and obtain the predicted activity set X n ;

[0010] S5: Construct contribution set P through S” n , through the contribution set P n From the predicted activity set X n Select the final predicted activity; let n=n+1, S=S", and return to step S1.

[0011] Preferably, step S1 specifically comprises:

[0012] S11: Get business log L={σ 1 , σ 2 , σ 3 ,…, σ λ ,…}, where λ represents the number of the process track, σ λ Indicates the λth process trajectory executed;

[0013] S12: Extract the process trajectory set Σ of the business log L in the time period T∈[(n-1)t,nt] n ={σ i ,σ i+1 ,…}, where i represents the number of the process trajectory.

[0014] Preferably, step S2 specifically comprises:

[0015] S21: Obtain the first prediction model set S={M j , M j+1 ,…,M n-1}, where j represents the number of the prediction model, M j represents the prediction model generated by the jth iteration;

[0016] S22: Calculate and obtain the prediction model M j Σ n The average prediction accuracy of each process trajectory in M j The prediction score of

[0017] S23: Repeat step S22 to calculate the prediction score of each prediction model in S, eliminate the prediction model with a prediction score less than a preset value, and obtain a second prediction model set S'.

[0018] Preferably, step S3 specifically comprises:

[0019] S31: constructing an initial prediction model;

[0020] S32: Set the process trajectory Σ nUse the training data set to train the initial prediction model and obtain the current prediction model M n , M n Put the last bit in S' to obtain the third prediction model set S".

[0021] Preferably, step S4 is specifically:

[0022] S41: Obtain the process trajectory σ being executed in the time period T∈[nt, (n+1)t] k = <e 1 , e 2 ,…, e m >, where k is the number of the process track, m is the number of the activity, and e m Indicates that in σ k The mth activity executed in ;

[0023] S42: Through the prediction model M j ∈S” predicts the process trajectory σ k Next activity (e' m+1 ) j , and (e' m+1 ) j As M j The prediction activity of, where j is the number of the prediction model;

[0024] S43: Repeat step S42 to obtain the prediction activities of each prediction model in S", and form all the prediction activities into a prediction activity set X n .

[0025] Preferably, step S5 is specifically:

[0026] S51: Get the u-th prediction model M added in S” u , calculate and obtain M u Contribution value p to the prediction result u n , the calculation formula is:

[0027] ,

[0028] Among them, λ is the initialization contribution rate;

[0029] S52: Repeat step S51 to calculate the contribution rate of each prediction model in S”, and form a contribution degree set P of all contribution rates. n ;

[0030] S53: Obtain contribution set P n The prediction model with the largest contribution in the prediction activity set X nThe prediction activity corresponding to the prediction model with the largest contribution is selected as the final prediction activity;

[0031] S54: Let n=n+1, S=S”, and return to step S1.

[0032] A storage medium stores instructions and data for implementing a process prediction method based on historical model integration.

[0033] A process prediction device based on historical model integration includes: a processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement the process prediction method based on historical model integration.

[0034] The present invention has the following beneficial effects:

[0035] The prediction model is screened by the set of process trajectories in the most recent time interval, so as to maintain a relatively constant prediction model. While avoiding the increase in training costs over time, future prediction activities can be monitored more accurately. In the constantly updated business process scenarios, changes in process characteristics can be perceived in a timely manner, providing enterprises with clearer and more accurate activity prediction information, thereby facilitating subsequent outcome prediction and anomaly detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a flow chart of a method according to an embodiment of the present invention;

[0037] Figure 2 This is a schematic diagram of business log segmentation;

[0038] Figure 3 This is a structural diagram of the device according to an embodiment of the present invention;

[0039] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0040] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0041] Reference Figure 1 The present application provides a process prediction method based on historical model integration, comprising the steps of:

[0042] S1: Get the process trajectory set Σ of the business log L in the time period T∈[(n-1)t,nt] n , where n is the iteration number and t is the time interval;

[0043] Furthermore, step S1 is specifically as follows:

[0044] S11: Get business log L={σ 1 , σ 2 , σ 3 ,…, σ λ ,…}, where λ represents the number of the process track, σ λ Indicates the λth process trajectory executed;

[0045] S12: Extract the process trajectory set Σ of the business log L in the time period T∈[(n-1)t,nt] n ={σ i ,σ i+1 ,…}, where i represents the number of the process trajectory.

[0046] Specifically, the business logs are divided according to time intervals, such as Figure 2 shown.

[0047] S2: Get the first prediction model set S, and calculate the prediction model in S in Σ n The prediction scores on the , eliminate the prediction models whose prediction scores are less than the preset value, and obtain the second prediction model set S';

[0048] Furthermore, step S2 is specifically as follows:

[0049] S21: Obtain the first prediction model set S={M j , M j+1 ,…,M n-1}, where j represents the number of the prediction model, M j represents the prediction model generated by the jth iteration;

[0050] S22: Calculate and obtain the prediction model M j Σ n The average prediction accuracy of each process trajectory in M j The prediction score of

[0051] Furthermore, step S22 is specifically as follows:

[0052] S221: Initialize the accuracy set Acc of the prediction model to 0; n ={σ i , σ i+1 ,…} are arranged in chronological order and traverse Σ n The process trajectory in

[0053] S222: For the process trajectory σ obtained by the qth traversal q = <e 1 , e 2 ,…, e w >, where e wIndicates that in σ q The wth activity performed in q The activities in get the set of activity subsequences φ q ={ <e 1 >, <e 1 , e 2 >,…, <e 1 , e 2 ,…, e w-1 >} and the corresponding set of next activities {e 2 ,e 3 ,…,e w};

[0054] S223: For the set φ q , input the activity subsequences into the prediction model M j The set of prediction results is obtained, and the set of prediction results and the next activity {e 2 ,e 3 ,…,e w}Compared with the calculation prediction model M j The prediction accuracy is calculated and the accuracy results are merged into the set Acc;

[0055] S224: Return to step S222 until all Σ n All process trajectories in the calculation prediction model M j The final prediction score is the average value (Acc) of the elements in the accuracy set Acc.

[0056] S23: Repeat step S22 to calculate the prediction score of each prediction model in S, eliminate the prediction model with a prediction score less than a preset value, and obtain a second prediction model set S'.

[0057] S3: Through the process trajectory set Σ n Train to obtain the current prediction model M n , M n Put in S' to obtain the third prediction model set S";

[0058] Furthermore, step S3 is specifically as follows:

[0059] S31: constructing an initial prediction model;

[0060] S32: Set the process trajectory Σ n Use the training data set to train the initial prediction model and obtain the current prediction model M n , M n Put the last bit in S' to obtain the third prediction model set S".

[0061] Further, step S32 is specifically as follows:

[0062] S321: Initialize the training data set D={}, the set of next activities corresponding to the training data T={}, and obtain the set Σ n The average length d of the process trajectory in;

[0063] S322: Σ n ={σ i , σ i+1 ,…} are arranged in chronological order and traverse Σ n The process trajectory in

[0064] S323: Set the process trajectory σ q = <e 1 , e 2 ,…, e w >∈Σ n Further partitioning to obtain the set of active subsequences φ q ={ <e 1 >, <e 1 , e 2 >,…, <e 1 , e 2 ,…, e w-1 >} and the corresponding set of next activities {e 2 ,e 3 ,…,e w}, where e w Indicates that in σ q The wth activity executed in ;

[0065] S324: φ q The active subsequences in are unified into a sequence of length d by truncating or filling the integer 0, and a new active subsequence set φ' is obtained. q ;

[0066] Specifically, when d=4, φ' q ={<0,0,0,e 1 >,<0,0,e 1 , e 2 >,…, <e w-4 ,…, e w-1 >};

[0067] S325: Update T = T ∪ {e 2 ,e 3 ,…,e w}, D = D ∪ φ' q , until all Σ are traversed n The process trajectory in the training dataset D and the corresponding next activity set T are obtained;

[0068] S326: The trajectories in the training data set D are input into the prediction model based on the two-layer long short-term memory network (LSTM), and the set of prediction results P is output, and the loss value loss=Crossentropy(P,T) is calculated;

[0069] S327: Optimize the model parameters based on the loss value, return to step S335, until the maximum number of iterations is exceeded or the loss value converges, and save the prediction model M corresponding to the minimum loss n .

[0070] Specifically, we choose to build an initial prediction model based on a two-layer long short-term memory network (LSTM) and transform the process trajectory set Σ n The entire trajectory in is divided into a set of activity subsequences and input into the initial prediction model. The cross entropy between the prediction result and the true value is calculated as the loss function. The model parameters are updated based on the loss value to obtain the current prediction model M n .

[0071] S4: Get the process trajectory σ being executed in the time period T∈[nt, (n+1)t] k , through each prediction model in S' k Perform activity prediction and obtain the predicted activity set X n ;

[0072] Furthermore, step S4 is specifically as follows:

[0073] S41: Obtain the process trajectory σ being executed in the time period T∈[nt, (n+1)t] k = <e 1 , e 2 ,…, e m >, where k is the number of the process track, m is the number of the activity, and e m Indicates that in σ k The mth activity executed in ;

[0074] S42: Through the prediction model M j ∈S” predicts the process trajectory σ k Next activity (e' m+1 ) j , and (e' m+1 ) j As M j The prediction activity of, where j is the number of the prediction model;

[0075] S43: Repeat step S42 to obtain the prediction activities of each prediction model in S", and form all the prediction activities into a prediction activity set X n .

[0076] S5: Construct contribution set P through S” n , through the contribution set P n From the predicted activity set X n Select the final predicted activity; let n=n+1, S=S", and return to step S1.

[0077] Furthermore, step S5 is specifically as follows:

[0078] S51: Get the u-th prediction model M added in S” u , calculate and obtain M u Contribution value p to the prediction result u n , the calculation formula is:

[0079] ,

[0080] Among them, λ is the initialization contribution rate;

[0081] S52: Repeat step S51 to calculate the contribution rate of each prediction model in S”, and form a contribution degree set P of all contribution rates. n ;

[0082] S53: Obtain contribution set P n The prediction model with the largest contribution in the prediction activity set X n The prediction activity corresponding to the prediction model with the largest contribution is selected as the final prediction activity;

[0083] Specifically, the prediction model M with the greatest contribution v =argmax(P n ), the prediction model M v The corresponding predicted activity (e' m+1 ) v As process trajectory σ k The next final forecast activity of ;

[0084] S54: Let n=n+1, S=S”, and return to step S1.

[0085] See also Figure 3 , Figure 3 It is a schematic diagram of the working of the hardware device of an embodiment of the present invention, and the hardware device specifically includes: a process prediction device 401 based on historical model integration, a processor 402 and a storage medium 403.

[0086] A process prediction device 401 based on historical model integration: The process prediction device 401 based on historical model integration implements the process prediction method based on historical model integration.

[0087] Processor 402: The processor 402 loads and executes the instructions and data in the storage medium 403 to implement the process prediction method based on historical model integration.

[0088] Storage medium 403: The storage medium 403 stores instructions and data; the storage medium 403 is used to implement the process prediction method based on historical model integration.

[0089] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0090] The serial numbers of the embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In a unit claim that lists several means, several of these means may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order and these words may be interpreted as identifiers.

[0091] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A process prediction method based on historical model integration, characterized in that: Includes steps: S1: Get the process trajectory set Σ of the business log L in the time period T∈[(n-1)t,nt] n , where n is the iteration number and t is the time interval; S2: Get the first prediction model set S, and calculate the prediction model in S in Σ n The prediction scores on the above are eliminated, and the prediction models with prediction scores less than the preset value are eliminated to obtain a second prediction model set S'; S3: Through the process trajectory set Σ n Train to obtain the current prediction model M n , M n Put in S' to obtain the third prediction model set S"; S4: Get the process trajectory σ being executed in the time period T∈[nt, (n+1)t] k , through each prediction model in S' k Perform activity prediction and obtain the predicted activity set X n ; S5: Construct contribution set P through S” n , through the contribution set P n From the predicted activity set X n Select the final predicted activity; let n=n+1, S=S", and return to step S1; Step S5 is specifically as follows: S51: Get the u-th prediction model M added in S” u , calculate and obtain M u Contribution value p to the prediction result u n , the calculation formula is: , Among them, a is the initialization contribution rate; S52: Repeat step S51 to calculate the contribution rate of each prediction model in S”, and form a contribution degree set P of all contribution rates. n ; S53: Obtain contribution set P n The prediction model with the largest contribution in the prediction activity set X n The prediction activity corresponding to the prediction model with the largest contribution is selected as the final prediction activity; S54: Let n=n+1, S=S”, and return to step S1.

2. The process prediction method based on historical model integration according to claim 1 is characterized in that: Step S1 is specifically as follows: S11: Get business log L={σ1, σ2,σ3,…,σ λ ,…}, where λ represents the number of the process track, σ λ represents the λth process trajectory executed; S12: Extract the process trajectory set Σ of the business log L in the time period T∈[(n-1)t,nt] n ={σ i ,σ i+1 ,…}, where i represents the number of the process trajectory.

3. The process prediction method based on historical model integration according to claim 1 is characterized in that: Step S2 is specifically as follows: S21: Obtain the first prediction model set S={M j , M j+1 ,…,M n-1 }, where j represents the number of the prediction model, M j represents the prediction model generated by the jth iteration; S22: Calculate and obtain the prediction model M j Σ n The average prediction accuracy of each process trajectory in M j The prediction score of S23: Repeat step S22 to calculate the prediction score of each prediction model in S, eliminate the prediction model with a prediction score less than a preset value, and obtain a second prediction model set S'.

4. A process prediction method based on historical model integration according to claim 1, characterized in that: Step S3 is specifically as follows: S31: constructing an initial prediction model; S32: Set the process trajectory Σ n Use the training data set to train the initial prediction model and obtain the current prediction model M n , M n Put the last bit in S' to obtain the third prediction model set S".

5. The process prediction method based on historical model integration according to claim 1 is characterized in that: Step S4 is specifically as follows: S41: Obtain the process trajectory σ being executed in the time period T∈[nt, (n+1)t] k = <e1, e2,…, e m >, where k is the number of the process track, m is the number of the activity, and e m Indicates that in σ k The mth activity executed in ; S42: Through the prediction model M j ∈S” predicts the process trajectory σ k Next activity (e' m+1 ) j , and (e' m+1 ) j As M j The prediction activity of, where j is the number of the prediction model; S43: Repeat step S42 to obtain the prediction activities of each prediction model in S", and form all the prediction activities into a prediction activity set X n .

6. A storage medium, characterized in that: The storage medium stores instructions and data for implementing a process prediction method based on historical model integration as described in any one of claims 1 to 5.

7. A process prediction device based on historical model integration, characterized in that: include: Processor and storage medium; the processor loads and executes instructions and data in the storage medium to implement a process prediction method based on historical model integration as described in any one of claims 1 to 5.