Inductive process mining method and system based on local structure optimization

Abstract

The invention discloses an inductive process mining method and system based on local structure optimization. The method comprises the following steps: acquiring an event log executed by an actual business process; mining an original process tree model from the event log, searching and identifying an inaccurate structure in the original process tree model, and extracting a corresponding sub-log; iteratively optimizing the inaccurate structure in the sub-log to obtain a new sub-structure set; the new substructure set is combined by a Cartesian product to obtain a new substructure combination, a new substructure in the new substructure combination replaces a corresponding inaccurate structure in the original process tree model to obtain a result process tree model, the result process tree model is converted into a Petri net model, quality indexes fit and precision of the model are evaluated, a comprehensive quality index F-measure is calculated, and a final process tree model is obtained. And finally, selecting a Petri net model with the highest F-measure as output. According to the method, the model mining efficiency is improved, possible losses are reduced, and it is guaranteed that the high-quality structure is not affected while the inaccurate structure in the process model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of process mining, and in particular to an inductive process mining method, system, storage medium and computing device based on local structure optimization. Background Art

[0002] With the development of the information industry, the log data stored in the software system is becoming increasingly large. It has become more difficult for process managers to obtain effective information from the log data to establish a process model. In this context, the concept of model mining has emerged. Compared with the traditional hand-drawn process model method, the process model obtained by model mining is more objective, accurate and efficient. Many researchers have conducted in-depth research, such as heuristic model mining methods and model mining methods based on genetic algorithms. Existing methods are based on the algorithm and processing mechanism of the mining method itself to directly obtain the result model by model mining the information obtained from the event log. This approach ignores the problem of inaccurate structure inside the process model due to the limitations of its own algorithm, making it still an unresolved problem to obtain high-quality process models through model mining. Summary of the invention

[0003] The first purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art and to provide an inductive process mining method based on local structure optimization, which solves the problem of low quality of process models obtained by traditional model mining methods and plays a key role in subsequent understanding of process information, analyzing process pain points, and guiding model optimization through process models.

[0004] The second object of the present invention is to provide an inductive process mining system based on local structure optimization.

[0005] A third object of the present invention is to provide a storage medium.

[0006] A fourth object of the present invention is to provide a computing device.

[0007] The first object of the present invention is achieved by the following technical solution: an inductive process mining method based on local structure optimization comprises the following steps:

[0008] 1) Obtain basic data, that is, event logs that record actual business process execution;

[0009] 2) Using the inductive model mining algorithm to mine the process model based on the event log in step 1), obtain the original process tree model, identify the local imprecise structure in the original process tree model, and extract the sub-logs corresponding to the imprecise structure;

[0010] 3) Using the inductive model mining algorithm, according to the preset threshold, iteratively mine new sub-models from the sub-logs extracted in step 2) as new sub-structures after the local imprecise structure is optimized. Each new sub-structure after the iterative optimization of the imprecise structure constitutes a set, which is called the new sub-structure set;

[0011] 4) In step 3), the new substructure sets are combined according to Cartesian products to obtain several new substructure combinations, and the new substructures in the new substructure combinations all come from different new substructure sets. The new substructures in the new substructure combinations are used to replace the corresponding imprecise structures in the original process tree model to obtain a result process tree model. The result process tree model is converted into a Petri net model, and the comprehensive quality indicator F-measure of the model is calculated by evaluating the quality indicators fitness and precision of the Petri net model, and finally a Petri net model with the highest F-measure is selected as output.

[0012] Furthermore, in step 1), the event log is information about user business execution captured and recorded by the information system, and is a collection of finite event sequences, each of which is called a trace.

[0013] Further, in step 2), the original process tree model is a directed acyclic graph, defined as a quintuple T = (V, E, A, l, r), wherein V is a finite set of nodes in the original process tree model, including leaf nodes and non-leaf nodes; E is a set of edges between nodes in the original process tree model; A is a set of non-silent activity labels; l represents the correspondence between nodes and activity labels in the original process tree model; r is the root node of the original process tree model; the inexact structure is a sub-process tree with a concurrent node as the root node, whose descendant nodes include a selection node or a loop node. When the first child node of the selection node is a silent node or the first child node of the loop node is an active node, the sub-process tree with the concurrent node as the root node is identified as an inexact structure.

[0014] Further, in step 2), the specific situation of identifying the local inaccurate structure in the original process tree model and extracting the corresponding sub-logs is as follows:

[0015] a. Before extracting the sub-logs corresponding to the local imprecise structure, it is necessary to align the activities contained in the event log with the activities corresponding to all leaf nodes in the original process tree model. According to the synchronization movement in the alignment result and the model movement containing invisible nodes, the connection between the activities contained in the event log and the activities corresponding to the leaf nodes in the original process tree model is established:

[0016] Where the alignment is a sequence of moves, the following three types of moves occur:

[0017] The first one is synchronous movement: the events recorded in the trace are consistent with the activities of the original process tree model execution sequence;

[0018] The second type is model movement: there is a deviation between the execution sequence of the trajectory and the original process tree model, and the execution activities in the original process tree model are skipped;

[0019] The third type is log movement: there is a deviation between the execution sequence of the trace and the original process tree model, and the events recorded in the log are skipped;

[0020] b. Based on the characteristics of the imprecise structure, start from the root node of the original process tree model and traverse from top to bottom and from left to right to find and identify the imprecise structure, and identify the substructure in the original process tree model that meets the characteristics of the imprecise structure as the imprecise structure;

[0021] c. Based on the activities corresponding to the leaf nodes in the imprecise structure identified in b and the activity connection between the event log obtained in a and the original process tree model, extract the sub-log corresponding to the imprecise structure from the event log.

[0022] Further, in step 3), the imprecise structure is optimized using a frequency-based inductive model mining algorithm to iteratively mine new sub-models in the sub-logs corresponding to the imprecise structure according to preset thresholds, wherein the preset thresholds are set to 0.2, 0.3, 0.4, 0.5 and 0.6, and the five new sub-structures obtained after each imprecise structure is iteratively optimized according to the above five preset thresholds constitute a new sub-structure set.

[0023] Further, in step 4), the Petri net model is a four-tuple ∑=(P,Tr,F,M0), satisfying: P={p1,p2,p3,...,p m} is a finite set of places, where p m is the mth place in the place set P; Tr = {Tr1, Tr2, Tr3, ..., Tr m} is a finite set of transitions, where Tr m is the mth transition in the transition set Tr; F is a finite set of relations between transitions and places; M0 is the initial mark; the Cartesian product is a combination method of a new substructure set, which combines new substructures in different new substructure sets into a new substructure combination, and the new substructures in the new substructure combination all come from different new substructure sets, that is, Tree1×Tree2×...×Tree m ={(Tree 1.i ,Tree 2.i ,...,Tree m.i )|Tree 1.i∈Tree1,Tree 2.i ∈Tree2,,...,Tree m.i ∈Tree m}, where Tree m is the mth imprecise structure identified in step 2), Tree m.i It is an inexact structure Tree m The i-th new substructure among the five new substructures obtained after iterative optimization according to five preset thresholds.

[0024] Further, in step 4), the root node of the new substructure in the new substructure combination replaces the root node of the corresponding inaccurate structure in the original process tree model to complete the inaccurate structure replacement, thereby obtaining a result process tree model; the obtained result process tree model is converted into a Petri net model, specifically, the corresponding structure conversion between the result process tree model and the Petri net model, and the following five types of conversion occur:

[0025] The first type is the conversion of leaf nodes: convert the leaf nodes in the result process tree model into a transition in the Petri net model, and add a place before and after the transition as the input place and output place respectively;

[0026] The second type is the conversion of sequential operators: in the Petri net model, the substructures corresponding to the sequential operators in the result process tree model are sequentially connected using the library;

[0027] The third type is the conversion of concurrent operators: in the Petri net model, the substructures corresponding to the concurrent operators in the result process tree model are connected to the same transition using the library.

[0028] The fourth type is the conversion of selection operators: in the Petri net model, the substructure corresponding to the selection operator in the result process tree model is connected using the same library;

[0029] The fifth type is the transformation of loop operators: in the Petri net model, additional places and transitions are used to construct paths to implement the loop execution of the substructure corresponding to the loop operator in the result process tree model;

[0030] Fitness, precision and F-measure are the quality indicators of the Petri net model. Fitness is used to evaluate the proportion of trajectories contained in the event log that can be reproduced by the Petri net model. The total number of trajectories contained in the event log is recorded as AT, and the number of trajectories that can be reproduced by the Petri net model is recorded as ATR. The calculation formula of fitness is:

[0031]

[0032] Precision is used to quantify the proportion of trajectories that can be generated by the Petri net model that do not appear in the event log. The total number of trajectories that can be generated by the Petri net model is recorded as GT, and the number of trajectories that appear in the event log is recorded as EGT. The calculation formula of precision is:

[0033]

[0034] In order to better balance the fitness and precision of the Petri net model, F-measure is used as the harmonic mean of the two to objectively evaluate the comprehensive quality of the model. The calculation formula of F-measure is:

[0035]

[0036] The second object of the present invention is achieved through the following technical solution: an inductive process mining system based on local structure optimization, used to implement the above-mentioned inductive process mining method based on local structure optimization, which includes:

[0037] The data acquisition module is used to obtain basic data, that is, the event log that records the actual business process execution;

[0038] The imprecise structure identification and extraction module is used to map the relationship between the activities contained in the event log and the activities contained in the nodes of the original process tree model, find and identify the imprecise structure in the process tree model, and extract the sub-logs corresponding to the imprecise structure;

[0039] An imprecise structure optimization module is used to iteratively optimize the imprecise structure identified in the original process tree model according to a preset threshold based on the frequency-based inductive model mining algorithm;

[0040] Substructure replacement and model screening module: used to replace the inaccurate structure in the original process tree model with the optimized new substructure, and to calculate the comprehensive quality indicator F-measure of the model by evaluating the quality indicators fitness and precision of the Petri net model, and select the Petri net model with the highest F-measure as the output.

[0041] The third purpose of the present invention is achieved through the following technical solution: a storage medium storing a program, which, when executed by a processor, implements the above-mentioned inductive process mining method based on local structure optimization.

[0042] The fourth purpose of the present invention is achieved through the following technical solution: a computing device, comprising a processor and a memory for storing processor executable programs, when the processor executes the program stored in the memory, the above-mentioned inductive process mining method based on local structure optimization is implemented.

[0043] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0044] 1. The present invention considers the result model of the inductive model mining algorithm for the first time, performs self-inspection and improvement on the internal structure of the result model, realizes targeted model structure optimization, and is beneficial to improving the efficiency and performance of model mining.

[0045] 2. The present invention optimizes the compliance logs corresponding to the inaccurate structure for the first time, avoiding the influence of the logs with deviations on the optimization model, which not only improves the quality of the optimized model, but also reduces the complexity of the optimized model and improves the readability of the model.

[0046] 3. The present invention adopts a flexible method of iterative optimization of imprecise structures for the first time, taking into account the impact of different new substructures obtained by mining at different thresholds on the overall quality of the original process model, and combining new substructures according to the idea of ​​Cartesian product to obtain all improvement schemes of the original model, thereby ensuring the comprehensiveness of the result model selection.

[0047] 4. The present invention has a wide range of applications in improving and enhancing model mining methods, and has broad prospects for improving the quality of process models obtained through model mining. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The figure is a logical flow diagram of the method of the present invention.

[0049] Figure 2 Schematic diagram of the system of the present invention.

[0050] Figure 3 This is a schematic diagram of the original process tree model.

[0051] Figure 4 Schematic diagram of the process tree model after optimization of the imprecise structure.

[0052] Figure 5 Schematic diagram of the Petri net model after optimization of the imprecise structure. DETAILED DESCRIPTION

[0053] The present invention will be further described below in conjunction with specific embodiments.

[0054] like Figure 1As shown, this embodiment discloses an inductive process mining method based on local structure optimization, firstly, the event log of the actual business process execution is obtained, then the inductive model mining algorithm is used to mine the original process tree model and the sub-log corresponding to the inaccurate structure is extracted after the inaccurate structure is identified, then the frequency-based inductive model mining algorithm method is used to iteratively mine new sub-models in the extracted sub-logs according to a preset threshold as new sub-structures after the inaccurate structure is optimized, each new sub-structure after iterative optimization of the inaccurate structure constitutes a new sub-structure set, then the new sub-structure set is combined according to the Cartesian product to obtain different new sub-structure combinations, each new sub-structure in the combination comes from a different new sub-structure set, the new sub-structure in the combination is used to replace the corresponding inaccurate structure in the original process tree model, to obtain a result process tree model, the result process tree model is converted into a Petri net model, the quality indicators fitness and precision of the model are evaluated to calculate the comprehensive quality indicator F-measure of the model, and finally a Petri net model with the highest F-measure is selected as the output, which includes the following steps:

[0055] 1) Obtain basic data, that is, record the event log of the actual business process execution; the event log is captured and recorded by the information management system when the user participates in the process execution, and is a collection of finite event sequences, each of which is called a trajectory.

[0056] 2) Using an inductive model mining algorithm to mine the process model according to the event log in step 1), obtain the original process tree model, identify the local imprecise structure in the original process tree model, and extract the sub-log corresponding to the imprecise structure; the original process tree model is a directed acyclic graph, defined as a five-tuple T = (V, E, A, l, r), where V is a finite set of nodes in the original process tree model, including leaf nodes and non-leaf nodes; E is a set of edges between nodes in the original process tree model; A is a set of non-silent activity labels; l represents the correspondence between nodes and activity labels in the original process tree model; r is the root node of the original process tree; the imprecise structure is a sub-process tree with a concurrent node as the root node, and its descendant nodes contain a selection node or a loop node. When the first child node of the selection node is a silent node or the first child node of the loop node is an active node, the sub-process tree with the concurrent node as the root node is identified as an imprecise structure.

[0057] The specific situation of identifying the local inaccurate structure in the original process tree model and extracting the corresponding sub-logs is as follows:

[0058] a. Before extracting the sub-logs corresponding to the local imprecise structure, it is necessary to align the activities contained in the event log with the activities corresponding to all leaf nodes in the original process tree model. According to the synchronous movement in the alignment result and the model movement containing invisible nodes, establish the connection between the activities contained in the event log and the activities corresponding to the leaf nodes in the original process tree model:

[0059] Where the alignment is a sequence of moves, the following three types of moves occur:

[0060] The first one is synchronous movement: the events recorded in the trace are consistent with the activities of the original process tree model execution sequence;

[0061] The second type is model movement: there is a deviation between the execution sequence of the trajectory and the original process tree model, and the execution activities in the original process tree model are skipped;

[0062] The third type is log movement: there is a deviation between the execution sequence of the trace and the original process tree model, and the events recorded in the log are skipped;

[0063] b. Based on the characteristics of the imprecise structure, start from the root node of the original process tree model and traverse from top to bottom and from left to right to find and identify the imprecise structure, and identify the substructure in the original process tree model that meets the characteristics of the imprecise structure as the imprecise structure;

[0064] c. Based on the activities corresponding to the leaf nodes in the imprecise structure identified in b and the activity connection between the event log obtained in a and the original process tree model, extract the sub-log corresponding to the imprecise structure from the event log.

[0065] Using the above steps, the event logs involved in process model mining are L = [<b,a,b> ,<d,d,a,d> ,<a,b,b,c,a> ] as an example; the process tree model T obtained by mining the event log L using the inductive model mining algorithm is as follows Figure 3 As shown in the figure, in the process tree model T node a, b, d, c represent activities, τ represents silent activities, ∧ is a concurrency operator, and × is a selection operator. is a loop operator; according to the imprecise structure feature, the imprecise structure is identified and the process tree model T is found to be an imprecise structure; the sub-log corresponding to the imprecise structure process tree model T is extracted as L s =[<b,a> ,<d,d,a,d> ,<a,b,c> ].

[0066] 3) Using a frequency-based inductive model mining algorithm, a new sub-model is iteratively mined from the sub-logs extracted in step 2) according to a preset threshold as a new sub-structure after optimization of the local imprecise structure; wherein the preset threshold is set to 0.2, 0.3, 0.4, 0.5 and 0.6, and the five new sub-structures obtained after each imprecise structure is iteratively optimized according to the above five preset thresholds constitute a set, which is called a new sub-structure set.

[0067] Using the above steps, the frequency-based inductive model mining algorithm is used to extract sub-logs L according to the preset thresholds of 0.2, 0.3, 0.4, 0.5 and 0.6. s The above iterative mining of new sub-models obtains five new process tree models as new sub-structures. These five new sub-structures constitute a new sub-structure set. In this new sub-structure set, the new sub-structure corresponding to the threshold of 0.2 is as follows Figure 4 The new substructure T1 is shown in Figure 1. The new substructures corresponding to thresholds 0.3, 0.4, 0.5, and 0.6 are shown in Figure 1. Figure 4 As shown in the new substructure T2, Figure 4 In the new substructures T1 and T2, a, b, d, and c in the nodes represent activities, τ represents silent activities, → is a sequential operator, ∧ is a concurrent operator, and × is a selection operator. Is the loop operator.

[0068] 4) The new substructure sets are combined according to Cartesian products to obtain several new substructure combinations. The new substructures in the new substructure combinations all come from different new substructure sets. The new substructures in the new substructure combinations are used to replace the corresponding imprecise structures in the original process tree model to obtain a result process tree model. The result process tree model is converted into a Petri net model. The comprehensive quality indicator F-measure of the model is calculated by evaluating the quality indicators fitness and precision of the Petri net model, and finally a Petri net model with the highest F-measure is selected as output.

[0069] The Petri net model is a four-tuple ∑=(P,Tr,F,M0), satisfying: P={p1,p2,p3,...,p m} is a finite set of places, where p m is the mth place in the place set P; Tr = {Tr1, Tr2, Tr3, ..., Tr m} is a finite set of transitions, where Tr mis the mth transition in the transition set Tr; F is a finite set of relations between transitions and places; M0 is the initial mark; the Cartesian product is a combination method of a new substructure set, which combines new substructures in different new substructure sets into a new substructure combination, and the new substructures in the new substructure combination all come from different new substructure sets, that is, Tree1×Tree2×...×Tree m ={(Tree 1.i ,Tree 2.i ,...,Tree m.i )|Tree 1.i ∈Tree1,Tree 2.i ∈Tree2,,...,Tree m.i ∈Tree m}, where Tree m is the mth imprecise structure identified in step 2), Tree m.i It is an inexact structure Tree m The i-th new substructure among the five new substructures obtained after iterative optimization according to five preset thresholds.

[0070] The root node of the new substructure in the new substructure combination replaces the root node of the corresponding inaccurate structure in the original process tree model to complete the inaccurate structure replacement, thereby obtaining a result process tree model; the result process tree model is converted into a Petri net model, specifically, the corresponding structural conversion between the result process tree model and the Petri net model, and the following five types of conversions occur:

[0071] The first type is the conversion of leaf nodes: convert the leaf nodes in the result process tree model into a transition in the Petri net model, and add a place before and after the transition as the input place and output place respectively;

[0072] The second type is the conversion of sequential operators: in the Petri net model, the substructures corresponding to the sequential operators in the result process tree model are sequentially connected using the library;

[0073] The third type is the conversion of concurrent operators: in the Petri net model, the substructures corresponding to the concurrent operators in the result process tree model are connected to the same transition using the library.

[0074] The fourth type is the conversion of selection operators: in the Petri net model, the substructure corresponding to the selection operator in the result process tree model is connected using the same library;

[0075] The fifth type is the transformation of loop operators: in the Petri net model, additional places and transitions are used to construct paths to implement the loop execution of the substructure corresponding to the loop operator in the result process tree model;

[0076] Fitness, precision and F-measure are the quality indicators of the Petri net model. Fitness is used to evaluate the proportion of trajectories contained in the event log that can be reproduced by the Petri net model. The total number of trajectories contained in the event log is recorded as AT, and the number of trajectories that can be reproduced by the Petri net model is recorded as ATR. The calculation formula of fitness is:

[0077]

[0078] Precision is used to quantify the proportion of trajectories that can be generated by the Petri net model that do not appear in the event log. The total number of trajectories that can be generated by the Petri net model is recorded as GT, and the number of trajectories that appear in the event log is recorded as EGT. The calculation formula of precision is:

[0079]

[0080] In order to better balance the fitness and precision of the Petri net model, F-measure is used as the harmonic mean of the two to objectively evaluate the comprehensive quality of the model. The calculation formula of F-measure is:

[0081]

[0082] Using the above steps, since the original process tree model T is an imprecise structure as a whole, the imprecise structure replacement is to replace the original process tree model T with a new substructure in the new substructure set. Therefore, the new substructure T1 corresponding to the threshold 0.2 and the new substructure T2 corresponding to the thresholds 0.3, 0.4, 0.5 and 0.6 are the original process tree T as the result process tree model after the imprecise structure optimization. The original process tree model T and the optimized result process tree models T1 and T2 are converted into Petri net models P, P1 and P2 respectively. Figure 5As shown in the figure, the hollow squares marked with a, b, c, and d in the Petri net model are transitions, the black solid squares are invisible transitions, the circles are places, and the black solid circles in the places are tokens. The model quality of the three Petri net models is calculated. The fitness of the Petri net model P is 0.806, and the precision is 0.094, so the F-measure is 0.168. The fitness of the Petri net model P1 is 0.838, and the precision is 0.561, so the F-measure is 0.672. The fitness of the Petri net model P2 is 0.727, and the precision is 0.733, so the F-measure is 0.730. It can be seen that the F-measure of Petri net P2 is the highest, so the Petri net model P2 is selected as the result model output.

[0083] Example 2

[0084] This embodiment discloses an inductive process mining system based on local structure optimization, which is used to implement the inductive process mining method based on local structure optimization described in Example 1. Figure 2 As shown, the system includes the following functional modules:

[0085] The data acquisition module is used to obtain basic data, that is, the event log that records the actual business process execution;

[0086] The imprecise structure identification and extraction module is used to map the relationship between the activities contained in the event log and the activities contained in the nodes of the original process tree model, find and identify the imprecise structure in the process tree model, and extract the sub-logs corresponding to the imprecise structure;

[0087] An imprecise structure optimization module is used to iteratively optimize the imprecise structure identified in the original process tree model according to a preset threshold based on the frequency-based inductive model mining algorithm;

[0088] Substructure replacement and model screening module: used to replace the inaccurate structure in the original process tree model with the optimized new substructure, calculate the comprehensive quality F-measure of the model by evaluating the quality indicators fitness and precision of the Petri net model, and select the Petri net model with the highest F-measure as the output.

[0089] Example 3

[0090] This embodiment discloses a storage medium storing a program. When the program is executed by a processor, the inductive process mining method based on local structure optimization described in Embodiment 1 is implemented.

[0091] The storage medium in this embodiment can be a disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), a USB flash drive, a mobile hard disk, or other media.

[0092] Example 4

[0093] This embodiment discloses a computing device, including a processor and a memory for storing a program executable by the processor. When the processor executes the program stored in the memory, the inductive process mining method based on local structure optimization described in Example 1 is implemented.

[0094] The computing device described in this embodiment may be a desktop computer, a laptop computer, a smart phone, a PDA handheld terminal, a tablet computer, a programmable logic controller (PLC), or other terminal devices with a processor function.

[0095] To sum up, after adopting the above scheme, the present invention provides a new method and system for mining business process models, that is, self-checking the model obtained by the first mining, and repairing the inaccurate structure detected to improve the overall quality of the process model. It can provide guidance for improving the mining ideas of process models, improve production efficiency, and reduce possible losses. It has practical promotion value and is worthy of promotion.

[0096] The above-described embodiments are only preferred embodiments of the present invention and are not intended to limit the scope of implementation of the present invention. Therefore, all changes made according to the shape and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An inductive process mining method based on local structure optimization, characterized by: The following steps are involved: 1) Obtain basic data, that is, event logs that record actual business process execution; 2) Using the inductive model mining algorithm to mine the process model based on the event log in step 1), obtain the original process tree model, identify the local imprecise structure in the original process tree model, and extract the sub-logs corresponding to the imprecise structure; 3) Using the inductive model mining algorithm, according to the preset threshold, iteratively mine new sub-models from the sub-logs extracted in step 2) as new sub-structures after the local imprecise structure is optimized. Each new sub-structure after the iterative optimization of the imprecise structure constitutes a set, which is called the new sub-structure set; 4) In step 3), the new substructure sets are combined according to Cartesian products to obtain several new substructure combinations, and the new substructures in the new substructure combinations all come from different new substructure sets. The new substructures in the new substructure combinations are used to replace the corresponding imprecise structures in the original process tree model to obtain a result process tree model. The result process tree model is converted into a Petri net model, and the comprehensive quality indicator F-measure of the model is calculated by evaluating the quality indicators fitness and precision of the Petri net model, and finally a Petri net model with the highest F-measure is selected as output.

2. The inductive process mining method based on local structure optimization according to claim 1 is characterized in that: In step 1), the event log is information about user business execution captured and recorded by the information system, and is a collection of finite event sequences, each of which is called a trace.

3. The inductive process mining method based on local structure optimization according to claim 2 is characterized in that: In step 2), the original process tree model is a directed acyclic graph, defined as a quintuple T = (V, E, A, l, r), where V is a finite set of nodes in the original process tree model, including leaf nodes and non-leaf nodes; E is a set of edges between nodes in the original process tree model; A is a set of non-silent activity labels; l represents the correspondence between nodes and activity labels in the original process tree model; r is the root node of the original process tree model; the inexact structure is a sub-process tree with a concurrent node as the root node, whose descendant nodes include a selection node or a loop node. When the first child node of the selection node is a silent node or the first child node of the loop node is an active node, the sub-process tree with the concurrent node as the root node is identified as an inexact structure.

4. The inductive process mining method based on local structure optimization according to claim 3 is characterized in that: In step 2), the specific situation of identifying the local inaccurate structure in the original process tree model and extracting the corresponding sub-logs is as follows: a. Before extracting the sub-logs corresponding to the local imprecise structure, it is necessary to align the activities contained in the event log with the activities corresponding to all leaf nodes in the original process tree model. According to the synchronization movement in the alignment result and the model movement containing invisible nodes, the connection between the activities contained in the event log and the activities corresponding to the leaf nodes in the original process tree model is established: Where the alignment is a sequence of moves, the following three types of moves occur: The first one is synchronous movement: the events recorded in the trace are consistent with the activities of the original process tree model execution sequence; The second type is model movement: there is a deviation between the execution sequence of the trajectory and the original process tree model, and the execution activities in the original process tree model are skipped; The third type is log movement: there is a deviation between the execution sequence of the trace and the original process tree model, and the events recorded in the log are skipped; b. Based on the characteristics of the imprecise structure, start from the root node of the original process tree model and traverse from top to bottom and from left to right to find and identify the imprecise structure, and identify the substructure in the original process tree model that meets the characteristics of the imprecise structure as the imprecise structure; c. Based on the activities corresponding to the leaf nodes in the imprecise structure identified in b and the activity connection between the event log obtained in a and the original process tree model, extract the sub-log corresponding to the imprecise structure from the event log.

5. The inductive process mining method based on local structure optimization according to claim 4 is characterized in that: In step 3), the imprecise structure is optimized using a frequency-based inductive model mining algorithm to iteratively mine new sub-models in the sub-logs corresponding to the imprecise structure according to preset thresholds, where the preset thresholds are set to 0.2, 0.3, 0.4, 0.5 and 0.6, and the five new sub-structures obtained after each imprecise structure is iteratively optimized according to the above five preset thresholds constitute a new sub-structure set.

6. The inductive process mining method based on local structure optimization according to claim 5 is characterized in that: In step 4), the Petri net model is a four-tuple ∑=(P,Tr,F,M0), satisfying: P={p1,p2,p3,...,p m } is a finite set of places, where p m is the mth place in the place set P; Tr = {Tr1, Tr2, Tr3, ..., Tr m } is a finite set of transitions, where Tr m is the mth transition in the transition set Tr; F is a finite set of relations between transitions and places; M0 is an initial mark; the Cartesian product is a combination method of a new substructure set, which combines new substructures in different new substructure sets into a new substructure combination, and the new substructures in the new substructure combination all come from different new substructure sets, that is, Tree1×Tree2×...×Tree m ={(Tree 1.i ,Tree 2.i ,...,Tree m.i )|Tree 1.i ∈Tree1,Tree 2.i ∈Tree2,,...,Tree m.i ∈Tree m }, where Tree m is the mth imprecise structure identified in step 2), Tree m.i It is an inexact structure Tree m The i-th new substructure among the five new substructures obtained after iterative optimization according to five preset thresholds.

7. The inductive process mining method based on local structure optimization according to claim 6 is characterized in that: In step 4), the root node of the new substructure in the new substructure combination replaces the root node of the corresponding inaccurate structure in the original process tree model to complete the inaccurate structure replacement, thereby obtaining a result process tree model; the result process tree model is converted into a Petri net model, specifically, the corresponding structure conversion between the result process tree model and the Petri net model, and the following five types of conversion occur: The first type is the conversion of leaf nodes: convert the leaf nodes in the result process tree model into a transition in the Petri net model, and add a place before and after the transition as the input place and output place respectively; The second type is the conversion of sequential operators: in the Petri net model, the substructures corresponding to the sequential operators in the result process tree model are sequentially connected using the library; The third type is the conversion of concurrent operators: in the Petri net model, the substructures corresponding to the concurrent operators in the result process tree model are connected to the same transition using the library. The fourth type is the conversion of selection operators: in the Petri net model, the substructure corresponding to the selection operator in the result process tree model is connected using the same library; The fifth type is the transformation of loop operators: in the Petri net model, additional places and transitions are used to construct paths to implement the loop execution of the substructure corresponding to the loop operator in the result process tree model; Fitness, precision and F-measure are the quality indicators of the Petri net model. Fitness is used to evaluate the proportion of trajectories contained in the event log that can be reproduced by the Petri net model. The total number of trajectories contained in the event log is recorded as AT, and the number of trajectories that can be reproduced by the Petri net model is recorded as ATR. The calculation formula of fitness is: Precision is used to quantify the proportion of trajectories that can be generated by the Petri net model that do not appear in the event log. The total number of trajectories that can be generated by the Petri net model is recorded as GT, and the number of trajectories that appear in the event log is recorded as EGT. The calculation formula of precision is: In order to better balance the fitness and precision of the Petri net model, F-measure is used as the harmonic mean of the two to objectively evaluate the comprehensive quality of the model. The calculation formula of F-measure is:

8. An inductive process mining system based on local structure optimization, characterized by: The inductive process mining method based on local structure optimization for implementing any one of claims 1 to 7 comprises: The data acquisition module is used to obtain basic data, that is, the event log that records the actual business process execution; The imprecise structure identification and extraction module is used to map the relationship between the activities contained in the event log and the activities contained in the nodes of the original process tree model, find and identify the imprecise structure in the process tree model, and extract the sub-logs corresponding to the imprecise structure; An imprecise structure optimization module is used to iteratively optimize the imprecise structure identified in the original process tree model according to a preset threshold based on the frequency-based inductive model mining algorithm; Substructure replacement and model screening module: used to replace the inaccurate structure in the original process tree model with the optimized new substructure, and to calculate the comprehensive quality indicator F-measure of the model by evaluating the quality indicators fitness and precision of the Petri net model, and select the Petri net model with the highest F-measure as the output.

9. A storage medium storing a program, characterized in that: When the program is executed by a processor, the inductive process mining method based on local structure optimization described in any one of claims 1 to 7 is implemented.

10. A computing device comprising a processor and a memory for storing a program executable by the processor, characterized in that: When the processor executes the program stored in the memory, the inductive process mining method based on local structure optimization described in any one of claims 1 to 7 is implemented.

Images