Airline task execution efficiency analysis method based on process mining

Abstract

The invention discloses an aviation task execution efficiency analysis method based on process mining, which comprises the following steps: acquiring task execution data from a plurality of aviation business systems, and constructing an initial event log, executing data standardization processing and generating a standardized event log according to four serial links of demonstration, planning, monitoring and redisk; performing flow mining and efficiency quantitative modeling based on a standardized event log, forming a feature vector according to the duration of each task case in four links to obtain a flow model, obtaining an efficiency grade division result by adopting a clustering algorithm, and obtaining a processing efficiency reference value; based on the efficiency grade division result and the processing efficiency reference value, performing real-time anomaly detection on the execution data of the newly-added task case, calculating a process model of the newly-added task case, and obtaining a deviation degree of a clustering center corresponding to the efficiency grade division result and / or a deviation degree with the processing efficiency reference value; and in response to early warning triggered by abnormal detection, positioning a flow bottleneck to implement an optimization measure.

Description

Technical Field

[0001] This invention belongs to the field of aviation mission execution efficiency analysis technology, specifically relating to an aviation mission execution efficiency analysis method based on process mining. Background Technology

[0002] Currently, the civil aviation industry is at a critical stage of digital transformation. The execution and management of aviation missions involves multiple heterogeneous business systems, such as operations control management, approval timeliness, flight monitoring, and debriefing management. These systems generate massive amounts of event log data during operation, including start / end timestamps for each stage of the mission, operator information, and mission status. This data provides the possibility for analyzing and optimizing mission execution processes.

[0003] In the early stages, management relied mainly on manual methods, with data recorded using paper or spreadsheets for post-event review. With the development of information technology, rudimentary digital solutions emerged, integrating data from multiple business systems and leveraging business intelligence (BI) tools for basic statistical analysis.

[0004] In recent years, process mining technology, as an emerging process management technique, has been successfully applied in manufacturing, finance, and other fields due to its ability to automatically discover, monitor, and improve actual business processes from event logs. Some researchers have begun to explore introducing general-purpose process mining tools into the field of civil aviation mission management, hoping to achieve more refined efficiency analysis.

[0005] However, when applying existing technologies to analyze the efficiency of aviation mission execution, a series of technical problems still need to be solved. First, at the analytical level, existing methods are either lagging behind and unable to achieve real-time monitoring; or even if general process mining tools are introduced, they fail to effectively combine the business logic of civil aviation missions to build process models that fit business characteristics. Furthermore, there is a lack of an objective and quantitative efficiency grading and benchmark system, making efficiency assessment too reliant on subjective experience.

[0006] Furthermore, in terms of early warning and intervention, existing technologies struggle to set scientific dynamic early warning thresholds, often leading to false alarms or missed alarms. At the same time, for identified inefficient processes, process optimization often becomes a one-off attempt, failing to achieve continuous iterative improvement.

[0007] Therefore, how to overcome the above-mentioned shortcomings and provide a task execution efficiency analysis method that can adapt to the characteristics of civil aviation data, realize dynamic quantitative evaluation of efficiency, provide accurate early warning, and form a closed-loop optimization has become an urgent technical problem that needs to be solved by those skilled in the art. Summary of the Invention

[0008] This invention provides a method for analyzing the execution efficiency of aviation missions based on process mining, which solves the technical problems in existing technologies for civil aviation mission efficiency analysis, such as poor data quality leading to unreliable analysis, lagging efficiency assessment and lack of objective standards, unscientific warning thresholds leading to false alarms and missed alarms, and the disconnect between optimization measures and effect verification, which prevents continuous iteration.

[0009] The technical solution adopted in this invention is as follows: A method for analyzing the execution efficiency of aviation missions based on process mining, comprising: Task execution data is collected from multiple aviation business systems, and initial event logs are constructed according to the four sequential steps of demonstration, planning, monitoring, and review. Data standardization processing is then performed to generate standardized event logs. Based on the standardized event logs, the execution process is mined and efficiency is quantified and modeled. The process model is obtained by forming a feature vector based on the duration of each task case in the four stages. The efficiency level classification result is obtained by using a clustering algorithm, and the processing efficiency benchmark value is obtained. Based on the efficiency level classification results and the processing efficiency benchmark value, real-time anomaly detection is performed on the execution data of the new task cases, the process model of the new task cases is calculated, and the deviation of the cluster center of the corresponding efficiency level classification results and / or the deviation from the processing efficiency benchmark value are obtained. In response to the warning triggered by the anomaly detection, the system locates process bottlenecks to implement optimization measures; and collects execution data of optimized task cases to update process mining and efficiency quantification modeling, and adjusts the anomaly detection threshold based on the updated model data.

[0010] The aviation mission execution efficiency analysis method based on process mining used in this invention also has the following additional technical features: The data standardization process includes deduplication based on composite business primary keys and unified timezone conversion of timestamps from multiple time zones, specifically: Identify the source timezone identifier of the timestamp; Query the time zone rules database to determine the offset from the source time zone to the target standard time zone and the daylight saving time rules; Apply the aforementioned offset and daylight saving time rules to convert the timestamp to the target standard time zone.

[0011] The efficiency ranking results were obtained using a clustering algorithm, specifically including: Use the elbow rule to determine the optimal number of clusters K; The K-means++ algorithm is used to cluster the feature vectors to obtain K clusters and their cluster centers. Based on the duration characteristics of each cluster center, the K clusters are labeled as high-efficiency, medium-efficiency, and low-efficiency levels, respectively.

[0012] The baseline value for processing efficiency is obtained as follows: The sum of squared errors of each of the four stages is calculated for the task clusters marked as high efficiency, and the average of the four sums of squared errors is used as the processing efficiency benchmark.

[0013] Based on the efficiency level classification results and the processing efficiency benchmark value, real-time anomaly detection is performed on the execution data of newly added task cases, specifically including: Based on the existing duration data of the newly added task cases, the corresponding efficiency level classification results are obtained, and the corresponding cluster centers are determined. Calculate the first deviation ratio and the second deviation ratio; the first deviation ratio is the ratio of the absolute difference between the process model of the task case and the corresponding cluster center of the current efficiency level classification result to the cluster center; the second deviation ratio is the ratio of the absolute difference between the process model of the task case and the processing efficiency benchmark value to the benchmark value. The first deviation ratio and / or the second deviation ratio are compared with a preset anomaly detection threshold, and a graded warning is triggered based on the comparison result.

[0014] A tiered warning is triggered based on the comparison results, specifically: If at least one of the first deviation ratio or the second deviation ratio is greater than or equal to the first threshold, a first-level warning is triggered; If both the first deviation ratio and the second deviation ratio are greater than or equal to the second threshold, a second-level warning is triggered; wherein the second threshold is greater than or equal to the first threshold.

[0015] Identify process bottlenecks, specifically: Based on the standardized event logs and corresponding process models, the deviation detection function of process mining is used to analyze the process models of task cases that trigger early warnings by anomaly detection, and process bottlenecks are determined according to the frequency of occurrence and / or time consumption.

[0016] This invention also discloses an aviation mission execution efficiency analysis system based on process mining, comprising: The data acquisition and log building module is used to collect task execution data from multiple aviation business systems and build initial event logs according to the four sequential stages of demonstration, planning, monitoring, and review. The data standardization module is used to perform data standardization processing and generate standardized event logs. The process modeling and efficiency quantification module is used to perform process mining and efficiency quantification modeling based on the standardized event logs. It obtains the process model by forming a feature vector based on the duration of each task case in the four stages, uses a clustering algorithm to obtain the efficiency level classification results, and obtains the processing efficiency benchmark value. The real-time detection and early warning module is used to perform real-time anomaly detection on the execution data of the new task case based on the efficiency level classification result and the processing efficiency benchmark value, calculate the process model of the new task case, and obtain the deviation of the cluster center of the corresponding efficiency level classification result and / or the deviation from the processing efficiency benchmark value. The implementation and verification module is optimized to respond to the early warning triggered by the anomaly detection, locate process bottlenecks, and implement optimization measures. The model iteration and closed-loop control module is used to collect the execution data of the optimized task cases to update the process mining and efficiency quantification model, and adjust the anomaly detection threshold based on the updated model data. The output of the model iteration and closed-loop control module is fed back to the process modeling and efficiency quantification module and the real-time detection and early warning module.

[0017] The present invention further discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described thereon.

[0018] The present invention also discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.

[0019] Due to the adoption of the above technical solution, the beneficial effects achieved by this invention are as follows: 1. In this invention, by collecting data from multiple business systems and performing data standardization processing, heterogeneous data sources can be automatically integrated, solving the problems of inconsistent original data formats and scattered sources. Standardized event logs are generated, laying a unified and standardized data foundation for all subsequent analyses, ensuring the consistency of analytical input, and thus significantly improving the reliability and accuracy of the entire analysis process.

[0020] By automatically constructing process models from real event logs through process mining, the actual task flow is objectively reflected, rather than a pre-set ideal process. Secondly, clustering algorithms are used to automatically classify historical task cases into efficiency levels and calculate benchmark values. Together, these establish an objective evaluation system based on data statistics, replacing the traditional method of subjective and qualitative judgments relying on managerial experience, making efficiency assessments more scientific and comparable.

[0021] Furthermore, based on the established quantitative model (efficiency level classification results and benchmark values), real-time comparative analysis is performed on tasks that are being executed or newly generated. This changes the traditional method's lagging mode of only conducting post-mortem analysis after all tasks have been completed, enabling dynamic and near real-time monitoring of task execution efficiency, allowing managers to promptly detect problems in their early stages or during the process.

[0022] Once an anomaly (deviation exceeding limits) is detected in real-time, an alert can be automatically triggered, guiding or assisting users in locating process bottlenecks. This transforms management from passively responding to known outcomes to proactively addressing process deviations, significantly shortening the cycle from problem discovery to intervention and improving the proactivity and timeliness of process management.

[0023] Furthermore, the effectiveness of optimization measures is validated and recorded by new execution data. This data, in turn, is used to update models and adjust thresholds, enabling the system to adapt to business changes, absorb optimization results, and continuously calibrate early warning sensitivity. This ensures that the entire analytical methodology is not one-off or static, but rather can continuously iterate and improve itself over time and as the company's operational level improves. Attached Figure Description

[0024] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart illustrating the aviation mission execution efficiency analysis method based on process mining according to one embodiment of the present invention. Detailed Implementation

[0025] To more clearly illustrate the overall concept of the present invention, a detailed description will be provided below with reference to the accompanying drawings and examples.

[0026] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0027] like Figure 1 As shown, a method for analyzing the execution efficiency of aviation missions based on process mining includes: S100: Collects task execution data from multiple aviation business systems, and constructs an initial event log according to the four sequential steps of demonstration, planning, monitoring, and review, performs data standardization processing, and generates a standardized event log.

[0028] This step lays the data foundation for the entire method. Its core purpose is to address the inherent multi-source heterogeneity and low quality issues of civil aviation mission data, and to build a unified, complete, and time-accurate high-quality event log for subsequent process mining and efficiency analysis.

[0029] Multi-source data acquisition is performed, proactively collecting raw data from distributed aviation business systems (such as business jet operation control management systems, mission approval and timeliness systems, flight monitoring systems, and mission execution log systems). Specifically, near real-time or scheduled data acquisition can be achieved through pre-configured API interfaces, database synchronization, or message middleware listening.

[0030] Based on the strongly sequential business logic of civil aviation missions—"Audit (A) - Plan (P) - Monitor (M) - Review (R)"—the collected raw data records (such as mission creation, approval, takeoff, landing, report completion, etc.) are categorized, sorted, and associated to construct an initial event log containing multiple ordered events for each independent mission case. Each event must contain at least the following key attributes: case ID, activity (stage) identifier, timestamp, and executor.

[0031] Data cleaning and standardization can include addressing data redundancy (duplicate records), time-series disorder (multiple time zone issues), and missing information (empty key fields).

[0032] Specifically, to address data redundancy (duplicate records): a hash-based deduplication strategy based on a composite business primary key is adopted. As described in the disclosure document, a composite primary key is constructed using fields such as FlightID (flight number), ArgID (argument ID), CustomerID (customer ID), and TaskID (task ID), generating a unique hash value (such as HashKey). Bloom filters and other technologies are then used to quickly identify and remove completely duplicate or highly similar records caused by system interface timeouts or duplicate submissions, ensuring log uniqueness.

[0033] For missing information (empty key fields): an intelligent completion strategy based on business context is adopted. For missing key values ​​such as processing time of missing steps, instead of simply using the global average, a sliding window is used to find historical task data of the same type (such as international charter flight tasks) within similar time periods. This data is then combined with business weights (such as task type and route attribute weights) for weighted calculation or interpolation, thereby maintaining the consistency and rationality of business logic to the greatest extent possible when completing the data.

[0034] After cleaning and transformation, the output is a standardized event log with a consistent structure, complete fields, a consistent time base, and no duplicate records. This log is a high-quality dataset that meets the input requirements of process mining tools.

[0035] In civil aviation mission scenarios, the diversity of data sources and the global nature of operations inevitably result in heterogeneous, time-zone-inconsistent, and noisy raw data. Skipping or simplifying this standardization step and directly using raw data for process mining will lead to distorted process models, unreliable efficiency calculation results, and inaccurate early warning mechanisms.

[0036] This step is a creative preprocessing procedure that deeply integrates the characteristics of civil aviation business and aims to systematically solve the quality problems of multi-source heterogeneous data. It provides high-quality data input, which can significantly improve the convergence speed and stability of subsequent clustering algorithms, and make the calculated efficiency benchmark values ​​(such as CE) and deviation indicators more reliable, thereby enhancing the analytical accuracy and decision support value of the entire system.

[0037] S200: Based on the standardized event log, the execution process is mined and efficiency is quantified and modeled. The process model is obtained by forming a feature vector based on the duration of each task case in the four stages. The efficiency level classification result is obtained by using a clustering algorithm, and the processing efficiency benchmark value is obtained.

[0038] The main objectives of this step are: First, to automatically discover and visualize the actual process model of aviation mission execution from the standardized event logs, so as to objectively reflect the actual flow relationship and variation between each stage of "demonstration-planning-monitoring-review"; Second, based on this process model, to quantitatively evaluate and automatically classify the execution efficiency of historical mission cases, and to establish a measurable efficiency benchmark value, thereby transforming subjective and vague efficiency evaluations into objective and calculable data indicators, providing a scientific reference system for subsequent real-time monitoring and accurate early warning.

[0039] First, process mining is performed to construct a process model. The standardized event logs are input into the process mining engine. Based on the case IDs, activity identifiers (i.e., stage identifiers, such as A, P, M, R) and timestamp sequences in a unified time zone from the event logs, this engine automatically constructs a process model representing the actual execution path of the task using process discovery algorithms such as the alpha algorithm and heuristic algorithms. This model graphically and intuitively displays the logical relationships such as sequence, parallelism, and selection between stages, and can identify the most frequently occurring (mainstream path) and deviation paths with additional waiting or rework. This process replaces the practice of manually drawing ideal flowcharts, ensuring a high degree of consistency between the model and actual business operations.

[0040] Secondly, feature extraction and efficiency quantification modeling are performed. Based on the constructed process model and processing time data in the event logs, the following sub-steps are executed: 1) Feature Vector Construction: For each completed historical task case, extract its actual processing time in the four key stages (A, P, M, R) to form a four-dimensional feature vector. For example, for case i, its feature vector is T_i=[T_Ai,T_Pi,T_Mi,T_Ri]. The feature vectors of all historical cases constitute the sample set used for cluster analysis.

[0041] 2) Cluster Analysis and Efficiency Level Classification: An unsupervised clustering algorithm, such as K-means, is used to perform cluster analysis on the sample set. After the algorithm runs, all historical task cases are divided into K clusters, each with a centroid vector (i.e., the typical value of the duration of each stage at that level). Based on the numerical characteristics of the centroid vectors of each cluster (e.g., clusters with generally shorter durations for each stage), a semantic label of efficiency level is assigned to each cluster, thereby completing the automated and objective classification of efficiency.

[0042] 3) Calculate the processing efficiency benchmark: To establish an efficiency benchmark for the entire process, clusters marked as "highly efficient" are selected as benchmarks. This benchmark (CE) is a comprehensive scalar that quantitatively reflects the overall process efficiency concentration and stability under the company's current best practices, providing a core reference point for measuring the degree of deviation from new tasks.

[0043] Without this step, the system only possesses cleaned data but cannot extract reusable process knowledge and efficiency standards, leaving subsequent anomaly detection without a basis for judgment. This step, through the automatically generated model from process mining, enables managers to gain a clear and objective understanding of the actual overall picture of task execution and potential bottlenecks for the first time, overcoming the limitations of cognition based on reports or partial experience.

[0044] Furthermore, by automatically classifying efficiency levels, an objective evaluation standard based on data statistics has been established, making efficiency assessments consistent and comparable, and providing a dynamic, data-driven comparison standard for real-time anomaly detection in subsequent steps.

[0045] S300: Based on the efficiency level classification results and the processing efficiency benchmark value, perform real-time anomaly detection on the execution data of the newly added task cases, calculate the process model of the newly added task cases, and obtain the deviation of the cluster center of the corresponding efficiency level classification results, and / or the deviation from the processing efficiency benchmark value.

[0046] The main purpose of this step is to apply the static efficiency model established in the previous steps to actual operations, enabling near real-time, automated efficiency monitoring of ongoing or new task cases. By quantitatively calculating the deviation of new tasks from historical efficiency patterns (their respective efficiency levels) and the company's optimal standards (processing efficiency benchmarks), data-driven decision-making is provided for precise and timely business intervention.

[0047] First, new task data is acquired and preprocessed in real time. The system continuously monitors or polls the connected aviation business systems. When a new task case is detected (such as new task creation or completion of a certain stage), the currently generated execution data is collected in real time. This data undergoes standardization processing (such as time zone conversion) to extract the actual processing time of the completed stages.

[0048] Efficiency level prediction and benchmark determination are performed. Based on the duration data already generated for this new task case, the system calculates its similarity with the cluster centers of each efficiency level (e.g., calculating the Euclidean distance between some of its feature vectors and some dimensions of each level's center). It is temporarily classified into the most similar efficiency level, and the cluster center vector corresponding to that level is recorded as its benchmark for comparison within the same category. Simultaneously, the obtained processing efficiency benchmark value is pre-loaded as its optimal benchmark.

[0049] Furthermore, a dual deviation calculation is performed to trigger early warnings for anomaly detection. The deviation of the cluster centers corresponding to the efficiency level classification results reflects the performance of this task in this stage, and the degree of deviation from the normal level of its peers (a group of tasks with similar efficiency levels), which is used to capture local and individual efficiency fluctuations.

[0050] The deviation from the aforementioned processing efficiency benchmark reflects the performance of this task at this stage, and the degree of deviation from the company's highest efficiency standard (best practice), used to assess its drag on the overall process's optimal goals.

[0051] This step, through real-time data access and computation, enables the system to proactively detect efficiency deviations during execution, changing the traditional, lagging model of relying on statistical analysis only after all tasks are completed. By introducing a dual deviation calculation strategy, it not only assesses the task's own anomalies relative to its peers but also evaluates its gap with the company's highest standards, making anomaly detection more comprehensive and accurate. It can detect both sudden deterioration of individual tasks and identify systemic shortcomings that deviate from optimal levels over a long period.

[0052] S400: In response to the warning triggered by the anomaly detection, locate the process bottleneck to implement optimization measures; and collect the execution data of the optimized task cases to update the process mining and efficiency quantification model, and adjust the anomaly detection threshold based on the updated model data.

[0053] The main purpose of this step is to transform identified efficiency anomalies into concrete improvement actions, and to refine the system itself through quantitative feedback on the effectiveness of these actions. This step not only achieves effective optimization after problem identification, but also enables the analysis system to be adaptive and continuously improved, ensuring that the method can adapt to business changes and efficiency improvements in a long-term and dynamic manner.

[0054] Phase 1: Bottleneck Identification and Optimization Implementation. This phase is initiated automatically or with manual assistance upon receiving a triggered alert. First, bottlenecks are precisely located based on process mining. The complete event sequence of the task case that triggered the alert is retrieved, and the deviation detection and compliance check functions of process mining technology are used to compare the actual execution path of the case with the discovered efficient process model.

[0055] By analyzing variations in the activity sequence (such as extra loops, unplanned waits), abnormal resource allocation, or a significant increase in the time consumed in specific steps, the most likely bottleneck node causing this inefficiency can be automatically identified, such as "the average waiting time in the planning approval process is too long due to cross-departmental coordination".

[0056] Secondly, compliance optimization measures are formulated and implemented. Based on the identified bottlenecks and combined with a pre-built civil aviation business rule base (such as safety regulations, qualification requirements, and coordination processes), targeted optimization solutions can be generated or assisted in generating, such as "optimizing the cross-departmental electronic countersigning process and pre-setting parallel approval paths." This solution is deployed and implemented while ensuring compliance with mandatory civil aviation compliance requirements.

[0057] Phase Two: Effect Verification and Model Iteration. After the optimization measures are implemented, this phase ensures that the optimization effects are scientifically evaluated and fed back into the system model. First, data is collected and the optimization effects are verified. The system continuously collects and observes the execution data of newly generated similar task cases after optimization. Statistical methods (such as comparing the average processing time of tasks at bottleneck stages before and after optimization, and calculating the statistical significance p-value) are used to quantitatively evaluate the actual effect of the optimization measures.

[0058] Secondly, dynamically update the model and thresholds. If the validation shows that the optimization effect is significant (e.g., the time is shortened and statistically significant), then: 1) Update the efficiency quantification model: take these new task cases with improved performance as positive samples, integrate them into the historical dataset, and trigger model reconstruction or incremental updates so that the efficient clustering centers and processing efficiency benchmarks can reflect the latest best practice level.

[0059] 2) Adjust the anomaly detection threshold: Based on the updated model and the data distribution characteristics of the optimized tasks, recalculate the threshold used for early warning to adapt it to the new efficiency level and avoid generating too many unnecessary early warnings under the new efficient operating state.

[0060] Without this step, the aforementioned analysis can only identify problems but cannot facilitate effective improvements, and the system model will gradually lag behind business development, leading to ineffective early warnings. This step uses process mining technology to locate bottlenecks, enabling optimization measures to be precise and compliant, avoiding blind optimization. It forms a data-driven continuous improvement loop, establishing a long-term improvement mechanism. By feeding back effective data from optimization to the model and adjusting thresholds, it maintains the accuracy and sensitivity of monitoring and early warning over the long term, thus overcoming the inherent limitation of static models that inevitably become ineffective as business evolves.

[0061] In a preferred embodiment of the present invention, the data standardization processing includes deduplication based on the composite business primary key and unified time zone conversion processing for timestamps in multiple time zones, specifically: Identify the source timezone identifier of the timestamp; Query the time zone rules database to determine the offset from the source time zone to the target standard time zone and the daylight saving time rules; Apply the aforementioned offset and daylight saving time rules to convert the timestamp to the target standard time zone.

[0062] The primary objective of this implementation method is to completely resolve the issue of inconsistent temporal logic caused by the globalization of aviation mission data sources (e.g., the use of Coordinated Universal Time (UTC) by the operations control system, Beijing Time (CST) by the domestic approval system, and local time for aircraft operation records at local airports). By performing precise and consistent time zone conversion on all timestamps, the method ensures the accuracy and reliability of the chronological order of activities recorded in the event log, facilitating accurate process mining, duration calculation, and timeliness analysis in subsequent processes.

[0063] Source time zone identification involves the system parsing the timestamp field and its associated metadata in each event record. First, it attempts to directly read the explicitly contained time zone identifier field from the data. If this field is missing or unclear, it intelligently infers the time zone based on other business attributes strongly associated with the record. For example, based on the destination / departure airport code of the flight mission, it searches for the standard time zone it belongs to in a pre-defined "airport-time zone" mapping table.

[0064] Dynamic time zone rule query and calculation: The system maintains or connects to an authoritative and up-to-date time zone rule database. Based on the identified source time zone identifier and the preset target standard time zone (usually unified as Beijing Time CST for easy domestic business analysis), the system dynamically queries the basic time zone offset between the two.

[0065] More importantly, this query needs to be combined with the specific date corresponding to the timestamp to calculate and determine whether daylight saving time rules applied at that time. This step ensures that the conversion not only considers the fixed time difference but also takes into account the dynamic ±1-hour adjustments caused by daylight saving time, achieving sub-second precision in time difference calculation.

[0066] Precise timestamp conversion and overwriting: The calculated total offset (base offset + daylight saving time adjustment) is used to perform arithmetic operations on the original timestamp, converting it to a new timestamp in the target standard time zone. After conversion, this new timestamp is used to overwrite the corresponding field in the original record, or stored in the new standard time field. Ultimately, all timestamps in the entire event log are unified to the same time zone base.

[0067] If time zone unification is ignored, the timestamps of task stages from different systems will be on different timelines, and the directly calculated stage duration and task flow interval will be completely distorted. Any process model and efficiency indicators built on this basis will become meaningless.

[0068] This implementation provides the process mining algorithm with a completely accurate event log of temporal relationships, enabling the discovered process model to truly reflect the execution logic and dependencies of cross-regional tasks. Furthermore, the processing time calculation for all tasks is based on the same time scale, ensuring fairness and scientific rigor in efficiency comparisons, cluster analysis, and benchmark setting.

[0069] As a preferred embodiment of the present invention, the efficiency level classification result is obtained by using a clustering algorithm, specifically including: Use the elbow rule to determine the optimal number of clusters K; The K-means++ algorithm is used to cluster the feature vectors to obtain K clusters and their cluster centers. Based on the duration characteristics of each cluster center, the K clusters are labeled as high-efficiency, medium-efficiency, and low-efficiency levels, respectively.

[0070] This implementation details the core process of objectively and automatically classifying the efficiency of aviation mission execution using clustering techniques in machine learning. Its main objective is to automatically categorize massive amounts of historical mission cases into several categories (levels) with clearly defined efficiency levels based on their execution duration characteristics, thereby constructing a data-driven, non-subjective efficiency evaluation system. This solves the problems of traditional methods relying on manual experience to define standards, resulting in strong subjectivity, poor consistency, and an inability to handle complex, multi-dimensional data. It provides a structured, quantitative benchmark for subsequent precise early warning and benchmarking management.

[0071] Determining the optimal number of clusters based on the elbow rule: First, it is necessary to determine whether dividing the task into several efficiency levels is scientifically reasonable. To this end, the elbow rule is used for automated decision-making. Specifically, the number of clusters K is incremented from 2 to a preset maximum value (e.g., 10). For each K value, subsequent clustering algorithms are run, and the sum of squared distances between all sample points and their respective cluster centers is calculated, i.e., the sum of squared errors.

[0072] As the K value increases, SSE decreases, but the rate of decrease gradually slows down. The system plots the SSE-K curve and automatically identifies the elbow point on the curve, which is the location where the rate of decrease of SSE decreases significantly. The K value corresponding to this point is the optimal number of clusters that balances the explanatory power and complexity of the model.

[0073] In aviation mission scenarios, this value is usually 3, which naturally corresponds to the three business understanding levels of "high efficiency", "medium efficiency" and "low efficiency".

[0074] Robust clustering is performed using the K-means++ algorithm: After determining the optimal K value, the K-means++ algorithm is used to cluster the feature vector sample set composed of the four stages of each task case. Preferably, instead of randomly selecting initial cluster centers, a probability distribution is used to make the initial center points as far apart as possible. This effectively avoids the risk of the traditional K-means algorithm getting trapped in local optima due to improper random initialization, ensuring the stability and repeatability of the clustering results.

[0075] The algorithm then iteratively performs two steps: assigning samples to the nearest center and recalculating the center point, until convergence. Finally, it outputs K clusters and the cluster center vector for each cluster. This center vector represents the typical time taken for this efficiency level across the four stages.

[0076] Efficiency-level semantic labeling based on cluster center features: After the algorithm completes clustering, it obtains K mathematical clusters. Then, based on the numerical features of each cluster center vector, it automatically assigns a business semantic label.

[0077] Specifically, the numerical values ​​of each central vector across various dimensions (stages) are compared. Clusters with relatively short stages are labeled as "efficient"; clusters with relatively long stages are labeled as "inefficient"; and the rest are labeled as "medium efficient." This process is entirely based on data distribution, requiring no manual pre-setting of thresholds, thus achieving complete automation and objectivity in efficiency grading.

[0078] This implementation method is entirely data-driven, eliminating human bias and ensuring the consistency and fairness of evaluation criteria, resulting in highly reproducible analysis results. Without needing to predefine what constitutes efficiency, the algorithm can automatically learn and summarize several typical efficiency patterns from historical data, potentially revealing efficiency levels that managers may not be consciously aware of.

[0079] As a preferred embodiment of this implementation, the processing efficiency benchmark value is obtained as follows: The sum of squared errors of each of the four stages is calculated for the task clusters marked as high efficiency, and the average of the four sums of squared errors is used as the processing efficiency benchmark.

[0080] The primary objective of this embodiment is to construct a single numerical indicator that can comprehensively and stably represent the company's current optimal operational level. This indicator is not a simple average over time, but rather measures the degree of consistency within high-performing task groups to characterize the efficiency concentration and stability of the entire process. This provides a scientific, dynamic, and robust benchmark for evaluating the deviation of any task or step from the company's optimal standards in subsequent steps.

[0081] Selection of benchmark group: The system selects task clusters that are automatically labeled as high-efficiency in the efficiency level classification (i.e., the cluster center of the cluster represents the shortest or relatively short time consumption pattern of each step). All task cases within this cluster are regarded as the best practice set for the company's current operational efficiency.

[0082] Calculate the sum of squared errors for each stage: For this efficient cluster, calculate the sum of squared errors for each of the four stages: argumentation (A), planning (P), monitoring (M), and review (R).

[0083] Specifically, for a certain stage (such as stage A), the calculation formula is: the sum of the squares of the differences between the duration of stage A for all task cases in the cluster and the cluster center value of stage A in the cluster.

[0084] SSE_A = Σ(T_Ai - C_Efficient_A)². The magnitude of this value intuitively reflects the degree of dispersion of the efficient group in the execution time of this stage; the smaller the SSE value, the more similar and stable the performance of efficient tasks in this stage.

[0085] Calculate the mean as the baseline: Take the arithmetic mean of the SSE values ​​(SSE_A, SSE_P, SSE_M, SSE_R) obtained from the above calculations for the four stages. That is: CE = (SSE_A + SSE_P + SSE_M + SSE_R) / 4. The final CE value is the baseline value for processing efficiency. It is a scalar representing the average internal dispersion (variance scale) of the high-efficiency task group across all stages of the entire process, comprehensively reflecting the overall precision and controllability of the company's best practices.

[0086] Simply using the average duration or historical shortest duration of high-efficiency clusters as a benchmark is easily affected by extreme values ​​and cannot reflect the stability of the process. This embodiment uses the mean of internal dispersion (SSE) as the benchmark, making the CE value insensitive to individual extreme high-efficiency cases and better reflecting the overall stability level of the high-efficiency process. Therefore, it is a more reliable and robust benchmark.

[0087] As a preferred embodiment of the present invention, based on the efficiency level classification results and the processing efficiency benchmark value, real-time anomaly detection is performed on the execution data of newly added task cases, specifically including: Based on the existing duration data of the newly added task cases, the corresponding efficiency level classification results are obtained, and the corresponding cluster centers are determined. Calculate the first deviation ratio and the second deviation ratio; the first deviation ratio is the ratio of the absolute difference between the process model of the task case and the corresponding cluster center of the current efficiency level classification result to the cluster center; the second deviation ratio is the ratio of the absolute difference between the process model of the task case and the processing efficiency benchmark value to the benchmark value. The first deviation ratio and / or the second deviation ratio are compared with a preset anomaly detection threshold, and a graded warning is triggered based on the comparison result.

[0088] This preferred embodiment aims to detail how to utilize an established efficiency quantification model to perform real-time, automated anomaly detection and tiered early warning for the execution efficiency of newly added task cases. Its main objective is to achieve a leap from static historical analysis to dynamic process monitoring. By calculating the dual quantitative deviation of a new task relative to its normal efficiency level and the company's optimal standard, efficiency deviations during execution can be promptly identified and transformed into actionable tiered early warning signals, thereby supporting early intervention and precise management of efficiency issues.

[0089] Dynamic Efficiency Level Prediction and Benchmark Matching: When a stage of a new task case is completed and time data is generated, the system does not wait for the entire process to finish, but immediately predicts the efficiency status. Based on the time data of the currently completed stages, the system calculates the distance (e.g., Euclidean distance) between the task and the cluster centers of each efficiency level (in the dimension of completed stages). The system then real-time categorizes the task into the nearest efficiency level and records the cluster center corresponding to that level as its benchmark. Simultaneously, the obtained processing efficiency benchmark (CE) remains available as the globally optimal benchmark.

[0090] Two standardized deviation ratios are calculated in parallel. The first deviation ratio (intra-stage deviation rate) is calculated as R_single = |t_i - C_match| / C_match. Here, t_i represents the actual processing time of the new task in that stage, and C_match represents the cluster center value of its currently predicted efficiency level within that stage. This ratio measures the current task's performance in that stage, reflecting its deviation from the average level of a group of tasks with similar efficiency levels, thus indicating local, individual abnormal fluctuations.

[0091] The second deviation ratio (overall deviation rate) is calculated as R_overall = |t_i - CE| / CE, where CE is the aforementioned company processing efficiency benchmark. This ratio measures the gap between the current task's performance at this stage and the stability standard represented by the company's best efficiency practices, assessing its drag on the overall optimal goal.

[0092] The first deviation ratio and / or the second deviation ratio are compared with a preset anomaly detection threshold, and a tiered warning is triggered based on the comparison result, specifically: If at least one of the first deviation ratio or the second deviation ratio is greater than or equal to the first threshold, a first-level warning is triggered; If both the first deviation ratio and the second deviation ratio are greater than or equal to the second threshold, a second-level warning is triggered; wherein the second threshold is greater than or equal to the first threshold.

[0093] The system pre-sets two key numerical thresholds: a first threshold (θ1) and a second threshold (θ2), where θ2 ≥ θ1. These two thresholds are not empirical values, but are derived from statistical analysis of inefficient or anomalous cases in historical task data (e.g., taking specific high quantiles of the first and second deviation ratios in historical anomalous cases, respectively). θ1 is typically set as the threshold that triggers general attention, while θ2 is set as a higher standard that triggers the assessment of serious problems.

[0094] The logic for a Level 1 warning (reminder-type warning) is (R_single≥θ1)OR(R_overall≥θ1). That is, as long as a new task case meets or exceeds the first threshold θ1 in either the first deviation ratio (deviation within the stage) or the second deviation ratio (overall deviation) at a certain stage, the trigger condition is met.

[0095] The first-level warning is automatically triggered. This level of warning falls under the alert and attention level. The implementation strategy typically involves automatically generating and sending a warning notification to the designated person responsible for that step via internal enterprise communication interfaces (such as email systems or instant messaging tools). The warning information includes the task ID, the abnormal step, the specific deviation value, and the exceeded threshold, prompting the responsible person to conduct immediate verification and initial intervention.

[0096] The logic for a Level 2 warning (action-oriented warning) is (R_single≥θ2)AND(R_overall≥θ2). That is, the trigger condition is met only if the first deviation ratio and the second deviation ratio of a newly added task case at a certain stage simultaneously reach or exceed the higher second threshold θ2.

[0097] The second-level warning is automatically triggered. This level of warning belongs to the assessment and action level. Its implementation strategy is usually as follows: 1) Simultaneous notification to management: In addition to the person responsible for the notification process, the warning information will be simultaneously pushed to a higher-level management dashboard or designated management personnel.

[0098] 2) Initiating the optimization evaluation process: The system can automatically create a process optimization evaluation task work order and associate it with the detailed context data of the abnormal task (such as complete event logs and deviation calculation details), providing a complete information entry point for the subsequent initiation of structured root cause analysis and process optimization procedures.

[0099] Without such a real-time, quantitative dual deviation detection mechanism, efficiency management will remain at the post-event statistical level, unable to detect problems in real time during task execution; while using only a single fixed threshold or a single dimension (such as comparing only with the planned time) for early warnings cannot distinguish the nature and severity of anomalies, which can easily lead to warning overload or ignore important issues.

[0100] This implementation method, through dual-ratio calculation, not only quantifies the magnitude of the deviation but also reveals its nature (whether it's an inherent fluctuation or a systemic lag), making the early warning information more business-oriented. The logic of classifying early warnings into Level 1 and Level 2 effectively distinguishes the severity and scope of the problem, guiding management resources to respond in a differentiated manner (such as self-inspection by responsible persons and process optimization assessment), improving the efficiency and targeting of management actions, and achieving precise governance.

[0101] As a preferred embodiment of the present invention, the process bottleneck is located specifically as follows: Based on the standardized event logs and corresponding process models, the deviation detection function of process mining is used to analyze the process models of task cases that trigger early warnings by anomaly detection, and process bottlenecks are determined according to the frequency of occurrence and / or time consumption.

[0102] The main purpose of this implementation method is to transform efficiency anomaly alarms issued during the anomaly detection process into specific and actionable bottleneck location diagnostic results, thereby providing precise targets for subsequent targeted optimization measures. It aims to overcome the shortcomings of traditional management methods that rely solely on experience to guess bottlenecks or engage in holistic, blind optimization, and achieve data-driven, precise problem localization.

[0103] The system retrieves complete standardized event logs of task cases that trigger warnings (especially level 2 warnings), and loads process models (i.e. efficient process models) generated from historical high-efficiency task mining as benchmarks.

[0104] The system utilizes the deviation detection and compliance check functions of process mining technology to conduct an in-depth comparative analysis of the actual execution path of the abnormal task (extracted from its event logs) and the benchmark process model. The analysis mainly focuses on two aspects: Activity sequence and frequency analysis: Examine whether there are additional activities outside the model, jumps or loops between activities, or missing key activities in the actual path. For example, it was found that an abnormal task frequently included an "additional coordination review" activity that did not appear in the efficient model between the "Plan (P)" and "Monitor (M)" stages.

[0105] Analysis of processing time and waiting time: Compare the actual processing time of abnormal tasks in each stage with the typical time (or cluster center value) of the corresponding stage in the efficient model. Identify specific stages with processing time significantly higher than the efficiency standard.

[0106] More importantly, analyzing the timestamp intervals in the event logs can identify whether there are abnormally long waiting times between specific activities, which often reflects hidden bottlenecks such as resource conflicts and approval stagnation.

[0107] Based on the results of the above deviation analysis, the system identifies one or more bottleneck nodes that cause the inefficiency, according to two core dimensions: frequency of occurrence (e.g., high frequency of a non-standard activity) and / or time consumption (e.g., abnormally long processing time of a certain step or unusually long waiting time between steps). For example, the system may determine that the bottleneck is: "The average waiting time in the planning approval process has increased by 35% due to the rigidity of the cross-departmental electronic countersigning sequence."

[0108] This implementation method ensures that any optimization action is based on a concrete insight into the root cause of the problem. By using process mining technology to objectively compare actual execution data with benchmark models, the specific steps and manifestations of the problem can be clearly identified based on data, achieving precise and objective problem localization. The identified bottlenecks directly indicate the direction of optimization, enabling subsequent optimization measures to be designed with a high degree of focus, greatly improving the effectiveness and implementation efficiency of the optimization plan.

[0109] This invention also provides an aviation mission execution efficiency analysis system based on process mining, comprising: The data acquisition and log building module is used to collect task execution data from multiple aviation business systems and build initial event logs according to the four sequential stages of demonstration, planning, monitoring, and review. The data standardization module is used to perform data standardization processing and generate standardized event logs. The process modeling and efficiency quantification module is used to perform process mining and efficiency quantification modeling based on the standardized event logs. It obtains the process model by forming a feature vector based on the duration of each task case in the four stages, uses a clustering algorithm to obtain the efficiency level classification results, and obtains the processing efficiency benchmark value. The real-time detection and early warning module is used to perform real-time anomaly detection on the execution data of the new task case based on the efficiency level classification result and the processing efficiency benchmark value, calculate the process model of the new task case, and obtain the deviation of the cluster center of the corresponding efficiency level classification result and / or the deviation from the processing efficiency benchmark value. The implementation and verification module is optimized to respond to the early warning triggered by the anomaly detection, locate process bottlenecks, and implement optimization measures. The model iteration and closed-loop control module is used to collect the execution data of the optimized task cases to update the process mining and efficiency quantification model, and adjust the anomaly detection threshold based on the updated model data. The output of the model iteration and closed-loop control module is fed back to the process modeling and efficiency quantification module and the real-time detection and early warning module.

[0110] Therefore, it can achieve any effect in the aviation mission execution efficiency analysis method based on process mining, which will not be elaborated here.

[0111] The present invention further provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements any of the methods described above.

[0112] Therefore, it can achieve any effect in the aviation mission execution efficiency analysis method based on process mining, which will not be elaborated here.

[0113] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the methods described herein.

[0114] Therefore, it can achieve any effect in the aviation mission execution efficiency analysis method based on process mining, which will not be elaborated here.

[0115] For any parts not mentioned in this invention, existing technologies can be used or referenced.

[0116] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0117] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for analyzing the execution efficiency of aviation missions based on process mining, characterized in that, include: Task execution data is collected from multiple aviation business systems, and initial event logs are constructed according to the four sequential steps of demonstration, planning, monitoring, and review. Data standardization processing is then performed to generate standardized event logs. Based on the standardized event logs, the execution process is mined and efficiency is quantified and modeled. The process model is obtained by forming a feature vector based on the duration of each task case in the four stages. The efficiency level classification result is obtained by using a clustering algorithm, and the processing efficiency benchmark value is obtained. Based on the efficiency level classification results and the processing efficiency benchmark value, real-time anomaly detection is performed on the execution data of the new task cases, the process model of the new task cases is calculated, and the deviation of the cluster center of the corresponding efficiency level classification results and / or the deviation from the processing efficiency benchmark value are obtained. In response to the early warning triggered by the anomaly detection, the process bottleneck is located and optimization measures are implemented; In addition, it collects execution data of optimized task cases to update process mining and efficiency quantification modeling, and adjusts the anomaly detection threshold based on the updated model data.

2. The method according to claim 1, characterized in that, The data standardization process includes deduplication based on composite business primary keys and unified timezone conversion of timestamps from multiple time zones, specifically: Identify the source timezone identifier of the timestamp; Query the time zone rules database to determine the offset from the source time zone to the target standard time zone and the daylight saving time rules; Apply the aforementioned offset and daylight saving time rules to convert the timestamp to the target standard time zone.

3. The method according to claim 1, characterized in that, The efficiency ranking results were obtained using a clustering algorithm, specifically including: Use the elbow rule to determine the optimal number of clusters K; The K-means++ algorithm is used to cluster the feature vectors to obtain K clusters and their cluster centers. Based on the duration characteristics of each cluster center, the K clusters are labeled as high-efficiency, medium-efficiency, and low-efficiency levels, respectively.

4. The method according to claim 3, characterized in that, The baseline value for processing efficiency is obtained as follows: The sum of squared errors of each of the four stages is calculated for the task clusters marked as high efficiency, and the average of the four sums of squared errors is used as the processing efficiency benchmark.

5. The method according to claim 1, characterized in that, Based on the efficiency level classification results and the processing efficiency benchmark value, real-time anomaly detection is performed on the execution data of newly added task cases, specifically including: Based on the existing duration data of the newly added task cases, the corresponding efficiency level classification results are obtained, and the corresponding cluster centers are determined. Calculate the first deviation ratio and the second deviation ratio; the first deviation ratio is the ratio of the absolute difference between the process model of the task case and the corresponding cluster center of the current efficiency level classification result to the cluster center; the second deviation ratio is the ratio of the absolute difference between the process model of the task case and the processing efficiency benchmark value to the benchmark value. The first deviation ratio and / or the second deviation ratio are compared with a preset anomaly detection threshold, and a graded warning is triggered based on the comparison result.

6. The method according to claim 5, characterized in that, A tiered warning is triggered based on the comparison results, specifically: If at least one of the first deviation ratio or the second deviation ratio is greater than or equal to the first threshold, a first-level warning is triggered; If both the first deviation ratio and the second deviation ratio are greater than or equal to the second threshold, a second-level warning is triggered. The second threshold is greater than or equal to the first threshold.

7. The method according to claim 1, characterized in that, Identify process bottlenecks, specifically: Based on the standardized event logs and corresponding process models, the deviation detection function of process mining is used to analyze the process models of task cases that trigger early warnings by anomaly detection, and process bottlenecks are determined according to the frequency of occurrence and / or time consumption.

8. A system for analyzing the efficiency of aviation mission execution based on process mining, characterized in that, include: The data acquisition and log building module is used to collect task execution data from multiple aviation business systems and build initial event logs according to the four sequential stages of demonstration, planning, monitoring, and review. The data standardization module is used to perform data standardization processing and generate standardized event logs. The process modeling and efficiency quantification module is used to perform process mining and efficiency quantification modeling based on the standardized event logs. It obtains the process model by forming a feature vector based on the duration of each task case in the four stages, uses a clustering algorithm to obtain the efficiency level classification results, and obtains the processing efficiency benchmark value. The real-time detection and early warning module is used to perform real-time anomaly detection on the execution data of the new task case based on the efficiency level classification result and the processing efficiency benchmark value, calculate the process model of the new task case, and obtain the deviation of the cluster center of the corresponding efficiency level classification result and / or the deviation from the processing efficiency benchmark value. The implementation and verification module is optimized to respond to the early warning triggered by the anomaly detection, locate process bottlenecks, and implement optimization measures. The model iteration and closed-loop control module is used to collect the execution data of the optimized task cases to update the process mining and efficiency quantification model, and adjust the anomaly detection threshold based on the updated model data. The output of the model iteration and closed-loop control module is fed back to the process modeling and efficiency quantification module and the real-time detection and early warning module.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

Images