A business process data analysis system and method based on distributed access

Abstract

The application discloses a kind of based on distributed access's service process data analysis system and method, it is related to big data analysis technical field, the present application is determined by analyzing historical abnormal record characteristic service node;Setting training period, construct characteristic index library and association library;Abnormal record is collected, and service score is calculated, and curve chart is drawn, and abnormal time is predicted;Real-time service process is collected, and real-time service score is calculated, and abnormal time is predicted, reduce invalid analysis range, improve abnormal detection efficiency and accuracy, significantly improve the depth and accuracy of abnormal diagnosis, effectively reduce artificial monitoring workload, improve operation efficiency.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, specifically a business process data analysis system and method based on distributed access. Background Technology

[0002] Existing business process data analysis systems mostly rely on distributed architecture. In a distributed business environment, the data generated by different nodes have significant differences in format structure, time precision and semantic expression, which makes it difficult to achieve unified parsing and effective alignment of data in subsequent analysis stages, affecting the accuracy and real-time performance of the overall analysis.

[0003] Furthermore, most existing process analysis and optimization methods are based on static models or fixed rules for evaluation, lacking the ability to dynamically model the dependencies between processes and to conduct cross-node collaborative analysis. Traditional methods typically calculate the running data of each node independently, ignoring the temporal correlation and task dependency between nodes. They cannot accurately identify the critical paths and potential bottlenecks that affect the overall business performance. When a process runs abnormally or its performance degrades, the system can often only provide post-event results, lacking the ability to trace the source of the abnormality and provide targeted optimization suggestions, which limits the realization of intelligent decision-making and proactive optimization.

[0004] Therefore, how to build a business process data analysis system that supports distributed access and real-time data fusion, realize the structured transformation and time synchronization of heterogeneous data from multiple business nodes, and on this basis, establish a multi-node collaborative analysis model based on process dependency graphs to achieve intelligent evaluation of process performance, bottleneck identification and anomaly detection, has become a technical problem that needs to be solved in the field of business process management and data intelligent analysis. Summary of the Invention

[0005] The purpose of this invention is to provide a business process data analysis system and method based on distributed access, so as to solve the problems raised in the prior art.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a business process data analysis method based on distributed access, the method comprising:

[0007] Step S100: Based on the historical execution records of the business process platform, analyze the historical anomaly records, count the frequency of occurrence of abnormal business nodes, and determine the set of characteristic business nodes for each business process.

[0008] Step S200: Set the training period, analyze the execution records, select the feature indicators of each business node and build a feature indicator library, analyze the similarity between business nodes based on the feature indicators of each business node, determine the correlation between each business node and establish a feature association library.

[0009] Step S300: Obtain abnormal records within the training period; based on the feature index library, collect the feature indexes corresponding to the abnormal business nodes within the analysis period; calculate the business score of the abnormal business nodes; based on the feature association library, obtain the business scores of the associated business nodes corresponding to the abnormal business nodes; calculate the feature scores; summarize the feature scores within each analysis period and draw a curve; extract the image features of the curve; and combine the duration to establish an abnormal time prediction model.

[0010] Step S400: Collect real-time business processes, determine the characteristic business nodes corresponding to the real-time business processes, calculate the real-time business score of each characteristic node, and calculate the real-time predicted abnormal time based on the abnormal time prediction model. Based on the real-time predicted abnormal time, determine whether to check.

[0011] Furthermore, step S100 includes:

[0012] Step S101: In the business process platform, the accessed business processes are identified, a business process library is preset, the business nodes corresponding to each business process are obtained, and each business node is coded. A preset time period is used as the analysis period. The start time of the analysis period is set as the initial time point, and the end time of the analysis period is set as the final time point. The business processes received by the business process platform are obtained, and the set of business nodes corresponding to the business processes is determined. During the execution of the business process, the running behavior data of the set of business nodes in each process analysis period is obtained. The boundary data of the running behavior data is extracted at the initial time point and the final time point respectively. The boundary data and the running behavior data are summarized to obtain the analysis data. The analysis data of each process analysis period are integrated to generate an execution record and upload it to the business process platform.

[0013] Step S102: Obtain historical execution records, collect the judgment results of each analysis cycle in the historical execution records, set the analysis cycle with the judgment result as an abnormal cycle, if there is an abnormal cycle in the historical execution records, mark the historical execution records as historical abnormal records, extract the business process corresponding to each historical abnormal record, and classify the historical abnormal records according to the business process;

[0014] Step S103: In the historical anomaly record set of a certain business process, collect the abnormal business nodes corresponding to the abnormal period in a certain historical anomaly record, count the number of times each abnormal business node appears in the historical anomaly record set, calculate the occurrence frequency of each abnormal business node, preset the occurrence frequency threshold, set the abnormal business nodes that exceed the occurrence frequency threshold as the feature business nodes of the business process, and summarize to generate the feature business node set of each business process.

[0015] By using a pre-defined business process library and business node coding mechanism, business process nodes can be quickly identified and accurately extracted from the accessed business processes, improving the efficiency of business process identification and facilitating the setting of subsequent analysis cycles and the collection of execution data, thereby providing basic data support for accurate analysis and decision-making.

[0016] By setting a preset time period as the analysis cycle and collecting operational behavior data of business nodes within each analysis cycle, it is possible to effectively capture the operational status of each node in the business process at different time periods, providing rich time series data for subsequent anomaly detection and helping to discover potential problems in a timely manner.

[0017] By acquiring historical execution records and identifying and classifying abnormal cycles, historical data can be effectively categorized, organized, and analyzed. This helps to discover potential recurring problems in historical execution and to take targeted preventative and optimization measures, thereby improving the stability and efficiency of the overall business process.

[0018] By statistically analyzing the frequency of abnormal business nodes during abnormal cycles and calculating the occurrence frequency of abnormal nodes, key characteristic business nodes in the business process can be identified. This helps to optimize business process design, provide early warning of potential bottlenecks and problem nodes, and thus provide a basis for continuous improvement of the business process.

[0019] Furthermore, step S200 includes:

[0020] Step S201: Select several consecutive days as the training period, summarize the execution records of each business process within the training period, and collect the analysis data of a certain execution record in each analysis period from the execution record set of a certain business process.

[0021] Step S202: Based on the analysis data of each analysis cycle, evaluate and extract features from the operational behavior data of each business node, select feature indicators from the candidate indicators, and establish a feature indicator library for each business node.

[0022] Step S203: Analyze the relationships between each business node based on the characteristic indicators of each business node, and establish a characteristic association library for each business node.

[0023] Furthermore, step S202 includes:

[0024] Step S202-1: By statistically analyzing the execution status of business nodes in each analysis period, the original indicators of the business node operation characteristics are extracted and summarized as a candidate indicator set. The original indicators include time-related indicators, load-related indicators, status-related indicators, path-related indicators, and scheduling-related indicators.

[0025] Step S202-2: Collect the values ​​of each original indicator in the candidate indicator set, and calculate the numerical distribution characteristics in each analysis period through statistical feature analysis. The numerical distribution characteristics include the mean, variance, dispersion, trend, fluctuation range, extreme value ratio and periodic stability of the original indicator in the analysis period. Normalize the numerical distribution characteristics, preset the weight of each numerical distribution characteristic, and perform weighted summation on the normalized numerical distribution characteristics to calculate the performance score of the original indicator. Preset the performance score threshold, set the original indicators below the performance score threshold as fluctuation indicators, and assign values ​​to the fluctuation indicators.

[0026] Step S202-3: Obtain the judgment result for each analysis period and assign values ​​to the judgment result. Using the candidate indicator set as input and the judgment result as output, use the Pearson correlation method to calculate the abnormal correlation value of each candidate indicator. Using the candidate indicator set as input and the fluctuation indicator as output, use a binary classification model for training and calculate the performance fluctuation correlation value of each candidate indicator.

[0027] Step S202-4: Normalize the abnormal correlation value and the performance fluctuation correlation value, preset the weights of the abnormal correlation and performance fluctuation correlation, and sum the normalized abnormal correlation value and the performance fluctuation correlation value to calculate the comprehensive score of each original indicator. Preset the comprehensive score threshold, and take the original indicators that exceed the comprehensive score threshold as feature indicators. Summarize the feature indicators of each business node and establish a feature indicator library.

[0028] By statistically analyzing the execution status of business nodes in each analysis cycle and extracting raw indicators from multiple dimensions such as time, load, status, path, and scheduling, the operational characteristics of business nodes can be comprehensively and granularly quantitatively described. This avoids the shortcomings of traditional methods that rely on only a single operational parameter, and helps to deeply characterize the true operational status of business nodes.

[0029] By calculating the distribution characteristics of the original indicators and using normalization and weighted scoring methods, indicators with significant fluctuation characteristics within the period can be accurately screened out. This can effectively identify key fluctuation indicators that have a significant impact on the stability of business nodes and improve the ability to detect anomalies in advance.

[0030] By using Pearson correlation analysis and binary classification model training, the abnormal correlation values ​​and performance fluctuation correlation values ​​of candidate indicators are calculated respectively. This allows us to simultaneously explore the intrinsic relationship between indicators and abnormal events from both the anomaly triggering level and the performance fluctuation level, thereby enhancing the credibility and interpretability of anomaly analysis.

[0031] By normalizing, weighting, and merging the abnormal correlation values ​​and performance fluctuation correlation values, a comprehensive scoring system is formed. This avoids the bias caused by single correlation evaluation and makes the final selected feature indicators more globally representative and stable, thereby ensuring that the construction of the feature indicator library has high accuracy and high reliability.

[0032] Furthermore, step S203 includes:

[0033] Step S203-1: Obtain the set of feature indicators corresponding to each business node, extract the numerical distribution characteristics of each feature indicator in different analysis periods, calculate the performance score of the feature indicator, and calculate the business score of the business node by weighted summation based on the performance fluctuation correlation value of the feature indicator. Summarize the business score and the numerical distribution characteristics of each feature indicator to construct the feature vector of each business node.

[0034] Step S203-2: Obtain the feature vectors of any two business nodes. Using the vector similarity method, with the feature vectors of the two business nodes as input, calculate the similarity between the business nodes in the overall feature space. Set a similarity threshold. Set two business nodes that exceed the similarity threshold as associated business nodes. Summarize the associated business nodes corresponding to each business node and establish a feature association library for each business node.

[0035] By summarizing the business scores and the numerical distribution characteristics of each feature indicator, feature vectors of business nodes can be constructed. This enables a unified vectorized expression of complex and multi-dimensional business operation characteristics, enhancing the integrity of the data structure and providing a high-quality input data foundation for subsequent similarity calculation, cluster analysis, and anomaly diagnosis.

[0036] The performance scores of the feature indicators are weighted and integrated based on their fluctuation correlation, so that the contribution of different indicators to the overall characteristics of the business nodes is more in line with the actual business impact, improving the distinguishability of the node feature vectors and making them more accurately reflect the comprehensive performance of the business nodes in terms of stability, load pressure, and behavior patterns in actual operation.

[0037] By calculating the vector similarity of the feature vectors of any two business nodes, a quantitative measure of the similarity of the operational characteristics between nodes can be achieved. Compared with the traditional association identification method based on experience judgment or limited rules, it can evaluate the similarity relationship between business nodes in a higher dimension and more refined feature space, thereby improving the reliability and objectivity of association mining.

[0038] Identifying business nodes with similarity exceeding a threshold as related business nodes and establishing a feature association database can form a structured and queryable network of business node associations. This helps to quickly locate related nodes that may experience linkage anomalies and can be used in various application scenarios such as business process optimization, node replacement suggestions, and risk propagation path analysis, thereby improving the overall analytical capabilities and maintainability of the system.

[0039] Furthermore, step S300 includes:

[0040] Step S301: Obtain the abnormal records of each business process within the training period, set the time point corresponding to the abnormal period in the abnormal record as the key time point, set the time period between the start time point of the abnormal record and the key time point as the feature time period, collect the abnormal business nodes corresponding to each abnormal record, classify the abnormal records according to different abnormal business nodes, and extract the analysis period contained in the feature time period.

[0041] Step S302: Based on the feature index library, collect the feature indicators corresponding to the abnormal business nodes in each analysis period, calculate the business score corresponding to each analysis period, obtain the associated business nodes corresponding to the abnormal business nodes based on the feature association library, extract the business scores of the associated business nodes, and calculate the feature score of the abnormal business nodes according to the following formula:

[0042] ;

[0043] Where A represents the feature score of the abnormal business node, B represents the business score of the abnormal business node, Ba represents the business score of the a-th associated business node, and Ca represents the similarity of the a-th associated business node.

[0044] Step S303: Sort the feature scores corresponding to the analysis period within the feature time period in chronological order, and draw a curve of the feature scores changing over time. Extract the image features of the curve, including slope, amplitude of change, and curvature. Normalize the image features, use the normalized image features as input, and the duration of the feature time period as output. Use a regression model for training to establish an abnormal time prediction model.

[0045] By extracting key time points and characteristic time periods from abnormal records, the important time windows before the occurrence of an anomaly can be clearly defined, so that the pre-operational state of abnormal behavior can be accurately captured. This avoids the problems of arbitrary time range selection and inconsistent window length in traditional anomaly analysis, and helps to improve the data quality and timeliness of anomaly modeling.

[0046] By combining the business scores of abnormal business nodes with the business scores of their related business nodes and weighting the similarity, a feature score that better reflects the upstream and downstream connections and linkage effects of the business is obtained. This quantifies the operation of the abnormal node itself, while also incorporating the impact of related nodes into the evaluation, making the feature representation of the abnormal node more comprehensive and effectively improving the accuracy of anomaly location and root cause analysis.

[0047] By sorting the feature scores within a specific time period by time series and plotting the change curves, the formation, development, and aggravation of abnormal states can be intuitively reflected. By extracting image features such as slope, change amplitude, and curvature, the abnormal process can be dynamically characterized from the time dimension and the change trend dimension. This makes the anomaly prediction process no longer dependent on single-point indicators, but based on the overall trend of the time series, significantly improving the reliability of anomaly monitoring.

[0048] By using normalized image features as input and feature time period duration as output, and training with a regression model, a more stable and predictive anomaly time prediction model can be established. This model can predict the duration or scope of anomalies based on the changing trends of anomaly feature scores, providing the system with early warning capabilities and significantly enhancing the initiative of fault management.

[0049] Furthermore, step S400 includes:

[0050] Step S401: Collect real-time business processes, determine the characteristic business nodes corresponding to the business processes, collect the characteristic indicators corresponding to each characteristic node based on the characteristic indicator library, and calculate the real-time business score of each characteristic node according to the numerical distribution characteristics of the characteristic indicators.

[0051] Step S402: Based on the feature association library, obtain the real-time business score of the associated business node corresponding to each feature node, calculate the real-time feature score of the feature node, draw a real-time curve according to the real-time feature score of each analysis period, extract the real-time image features of the real-time curve, and input them into the anomaly time prediction model to calculate the real-time predicted anomaly time of each feature node. If the real-time predicted anomaly time of a certain feature node is less than or equal to the analysis period duration, the staff is reminded to check the feature node. If the real-time predicted anomaly time of all feature nodes is greater than the analysis period duration, then continue to execute step S401.

[0052] By acquiring real-time business processes and calculating real-time business scores for each characteristic business node, the system can continuously and dynamically quantify the node status during business operations. Compared with traditional methods that rely on offline data or periodic inspections, the real-time scoring mechanism can significantly shorten the delay in anomaly detection and improve the system's real-time perception capabilities.

[0053] Based on the feature association library, real-time business scores of associated business nodes are obtained, and comprehensive feature scores are performed on feature nodes. The evaluation results reflect the linkage between the node's own status and its associated nodes, which can effectively identify the spread trend of anomalies in the business link and improve the accuracy and stability of real-time anomaly identification.

[0054] By plotting real-time curves based on real-time feature scores and extracting real-time image features, it is possible to capture trend information of changes in the status of business nodes, rather than relying solely on data from a single moment. This can sensitively reflect signs of abnormal development, enabling the system to identify trends in changes in business status, thereby significantly improving the ability to detect early signs of anomalies.

[0055] By inputting real-time image features into the anomaly time prediction model, the system can predict the future time of anomalies in real time. When the predicted anomaly time is less than or equal to the analysis cycle length, the system can generate an early warning prompt in advance, realize the pre-detection of potential faults, avoid the spread of anomalies and cause greater system risks, and transform from "post-event processing" to "pre-intervention", significantly improving the system's proactive protection capabilities.

[0056] When the real-time predicted anomaly time of all feature nodes is greater than the analysis cycle duration, the system automatically executes the next monitoring cycle, realizing the automatic loop and continuous operation of the monitoring process, effectively reducing the false alarm rate and improving the reliability of the system's early warning without affecting the real-time monitoring performance.

[0057] Real-time feature scoring and anomaly time prediction results provide operation and maintenance personnel with intuitive and quantitative risk information, enabling them to quickly locate high-risk nodes and conduct targeted inspections. This provides a solid data foundation for intelligent operation and maintenance scheduling, automated processing strategies, and business process optimization, significantly improving the overall operation and maintenance efficiency and intelligence level of the system.

[0058] To better implement the above methods, a business process data analysis system based on distributed access is also proposed. The system includes a feature business node module, a database establishment module, an anomaly time prediction model module, and a real-time analysis module.

[0059] Featured Business Node Module: Based on the historical execution records of the business process platform, analyze historical anomaly records, count the frequency of occurrence of anomaly business nodes, and determine the set of featured business nodes for each business process;

[0060] The library module sets a training cycle, analyzes the execution records, selects the feature indicators of each business node and builds a feature indicator library. Based on the feature indicators of each business node, it analyzes the similarity between business nodes, determines the correlation between each business node and establishes a feature association library.

[0061] Anomaly time prediction model module: acquires anomaly records within the training period, collects feature indicators corresponding to the abnormal business nodes within the analysis period based on the feature indicator library, calculates the business score of the abnormal business nodes, acquires the business scores of the associated business nodes corresponding to the abnormal business nodes based on the feature association library, calculates the feature scores, summarizes the feature scores within each analysis period and plots a curve, extracts the image features of the curve, and establishes an anomaly time prediction model in combination with the duration.

[0062] Real-time analysis module: Collects real-time business processes, identifies the characteristic business nodes corresponding to the real-time business processes, calculates the real-time business score for each characteristic node, and calculates the real-time predicted abnormal time based on the abnormal time prediction model. Based on the real-time predicted abnormal time, it determines whether to check.

[0063] Furthermore, the feature service node module includes an execution record unit and a feature service node determination unit:

[0064] Execution Record Unit: In the business process platform, the unit identifies the accessed business processes, presets a business process library, obtains the business nodes corresponding to each business process, encodes each business node, presets a time period as the analysis cycle, sets the start time of the analysis cycle as the initial time point, and sets the end time of the analysis cycle as the final time point. It obtains the business processes received by the business process platform, determines the set of business nodes corresponding to each business process, and during the execution of the business process, obtains the operational behavior data of the set of business nodes in each process analysis cycle. It extracts the boundary data of the operational behavior data at the initial time point and the final time point, summarizes the boundary data and operational behavior data to obtain analysis data, integrates the analysis data of each process analysis cycle, generates an execution record, and uploads it to the business process platform.

[0065] The feature business node unit is defined as follows: Historical execution records are acquired, and the judgment results of each analysis cycle in the historical execution records are collected. Analysis cycles with abnormal judgment results are designated as abnormal cycles. If an abnormal cycle exists in the historical execution records, the historical execution records are marked as historical abnormal records. The business process corresponding to each historical abnormal record is extracted, and the historical abnormal records are classified according to the business process. In the historical abnormal record set of a certain business process, the abnormal business node corresponding to the abnormal cycle in a certain historical abnormal record is collected. The occurrence frequency of each abnormal business node in the historical abnormal record set is counted, and the occurrence frequency of each abnormal business node is calculated. A preset occurrence frequency threshold is set, and abnormal business nodes exceeding the occurrence frequency threshold are designated as feature business nodes of the business process. A feature business node set for each business process is then generated.

[0066] Furthermore, the anomaly time prediction model module includes a unit for determining characteristic time periods and a unit for establishing the anomaly time prediction model:

[0067] Determine the characteristic time period unit: Obtain the abnormal records of each business process within the training period, set the time point corresponding to the abnormal period in the abnormal record as the key time point, set the time period between the start time point of the abnormal record and the key time point as the characteristic time period, collect the abnormal business node corresponding to each abnormal record, classify the abnormal records according to different abnormal business nodes, and extract the analysis period contained in the characteristic time period.

[0068] An abnormal time prediction model unit is established as follows: Based on the feature index library, feature indicators corresponding to abnormal business nodes within each analysis period are collected, and the business score corresponding to each analysis period is calculated. Based on the feature association library, the associated business nodes corresponding to the abnormal business nodes are obtained, and the business scores of the associated business nodes are extracted. The feature scores of the abnormal business nodes are calculated, and the feature scores corresponding to the analysis periods within the feature time period are sorted in chronological order. A curve graph of the feature scores changing over time is plotted, and the image features of the curve graph are extracted. The image features include slope, amplitude of change, and curvature. The image features are normalized. The normalized image features are used as input, and the duration of the feature time period is used as output. A regression model is used for training to establish an abnormal time prediction model.

[0069] Compared with the prior art, the beneficial effects of the present invention are: based on the frequency of anomalies in business nodes according to historical anomaly records, a set of characteristic business nodes is constructed, so that the analysis focuses on key nodes that are more likely to be abnormal, reducing the scope of invalid analysis and improving the efficiency and accuracy of anomaly detection;

[0070] The system extracts raw indicators from business nodes from multiple dimensions and filters feature indicators through statistical distribution, volatility and anomaly correlation. At the same time, it analyzes the relationship between business nodes through vector similarity and builds a feature association library, enabling the system to expand from single-node detection to multi-node correlation anomaly analysis, significantly improving the depth and accuracy of anomaly diagnosis.

[0071] The time series curve formed by the business node feature score is used as image feature for modeling, and the abnormal time prediction model is trained in combination with the feature time period. This enables the model to not only determine whether there is an abnormal trend, but also to predict when the abnormality may occur, with stronger prediction ability and more advance warning.

[0072] Automated prediction is achieved through real-time feature scoring of feature nodes, scoring of associated nodes, and calculation by prediction models. When the predicted anomaly time is less than the analysis cycle, an inspection request is automatically triggered, effectively reducing the workload of manual monitoring and improving operation and maintenance efficiency. Attached Figure Description

[0073] Figure 1 This is a flowchart illustrating a business process data analysis method based on distributed access according to the present invention.

[0074] Figure 2 This is a schematic diagram of the structure of a business process data analysis system based on distributed access according to the present invention. Detailed Implementation

[0075] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0076] Please see Figure 1 and Figure 2 This invention provides a technical solution: a business process data analysis method based on distributed access, the method comprising:

[0077] Step S100: Based on the historical execution records of the business process platform, analyze the historical anomaly records, count the frequency of occurrence of abnormal business nodes, and determine the set of characteristic business nodes for each business process.

[0078] Step S100 includes:

[0079] Step S101: In the business process platform, the accessed business processes are identified, a business process library is preset, the business nodes corresponding to each business process are obtained, and each business node is coded. A preset time period is used as the analysis period. The start time of the analysis period is set as the initial time point, and the end time of the analysis period is set as the final time point. The business processes received by the business process platform are obtained, and the set of business nodes corresponding to the business processes is determined. During the execution of the business process, the running behavior data of the set of business nodes in each process analysis period is obtained. The boundary data of the running behavior data is extracted at the initial time point and the final time point respectively. The boundary data and the running behavior data are summarized to obtain the analysis data. The analysis data of each process analysis period are integrated to generate an execution record and upload it to the business process platform.

[0080] Step S102: Obtain historical execution records, collect the judgment results of each analysis cycle in the historical execution records, set the analysis cycle with the judgment result as an abnormal cycle, if there is an abnormal cycle in the historical execution records, mark the historical execution records as historical abnormal records, extract the business process corresponding to each historical abnormal record, and classify the historical abnormal records according to the business process;

[0081] Step S103: In the historical anomaly record set of a certain business process, collect the abnormal business nodes corresponding to the abnormal period in a certain historical anomaly record, count the number of times each abnormal business node appears in the historical anomaly record set, calculate the occurrence frequency of each abnormal business node, preset the occurrence frequency threshold, set the abnormal business nodes that exceed the occurrence frequency threshold as the feature business nodes of the business process, and summarize to generate the feature business node set of each business process.

[0082] For example, the business process platform includes: Process F1: Monthly expense reimbursement approval process, Process F2: Budget application process, and Process F3: Supplier payment process;

[0083] Taking process F1: Monthly expense reimbursement approval process as an example:

[0084] Process F1 typically includes the following business nodes: Node N1 is the employee initiating the expense report, Node N2 is the department head approving the expense report, Node N3 is the initial financial review, Node N4 is the financial final review, and Node N5 is the cashier making the payment.

[0085] The preset analysis period is 2 hours. An example of the running behavior data is shown below:

[0086] (Boundary) (Boundary) Data running within the cycle

[0087] During the analysis period from 09:00 to 11:00, the initial data queue length was 12 orders, and the final data queue length was 7 orders. The average data review time during the period was 180 seconds, and the number of errors was 0.

[0088] During the analysis period of 11:00–13:00, the initial data queue length was 7 orders, and the final data queue length was 15 orders. The average data review time during the period was 240 seconds, and the number of errors was 1.

[0089] During the analysis period of 13:00–15:00, the initial data queue length was 15 orders, and the final data queue length was 14 orders. The average data review time during the period was 260 seconds, and the number of errors was 2.

[0090] Step S200: Set the training period, analyze the execution records, select the feature indicators of each business node and build a feature indicator library, analyze the similarity between business nodes based on the feature indicators of each business node, determine the correlation between each business node and establish a feature association library.

[0091] Step S200 includes:

[0092] Step S201: Select several consecutive days as the training period, summarize the execution records of each business process within the training period, and collect the analysis data of a certain execution record in each analysis period from the execution record set of a certain business process.

[0093] Step S202: Based on the analysis data of each analysis cycle, evaluate and extract features from the operational behavior data of each business node, select feature indicators from the candidate indicators, and establish a feature indicator library for each business node.

[0094] Step S202 includes:

[0095] Step S202-1: By statistically analyzing the execution status of business nodes in each analysis period, the original indicators of the business node operation characteristics are extracted and summarized as a candidate indicator set. The original indicators include time-related indicators, load-related indicators, status-related indicators, path-related indicators, and scheduling-related indicators.

[0096] Step S202-2: Collect the values ​​of each original indicator in the candidate indicator set, and calculate the numerical distribution characteristics in each analysis period through statistical feature analysis. The numerical distribution characteristics include the mean, variance, dispersion, trend, fluctuation range, extreme value ratio and periodic stability of the original indicator in the analysis period. Normalize the numerical distribution characteristics, preset the weight of each numerical distribution characteristic, and perform weighted summation on the normalized numerical distribution characteristics to calculate the performance score of the original indicator. Preset the performance score threshold, set the original indicators below the performance score threshold as fluctuation indicators, and assign values ​​to the fluctuation indicators.

[0097] Step S202-3: Obtain the judgment result for each analysis period and assign values ​​to the judgment result. Using the candidate indicator set as input and the judgment result as output, use the Pearson correlation method to calculate the abnormal correlation value of each candidate indicator. Using the candidate indicator set as input and the fluctuation indicator as output, use a binary classification model for training and calculate the performance fluctuation correlation value of each candidate indicator.

[0098] Step S202-4: Normalize the abnormal correlation value and the performance fluctuation correlation value, preset the weights of the abnormal correlation and performance fluctuation correlation, and sum the normalized abnormal correlation value and the performance fluctuation correlation value to calculate the comprehensive score of each original indicator. Preset the comprehensive score threshold, and take the original indicators that exceed the comprehensive score threshold as feature indicators. Summarize the feature indicators of each business node and establish a feature indicator library.

[0099] For example, if a 14-day period is selected as the training period, the company will run a total of 82,300 execution records of process F1 during the training period;

[0100] The following is a partial sample of node N3's data during the training period:

[0101] The average queue length for analysis period C21 was 13, the average processing time was 210, the number of errors was 0, the scheduling wait time was 35, and the number of node state fluctuations was 2.

[0102] The average queue length for analysis period C22 was 18, the average processing time was 260, the number of errors was 1, the scheduling wait time was 60, and the number of node state fluctuations was 3.

[0103] The average queue length for analysis period C23 was 16, the average processing time was 240, the number of errors was 0, the scheduling wait time was 55, and the number of node state fluctuations was 1.

[0104] The average queue length for analysis period C24 was 22, the average processing time was 310, the number of errors was 2, the scheduling wait time was 80, and the number of node state fluctuations was 4.

[0105] The system performs statistical analysis on all training cycle data for node N3 and extracts five categories of raw metrics:

[0106] Time-related metrics: average processing time is 200-350, total cycle processing time is 5000-16000, and scheduling waiting time is 20-120.

[0107] Load-related metrics: average queue length is 8-32, queue growth rate is -10% to +35%;

[0108] Status metrics: Node status transition count is 1-6, error count is 0-4;

[0109] Path-related metrics: The occurrence rate of successor nodes of this node is fixed at 100%, pointing to N4; the proportion of branch paths is 0%.

[0110] Scheduling metrics: 1-5 scheduling attempts, 0-2 scheduling failures;

[0111] After summarizing, there are a total of 12 candidate indicators for node N3;

[0112] The system calculates seven distribution characteristics for each raw indicator: mean, variance, dispersion, trend, fluctuation range, extreme value proportion, and periodic stability.

[0113] For example, consider the "average processing time":

[0114] The mean value is 265, and the normalized value is 0.62;

[0115] The variance is 91, and the normalized value is 0.41.

[0116] The dispersion value is 0.28, and the normalized value is 0.65;

[0117] The trend value is +4.2%, and the normalized value is 0.58;

[0118] The fluctuation range is 30, and the normalized value is 0.47;

[0119] The extreme value ratio is 7.4%, and the normalized value is 0.39;

[0120] The stability value is 0.71, and the normalized value is 0.71.

[0121] Preset weights: mean 0.15, variance 0.15, dispersion 0.15, trend 0.1, fluctuation range 0.15, extreme values ​​0.1, stability 0.2;

[0122] The performance score obtained by weighted summation is 0.559, and the preset performance score threshold is 0.40, which is marked as a feature index.

[0123] Step S203: Based on the characteristic indicators of each business node, analyze the relationship between each business node and establish a characteristic association library for each business node;

[0124] Step S203 includes:

[0125] Step S203-1: Obtain the set of feature indicators corresponding to each business node, extract the numerical distribution characteristics of each feature indicator in different analysis periods, calculate the performance score of the feature indicator, and calculate the business score of the business node by weighted summation based on the performance fluctuation correlation value of the feature indicator. Summarize the business score and the numerical distribution characteristics of each feature indicator to construct the feature vector of each business node.

[0126] Step S203-2: Obtain the feature vectors of any two business nodes. Using the vector similarity method, with the feature vectors of the two business nodes as input, calculate the similarity between the business nodes in the overall feature space. Set a similarity threshold. Set two business nodes that exceed the similarity threshold as associated business nodes. Summarize the associated business nodes corresponding to each business node and establish a feature association library for each business node.

[0127] Step S300: Obtain abnormal records within the training period; based on the feature index library, collect the feature indexes corresponding to the abnormal business nodes within the analysis period; calculate the business score of the abnormal business nodes; based on the feature association library, obtain the business scores of the associated business nodes corresponding to the abnormal business nodes; calculate the feature scores; summarize the feature scores within each analysis period and draw a curve; extract the image features of the curve; and combine the duration to establish an abnormal time prediction model.

[0128] Step S300 includes:

[0129] Step S301: Obtain the abnormal records of each business process within the training period, set the time point corresponding to the abnormal period in the abnormal record as the key time point, set the time period between the start time point of the abnormal record and the key time point as the feature time period, collect the abnormal business nodes corresponding to each abnormal record, classify the abnormal records according to different abnormal business nodes, and extract the analysis period contained in the feature time period.

[0130] Step S302: Based on the feature index library, collect the feature indicators corresponding to the abnormal business nodes in each analysis period, calculate the business score corresponding to each analysis period, obtain the associated business nodes corresponding to the abnormal business nodes based on the feature association library, extract the business scores of the associated business nodes, and calculate the feature score of the abnormal business nodes according to the following formula:

[0131] ;

[0132] Where A represents the feature score of the abnormal business node, B represents the business score of the abnormal business node, Ba represents the business score of the a-th associated business node, and Ca represents the similarity of the a-th associated business node.

[0133] Step S303: Sort the feature scores corresponding to the analysis period within the feature time period in chronological order, and draw a curve of the feature scores changing over time. Extract the image features of the curve, including slope, amplitude of change, and curvature. Normalize the image features, use the normalized image features as input, and the duration of the feature time period as output. Use a regression model for training to establish an abnormal time prediction model.

[0134] Step S400: Collect real-time business processes, determine the characteristic business nodes corresponding to the real-time business processes, calculate the real-time business score of each characteristic node, and calculate the real-time predicted abnormal time based on the abnormal time prediction model. Based on the real-time predicted abnormal time, determine whether to check.

[0135] Step S400 includes:

[0136] Step S401: Collect real-time business processes, determine the characteristic business nodes corresponding to the business processes, collect the characteristic indicators corresponding to each characteristic node based on the characteristic indicator library, and calculate the real-time business score of each characteristic node according to the numerical distribution characteristics of the characteristic indicators.

[0137] Step S402: Based on the feature association library, obtain the real-time business score of the associated business node corresponding to each feature node, calculate the real-time feature score of the feature node, draw a real-time curve according to the real-time feature score of each analysis period, extract the real-time image features of the real-time curve, and input them into the anomaly time prediction model to calculate the real-time predicted anomaly time of each feature node. If the real-time predicted anomaly time of a certain feature node is less than or equal to the analysis period duration, the staff is reminded to check the feature node. If the real-time predicted anomaly time of all feature nodes is greater than the analysis period duration, then continue to execute step S401.

[0138] To better implement the above methods, a business process data analysis system based on distributed access is also proposed. The system includes a feature business node module, a database establishment module, an anomaly time prediction model module, and a real-time analysis module.

[0139] Featured Business Node Module: Based on the historical execution records of the business process platform, analyze historical anomaly records, count the frequency of occurrence of anomaly business nodes, and determine the set of featured business nodes for each business process;

[0140] The feature service node module includes an execution record unit and a feature service node determination unit:

[0141] Execution Record Unit: In the business process platform, the unit identifies the accessed business processes, presets a business process library, obtains the business nodes corresponding to each business process, encodes each business node, presets a time period as the analysis cycle, sets the start time of the analysis cycle as the initial time point, and sets the end time of the analysis cycle as the final time point. It obtains the business processes received by the business process platform, determines the set of business nodes corresponding to each business process, and during the execution of the business process, obtains the operational behavior data of the set of business nodes in each process analysis cycle. It extracts the boundary data of the operational behavior data at the initial time point and the final time point, summarizes the boundary data and operational behavior data to obtain analysis data, integrates the analysis data of each process analysis cycle, generates an execution record, and uploads it to the business process platform.

[0142] The feature business node unit is defined as follows: Historical execution records are acquired, and the judgment results of each analysis cycle in the historical execution records are collected. Analysis cycles with abnormal judgment results are designated as abnormal cycles. If an abnormal cycle exists in the historical execution records, the historical execution records are marked as historical abnormal records. The business process corresponding to each historical abnormal record is extracted, and the historical abnormal records are classified according to the business process. In the historical abnormal record set of a certain business process, the abnormal business node corresponding to the abnormal cycle in a certain historical abnormal record is collected. The occurrence frequency of each abnormal business node in the historical abnormal record set is counted, and the occurrence frequency of each abnormal business node is calculated. A preset occurrence frequency threshold is set, and abnormal business nodes exceeding the occurrence frequency threshold are designated as feature business nodes of the business process. A feature business node set for each business process is then generated.

[0143] The library module sets a training cycle, analyzes the execution records, selects the feature indicators of each business node and builds a feature indicator library. Based on the feature indicators of each business node, it analyzes the similarity between business nodes, determines the correlation between each business node and establishes a feature association library.

[0144] Anomaly time prediction model module: acquires anomaly records within the training period, collects feature indicators corresponding to the abnormal business nodes within the analysis period based on the feature indicator library, calculates the business score of the abnormal business nodes, acquires the business scores of the associated business nodes corresponding to the abnormal business nodes based on the feature association library, calculates the feature scores, summarizes the feature scores within each analysis period and plots a curve, extracts the image features of the curve, and establishes an anomaly time prediction model in combination with the duration.

[0145] The abnormal time prediction model module includes a unit for determining characteristic time periods and a unit for establishing an abnormal time prediction model.

[0146] Determine the characteristic time period unit: Obtain the abnormal records of each business process within the training period, set the time point corresponding to the abnormal period in the abnormal record as the key time point, set the time period between the start time point of the abnormal record and the key time point as the characteristic time period, collect the abnormal business node corresponding to each abnormal record, classify the abnormal records according to different abnormal business nodes, and extract the analysis period contained in the characteristic time period.

[0147] An abnormal time prediction model unit is established as follows: Based on the feature index library, feature indicators corresponding to abnormal business nodes within each analysis period are collected, and the business score corresponding to each analysis period is calculated. Based on the feature association library, the associated business nodes corresponding to the abnormal business nodes are obtained, and the business scores of the associated business nodes are extracted. The feature scores of the abnormal business nodes are calculated, and the feature scores corresponding to the analysis periods within the feature time period are sorted in chronological order. A curve graph of the feature scores changing over time is plotted, and the image features of the curve graph are extracted. The image features include slope, amplitude of change, and curvature. The image features are normalized. The normalized image features are used as input, and the duration of the feature time period is used as output. A regression model is used for training to establish an abnormal time prediction model.

[0148] Real-time analysis module: Collects real-time business processes, identifies the characteristic business nodes corresponding to the real-time business processes, calculates the real-time business score for each characteristic node, and calculates the real-time predicted abnormal time based on the abnormal time prediction model. Based on the real-time predicted abnormal time, it determines whether to check.

[0149] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A business process data analysis method based on distributed access, characterized in that, The methods include: Step S100: Based on the historical execution records of the business process platform, analyze the historical anomaly records, count the frequency of occurrence of abnormal business nodes, and determine the set of characteristic business nodes for each business process. Step S200: Set the training period, analyze the execution records, select the feature indicators of each business node and build a feature indicator library, analyze the similarity between business nodes based on the feature indicators of each business node, determine the correlation between each business node and establish a feature association library. Step S300: Obtain abnormal records within the training period; based on the feature index library, collect the feature indexes corresponding to the abnormal business nodes within the analysis period; calculate the business score of the abnormal business nodes; based on the feature association library, obtain the business scores of the associated business nodes corresponding to the abnormal business nodes; perform weighted calculation on the business scores of the abnormal business nodes and the associated business nodes corresponding to the abnormal business nodes to calculate the feature score; summarize the feature scores within each analysis period and draw a curve; extract the image features of the curve; and combine the duration to establish an abnormal time prediction model. Step S400: Collect real-time business processes, determine the characteristic business nodes corresponding to the real-time business processes, calculate the real-time business score of each characteristic node, and calculate the real-time predicted abnormal time based on the abnormal time prediction model. Based on the real-time predicted abnormal time, determine whether to check. Step S400 includes the following steps: Step S401: Collect real-time business processes, determine the characteristic business nodes corresponding to the business processes, collect the characteristic indicators corresponding to each characteristic node based on the characteristic indicator library, and calculate the real-time business score of each characteristic node according to the numerical distribution characteristics of the characteristic indicators. Step S402: Based on the feature association library, obtain the real-time business score of the associated business node corresponding to each feature node, calculate the real-time feature score of the feature node, draw a real-time curve according to the real-time feature score of each analysis period, extract the real-time image features of the real-time curve, and input them into the anomaly time prediction model to calculate the real-time predicted anomaly time of each feature node. If the real-time predicted anomaly time of a certain feature node is less than or equal to the analysis period duration, the staff is reminded to check the feature node. If the real-time predicted anomaly time of all feature nodes is greater than the analysis period duration, then continue to execute step S401.

2. The business process data analysis method based on distributed access according to claim 1, characterized in that, Step S100 includes the following steps: Step S101: In the business process platform, the accessed business processes are identified, a business process library is preset, the business nodes corresponding to each business process are obtained, and each business node is coded. A preset time period is used as the analysis period. The start time of the analysis period is set as the initial time point, and the end time of the analysis period is set as the final time point. The business processes received by the business process platform are obtained, and the set of business nodes corresponding to the business processes is determined. During the execution of the business process, the running behavior data of the set of business nodes in each process analysis period is obtained. The boundary data of the running behavior data is extracted at the initial time point and the final time point respectively. The boundary data and the running behavior data are summarized to obtain the analysis data. The analysis data of each process analysis period are integrated to generate an execution record and upload it to the business process platform. Step S102: Obtain historical execution records, collect the judgment results of each analysis cycle in the historical execution records, set the analysis cycle with the judgment result as an abnormal cycle, if there is an abnormal cycle in the historical execution records, mark the historical execution records as historical abnormal records, extract the business process corresponding to each historical abnormal record, and classify the historical abnormal records according to the business process; Step S103: In the historical anomaly record set of a certain business process, collect the abnormal business nodes corresponding to the abnormal period in a certain historical anomaly record, count the number of times each abnormal business node appears in the historical anomaly record set, calculate the occurrence frequency of each abnormal business node, preset the occurrence frequency threshold, set the abnormal business nodes that exceed the occurrence frequency threshold as the feature business nodes of the business process, and summarize to generate the feature business node set of each business process.

3. The business process data analysis method based on distributed access according to claim 2, characterized in that, Step S200 includes the following steps: Step S201: Select several consecutive days as the training period, summarize the execution records of each business process within the training period, and collect the analysis data of a certain execution record in each analysis period from the execution record set of a certain business process. Step S202: Based on the analysis data of each analysis cycle, evaluate and extract features from the operational behavior data of each business node, select feature indicators from the candidate indicators, and establish a feature indicator library for each business node. Step S203: Analyze the relationships between each business node based on the characteristic indicators of each business node, and establish a characteristic association library for each business node.

4. The business process data analysis method based on distributed access according to claim 3, characterized in that, The step S202, which establishes a feature index library for each business node, includes the following steps: Step S202-1: By statistically analyzing the execution status of business nodes in each analysis period, the original indicators of the business node operation characteristics are extracted and summarized as a candidate indicator set. The original indicators include time-related indicators, load-related indicators, status-related indicators, path-related indicators, and scheduling-related indicators. Step S202-2: Collect the values ​​of each original indicator in the candidate indicator set, and calculate the numerical distribution characteristics in each analysis period through statistical feature analysis. The numerical distribution characteristics include the mean, variance, dispersion, trend, fluctuation range, extreme value ratio and periodic stability of the original indicator in the analysis period. Normalize the numerical distribution characteristics, preset the weight of each numerical distribution characteristic, and perform weighted summation on the normalized numerical distribution characteristics to calculate the performance score of the original indicator. Preset the performance score threshold, set the original indicators below the performance score threshold as fluctuation indicators, and assign values ​​to the fluctuation indicators. Step S202-3: Obtain the judgment result for each analysis period and assign values ​​to the judgment result. Using the candidate indicator set as input and the judgment result as output, use the Pearson correlation method to calculate the abnormal correlation value of each candidate indicator. Using the candidate indicator set as input and the fluctuation indicator as output, use a binary classification model for training and calculate the performance fluctuation correlation value of each candidate indicator. Step S202-4: Normalize the abnormal correlation values ​​and performance fluctuation correlation values, preset the weights of abnormal correlation and performance fluctuation correlation, and sum the normalized abnormal correlation values ​​and performance fluctuation correlation values ​​to calculate the comprehensive score of each original indicator. Preset the comprehensive score threshold, and take the original indicators that exceed the comprehensive score threshold as feature indicators. Summarize the feature indicators of each business node and establish a feature indicator library.

5. The business process data analysis method based on distributed access according to claim 3, characterized in that, The step S203, which establishes a feature association library for each business node, includes the following steps: Step S203-1: Obtain the set of feature indicators corresponding to each business node, extract the numerical distribution characteristics of each feature indicator in different analysis periods, calculate the performance score of the feature indicator, and calculate the business score of the business node by weighted summation based on the performance fluctuation correlation value of the feature indicator. Summarize the business score and the numerical distribution characteristics of each feature indicator to construct the feature vector of each business node. Step S203-2: Obtain the feature vectors of any two business nodes. Using the vector similarity method, with the feature vectors of the two business nodes as input, calculate the similarity between the business nodes in the overall feature space. Set a similarity threshold. Set two business nodes that exceed the similarity threshold as associated business nodes. Summarize the associated business nodes corresponding to each business node and establish a feature association library for each business node.

6. The business process data analysis method based on distributed access according to claim 1, characterized in that, Step S300 includes the following steps: Step S301: Obtain the abnormal records of each business process within the training period, set the time point corresponding to the abnormal period in the abnormal record as the key time point, set the time period between the start time point of the abnormal record and the key time point as the feature time period, collect the abnormal business nodes corresponding to each abnormal record, classify the abnormal records according to different abnormal business nodes, and extract the analysis period contained in the feature time period. Step S302: Based on the feature index library, collect the feature indicators corresponding to the abnormal business nodes in each analysis period, calculate the business score corresponding to each analysis period, obtain the associated business nodes corresponding to the abnormal business nodes based on the feature association library, extract the business scores of the associated business nodes, and calculate the feature score of the abnormal business nodes according to the following formula: ； Where A represents the feature score of the abnormal business node, B represents the business score of the abnormal business node, Ba represents the business score of the a-th associated business node, and Ca represents the similarity of the a-th associated business node. Step S303: Sort the feature scores corresponding to the analysis period within the feature time period in chronological order, and draw a curve of the feature scores changing over time. Extract the image features of the curve, including slope, amplitude of change, and curvature. Normalize the image features, use the normalized image features as input, and the duration of the feature time period as output. Use a regression model for training to establish an abnormal time prediction model.

7. A business process data analysis system based on distributed access, used to implement the business process data analysis method based on distributed access as described in any one of claims 1-6, characterized in that, The system includes a feature business node module, a database establishment module, an anomaly time prediction model module, and a real-time analysis module. The feature business node module: Based on the historical execution records of the business process platform, it analyzes historical anomaly records, counts the frequency of occurrence of anomaly business nodes, and determines the feature business node set for each business process; The library building module: sets a training period, analyzes the execution records, selects the feature indicators of each business node and builds a feature indicator library, analyzes the similarity between business nodes based on the feature indicators of each business node, determines the correlation between each business node and builds a feature association library. The abnormal time prediction model module: acquires abnormal records within the training period, collects and analyzes the feature indicators corresponding to abnormal business nodes within the analysis period based on the feature indicator library, calculates the business score of the abnormal business node, acquires the business score of the associated business node corresponding to the abnormal business node based on the feature association library, calculates the feature score by weighting the business score of the abnormal business node and the business score of the associated business node corresponding to the abnormal business node, summarizes the feature scores within each analysis period and plots a curve, extracts the image features of the curve, and establishes an abnormal time prediction model by combining the duration. The real-time analysis module collects real-time business processes, identifies the characteristic business nodes corresponding to the real-time business processes, calculates the real-time business score for each characteristic node, calculates the real-time predicted abnormal time based on the abnormal time prediction model, and determines whether to check based on the real-time predicted abnormal time.

8. A business process data analysis system based on distributed access according to claim 7, characterized in that, The feature service node module includes an execution record unit and a feature service node determination unit: The execution record unit: In the business process platform, it identifies the accessed business processes, presets a business process library, obtains the business nodes corresponding to each business process, encodes each business node, presets a time period as the analysis cycle, sets the start time of the analysis cycle as the initial time point, sets the end time of the analysis cycle as the final time point, obtains the business processes received by the business process platform, determines the set of business nodes corresponding to the business processes, obtains the running behavior data of the set of business nodes in each process analysis cycle during the execution of the business process, extracts the boundary data of the running behavior data at the initial time point and the final time point respectively, summarizes the boundary data and running behavior data to obtain analysis data, integrates the analysis data of each process analysis cycle, generates an execution record and uploads it to the business process platform; The feature business node determination unit: acquires historical execution records, collects the judgment results of each analysis cycle in the historical execution records, sets the analysis cycles with abnormal judgment results as abnormal cycles, marks the historical execution records as historical abnormal records if there are abnormal cycles in the historical execution records, extracts the business process corresponding to each historical abnormal record, classifies the historical abnormal records according to the business process, collects the abnormal business nodes corresponding to the abnormal cycles in a certain historical abnormal record in the historical abnormal record set of a certain business process, counts the number of times each abnormal business node appears in the historical abnormal record set, calculates the occurrence frequency of each abnormal business node, presets an occurrence frequency threshold, sets abnormal business nodes that exceed the occurrence frequency threshold as feature business nodes of the business process, and summarizes and generates a feature business node set for each business process.

9. A business process data analysis system based on distributed access according to claim 7, characterized in that, The abnormal time prediction model module includes a unit for determining characteristic time periods and a unit for establishing an abnormal time prediction model: The defined feature time period unit: acquires the abnormal records of each business process within the training period, sets the time point corresponding to the abnormal period in the abnormal record as the key time point, sets the time period between the start time point of the abnormal record and the key time point as the feature time period, collects the abnormal business nodes corresponding to each abnormal record, classifies the abnormal records according to different abnormal business nodes, and extracts the analysis period contained in the feature time period. The abnormal time prediction model unit: Based on the feature index library, it collects the feature indicators corresponding to the abnormal business nodes in each analysis period, calculates the business score corresponding to each analysis period, obtains the associated business nodes corresponding to the abnormal business nodes based on the feature association library, extracts the business scores of the associated business nodes, calculates the feature scores of the abnormal business nodes, sorts the feature scores corresponding to the analysis periods in the feature time period according to the time sequence, and draws a curve of the feature scores changing with time. It extracts the image features of the curve, including slope, change amplitude, and curvature, and performs normalization calculation on the image features. Using the normalized image features as input and the duration of the feature time period as output, it uses a regression model for training to establish an abnormal time prediction model.

Images