Communication data fusion analysis method for abnormal fluctuation of enterprise on-duty personnel in chemical industry park

Abstract

The embodiment of the invention provides a communication data fusion analysis method for abnormal fluctuation of enterprise on-duty personnel in a chemical industry park, which relates to the technical field of communication data analysis, and comprises the following steps: collecting a plurality of signaling data streams of a plurality of distributed communication base stations in real time by a plurality of edge stream processing engines in a dynamic geofence range, performing privacy-enhanced real-time hash matching based on the enterprise employee white list library, and generating a plurality of employee on-duty state vector sets; after the cloud analysis center receives a plurality of employee on-duty state vector sets, time sequence dynamic aggregation is carried out based on enterprise affiliation, M enterprise-level on-duty personnel time sequence maps corresponding to M park enterprises are constructed to execute on-duty personnel level abnormal fluctuation analysis, M on-duty abnormal risk factors are output, and the M on-duty abnormal risk factors are analyzed; the technical problems that in the prior art, counting precision of on-duty personnel of an enterprise is low, and safe production management decisions cannot be supported are solved. The effects of radically treating data misjudgment, accurately mapping the human fluctuation rule and providing a reliable decision basis for safe production management of the park are achieved.

Description

Technical Field

[0001] This invention relates to the field of communication data analysis technology, and in particular to a method for communication data fusion analysis of abnormal fluctuations in on-duty personnel in chemical industrial park enterprises. Background Technology

[0002] Existing technologies, when using a single enterprise as the smallest monitoring unit, inevitably extend the signal coverage of the base station beyond the enterprise's physical boundaries. This forces irrelevant signals such as nearby mobile personnel and employees of neighboring enterprises into the statistical scope. Furthermore, the raw signaling data lacks an effective identity filtering mechanism, resulting in a large number of non-target personnel being mistakenly identified as on-duty personnel.

[0003] Meanwhile, the spatial correspondence between base stations and enterprises is chaotic with many-to-many mapping. When the same base station serves multiple enterprises or a single enterprise is covered by multiple overlapping base stations, the signal attribution cannot be accurately decoupled, resulting in uncontrollable deviations in personnel location determination.

[0004] This deficiency in underlying data collection further leads to output results that seriously deviate from the actual business situation. The number of personnel frequently shows orders of magnitude errors and fluctuations that completely lose their regularity, resulting in the monitoring data being unusable for a long time. It cannot reflect the real production regularity characteristics such as shift handover and job rotation, nor does it have the ability to identify key scenarios such as overloaded production or abnormal shutdowns.

[0005] In summary, the existing technology suffers from the problem that the accuracy of counting on-duty personnel in enterprises is low because a single base station covers multiple enterprises. The obtained monitoring data is of poor reference value for analyzing the actual fluctuation patterns of human resources in enterprises, and cannot support safety production management decisions.

[0006] It should be noted that the information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0007] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a communication data fusion analysis method for abnormal fluctuations in on-duty personnel in chemical industrial parks. This method solves the technical problems of existing technologies where a single base station covering multiple enterprises results in low accuracy in counting on-duty personnel, leading to poor reference value for analyzing the actual fluctuation patterns of on-duty personnel and hindering support for safety production management decisions. It achieves the effect of eradicating data misjudgment, consistently matching the actual on-duty scale over the long term, and accurately mapping the fluctuation patterns of on-duty personnel, thus providing a reliable basis for decision-making in the park's safety production management. The specific technical solution is as follows:

[0008] This invention provides a method for fusion analysis of communication data related to abnormal fluctuations in on-duty personnel in chemical industrial park enterprises. The method includes:

[0009] A dynamic geofence is constructed by extending the geographical boundary of the target park by a preset outward distance. Multiple edge stream processing engines are configured for multiple distributed communication base stations within the dynamic geofence area. A communication monitoring network is constructed by establishing bidirectional communication connections between the multiple edge stream processing engines and the cloud analysis center using a low-latency message queue. The multiple edge stream processing engines collect multiple signaling data streams from the multiple distributed communication base stations in real time, perform privacy-enhanced real-time hash matching based on an enterprise employee whitelist database, and generate multiple employee on-duty status vector sets. After receiving the multiple employee on-duty status vector sets, the cloud analysis center performs time-series dynamic aggregation based on enterprise affiliation to construct M enterprise-level on-duty personnel time-series maps corresponding to M park enterprises. Based on the M enterprise-level on-duty personnel time-series maps, an on-duty personnel hierarchical anomaly fluctuation analysis is performed, outputting M on-duty anomaly risk factors. The hierarchical anomaly fluctuation analysis covers enterprise-level fluctuation analysis, job-level risk analysis, and employee-level behavior analysis.

[0010] In one implementation, the multiple edge stream processing engines collect multiple signaling data streams from the multiple distributed communication base stations in real time, perform privacy-enhancing real-time hash matching based on the enterprise employee whitelist database, generate multiple employee on-duty status vector sets, and also perform the following processing:

[0011] The first edge stream processing engine continuously monitors the bound first distributed communication base station and captures raw signaling data packets to obtain a first signaling data stream: S1: Extract protocol parsing fields from multiple time-series signaling data packets in the first signaling data stream to obtain multiple raw signaling tuples, wherein each signaling tuple includes an encrypted device identifier, a signaling timestamp, and a signaling event type; S2: Add Laplace noise to the multiple encrypted device identifiers of the multiple raw signaling tuples to perform differential privacy protection, resulting in multiple anonymous signaling tuples; S3: Use the multiple anonymous device identifiers of the multiple anonymous signaling tuples to traverse and compare with the enterprise employee whitelist database to perform local sensitive hash comparison, so as to match and filter P anonymous employee IDs and P enterprise IDs of P whitelisted employees; S4: Construct P employee on-duty status vectors based on the P anonymous signaling tuples, P anonymous employee IDs, and P enterprise IDs of the P whitelisted employees, forming a first employee on-duty status vector set.

[0012] In one implementation, P employee on-duty status vectors are constructed based on P anonymous signaling tuples, P anonymous employee IDs, and P enterprise IDs of the P whitelisted employees, forming a first employee on-duty status vector set. The following processing is then performed:

[0013] Extract P signaling timestamps and P signaling event types from P signaling tuples; map P employee action types according to the P signaling event types; use the P signaling timestamps as the starting point reference, perform downward rounding and alignment with a preset granularity to generate P time windows; construct the P employee on-duty status vectors by structurally combining the P anonymous employee IDs, P enterprise IDs, P employee action types, and P time windows.

[0014] In one implementation, after receiving the multiple sets of employee on-duty status vectors, the cloud-based analysis center performs time-series dynamic aggregation based on enterprise affiliation to construct M enterprise-level on-duty personnel time-series maps corresponding to M park enterprises, and further performs the following processing:

[0015] Using the enterprise ID as the basis for enterprise affiliation, the enterprise dimension is grouped into multiple employee on-duty status vector sets to obtain M enterprise vector sets corresponding to the M park enterprises; cross-window status transfer is performed on the M enterprise vector sets to construct M time-series window vector sequences; incremental updates of the M historical on-duty personnel time-series graphs are performed using the M time-series window vector sequences to output the M enterprise-level on-duty personnel time-series graphs.

[0016] In one implementation, cross-window state transfer is performed on the M enterprise vector sets to construct M time-series window vector sequences, and the following processing is also performed:

[0017] The first enterprise vector set is reconstructed based on ascending alignment of time windows to obtain a time window state sequence; the cumulative number of employees on duty is initialized according to the first time window state in the time window state sequence to obtain a first time window vector; with the first time window vector as the initial state, the cross-window state is passed through the time window state sequence by dynamically accumulating the number of employees on duty, and the first time window vector sequence is output, wherein the first time window vector includes the time window interval and the net number of employees on duty.

[0018] In one implementation, based on the time series graphs of the M enterprise-level on-the-job personnel, an abnormal fluctuation analysis of the on-the-job personnel hierarchy is performed, outputting M on-the-job abnormality risk factors, and the following processing is also performed:

[0019] The following steps are performed: First, calculate the moving average of the M historical on-the-job personnel time-series graphs over a preset production cycle to construct M enterprise-level on-the-job baseline curves. Second, construct M enterprise-level on-the-job incremental curves based on the M enterprise-level on-the-job personnel time-series graphs. Third, align the M enterprise-level on-the-job baseline curves and M enterprise-level on-the-job incremental curves according to the preset production cycle, and perform window-level on-the-job deviation calculations to output M enterprise-level on-the-job personnel deviations. Fourth, decompose the M enterprise-level on-the-job personnel time-series graphs based on anonymous employee IDs, perform fine-grained deviation hierarchical analysis, and output M job-level vacancy ratio deviations and M employee-level behavioral time-series deviations. Fifth, integrate the M enterprise-level on-the-job personnel deviations, M job-level vacancy ratio deviations, and M employee-level behavioral time-series deviations to output the M on-the-job abnormality risk factors.

[0020] In one implementation, the following processing is also performed:

[0021] Based on anonymous employee IDs, the time series graph of on-duty personnel at the first enterprise level is decomposed into employee dimensions to generate multiple employee-level behavioral time series graphs corresponding to multiple employees of the first enterprise; multiple behavioral time series baselines are constructed based on the shift scheduling information generated by the first enterprise; multiple behavioral time series graphs and multiple behavioral time series baselines are compared to obtain multiple behavioral deviation value sequences; based on the historical violation frequency of the multiple employees of the first enterprise, the multiple behavioral deviation value sequences are weighted and fused to serve as the first employee-level behavioral time series deviation.

[0022] In one implementation, the following processing is also performed:

[0023] Based on the job title and grade, aggregate the multiple employee-level behavioral time-series graphs to output the job title and grade on-duty personnel time-series graph; access the job proportion time-series requirements and traverse the job title and grade on-duty personnel time-series graphs to output the time-series vacancy ratio; classify the time-series vacancy ratio by risk and output the time-series risk level; based on the preset high-risk level threshold, traverse the time-series risk levels to perform high-risk level percentage statistics and output the first job title and grade vacancy ratio deviation.

[0024] Beneficial effects of the embodiments of the present invention:

[0025] In the solution provided by this invention, a dynamic geofence is constructed by extending the geographical boundary of the target park by a preset outward distance to ensure coverage of commuting routes and related areas. Within the fenced area, an edge stream processing engine is configured for distributed communication base stations, and a bidirectional communication connection is established between the engine and the cloud analysis center through a low-latency message queue, forming a distributed and collaborative communication monitoring network. The edge stream processing engine collects base station signaling data streams in real time, performs privacy-enhanced real-time hash matching based on the enterprise employee whitelist database, and generates a vector set of employee on-duty status. After receiving the vector set, the cloud analysis center performs time-series dynamic aggregation based on the enterprise ID as the attribution basis to construct M enterprise-level on-duty personnel time-series maps. Based on these time-series maps, hierarchical abnormal fluctuation analysis is performed: an on-duty baseline curve is constructed by using a moving average and aligned with the real-time incremental curve to calculate the enterprise-level on-duty personnel deviation. At the same time, the deviation of the position-level staffing ratio is statistically analyzed and the employee-level behavioral time-series deviation is fused based on the anonymous employee ID decomposition map. Finally, the three-layer deviation is fused to output M on-duty abnormal risk factors, realizing a full-dimensional quantitative early warning of the overall enterprise staffing scale, the continuous shortage of key positions, and employee behavioral deviations. This approach effectively eliminates data misjudgment, maintains a stable long-term match with actual on-duty personnel, and accurately maps human resource fluctuation patterns, providing a reliable basis for decision-making in park safety production management. Of course, implementing any product or method of this invention does not necessarily require achieving all of the advantages described above simultaneously. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 The diagram illustrates the flowchart of the communication data fusion analysis method for abnormal fluctuations in on-duty personnel in chemical industrial park enterprises provided by the present invention.

[0028] Figure 2 This paper illustrates a flowchart of the process for constructing an employee on-duty status vector set in the communication data fusion analysis method for abnormal fluctuations in on-duty personnel in chemical industrial park enterprises provided by the present invention. Detailed Implementation

[0029] To facilitate understanding of the present invention, a more complete description of the invention will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein; rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the invention.

[0030] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0031] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0032] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0033] The communication data fusion analysis method for abnormal fluctuations in on-duty personnel in chemical industrial park enterprises provided by this invention is used to solve the technical problem in the prior art where the accuracy of on-duty personnel counting is low due to a single base station covering multiple enterprises, and the obtained monitoring data has poor reference value for analyzing the actual fluctuation pattern of human resources in enterprises, thus failing to support safety production management decisions.

[0034] Example: See Figure 1 The flowchart of the communication data fusion analysis method for abnormal fluctuations in on-duty personnel in chemical industrial park enterprises provided in this embodiment of the invention includes:

[0035] S100: Extends the geographical boundary of the target park by a preset outward extension distance to construct a dynamic geofence.

[0036] Specifically, in this embodiment, the preset outward extension distance refers to a pre-set distance value, usually 500 meters to 1 kilometer, used to extend the original geographical boundary of the target park outward. By extending the original boundary of the park through the outward extension distance, a dynamic electronic fence is formed to ensure coverage of employee commuting routes and surrounding related areas, such as park entrance roads and adjacent parking lots, thereby solving the monitoring blind spot problem caused by incomplete base station signal coverage and providing a complete spatial reference for subsequent personnel on-duty monitoring.

[0037] S200: Configure multiple edge stream processing engines for multiple distributed communication base stations within the scope of the dynamic geofence.

[0038] S300: Employs low-latency message queues for bidirectional communication between the multiple edge stream processing engines and the cloud analytics hub, thereby constructing a communication monitoring network.

[0039] In this embodiment, the distributed communication base station refers to the discrete signal transceiver node deployed in the physical space of the park, which is responsible for capturing mobile device signaling, and the edge stream processing engine refers to the lightweight real-time computing module deployed on the base station side, which has the ability to process streaming data in real time.

[0040] An edge stream processing engine is independently configured on all distributed base stations covered by dynamic geofencing, enabling each base station to process raw signaling locally in real time. This avoids data transmission delays in traditional centralized processing models and provides an edge computing foundation for subsequent privacy-enhancing matching.

[0041] The cloud-based analytics hub is a pre-built centralized data analysis platform. In this embodiment, a bidirectional data transmission channel is established between the edge stream processing engine and the cloud-based analytics hub through a low-latency message queue. The edge side uploads the processed on-duty status vector in real time, and the cloud side dynamically distributes configuration policies, forming a distributed and collaborative communication and monitoring network to ensure data real-time performance and system scalability.

[0042] For example, low-latency message queues serve as high-throughput communication middleware based on a publish-subscribe model, preferably Kafka or Pulsar.

[0043] S400: The multiple edge stream processing engines collect multiple signaling data streams from the multiple distributed communication base stations in real time, perform privacy-enhancing real-time hash matching based on the enterprise employee whitelist database, and generate multiple employee on-duty status vector sets.

[0044] In one implementation, the multiple edge stream processing engines collect multiple signaling data streams from the multiple distributed communication base stations in real time, perform privacy-enhancing real-time hash matching based on the enterprise employee whitelist database, and generate multiple employee on-duty status vector sets. Step S400 may further include:

[0045] The first edge stream processing engine continuously monitors the bound first distributed communication base station, captures raw signaling data packets, and obtains the first signaling data stream:

[0046] S1: Extract protocol parsing fields from multiple time-series signaling data packets in the first signaling data stream to obtain multiple original signaling tuples, wherein each signaling tuple includes an encryption device identifier, a signaling timestamp, and a signaling event type.

[0047] S2: Add Laplace noise to the multiple encrypted device identifiers of the multiple original signaling tuples to perform differential privacy protection, thereby obtaining multiple anonymous signaling tuples.

[0048] S3: Use the multiple anonymous device identifiers of the multiple anonymous signaling tuples to perform local sensitive hash comparison with the enterprise employee whitelist database to match and filter the P anonymous employee IDs and P enterprise IDs of the P whitelist employees.

[0049] S4: Based on the P anonymous signaling tuples, P anonymous employee IDs, and P enterprise IDs of the P whitelisted employees, construct P employee on-duty status vectors to form the first employee on-duty status vector set.

[0050] In one implementation, see Figure 2 Based on the P anonymous signaling tuples, P anonymous employee IDs, and P enterprise IDs of the P whitelisted employees, P employee on-duty status vectors are constructed to form the first employee on-duty status vector set. Step S4 may further include:

[0051] S4-1: Extract P signaling timestamps and P signaling event types from P signaling tuples.

[0052] S4-2: Map P employee action types according to the P signaling event types.

[0053] S4-3: Using the P signaling timestamps as the starting point, perform downward rounding and alignment with a preset granularity to generate P time windows.

[0054] S4-4: Construct the P employee on-duty status vectors by structurally combining the P anonymous employee IDs, P enterprise IDs, P employee action types, and P time windows.

[0055] In this embodiment, multiple edge stream processing engines synchronously collect real-time signaling data streams from the bound base stations, filter out employee signaling belonging to the enterprise's whitelist through privacy protection technology, and finally generate a structured on-duty status vector set to provide standardized input for aggregation and analysis by the cloud analysis center.

[0056] Each edge engine independently monitors its assigned base station, captures unprocessed raw signaling data packets in real time, and forms a continuous data stream to ensure the timeliness and completeness of signaling acquisition. Since the processing logic of each edge stream processing engine is consistent, this embodiment takes the first edge stream processing engine as an example to elaborate on the technical solution in detail.

[0057] The first edge stream processing engine continuously and independently monitors the first distributed communication base station it is bound to, captures unprocessed raw signaling data packets in real time and forms a continuous data stream to obtain the first signaling data stream, ensuring the timeliness and completeness of signaling collection.

[0058] Protocol parsing field extraction refers to stripping key fields from the communication protocol of the original signaling data packet. The original signaling tuple is a structured data unit containing an encrypted device identifier, a signaling timestamp, and a signaling event type. The encrypted device identifier is an encrypted value of the International Mobile Equipment Identity (IMEI) or the Mobile Subscriber Identity (MSISDN), used to uniquely identify the user device but requiring privacy protection. The signaling timestamp is the precise time point of the signaling event, in Unix or ISO 8601 format, used to locate the absolute time reference of employee actions. The signaling event type includes device connection events and device disconnection events, used to map employee entry or exit behavior states.

[0059] In this embodiment, multiple time-series signaling data packets in the first signaling data stream are parsed packet by packet to extract fields and obtain multiple original signaling tuples.

[0060] Laplace noise refers to a mathematical noise model that satisfies differential privacy requirements. Privacy protection is achieved by injecting random perturbations into the encrypted device identifier. In this embodiment, Laplace noise is added to the encrypted device identifier of each original signaling tuple to generate an anonymous device identifier, making the device impossible to be directly tracked. This generates an anonymous signaling tuple with privacy protection, thus solving the risk of employee identity leakage.

[0061] The pre-built enterprise employee whitelist database stores differentially privatized anonymous employee IDs and enterprise IDs for all employees in the park. The enterprise IDs can be decoded to obtain the specific job information of the employees. The object of the local sensitive hash comparison of the enterprise employee whitelist database is the hash value of the anonymous employee ID in the database, which is traversed and compared with the multiple anonymous device identifiers of the multiple anonymous signaling tuples. If the match is successful, the corresponding anonymous employee ID and enterprise ID are extracted. Finally, P whitelist employees belonging to the park's enterprise employees are selected, and their corresponding P anonymous employee IDs and P enterprise IDs are extracted to achieve efficient identity matching.

[0062] P signaling timestamps and P signaling event types are extracted from P signaling tuples to provide basic data for action mapping and time alignment. The P signaling event types are mapped and converted into employee behavior semantics to obtain P employee action types, such as mapping a device connection event to an "entry" action and a disconnection event to an "exit" action, thereby realizing business logic abstraction.

[0063] The preset granularity refers to the set time segment unit, and rounding down refers to truncating the timestamp to the nearest time window start point. In this embodiment, the P signaling timestamps are used as the starting point reference, and the preset fixed time granularity is used for rounding down to generate standardized P time window intervals to provide a unified timing framework for state aggregation.

[0064] The four types of fields, namely P anonymous employee IDs, P enterprise IDs, P employee action types, and P time windows, are combined into a single state vector to form a standardized data unit integrating "employee-enterprise-action-time". The P employee on-duty state vectors are constructed to form the first employee on-duty state vector set.

[0065] Similarly, the multiple edge stream processing engines collect multiple signaling data streams from the multiple distributed communication base stations in real time, perform privacy-enhancing real-time hash matching based on the enterprise employee whitelist database, and generate multiple employee on-duty status vector sets.

[0066] This embodiment achieves the technical effect of accurately identifying the identity of on-duty employees and ensuring privacy and security, providing standardized and traceable input data for subsequent enterprise-level time-series aggregation and job-level anomaly analysis.

[0067] S500: After receiving the multiple employee on-duty status vector sets, the cloud-based analysis center performs time-series dynamic aggregation based on enterprise affiliation to construct M enterprise-level on-duty personnel time-series maps corresponding to M park enterprises.

[0068] In one implementation, after receiving the multiple employee on-duty status vector sets, the cloud-based analysis center performs time-series dynamic aggregation based on enterprise affiliation to construct M enterprise-level on-duty personnel time-series maps corresponding to M park enterprises. Step S500 may further include:

[0069] S510: Using the enterprise ID as the basis for enterprise affiliation, group the multiple employee on-duty status vector sets by enterprise dimension to obtain M enterprise vector sets corresponding to the M park enterprises.

[0070] S520: Perform cross-window state transfer on the M enterprise vector sets to construct M time-series window vector sequences.

[0071] S530: Use the M time-series window vector sequences to incrementally update the time-series graphs of M historical on-the-job personnel, and output the M enterprise-level on-the-job personnel time-series graphs.

[0072] In one implementation, cross-window state transfer is performed on the M enterprise vector sets to construct M time-series window vector sequences. Step S520 may further include:

[0073] S521: Reconstruct the first enterprise vector set based on ascending alignment of the time window to obtain the time window state sequence.

[0074] S522: Initialize the cumulative number of employees based on the first time window state in the time window state sequence to obtain the first time window vector.

[0075] S523: Using the first time-series window vector as the initial state, the time-series window state sequence is traversed by dynamically accumulating the number of employees on duty to perform cross-window state transfer, and the first time-series window vector sequence is output. The first time-series window vector includes the time window interval and the net number of employees on duty.

[0076] Specifically, after receiving the multiple employee on-duty status vector sets, the cloud-based analysis center uses the enterprise ID as the basis for enterprise affiliation and groups and aggregates all employee status vectors according to the enterprise ID value. For example, all entry / exit event vectors corresponding to a certain enterprise ID are grouped into the same set, forming a discrete status dataset for that enterprise. Ultimately, this results in M ​​enterprise vector sets corresponding to the M park enterprises, achieving data isolation at the enterprise level and laying the foundation for subsequent time series construction.

[0077] Ascending alignment of time windows is a prerequisite for time series processing. The first enterprise vector set is sorted and reorganized according to the time window value from smallest to largest. For example, window events such as 08:00, 08:05, and 08:10 are strictly arranged in order to form a time-dimensional regular time series window state sequence, so as to eliminate the problem of disordered time in the original data and provide a correct time series basis for cumulative calculation.

[0078] Select all events in the first time window of the time series to calculate the initial number of people on duty. For example, if 10 people enter and 2 people leave within the first window from 08:00 to 08:05, the initial net number of people on duty is 8, providing an accurate starting point for subsequent dynamic accumulation.

[0079] Using the first time-series window vector as the initial state, and taking the net number of employees in the previous window as a benchmark, the changes in events in the current window are added. For example, if there are 8 employees in window T1 and 3 new employees enter and 1 employee leaves in window T2, then there are 10 employees in window T2. By dynamically accumulating the number of employees on duty, the time-series window state sequence is traversed to perform cross-window state transfer, generating a complete vector sequence containing the time window interval and the net number of employees on duty, and outputting the first time-series window vector sequence.

[0080] The newly generated M time-series window vector sequences are fused with M historical on-duty personnel time-series graphs using a sliding window method. Only the data for the changing time periods are updated, and the M enterprise-level on-duty personnel time-series graphs are output to ensure that the time-series graphs reflect the latest trends in real time, while reducing the cloud computing load.

[0081] This embodiment achieves the technical effect of eliminating external interference and generating continuous and stable enterprise-level time series graphs, providing a high-precision, low-latency trend benchmark for subsequent multi-level abnormal fluctuation analysis.

[0082] S600: Based on the M enterprise-level on-the-job personnel time series maps, perform on-the-job personnel hierarchical abnormal fluctuation analysis and output M on-the-job abnormal risk factors. The hierarchical abnormal fluctuation analysis covers enterprise-level fluctuation analysis, job-level risk analysis and employee-level behavior analysis.

[0083] In one implementation, based on the M enterprise-level on-the-job personnel time series maps, an abnormal fluctuation analysis of on-the-job personnel hierarchy is performed, and M on-the-job abnormality risk factors are output. Step S600 may further include:

[0084] S610: Calculate the moving average of the M historical on-the-job personnel time series graphs for a preset production cycle to construct M enterprise on-the-job baseline curves.

[0085] S620: Construct M enterprise on-the-job incremental curves based on the time series maps of the M enterprise-level on-the-job personnel.

[0086] S630: After aligning the on-the-job baseline curves and the on-the-job incremental curves of the M enterprises according to the preset production cycle, perform window-level on-the-job deviation calculation and output the on-the-job deviation of the M enterprises.

[0087] S640: Based on the anonymous employee ID, decompose the time series graphs of the M enterprise-level on-the-job personnel, perform fine-grained deviation hierarchical analysis, and output the M job-level vacancy ratio deviations and M employee-level behavioral time series deviations.

[0088] S650: Integrate the M enterprise-level attendance deviations, M job-level vacancy ratio deviations, and M employee-level behavioral timing deviations to output the M attendance anomaly risk factors.

[0089] One implementation also includes:

[0090] S641: Based on the anonymous employee ID, the time series graph of the on-the-job personnel of the first enterprise is decomposed into employee dimensions to generate multiple employee-level behavioral time series graphs corresponding to multiple employees of the first enterprise.

[0091] S642: Construct multiple behavioral timeline baselines for the employees of the multiple first enterprises based on the scheduling information generated by the first enterprise.

[0092] S643: By comparing the multiple employee-level behavioral time series maps and multiple behavioral time series baselines, multiple behavioral deviation value sequences are obtained.

[0093] S644: Based on the historical violation frequency of the multiple first enterprise employees, perform weighted fusion of the multiple behavioral deviation value sequences to obtain the first employee-level behavioral time sequence deviation.

[0094] One implementation also includes:

[0095] S645: Aggregate the multiple employee-level behavioral time series graphs based on job level and output the job-level on-the-job personnel time series graph.

[0096] S646: Access the job position ratio time sequence requirement, traverse the time sequence graph of the on-the-job personnel at the job level, and output the time sequence vacancy ratio.

[0097] S647: Perform risk classification on the aforementioned time-series staff shortage ratio and output the time-series risk level.

[0098] S648: Based on the preset high-risk level threshold, traverse the time-series risk levels to perform high-risk level proportion statistics and output the first position-level vacancy ratio deviation.

[0099] In this embodiment, the cloud-based analysis center performs multi-level anomaly detection on the enterprise-level time series graph, covering three dimensions: overall enterprise fluctuations, job risk distribution, and individual employee behavior. By comprehensively calculating enterprise-level deviations, job vacancy rates, and employee behavior deviations, it generates quantitative risk factors, achieving comprehensive risk warnings from macro to micro levels and providing decision-making basis for park safety management.

[0100] Specifically, this embodiment uses the M historical on-duty personnel time series data as continuous on-duty personnel time series data within historical periods. Based on the enterprise's production cycle, the average historical on-duty personnel is calculated on a rolling basis to form a smoothed baseline curve, thus constructing M enterprise on-duty baseline curves. For example, a manufacturing enterprise calculates the average on-duty personnel during peak hours each day using a 7-day cycle as a stability reference benchmark, effectively filtering out the impact of occasional events and establishing a reliable trend baseline.

[0101] Based on the time-series graphs of the M enterprise-level on-duty personnel, M enterprise on-duty increment curves are constructed. These curves capture instantaneous fluctuations such as sudden staff shortages or abnormal clustering, providing a real-time data source for deviation calculation. For example, the real-time monitoring value of the number of on-duty personnel in a company with 160 employees forms the vertical axis of the curve, and the horizontal axis is a continuous time window, capturing instantaneous fluctuations.

[0102] To ensure window-level alignment and comparability, the time axes of the incremental on-duty curves for the M enterprises are truncated and aligned according to the period of the baseline on-duty curves for the M enterprises. Then, the difference between the actual value and the baseline value is calculated window by window, outputting the on-duty deviation for the M enterprises. The on-duty deviation for the M enterprises is a time-series record of on-duty personnel deviation data. For example, if a chemical company's baseline for the morning shift is 95 people and the real-time value is 80 people, the deviation is -15 people. This quantifies the scale of on-duty personnel at the enterprise level and forms the first-level risk indicator.

[0103] The cloud-based analytics hub breaks down the enterprise-level time-series graph layer by layer using anonymous employee IDs. First, it separates the behavioral time-series chain of each individual employee based on a unique anonymous identifier, forming a micro-behavioral trajectory. Then, it aggregates the behavioral data of similar employees by job position and level, generating a meso-level job position-level human resource distribution curve.

[0104] Based on this dual decomposition, the real-time vacancy rate of positions and the deviation of employee behavior compliance are calculated in parallel, realizing fine-grained risk quantification from group manpower shortage to individual behavioral deviations, providing enterprises with accurate two-dimensional risk profiles down to positions and employees, and supporting targeted management intervention.

[0105] Specifically, using anonymous employee IDs as index keys, the time-series graph of on-duty personnel at the enterprise level is broken down into independent individual behavior sequences at the employee level, generating multiple employee-level behavior time-series graphs corresponding to multiple employees of the enterprise. For example, the timestamps and action sequences corresponding to anonymous employee ID 007 {"08:03: Enter", "12:00: Leave", "13:30: Enter", "17:05: Leave"} form a unique behavior graph, which fully records the type of action, the time of occurrence, and the duration of the individual employee's actions within the monitoring period, providing atomic-level data units for behavioral compliance analysis.

[0106] Based on the scheduling information generated by the first company, multiple behavioral timeline baselines are constructed for the employees of the multiple first companies. For example, if employee ID007's schedule is "08:00-12:00, 13:30-17:00 on duty", then the baseline sequence {"08:00: Enter", "12:00: Leave", "13:30: Enter", "17:00: Leave"} is generated. This baseline strictly defines the expected time points and types of employee actions as an objective standard for judging behavioral compliance.

[0107] The spatiotemporal deviation of the multiple employee-level behavioral time series maps and multiple behavioral time series baselines is calculated by event-by-event mapping. For each action record, the absolute difference between it and the baseline timestamp is extracted to obtain multiple behavioral deviation value sequences.

[0108] For example, if the actual entry time is 08:03 and the baseline is 08:00, a lateness deviation value of 180 seconds is generated; if the actual departure time is 17:05 and the baseline is 17:00, an early departure deviation value of 300 seconds is generated. Finally, a behavioral deviation sequence for the employee throughout the day is formed, quantifying all abnormal action points.

[0109] The system statistically analyzes the historical violation frequency of multiple employees of the first enterprise, such as the number of violations by an employee in the past 30 days. It calculates the current weight coefficient through an index weight model and performs a weighted fusion calculation on the current multiple behavioral deviation value sequences. For example, if employee ID007 has an entry deviation of 180 seconds and an exit deviation of 300 seconds on the current day, after weighting and amplification, the fusion results in a comprehensive deviation value of (180×1.2+300×1.2)=576 seconds. The system outputs the standardized first-employee-level behavioral time sequence deviation of the first enterprise. The first-employee-level behavioral time sequence deviation quantifies the degree of abnormality in employee behavior compliance at the enterprise level.

[0110] Aggregation operations are performed on multiple employee-level behavioral time-series graphs based on the job title field. All employee behavioral records belonging to the same job code are merged and calculated to generate a job-level on-duty personnel time-series graph. This graph uses continuous time windows as the horizontal axis, and the aggregated calculation of the actual number of employees on duty for each window as the vertical axis value, forming a time-series curve describing the overall human resource distribution for the job.

[0111] For example, if the control room has 10 employees, and their entry actions are triggered a total of 7 times during the 08:00-08:05 window, then the number of employees on duty during that window is recorded as 7. If 2 more employees enter during the 08:05-08:10 window, the number of employees on duty is updated to 9. The final output is a structured job-level time series graph containing the time window interval and the real-time number of employees on duty, providing a basic data carrier for job risk quantification.

[0112] The system accesses a pre-defined job ratio time-series demand model. This model defines the standard number of staff for each job according to time windows. The system then uses the job ratio time-series demand model to traverse each time window in the job-level on-the-job personnel time-series graph, extracts the actual on-the-job personnel and the required number of staff to perform real-time gap calculation, and outputs a time-series vacancy ratio sequence.

[0113] The specific calculation logic is as follows: For a single time window, the staffing shortage ratio = (required staffing number - actual staffing number) / required staffing number × 100%.

[0114] For example, if the control room position requires 10 people at the window from 08:00 to 08:05, but only 7 are actually on duty, then the staff shortage rate is 30%; if the position requires 10 people at the window from 08:05 to 08:10, but only 9 are actually on duty, then the staff shortage rate is 10%. This generates a continuous sequence containing time window labels and staff shortage percentages, accurately quantifying the fluctuations in staff shortages for each position.

[0115] The time-series vacancy rate sequence is discretized for risk classification, and continuous percentage values ​​are mapped to discrete risk level labels according to preset classification threshold rules.

[0116] The grading rules are set as follows: when the staff shortage rate is greater than 20%, it is marked as high-risk level (Level 3); when the staff shortage rate is between 10% and 20%, it is marked as medium-risk level (Level 2); and when the staff shortage rate is less than 10%, it is marked as low-risk level (Level 1).

[0117] The system performs hierarchical mapping window by window. For example, if the control room position has a 30% staff shortage at the window from 08:00 to 08:05, it is marked as a high-risk position. If the position has a 10% staff shortage at the window from 08:05 to 08:10, it is marked as a medium-risk position. The system outputs a time-series risk level sequence composed of time windows and risk level labels, thus enabling the visualization and labeling of the risk status of the position.

[0118] Based on the preset high-risk level threshold, i.e. Level 3, we iterate through each time window in the time-series risk level sequence and count the frequency of windows that meet the high-risk level conditions; we calculate the percentage of high-risk windows to the total number of windows in the monitoring period and output the first-level vacancy ratio deviation.

[0119] The specific calculation process is as follows: Suppose that the monitoring period contains N time windows, such as 7 days × 24 hours × 12 5-minute windows = 2016 windows. The number of windows marked as Level 3 in the statistical risk level sequence is M. Then the deviation of the first-level vacancy rate is M / N × 100%.

[0120] Using the same method, we calculated and quantified M enterprise-level attendance deviations, M job-level vacancy ratio deviations, and M employee-level behavioral time-series deviations for each park enterprise. Among them, the enterprise-level attendance deviation reflects the overall vacancy scale of the enterprise. For example, a chemical enterprise with 80 employees on duty at the moment has a deviation of -15 employees compared to a baseline of 95 employees. The job-level vacancy ratio deviation quantifies the persistence of manpower shortages in key positions. For example, the control room position was at high risk of vacancy for 42.9% of the time within 7 days. The employee-level behavioral time-series deviation provides early warning of individual deviant behavior. For example, the weighted deviation of the lateness deviation of frequently violating employees reaches 576 seconds.

[0121] The deviations are normalized and calculated according to the preset weights of enterprise level, job level, and employee level. Enterprise-level deviations are converted into percentages for calculation, job-level deviations are directly entered as percentage values, and employee-level deviations are mapped to a risk coefficient of 0-1 through the number of seconds.

[0122] The fusion formula is: Risk Factor = (Enterprise Deviation Value × Enterprise-level Weight) + (Job Deviation Value × Job-level Weight) + (Employee Deviation Coefficient × Employee-level Weight). For example, the calculated value for a certain enterprise is (-15.8% × 0.4) + (42.9% × 0.35) + (0.8 × 0.25) = 0.318, which is standardized to 0.632 in the 0-1 interval after linear transformation.

[0123] By integrating the M enterprise-level on-site staffing deviations, M job-level staffing shortage ratio deviations, and M employee-level behavioral timing deviations, the M on-site staffing abnormality risk factors are output.

[0124] When the risk factor of abnormal attendance exceeds the threshold of 0.6, an enterprise-level early warning is triggered, realizing the full-dimensional risk quantification covering the macro scale of staff shortage, the meso level of job risk exposure, and the micro level of employee misconduct, providing a decision-making basis for graded response for park safety management.

[0125] The technical effects achieved by this embodiment are as follows:

[0126] 1. Eliminate misjudgments caused by base station coverage of multiple enterprises, and consistently match the actual on-duty scale in various types of enterprises, such as those in normal operation and those that have been shut down for a long time. The trend curve accurately reflects the real fluctuation patterns of manpower, such as shift changes.

[0127] 2. By directly triggering early warnings through the integration of three risk factors, the system can simultaneously quantify overall staff shortages, persistent gaps in key positions, and employee misconduct, supporting precise intervention from the overall park level to high-risk positions.

[0128] 3. Verify the ability to operate stably across all scenarios and scales, eliminate instantaneous fluctuations and interference, and ensure the reliability of monitoring results under different operating conditions.

[0129] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0130] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. A method for fusion analysis of communication data related to abnormal fluctuations in on-the-job personnel in chemical industrial park enterprises, characterized in that, The method includes: A dynamic geofence is constructed by extending the geographical boundary of the target park by a preset outward expansion distance. Configure multiple edge stream processing engines for multiple distributed communication base stations within the scope of the dynamic geofence; A communication monitoring network is constructed by using low-latency message queues to establish bidirectional communication connections between the multiple edge stream processing engines and the cloud analysis center. The multiple edge stream processing engines collect multiple signaling data streams from the multiple distributed communication base stations in real time, perform privacy-enhancing real-time hash matching based on the enterprise employee whitelist database, and generate multiple employee on-duty status vector sets. After receiving the multiple employee on-duty status vector sets, the cloud-based analysis center performs time-series dynamic aggregation based on enterprise affiliation to construct M enterprise-level on-duty personnel time-series maps corresponding to M park enterprises. Based on the time series graphs of the M enterprise-level on-the-job personnel, perform anomaly fluctuation analysis at the on-the-job personnel level and output M on-the-job abnormality risk factors. The anomaly fluctuation analysis at the level covers enterprise-level fluctuation analysis, job-level risk analysis, and employee-level behavior analysis.

2. The communication data fusion analysis method for abnormal fluctuations in on-the-job personnel in chemical industrial park enterprises as described in claim 1, characterized in that, The multiple edge stream processing engines collect multiple signaling data streams from the multiple distributed communication base stations in real time, perform privacy-enhancing real-time hash matching based on the enterprise employee whitelist database, and generate multiple employee on-duty status vector sets, including: The first edge stream processing engine continuously monitors the bound first distributed communication base station, captures raw signaling data packets, and obtains the first signaling data stream: S1: Extract protocol parsing fields from multiple time-series signaling data packets in the first signaling data stream to obtain multiple original signaling tuples, wherein each signaling tuple includes an encrypted device identifier, a signaling timestamp, and a signaling event type; S2: Add Laplace noise to the multiple encrypted device identifiers of the multiple original signaling tuples to perform differential privacy protection, and obtain multiple anonymous signaling tuples; S3: Use the multiple anonymous device identifiers of the multiple anonymous signaling tuples to perform local sensitive hash comparison with the enterprise employee whitelist database to match and filter the P anonymous employee IDs and P enterprise IDs of the P whitelist employees; S4: Based on the P anonymous signaling tuples, P anonymous employee IDs, and P enterprise IDs of the P whitelisted employees, construct P employee on-duty status vectors to form the first employee on-duty status vector set.

3. The communication data fusion analysis method for abnormal fluctuations in on-the-job personnel in chemical industrial park enterprises as described in claim 2, characterized in that, Based on the P anonymous signaling tuples, P anonymous employee IDs, and P enterprise IDs of the P whitelisted employees, P employee on-duty status vectors are constructed, forming the first employee on-duty status vector set, including: Extract P signaling timestamps and P signaling event types from P signaling tuples; P employee action types are mapped to the P signaling event types; Using the P signaling timestamps as the starting point, perform downward rounding and alignment with a preset granularity to generate P time windows; By structurally combining the P anonymous employee IDs, P enterprise IDs, P employee action types, and P time windows, the P employee on-duty status vectors are constructed.

4. The communication data fusion analysis method for abnormal fluctuations in on-the-job personnel in chemical industrial park enterprises as described in claim 1, characterized in that, After receiving the multiple employee on-duty status vector sets, the cloud-based analysis center performs time-series dynamic aggregation based on enterprise affiliation to construct M enterprise-level on-duty personnel time-series maps corresponding to M park enterprises, including: Using the enterprise ID as the basis for enterprise affiliation, the enterprise dimension is grouped into the multiple employee on-duty status vector sets to obtain M enterprise vector sets corresponding to the M park enterprises; Cross-window state transfer is performed on the M enterprise vector sets to construct M time-series window vector sequences; The M time-series window vector sequences are used to incrementally update the time-series graphs of M historical on-the-job personnel, and the M enterprise-level on-the-job personnel time-series graphs are output.

5. The communication data fusion analysis method for abnormal fluctuations in on-the-job personnel in chemical industrial park enterprises as described in claim 4, characterized in that, Cross-window state transfer is performed on the M enterprise vector sets to construct M time-series window vector sequences, including: The first enterprise vector set is reconstructed based on ascending alignment of the time window to obtain the time window state sequence; The cumulative number of employees on duty is initialized based on the first time window state in the time window state sequence, and the first time window vector is obtained. Using the first time-series window vector as the initial state, the system performs cross-window state transfer by dynamically accumulating the number of employees on duty and traversing the time-series window state sequence, and outputs the first time-series window vector sequence, wherein the first time-series window vector includes the time window interval and the net number of employees on duty.

6. The communication data fusion analysis method for abnormal fluctuations in on-the-job personnel in chemical industrial park enterprises as described in claim 4, characterized in that, Based on the time series graphs of the M enterprise-level on-the-job personnel, perform anomaly fluctuation analysis of on-the-job personnel hierarchy, and output M on-the-job anomaly risk factors, including: The moving average of the time series graphs of the M historical on-the-job personnel is calculated for a preset production cycle to construct the M enterprise on-the-job baseline curves; Based on the time series graphs of the M enterprise-level on-the-job personnel, construct the M enterprise on-the-job incremental curves; After aligning the on-duty baseline curves of the M enterprises and the on-duty incremental curves of the M enterprises according to the preset production cycle, perform window-level on-duty deviation calculation and output the on-duty personnel deviation of the M enterprises. Based on the anonymous employee IDs, the time series graphs of the M enterprise-level on-the-job personnel are decomposed, and fine-grained deviation hierarchical analysis is performed to output the M job-level vacancy ratio deviations and the M employee-level behavioral time series deviations. By integrating the M enterprise-level on-site staffing deviations, M job-level staffing shortage ratio deviations, and M employee-level behavioral timing deviations, the M on-site staffing abnormality risk factors are output.

7. The communication data fusion analysis method for abnormal fluctuations in on-the-job personnel in chemical industrial park enterprises as described in claim 6, characterized in that, Also includes: Based on anonymous employee IDs, the time series graph of on-the-job personnel in the first enterprise is decomposed into employee dimensions to generate multiple employee-level behavioral time series graphs corresponding to multiple employees of the first enterprise; Based on the shift scheduling information generated by the first enterprise, construct multiple behavioral timeline baselines for the employees of the multiple first enterprises; By comparing the multiple employee-level behavioral time series maps and multiple behavioral time series baselines, multiple behavioral deviation value sequences are obtained; Based on the historical violation frequencies of the multiple first-enterprise employees, the multiple behavioral deviation value sequences are weighted and fused to form the first-employee level behavioral time sequence deviation.

8. The communication data fusion analysis method for abnormal fluctuations in on-the-job personnel in chemical industrial park enterprises as described in claim 7, characterized in that, Also includes: Based on the job title and grade, aggregate the multiple employee-level behavioral time series graphs and output the job title and grade on-the-job personnel time series graph; The time-series requirement for accessing job positions is to traverse the time-series graph of on-the-job personnel at the job level and output the time-series vacancy rate. The time-series staff shortage ratio is classified into risk levels, and the time-series risk level is output. Based on a preset high-risk level threshold, the time-series risk levels are iterated to perform high-risk level percentage statistics, and the deviation of the first-level vacancy ratio is output.

Images