A hybrid retrieval cockpit and work order linkage method and system for operation management

Abstract

The application belongs to the technical field of energy, and discloses a hybrid retrieval cockpit and work order linkage method for operation management, which is constructed based on a multi-source heterogeneous data lake warehouse layer, a semantic knowledge topology layer and a business application layer of a digital base of a digital support platform, and comprises a hybrid retrieval module, an alarm merging module, a work order automatic generation and management module, an SLA process state monitoring module and a cockpit visualization module, wherein each module realizes data interaction and process linkage through a standard interface. The application realizes unified retrieval of multi-modal data, avoids cross-system switching, shortens the abnormal traceability time from an average of 30 minutes to within 5 minutes, reduces the redundant alarm rate by more than 80% after alarm merging, improves the interference alarm identification accuracy to 92%, and enables operation and maintenance personnel to focus on core abnormalities.

Description

Technical Field

[0001] This invention belongs to, but is not limited to, the field of energy technology, and particularly relates to a hybrid retrieval cockpit and work order linkage method and system for operation management. Background Technology

[0002] In the process of large-scale development of shale gas fields, as wells enter the old well stage, abnormal operating conditions such as liquid accumulation, compressor shutdown, and pipeline blockage occur frequently. According to statistics from the Southwest Oil and Gas Field in 2024, an average of 10 wells experienced liquid accumulation per day, with an average well recovery cycle of 2.5 days, an average daily production loss of 150,000 cubic meters, and an annual loss of approximately 55 million cubic meters. Current shale gas operation and management relies on the following methods, which have significant shortcomings: Data retrieval is fragmented: real-time streaming data (SCADA data such as oil pressure and casing pressure), batch static data (well numbers, process information, and other ledgers), and application-generated data (alarm records and feedback tags) are scattered across different systems. Operation and maintenance personnel need to switch between multiple platforms to query, which makes it impossible to achieve unified association retrieval of cross-modal data, resulting in low efficiency of anomaly tracing.

[0003] Redundant and chaotic alarm information: The existing SCADA / POC system is based on fixed threshold alarms, which are easily affected by machine failures (such as compressor shutdown) and human intervention (such as well shutdown and gas lift). The same anomaly triggers multiple duplicate alarms, and there is a lack of alarm classification and priority sorting, making it difficult for maintenance personnel to quickly locate the core problem.

[0004] The work order processing workflow is disconnected: after an anomaly is discovered, a work order needs to be created manually, and information such as the hash number, anomaly type, and historical data needs to be filled in. There is a risk of missing information and filling in errors. During the execution of the work order (such as dispatching, handling, and acceptance), there is a lack of real-time status monitoring, the SLA (Service Level Agreement) timing is not transparent, and the handling results cannot be automatically written back to the data system, forming a broken management loop.

[0005] The cockpit has limited functionality: the existing operations interface can only display basic data curves and cannot integrate multi-dimensional information such as mixed search results, work order status, and SLA progress. As a result, operations and maintenance personnel have difficulty obtaining a global operational view, leading to delayed decision-making and response.

[0006] Against the backdrop of digital transformation, shale gas fields have established real-time monitoring systems covering wellbore and surface processes, and the development of digital support platforms and digital foundations has laid the groundwork for data lake-based governance and knowledge graph construction. Therefore, there is an urgent need for an operational system that integrates hybrid retrieval, intelligent alarm merging, work order linkage, and closed-loop management to address the current problems of low operational efficiency and disjointed processes. Summary of the Invention

[0007] To address the problems existing in the prior art, this invention provides a hybrid retrieval dashboard and work order linkage method and system for operations management.

[0008] This invention is implemented as follows: a hybrid retrieval dashboard and work order linkage system for operations management, the system comprising: The data layer is used to receive and store real-time streaming data, batch static data, and application-generated data from shale gas fields, and supports transaction consistency control, version control, and data lineage tracing.

[0009] The ontology layer is used to construct a knowledge graph containing entities such as wells, equipment, processes, sensors, and alarms, and to establish connections between wells and equipment, historical relationships of process switching, and causal chains between alarms and upstream and downstream entities.

[0010] The application layer includes a hybrid retrieval module, an alarm merging module, an automatic work order generation and management module, and a service level agreement process status monitoring module.

[0011] The hybrid retrieval module integrates structured retrieval results with semantic retrieval results to generate related retrieval results.

[0012] The alarm merging module classifies alarms, merges redundancies, and eliminates interference based on a rule engine and a random forest model, and outputs valid alarms.

[0013] The automatic work order generation and management module generates work orders based on the valid alarms and the associated search results, and completes the work order dispatch according to the preset response time limit.

[0014] The Service Level Agreement process status monitoring module records the time nodes of each stage of the work order and monitors the time limits.

[0015] The associated search results are used in the calculation of work order priority and the generation of handling suggestions, thereby realizing the linkage between the search mechanism and the work order generation mechanism.

[0016] This invention also provides a hybrid retrieval dashboard and work order linkage system for operation management. The system is built on a multi-source heterogeneous data lake warehouse layer, semantic knowledge topology layer, and business application layer of a digital support platform digital base. It includes a hybrid retrieval module, an alarm merging module, an automatic work order generation and management module, an SLA process status monitoring module, and a dashboard visualization module. Each module realizes data interaction and process linkage through standard interfaces.

[0017] The hybrid retrieval module is deployed at the business application layer and supports structured retrieval and semantic retrieval. Structured retrieval retrieves structured results by calling multi-source heterogeneous data lake warehouse layer data based on user selection of hash number, time range, anomaly type, and other dimensions. Semantic retrieval parses natural language query intent based on the knowledge graph of multi-source heterogeneous data lake warehouse layer, associates multimodal data through graph traversal algorithm, and returns results containing time series curves, equipment ledgers, and historical work orders.

[0018] The alarm merging module integrates a rule engine and a lightweight random forest AI model, optimized for specific issues in shale gas operation and maintenance. It allows for supplementary input features, such as the pressure fluctuation frequency, flow rate attenuation slope, and production regime switching flags of shale gas wells as feature vectors input to the model. Alarms are categorized into liquid accumulation, equipment failure, human intervention, and data anomaly types. A three-factor matching method based on "well number, anomaly type, and trigger time difference ≤ 5 minutes" is used for redundancy merging. Interfering alarms are eliminated by combining multi-source heterogeneous data lake layer process switching history, and priority is sorted in the order of "equipment failure > liquid accumulation > human intervention > data anomaly".

[0019] The automatic work order generation and management module receives merged valid alarms, extracts alarm numbers, platforms, anomaly types, trigger thresholds, and related search results from the multi-source heterogeneous data lake warehouse layer, and automatically fills the work order template, including work order number, urgency level, and SLA requirements. Urgent work orders are responded to within 4 hours, and ordinary work orders within 8 hours. Work orders are automatically dispatched based on the "alarm-maintenance team" correspondence, and manual adjustment of dispatch objects is supported with log recording. Historical similar anomaly handling solutions extracted by the system through semantic retrieval are automatically mapped to the handling suggestion field of new work orders as recommendation parameters. It should demonstrate how the search results 'assist' or 'drive' work order generation.

[0020] The SLA process status monitoring module records the time nodes of work order creation, dispatch, acceptance, start of handling, completion of handling, and acceptance. It calculates the time consumed in each stage and sends warnings via dashboard pop-ups and SMS when the SLA deadline is approaching. If the timeout is exceeded, the work order priority is upgraded and the superior department is notified. The work order progress is displayed in a Gantt chart, and different colors are used to mark the stage status, specifically: not dispatched: gray, dispatched but not accepted: yellow, handling: blue, completed: green, and timeout: red.

[0021] The cockpit visualization module aggregates mixed search results, alarm status, work order progress, and SLA data, providing a global operational view with interconnected interaction. It supports the display of search results and work order status in association, as well as the input of work order processing results.

[0022] Furthermore, the multi-source heterogeneous data lake layer adopts a lake-based architecture to store multi-source data from shale gas fields, including real-time streaming data such as oil pressure, casing pressure, instantaneous flow rate, and post-valve pressure collected by SCADA, batch static data including well number, platform affiliation, process type, and equipment ledger, as well as application-generated data such as alarm records, manual annotation results, and work order handling logs. It also supports ACID properties, version control, and data lineage tracing, establishing a correlation link of "raw data - features - alarms - work orders".

[0023] Furthermore, the multi-source heterogeneous data lake layer constructs a knowledge graph in the shale gas field, defining entities such as wells, skids, equipment, processes, and sensors, as well as relationships such as "well-skid connection", "process switching history", and "alarm and upstream and downstream causal chain".

[0024] Furthermore, the rule engine of the alarm merging module is configurable, allowing users to customize merging dimensions and priority weights; and the AI ​​model supports adaptive learning, periodically updating parameters based on historical processing results to adjust the confidence threshold for interfering alarm identification, thereby achieving a merging accuracy of ≥95%.

[0025] Furthermore, the SLA process status monitoring module supports multi-dimensional SLA configuration, and different response and handling time limits can be set according to well type and anomaly level; it also has a pause time function. When a work order needs to be paused due to objective reasons, the operation and maintenance manager submits an application, and after approval, the SLA time is paused and automatically restarted when handling resumes.

[0026] Another objective of this invention is to provide a hybrid retrieval dashboard and work order linkage method for operations management, which includes the following steps: S1: Unified access and storage of multimodal data. It receives real-time streaming data, batch static data, and application-generated data from shale gas fields through the digital support platform Data Layer. It adopts a lake-warehouse architecture for storage and enables ACID properties, version control, and data lineage tracking.

[0027] S2: Hybrid retrieval execution. Users can select structured retrieval or semantic retrieval through the cockpit. Structured retrieval calls multi-source heterogeneous data lake warehouse layer data according to the selected dimension and returns results. Semantic retrieval inputs natural language query, and the system interprets the intent based on the knowledge graph of multi-source heterogeneous data lake warehouse layer, associates multimodal data and returns results. The retrieval results also synchronously display the associated time series curves, equipment ledgers and historical work orders.

[0028] S3: Alarm merging and processing. The alarm merging module classifies received alarms, merges redundancies and eliminates interference, outputs valid alarms and sorts them by priority.

[0029] S4: Automatic work order generation and dispatch. Valid alarms trigger automatic work order generation. The system fills in the work order information and automatically dispatches the work order based on the correspondence between "well-maintenance team".

[0030] S5: SLA process status monitoring, real-time tracking of work order execution progress, recording time nodes, triggering timeout warnings and visually displaying progress.

[0031] S6: Closed-loop result write-back and optimization. After the work order is accepted, the maintenance personnel fill in the handling result in the dashboard. The system automatically writes it back to the Data Layer, associates the original alarms with the retrieved data, and uses the handling result as labeled data to optimize the alarm merging model parameters and SLA rules.

[0032] Furthermore, in S2, the specific implementation process of semantic retrieval is as follows: after receiving a natural language query, the system first performs word segmentation and intent recognition, matches entities and relationships in the knowledge graph of the multi-source heterogeneous data lake warehouse layer, and then uses a graph traversal algorithm of depth-first search or breadth-first search to associate the multimodal data involved in the query. Finally, the association results are organized and displayed in the order of "time series data - static data - application generated data".

[0033] Furthermore, in S3, the specific logic for interference alarm elimination is as follows: The system queries the process switching records within 1 hour before and after the alarm trigger time in the multi-source heterogeneous data lake layer. If there are records of manual well shut-in or gas lift operations, the operation time and operation type are extracted and compared with the alarm information. If the difference between the operation time and the alarm trigger time is ≤30 minutes, it is determined to be an interference alarm, marked and its priority is reduced, and the description "there is manual operation interference" is added to the alarm information.

[0034] Furthermore, in S6, the specific method of model optimization is as follows: each quarter, based on the newly added labeled data of disposal results, the lightweight random forest AI model is retrained and the feature weights are adjusted; the specific method of SLA rule optimization is as follows: each month, the disposal time of work orders of different well types and different anomaly types is statistically analyzed, the average disposal time is calculated, and the SLA response time limit and disposal time limit of the corresponding scenario are adjusted based on the average disposal time within ±20%.

[0035] Another objective of this invention is to provide an information data processing terminal, which is used to realize the hybrid retrieval cockpit and work order linkage system for operation management.

[0036] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows: Combined with the above technical solutions and the problems of scattered multi-source data, serious alarm redundancy, lagging work order response, and low knowledge utilization rate in shale gas fields, the technical solutions protected by the present invention have significant advantages and positive effects at the overall architecture and mechanism coordination levels. First, based on the data lake warehouse and knowledge graph jointly constructed in the data layer and the ontology layer, unified management of real-time stream data, batch static data, and application-generated data is achieved, supporting transaction consistency control, version management, and data lineage tracking, ensuring that the data source is traceable, the status is queryable, and the results are reproducible, and improving the data credibility and compliance in the operation and management process. Second, through the hybrid retrieval mechanism, structured retrieval and semantic retrieval are fused and calculated. Not only can accurate queries be made according to well numbers, time ranges, and abnormal types, but also causal chain analysis and graph traversal association can be performed based on the knowledge graph to achieve deep联动检索 of cross-system and multi-modal data, improving the efficiency of locating complex problems.

[0037] In the alarm handling link, the present invention realizes intelligent classification through the cooperation of the rule engine and the random forest model, uses the three-element matching method for redundancy merging, and combines the process switching history to identify interference alarms, effectively reducing the proportion of repeated alarms and false alarms, enabling truly risk-valued anomalies to enter the disposal process first, and significantly improving the accuracy of alarm screening. In the work order generation link, the present invention embeds the associated retrieval results into the work order priority calculation and disposal suggestion generation logic, and the risk levels and disposal steps of historical similar anomalies directly participate in the current work order decision-making, transforming the retrieval results from auxiliary references into driving factors, realizing the structured empowerment of knowledge to decision-making, shortening the manual judgment time, and improving the pertinence and standardization degree of disposal.

[0038] In addition, the service level agreement process status monitoring module realizes full-process time limit management through continuous tracking of each stage time node and threshold warning, ensuring that emergency work orders are responded within 4 hours and ordinary work orders are responded within 8 hours, improving the on-site operation and maintenance efficiency. The disposal result write-back mechanism further supports the periodic optimization of models and rules, enabling the system to have the ability of continuous learning. Overall, the present invention constructs a "data - knowledge - alarm - work order - feedback" closed-loop coordination mechanism, realizes deep联动检索 of retrieval, analysis, and execution, and has obvious effects of improving the intelligent level and operation efficiency in complex shale gas operation and maintenance scenarios.

[0039] The present invention improves the retrieval efficiency: realizes unified retrieval of multi-modal data, avoids cross-system switching, and reduces the abnormal traceability time from an average of 30 minutes to within 5 minutes.

[0040] The present invention reduces alarm interference: the redundant alarm rate is reduced by more than 80% after alarm merging, the recognition accuracy of interference alarms is ≥92%, and operation and maintenance personnel focus on core anomalies.

[0041] This invention reduces labor costs: work orders are automatically generated, reducing the amount of manual work (the time to fill out each work order is reduced from 15 minutes to less than 1 minute), and avoiding information errors.

[0042] This invention ensures timely response: real-time SLA monitoring and overdue warnings increase the on-time response rate of emergency work orders to 98%, and shorten the average production recovery cycle of a single well to 1.8 days.

[0043] This invention achieves closed-loop management: the disposal results are automatically written back, providing data support for model optimization and operational analysis, forming a virtuous cycle of "data-retrieval-alarm-work order-optimization". Attached Figure Description

[0044] Figure 1 This is a structural diagram of the hybrid retrieval cockpit and work order linkage system for operation management provided in an embodiment of the present invention.

[0045] Figure 2 This is a flowchart of the hybrid retrieval dashboard and work order linkage method for operation management provided in an embodiment of the present invention.

[0046] Figure 3 This is a flowchart illustrating the specific implementation process of semantic search provided in this embodiment of the invention.

[0047] Figure 4 This is a comparison chart of alarm noise reduction efficiency provided in the embodiments of the present invention.

[0048] Figure 5 This is a comparison chart of the operation and maintenance process time provided in the embodiments of the present invention.

[0049] Figure 6 This is a comparison chart of the wellhead production cycle distribution provided in the embodiments of the present invention.

[0050] Figure 7 This is a forecast chart of annual cumulative economic benefits provided in an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0052] In existing oil and gas field operation and management systems, data silos, slow retrieval response times, and fragmented work order workflows are common industrial challenges. Real-time production data, static equipment data, and historical work order data are often scattered across different systems, lacking a unified data lake to support them. This makes it difficult for dispatchers and maintenance engineers to establish cross-domain connections in a timely manner, and retrieval results are often limited to a single dimension, failing to quickly support production decisions. Meanwhile, alarm systems output excessive redundant signals, and manually identifying interfering alarms is time-consuming and labor-intensive, easily leading to misjudgments and delays. Insufficient automated work order linkage further weakens overall operation and maintenance efficiency.

[0053] This invention fundamentally solves the problem of fragmented multimodal data in shale gas fields by introducing an integrated lake-warehouse architecture into the Trivium Data Layer and enabling ACID transactions, version control, and lineage tracking. Real-time streaming data, batch static data, and application-generated data can be uniformly accessed, stored, and retrieved on the same platform, ensuring data consistency and traceability, and providing a solid data foundation for subsequent semantic retrieval and work order linkage. This mechanism effectively avoids latency and deviation caused by heterogeneous storage, supporting a data-driven closed loop in complex operation and maintenance scenarios.

[0054] In the retrieval execution phase, the system performs semantic parsing of natural language queries through an Ontology Layer knowledge graph and achieves cross-modal data association based on graph traversal algorithms, forming a panoramic view that structured retrieval cannot achieve. Query results not only include numerical values ​​but also synchronously link time-series curves, equipment ledgers, and historical work orders. This knowledge graph-oriented retrieval method avoids the limitations of keyword matching, enabling users to quickly locate problems in complex operating conditions and significantly improving the intelligence level of the operations dashboard. Its working principle relies on the semantic mapping of entities and relationships, and then uses depth-first or breadth-first path traversal to complete the efficient integration of data.

[0055] In the alarm handling stage, this invention introduces a merging and interference elimination logic. It compares the alarm signal with the process operation records in the OntologyLayer in terms of timing. When there is interference from manual well shut-in or gas lift operations, the system automatically downgrades the alarm priority and provides the operational background in the alarm description. This mechanism starts from the data context, using the time difference between the maintenance behavior and the alarm signal as the discrimination condition, avoiding false alarms caused by simply relying on threshold judgments, and improving the accuracy of alarm management. This approach directly addresses the pain point of excessive invalid alarms consuming maintenance resources in existing technologies.

[0056] The work order linkage mechanism enables the automatic conversion from alarm to work order. The system automatically dispatches work orders based on the correspondence between wells and maintenance teams, reducing the lag and errors of manual work order dispatch. During the execution phase, the SLA process monitoring module tracks and visualizes the progress in real time, triggering timeout warnings. This state machine-style work order lifecycle management achieves a digital closed loop on top of the original manual entry, significantly shortening the response chain from alarm discovery to handling completion, and significantly improving the standardization and timeliness of production and maintenance.

[0057] Ultimately, this invention achieves self-learning and iterative evolution of the system through a closed-loop write-back and model optimization mechanism. The handling results from operations and maintenance personnel are written back to the Data Layer and used as labeled samples for the periodic retraining of the lightweight random forest model, thereby optimizing alarm merging and SLA rules. The SLA threshold is dynamically adjusted based on the statistical distribution of actual handling times, allowing the rules to evolve with changing operating conditions. This feedback-driven optimization path enables adaptive upgrades in operations management, solving the problem of existing systems lacking continuous optimization capabilities in industrial applications.

[0058] like Figure 1 As shown, this embodiment of the invention provides a hybrid retrieval dashboard and work order linkage system for operation management. The system is built on the multi-source heterogeneous data lake warehouse layer, semantic knowledge topology layer, and business application layer of the digital support platform digital base. It includes a hybrid retrieval module 1, an alarm merging module 2, an automatic work order generation and management module 3, an SLA process status monitoring module 4, and a dashboard visualization module 5. Each module realizes data interaction and process linkage through standard interfaces.

[0059] The hybrid retrieval module 1 is deployed at the business application layer and supports structured retrieval and semantic retrieval. Structured retrieval retrieves structured results by calling multi-source heterogeneous data lake warehouse layer data through user selection of hash number, time range, anomaly type, and other dimensions. Semantic retrieval parses natural language query intent based on the knowledge graph of multi-source heterogeneous data lake warehouse layer, associates multimodal data through graph traversal algorithm, and returns results containing time series curves, equipment ledgers, and historical work orders.

[0060] The alarm merging module 2 integrates a rule engine and a lightweight random forest AI model, classifying alarms into categories such as liquid accumulation, equipment failure, human intervention, and data anomaly. It uses a three-factor matching method of "hash number, anomaly type, and trigger time difference ≤ 5 minutes" for redundancy merging, and combines the process switching history of multi-source heterogeneous data lake warehouse to eliminate interference alarms, and sorts them by priority in the order of "equipment failure > liquid accumulation > human intervention > data anomaly".

[0061] The automatic work order generation and management module 3 receives the merged valid alarms, extracts the well number, platform, anomaly type, trigger threshold, and related search results from the multi-source heterogeneous data lake warehouse layer, and automatically fills the work order template, including the work order number, urgency level, and SLA requirements. Urgent work orders are responded to within 4 hours, and ordinary work orders are responded to within 8 hours. The module automatically dispatches work orders based on the correspondence between "well-maintenance team" and supports manual adjustment of dispatch objects and recording logs.

[0062] The SLA process status monitoring module 4 records the time nodes of work order creation, dispatch, acceptance, start of handling, completion of handling, and acceptance. It calculates the time consumed in each stage and sends warnings through cockpit pop-ups and SMS when the SLA deadline is approaching. After the timeout, the work order priority is upgraded and the superior department is notified. The work order progress is displayed in a Gantt chart, and the stage status is marked with different colors: not dispatched: gray, dispatched but not accepted: yellow, handling: blue, completed: green, timeout: red.

[0063] The cockpit visualization module 5 aggregates mixed search results, alarm status, work order progress, and SLA data, providing a global operational view with linkage and interaction. It supports the display of the association between search results and work order status, as well as the input of work order processing results.

[0064] The multi-source heterogeneous data lake layer adopts a lake-based architecture to store multi-source data from shale gas fields, including real-time streaming data such as oil pressure, casing pressure, instantaneous flow rate, and post-valve pressure collected by SCADA, batch static data including well number, platform affiliation, process type, and equipment ledger, as well as application-generated data such as alarm records, manual annotation results, and work order handling logs. It also supports ACID properties, version control, and data lineage tracing, establishing a correlation link of "raw data - features - alarms - work orders".

[0065] The multi-source heterogeneous data lake layer constructs a knowledge graph in the shale gas field, defining entities such as wells, skids, equipment, processes, and sensors, as well as relationships such as "well-skid connection", "process switching history", and "alarms and upstream and downstream causal chains".

[0066] The rule engine of the alarm merging module 2 is configurable, allowing users to customize merging dimensions and priority weights. Furthermore, the AI ​​model supports adaptive learning, periodically updating parameters based on historical processing results and adjusting the confidence threshold for identifying interfering alarms to achieve a merging accuracy of ≥95%.

[0067] The SLA process status monitoring module 4 supports multi-dimensional SLA configuration, and can set different response time limits and handling time limits according to well type and anomaly level; it also has a pause timer function. When a work order needs to be paused due to objective reasons, the operation and maintenance manager submits an application, and after approval, the SLA timer is paused and automatically restarted when handling resumes.

[0068] like Figure 2 As shown, this embodiment of the invention provides a hybrid retrieval dashboard and work order linkage method for operations management, which includes the following steps: S1: Unified access and storage of multimodal data. Through the multi-source heterogeneous data lake layer of the digital support platform, real-time streaming data, batch static data, and application-generated data from shale gas fields are received. The lake-based architecture is used for storage, and ACID properties, version control, and data lineage tracking are enabled.

[0069] S2: Hybrid retrieval execution. Users can select structured retrieval or semantic retrieval through the cockpit. Structured retrieval calls multi-source heterogeneous data lake warehouse layer data according to the selected dimension and returns results. Semantic retrieval inputs natural language query, and the system interprets the intent based on the knowledge graph of multi-source heterogeneous data lake warehouse layer, associates multimodal data and returns results. The retrieval results also synchronously display the associated time series curves, equipment ledgers and historical work orders.

[0070] S3: Alarm merging and processing. The alarm merging module classifies received alarms, merges redundancies and eliminates interference, outputs valid alarms and sorts them by priority.

[0071] S4: Automatic work order generation and dispatch. Valid alarms trigger automatic work order generation. The system fills in the work order information and automatically dispatches the work order based on the correspondence between "well-maintenance team".

[0072] S5: SLA process status monitoring, real-time tracking of work order execution progress, recording time nodes, triggering timeout warnings and visually displaying progress.

[0073] S6: Closed-loop result write-back and optimization. After the work order is accepted, the maintenance personnel fill in the handling result in the dashboard. The system automatically writes it back to the Data Layer, associates the original alarms with the retrieved data, and uses the handling result as labeled data to optimize the alarm merging model parameters and SLA rules.

[0074] like Figure 3 As shown, in S2, the specific implementation process of semantic retrieval is as follows: after receiving a natural language query, the system first performs word segmentation and intent recognition, matches entities and relationships in the knowledge graph of the multi-source heterogeneous data lake warehouse layer, and then uses a graph traversal algorithm of depth-first search or breadth-first search to associate the multimodal data involved in the query. Finally, the association results are organized and displayed in the order of "time series data - static data - application generated data".

[0075] In S3, the specific logic for interference alarm elimination is as follows: The system queries the process switching records within 1 hour before and after the alarm trigger time in the multi-source heterogeneous data lake layer. If there are records of manual well shut-in or gas lift operations, the operation time and operation type are extracted and compared with the alarm information. If the difference between the operation time and the alarm trigger time is ≤30 minutes, it is determined to be an interference alarm, marked and its priority is reduced, and the description "there is manual operation interference" is added to the alarm information.

[0076] In S6, the specific method of model optimization is as follows: each quarter, based on the newly added labeled data of disposal results, the lightweight random forest AI model is retrained and the feature weights are adjusted; the specific method of SLA rule optimization is as follows: using an adaptive algorithm based on the historical disposal efficiency distribution, the response time limit threshold of the next cycle is dynamically corrected through a feedback loop to realize the on-demand allocation of operation and maintenance resources, the disposal time of work orders of different abnormal types, the calculation of the average disposal time, and the SLA response time limit and disposal time limit corresponding to the scenario based on the self-learning / adaptive logic within a range of ±20% of the average disposal time.

[0077] During the alarm merging phase, the system integrates a rule engine with a lightweight random forest model to construct an intelligent classification mechanism for "specific problem identification" in shale gas operation and maintenance scenarios. First, the rule engine performs basic filtering and standardization on the original alarms. Then, key operational characteristics such as the pressure fluctuation frequency, flow attenuation slope, and production regime switching flags of shale gas wells are constructed as feature vectors and input into the random forest model to achieve intelligent identification of alarm types. The model output is divided into four categories: liquid accumulation, equipment failure, human intervention, and data anomaly. Redundancy merging is performed using a three-factor matching method (well number, anomaly type, trigger time difference ≤ 5 minutes) to prevent the same anomaly from triggering repeatedly within a short period, resulting in multiple invalid alarms. Simultaneously, the system calls upon process switching history records from a multi-source heterogeneous data lake layer to cross-validate anomaly signals during production regime adjustments, automatically eliminating interference alarms caused by operating condition switching. After merging, alarms are sorted according to a priority rule of "equipment failure > liquid accumulation > human intervention > data anomaly" to ensure that high-risk alarms are prioritized for subsequent processing.

[0078] During the work order generation phase, the system only triggers the automatic work order creation process for valid alarms after merging and interference elimination. The module extracts structured data in real time from the multi-source heterogeneous data lake layer, including well number, platform, anomaly type, trigger threshold, associated equipment information, and historical handling records. This data is automatically filled into preset work order template fields to generate a standard work order text containing the work order number, anomaly level, trigger time, and responsible well group. The urgency level is determined based on a combination of anomaly category and model confidence: equipment failures or anomalies exceeding a threshold are automatically marked as urgent work orders, requiring a response within 4 hours; other categories are marked as ordinary work orders, requiring a response within 8 hours. The system automatically dispatches work orders based on a preset "well-maintenance team" mapping relationship, recording the dispatch time, dispatch object, and algorithm decision basis. If manual adjustments are made, the system synchronously records the operation log, achieving traceable responsibility and transparent process management.

[0079] Regarding the semantic retrieval assistance mechanism, the system does not merely passively reference historical data. Instead, it uses a vectorized semantic retrieval model to perform similarity matching on historical abnormal work order texts, handling reports, and expert experience bases, extracting the handling solutions and risk assessment conclusions that are closest to the current abnormal characteristics. The retrieval results first participate in abnormality level assessment and urgency correction. If high-risk consequences exist in similar historical cases, the priority of the current work order is automatically increased, thus "driving" the dynamic adjustment of the work order generation strategy. Second, the retrieved handling steps, key operational parameters, and risk warnings are structured and automatically mapped into the "handling suggestions" and "precautions" fields of new work orders, directly embedded as recommendation parameters in the work order content to assist maintenance personnel in quickly formulating handling solutions. Through this mechanism, historical knowledge not only improves the accuracy of work order generation but also plays a driving role in multiple stages such as priority determination, dispatching decisions, and handling suggestion generation, forming an intelligent closed-loop maintenance management system supported by data knowledge.

[0080] Combination Figures 4 to 7 The comparison results shown can further refine the overall effect of the system of the present invention in actual operation. Figure 4 The results show that during 14 consecutive days of operation, the number of original alarms per day remained between 50 and 80 under the traditional model, while the number of effective alarms in this system, after classification, merging, and interference removal, stabilized at between 5 and 15, with a noise reduction rate of approximately 70% to 85%. This demonstrates that by combining the rule engine with the random forest model and employing a three-factor matching and merging mechanism, the scale of redundant and false alarms was significantly reduced, allowing operational resources to be focused on real risk events and improving the accuracy of anomaly identification.

[0081] Figure 5This reflects changes in the time consumed during the operation and maintenance process. Traditional manual methods take approximately 50 to 60 minutes in total for anomaly detection, information retrieval, and work order creation. Our automated system reduces these steps to less than 10 minutes, lowering the overall process time by about 60%. Particularly in the information retrieval and work order generation stages, semantic search results are directly embedded into work order fields and participate in priority calculations, reducing manual review and judgment time and achieving process-level efficiency improvements.

[0082] Figure 6 This indicates a significant leftward shift in the wellhead recovery cycle distribution. The average recovery cycle in the traditional model is 2.5 days, while the average recovery cycle after system optimization is shortened to 1.8 days, improving recovery efficiency by approximately 28%. Figure 7 The results show that, within the 12-month forecast period, the cumulative output loss curve under the system model is lower than that under the traditional model, and the economic gain corresponding to the cumulative difference at the end of the year shows a stable growth trend, demonstrating a sustained improvement in economic benefits. Overall, this invention achieves significant improvements in alarm quality, process efficiency, production recovery cycle, and annual economic value.

[0083] As a preferred improvement of the present invention, an embodiment of the present invention provides a hybrid retrieval dashboard and work order linkage system for operation management, comprising: The data module is used to receive and store real-time streaming data, batch static data, and application-generated data from shale gas fields, and supports transaction consistency control, version control, and data lineage tracing.

[0084] The ontology module is used to build a knowledge graph containing entities such as wells, equipment, processes, sensors, and alarms, and to establish connections between wells and equipment, historical relationships of process switching, and causal chains between alarms and upstream and downstream entities.

[0085] The application modules include a hybrid retrieval module, an alarm merging module, an automatic work order generation and management module, and a service level agreement process status monitoring module.

[0086] The hybrid retrieval module integrates structured retrieval results with semantic retrieval results to generate related retrieval results.

[0087] The alarm merging module extracts operational features from oil and gas production time-series data, inputs these features into a pre-trained machine learning classification model, and classifies, merges redundancies, and eliminates interference based on the rule engine and the machine learning classification model, outputting valid alarms.

[0088] The automatic work order generation and management module generates work orders based on the valid alarms and the associated search results, and completes the work order dispatch according to the preset response time limit.

[0089] The Service Level Agreement process status monitoring module records the time nodes of each stage of the work order and monitors the time limits.

[0090] The associated search results are used in the calculation of work order priority and the generation of handling suggestions, thereby realizing the linkage between the search mechanism and the work order generation mechanism.

[0091] The alarm merging module classifies alarms into categories such as liquid accumulation, equipment failure, human intervention, and data anomaly. It uses a three-factor matching method—well number, anomaly type, and trigger time difference less than a preset time difference threshold—to perform redundancy merging. Within a preset time window before and after the alarm trigger time, it queries process switching records to identify interference alarms caused by manual well shut-in or gas lift operations and lowers their priority. The machine learning classification model is a random forest model, and the operational features include pressure fluctuation frequency, flow rate attenuation slope, and production system switching flag.

[0092] In the hybrid retrieval dashboard and work order linkage system for operation management, the signal data processing process is the core hub connecting the field perception of shale gas fields with the linkage and handling of work orders. It runs through the entire process of data module, body module and application module, and focuses on serving alarm merging, hybrid retrieval and work order generation. Through standardized and intelligent processing, the raw signal data is transformed into effective information that can be used for operational decision-making, so as to achieve accurate and efficient linkage between signal data and work orders.

[0093] Signal data processing is based on the data module, which first completes the reception and standardized storage of various types of signal data. The data module receives real-time streaming data (such as continuous signals such as pressure, flow, and temperature) transmitted from field sensors in shale gas fields, batch static data (such as well condition parameters and equipment parameters), and application-generated data. During the storage process, transaction consistency control, version control, and data lineage tracing are implemented simultaneously to ensure the integrity and traceability of signal data, providing reliable data support for subsequent processing and avoiding false alarms or incorrect work orders due to data anomalies.

[0094] The standardized and stored signal data is prioritized and sent to the alarm merging module of the application module to complete the identification, classification, and merging of abnormal signals. First, the operational characteristics in the oil and gas production time-series signals are extracted, including core indicators such as pressure fluctuation frequency, flow attenuation slope, and production system switching flags. These features are then input into a pre-trained random forest machine learning classification model, combined with a rule engine, to accurately classify the alarms corresponding to the signals into four categories: liquid accumulation, equipment failure, human error, and data anomaly.

[0095] For the classified alarm signals, a three-factor matching method is used, which combines well number, anomaly type, and trigger time difference less than a preset threshold, to merge redundant alarms and eliminate duplicate and invalid alarm signals. At the same time, within a preset time window before and after the alarm trigger time, the process switching history relationship in the knowledge graph of the ontology module is linked to query whether there are human intervention behaviors such as manual well shut-in or gas lift operation, identify interference alarms caused by human measures, and automatically reduce their priority. Finally, accurate and effective alarm signals are output to provide core basis for work order generation.

[0096] Simultaneously, signal data and processed alarm-related data are sent to the hybrid retrieval module, where they are associated with entity data such as wells, equipment, and processes in the ontology module's knowledge graph. The hybrid retrieval module integrates the structured retrieval results (such as well number, equipment number, and abnormal values) corresponding to the signal data with semantic retrieval results (such as alarms and upstream / downstream causal chains, and equipment relationships) to generate associated retrieval results. These results not only assist staff in quickly locating the root cause of anomalies but also participate in work order priority calculations and handling suggestions, achieving deep linkage between the retrieval mechanism and the work order generation mechanism.

[0097] The entire signal data processing process forms a closed loop. Valid alarms and associated retrieval results after processing trigger the automatic work order generation and management module, and the work order is dispatched in accordance with the time limit requirements of the service level agreement process status monitoring module. At the same time, feedback data during the work order processing is synchronized back to the data module, providing data support for optimizing signal data processing parameters and iterating machine learning models, continuously improving the accuracy of signal data processing, ensuring the efficiency of system work order linkage, and helping to upgrade the intelligent operation and management of shale gas fields.

[0098] The specific application areas or related products of this invention.

[0099] 1. Specific application areas This invention is primarily applied in the fields of energy and industrial internet technology, particularly for the development and operation management of oil and gas resources under complex operating conditions. Shale gas field large-scale production management: used to monitor and resolve abnormal operating conditions such as liquid accumulation, compressor shutdown, and pipeline blockage that frequently occur after shale gas wells enter the old well stage.

[0100] Multi-source heterogeneous data integration and maintenance: Applicable to industrial maintenance scenarios that require unified association of real-time streaming data (such as oil pressure, casing pressure, and flow), batch static data (such as well number ledgers), and application-generated data (such as alarm records and work order logs).

[0101] Unmanned / Minimally Manned Station Monitoring: By merging alarms and automatic dispatching logic, the frequency of manual inspections is reduced, and the timeliness of anomaly response at remote wellhead stations is improved.

[0102] Enterprise-level digital production scheduling: Provides energy companies' scheduling centers with a global operational view, enabling closed-loop management of production anomalies from discovery, retrieval, handling to feedback.

[0103] 2. Related product forms The system and method described in this invention can be converted into the following product forms: Energy Operation Intelligent Dashboard: A graphical user interface product that integrates hybrid search, visual monitoring, and work order status tracking.

[0104] Digital Work Order Management System (EAM / CMMS Upgrade): Intelligent work order dispatching software with automatic filling function, SLA dynamic monitoring, and deep linkage with real-time production data.

[0105] Industrial Big Data Analysis and Decision Support Terminal: An information data processing terminal deployed on a server or industrial PC, which provides alarm diagnosis and decision support by integrating a random forest AI model and a knowledge graph.

[0106] Intelligent production scheduling and command base: As the central module in the digital transformation of oil and gas field enterprises, it connects to the SCADA system at the bottom and supports business applications at the top, realizing semantic governance of data assets.

[0107] To verify the practical application effect of the "Hybrid Retrieval Cockpit and Work Order Linkage Method and System for Operation Management" of the present invention, this embodiment selects the production data of a shale gas development block from January to February 2026 for comparative analysis.

[0108] 1. Alarm accuracy and redundancy elimination effect According to the original monitoring log of the block, before the system of this invention was adopted, invalid alarms were frequently triggered during the production monitoring process. (1) Current status statistics: According to the production weekly summary records from January 26 to February 9, 2026, there were a large number of "early warning errors". The main reasons for the errors were "abnormal data sampling", "serious human operation interference" and "sensor reading deviation". (2) Implementation effect: After applying the alarm merging module of this invention, the redundant alarm rate was reduced by more than 80% through the three-element matching algorithm and the process switching history exclusion logic. For interference alarms caused by "human well shut-in" or "gas lift operation", the identification accuracy rate reached more than 92%, enabling maintenance personnel to accurately focus on the real abnormal liquid accumulation wells (such as effective abnormal wells such as Changning H23-5).

[0109] 2. Operation and maintenance response and work order processing efficiency (1) Automated work order filling: In the traditional mode, after an anomaly is discovered, maintenance personnel need to manually search the production ledger across systems and manually fill in the handling measures. Each work order takes an average of about 15 minutes. (2) Speed ​​improvement: This system automatically extracts anomaly characteristics through the data storage layer of the digital base and automatically injects them into the work order template, reducing the time for generating a single work order to less than 1 minute. (3) Improved response timeliness: Through automatic monitoring of the Service Level Agreement (SLA) process, the on-time response rate for high-priority anomalies (such as the flooded well Changning H11-3) has increased from 85% to 98%.

[0110] 3. Analysis of Production Recovery Cycle and Economic Benefits Based on continuous tracking data from the end of January to the beginning of February 2026, the comparative indicators before and after the implementation of this invention are as follows: Average time for anomaly tracing: reduced from 30 minutes to 5 minutes, an improvement of 83.3%.

[0111] The average production recovery cycle for a single well has been shortened from 2.5 days to 1.8 days, an improvement of 28.0%.

[0112] Emergency work order response timeliness rate: increased from 85% to 98%, an increase of 13.0%.

[0113] Economic benefit estimation: Taking 10 wells with an average daily liquid accumulation in the block as an example, based on production statistics, the average daily production loss per well is about 15,000 cubic meters. (1) Production recovery calculation: Since the average recovery cycle has been shortened by 0.7 days (from 2.5 days to 1.8 days), the production loss can be reduced by 0.7 days multiplied by 15,000 cubic meters / day = 10,500 cubic meters. (2) Annualized economic benefit: If the block experiences about 3,650 wells with similar liquid accumulation anomalies each year, the annual production loss can be recovered by 3,650 wells multiplied by 10,500 cubic meters / well = 38,325,000 cubic meters. (3) Production value estimation: Based on the natural gas market price of 2.0 yuan / cubic meter, it is estimated that the annual production value can be directly increased by about 76.65 million yuan.

[0114] 4. Closed-Loop Management and Continuous Optimization: Tracking records from January 30, 2026, show that after an anomaly was detected, well Ning 209H64-5 quickly achieved "resumed production" thanks to the historical handling solutions automatically recommended by the system's semantic association layer. The handling results were automatically written back to the digital platform, forming a complete closed loop of "data-alarm-work order-feedback-model optimization," significantly reducing long-term operation and maintenance costs.

[0115] A specific implementation scheme for signal data processing of a hybrid retrieval cockpit and work order linkage system for operation management.

[0116] This solution is designed for shale gas field operation and management scenarios. It focuses on the entire process of signal data acquisition, storage, processing, and linkage, and combines the functions of each module of the system to provide a practical signal data processing solution. This ensures accurate and efficient data processing, achieves seamless integration with work orders, meets the daily operation alarm handling and work order management needs of shale gas fields, and is adapted to the system architecture and fits the actual application scenarios.

[0117] Data module implementation: A distributed database architecture is adopted, deploying a real-time stream processing engine and a batch data processing module. It receives real-time streaming data (sampling frequency set to once per minute, covering core signals such as pressure, flow, and temperature) transmitted from wellheads and equipment sensors in the shale gas field, as well as batch static data (updated monthly for well conditions and equipment parameters) and application-generated data. Transaction consistency control is enabled to ensure complete data writing, version control retains data versions from the past 6 months, and a data lineage tracing tool records the entire data trajectory from acquisition to processing, facilitating anomaly tracing. Storage capacity is planned for 100 wells and 3 years of data volume to ensure data storage security.

[0118] Ontology module support: Based on Neo4j, a knowledge graph is built, and entity information of 100 shale gas wells, 300 core equipment, 20 production processes and 500 sensors is entered. The connection relationship between wells and equipment, process switching history (accurate to the minute) and alarm upstream and downstream causal chain are established to provide knowledge support for signal data association retrieval and interference alarm identification. The knowledge graph is updated once a month to ensure data timeliness.

[0119] The application module is implemented as follows: The alarm merging module extracts three types of operational features from the time-series signal: pressure fluctuation frequency (unit: times / hour), flow attenuation slope (unit: m³ / d·h), and production system switching flag. The random forest model is trained with 100,000 historical alarm data (accuracy ≥95%) and combined with the rule engine to complete the classification into four categories, including liquid accumulation. In the three-factor matching method, the time difference threshold is set to 5 minutes, the time window is set to 10 minutes before and after the alarm trigger, and the knowledge graph is linked to query process operation records to complete the redundancy merging and interference elimination.

[0120] The hybrid retrieval module adopts a "structured + semantic" fusion algorithm. The structured retrieval extracts the well number, abnormal value, etc. corresponding to the signal, while the semantic retrieval associates the device relationships and alarm causal chains in the knowledge graph. After fusion, the associated retrieval results are generated. The automatic work order generation module automatically generates work orders based on the effective alarm level (level 1-4) and the associated retrieval results, with preset response time limits (level 1 30 minutes, level 2 1 hour, level 3 2 hours, level 4 4 hours). The system automatically dispatches the work order to the corresponding maintenance personnel.

[0121] The Service Level Agreement (SLA) monitoring module records the time of each node in the work order process—receiving, processing, and closing—in real time, automatically triggering a secondary alarm upon timeout. Work order processing feedback data is back-synchronized to the data module to optimize random forest model parameters and adjust alarm merging thresholds, forming a closed-loop optimization. This solution requires no complex modifications to existing equipment and can be directly deployed, effectively improving signal data processing efficiency and work order linkage accuracy.

[0122] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0123] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A hybrid retrieval dashboard and work order linkage system for operations management, characterized in that, include: The data module is used to receive and store real-time streaming data, batch static data and application-generated data from shale gas fields, and supports transaction consistency control, version control and data lineage tracing. The ontology module is used to build a knowledge graph containing entities such as wells, equipment, processes, sensors, and alarms, and to establish the connection relationship between wells and equipment, the historical relationship of process switching, and the causal chain relationship between alarms and upstream and downstream. The application modules include a hybrid retrieval module, an alarm merging module, an automatic work order generation and management module, and a service level agreement process status monitoring module; The hybrid retrieval module integrates structured retrieval results with semantic retrieval results to generate related retrieval results. The alarm merging module extracts operational features from oil and gas production time-series data, inputs these features into a pre-trained machine learning classification model, and classifies, redundantly merges, and eliminates interference based on the rule engine and the machine learning classification model, outputting valid alarms. The automatic work order generation and management module generates work orders based on the valid alarms and the associated search results, and completes the work order dispatch according to the preset response time limit; The Service Level Agreement process status monitoring module records the time nodes of each stage of the work order and monitors the time limit. The associated search results are used in the calculation of work order priority and the generation of handling suggestions, thereby realizing the linkage between the search mechanism and the work order generation mechanism.

2. The system according to claim 1, characterized in that, The semantic retrieval module performs word segmentation and intent recognition on the input text, matches knowledge graph entities and relationships, associates multimodal data through depth-first or breadth-first graph traversal algorithms, and performs weighted fusion of structured retrieval results and semantic retrieval results. The weighted fusion calculates the comprehensive association score by multiplying the weight coefficient by the corresponding retrieval similarity and then summing them.

3. The system according to claim 1, characterized in that, The alarm merging module classifies alarms into liquid accumulation, equipment failure, human intervention, and data anomaly categories. It uses a three-factor matching method—well number, anomaly type, and trigger time difference less than a preset time difference threshold—to perform redundancy merging. It also queries process switching records within a preset time window before and after the alarm trigger time to identify interference alarms caused by manual well shut-in or gas lift operations and lower their priority. The machine learning classification model is a random forest model, and the operational features include pressure fluctuation frequency, flow attenuation slope, and production system switching flag.

4. The system according to claim 1, characterized in that, The automatic work order generation and management module calculates the work order priority based on the valid alarm category, the confidence level output by the random forest model, and the historical risk level in the associated retrieval results. The priority calculation method is to multiply the anomaly category weight by the sum of the model confidence level and the historical risk coefficient, and determine the urgency level according to the numerical value. The response time limit for emergency work orders is 4 hours, and the response time limit for ordinary work orders is 8 hours.

5. The system according to claim 1, characterized in that, The Service Level Agreement process status monitoring module records the time nodes for work order creation, dispatch, acceptance, start of handling, completion of handling, and acceptance. It calculates the time consumed in each stage and sends an early warning when the time consumed reaches the preset proportional threshold of the corresponding response time limit. If the response time limit is exceeded, the work order priority is automatically increased and a notification is sent.

6. An operation management method based on hybrid retrieval and work order linkage, characterized in that, Includes the following steps: Step 1: Receive and uniformly store real-time streaming data, batch static data, and application-generated data; Step two: Perform structured search or semantic search. Semantic search uses knowledge graphs to analyze intent and uses graph traversal algorithms to associate multimodal data to generate related search results. Step 3: Classify alarms, merge redundancies and eliminate interference, and sort and output valid alarms according to preset priority rules; Step 4: Generate a work order and automatically dispatch the work order based on the valid alarms and the associated search results; Step 5: Track the progress of work orders and monitor the service level agreement time limits; Step six: Write the processing results back to the data layer and use them for model and rule optimization; In step four, the associated search results are used in the calculation of work order priority and the generation of the handling suggestion field, thereby enabling the search results to drive the generation of work orders.

7. The method according to claim 6, characterized in that, In step three, when the time difference between the alarm trigger time and the time of manual well shut-in or gas lift operation is less than a preset time threshold, the alarm is marked as an interference alarm and its ranking weight is reduced.

8. The method according to claim 6, characterized in that, In step four, the similarity between the current alarm feature vector and the historical anomaly handling record vector is calculated through vectorized semantic matching. When the similarity is higher than a preset threshold, the corresponding historical handling steps and risk levels are extracted and mapped to the handling suggestion field of the new work order. At the same time, the work order priority is adjusted according to the historical risk level.

9. The method according to claim 6, characterized in that, In step five, the system calculates the difference between the current time and the work order creation time. When the difference reaches a preset proportional threshold of the response time limit, an early warning is triggered. When the response time limit is exceeded, the work order level is automatically upgraded.

10. The method according to claim 6, characterized in that, In step six, the random forest model is retrained and the feature weights are adjusted based on the newly added handling result data according to the preset update cycle. The handling time of work orders of different well types and anomaly types is statistically analyzed, the average value is calculated, and the response time limit of the corresponding scenario is dynamically adjusted according to the preset proportional threshold range above and below the average value.

Images