Abnormal resource occupation self-diagnosis and online recovery method for downhole power supply monitoring system

Abstract

The present application relates to the underground power supply monitoring system abnormal resource occupation self-diagnosis and online recovery method, technology includes collecting abnormal resource occupation operation fact data;According to the operation fact data, determine that the resource instance enters the non-output type resource occupation and keep judging, and freeze the corresponding abnormal field;Based on the abnormal field after freezing, trace the resource instance, holding unit, waiting condition, blocking source and unreleased reason, form an abnormal resource occupation responsibility chain, carry out responsibility closure judgment and determine the responsibility closure endpoint, the present application is for the internal operation state of the underground power supply monitoring platform, around the resource instance to establish input boundary identification, processing stage promotion identification and result boundary identification, so as to distinguish the accumulation type non-output type resource occupation and the stagnation type non-output type resource occupation, improve the pertinence and accuracy of abnormal identification.

Description

Technical Field

[0001] This invention relates to a method for self-diagnosis and online recovery of abnormal resource occupancy, specifically a method for self-diagnosis and online recovery of abnormal resource occupancy in an underground power supply monitoring system. Background Technology

[0002] Coal mine underground power supply monitoring systems typically handle the centralized access, processing, alarming, and dissemination of high-voltage feeder switches, substation equipment, protection units, and various telemetry and telecommunication links. Their operational continuity directly impacts underground power supply safety and production efficiency. Existing technologies for improving coal mine power supply monitoring platforms largely focus on deploying the sensing layer, optimizing the communication architecture, and enhancing electrical fault early warning capabilities. For example, Chinese patent document CN207408826U discloses a mine power monitoring system. This system alleviates the bandwidth consumption problem of RS485 serial communication on Ethernet by adding an embedded industrial control computer to the monitoring and communication substation to store, package, and forward protection unit data. The key focus is on addressing communication link congestion and transmission problems. For example, Chinese patent document CN121035937A discloses an IoT-based intelligent monitoring system for coal mine power supply. This solution revolves around a closed loop of edge sensing, edge analysis, cloud modeling, and fault early warning, focusing on improving overload prediction, leakage location, and energy consumption optimization capabilities through multi-type sensors, edge computing, and multi-source data fusion. It is evident that existing coal mine power supply monitoring technologies primarily focus on electrical quantity acquisition, equipment health assessment, network transmission, and monitoring decision efficiency. However, they lack dedicated and fine-grained diagnostic and recovery mechanisms for software runtime issues such as abnormal resource instance occupancy, thread waiting coupling, and result publication stagnation within the monitoring platform's internal processing links.

[0003] Furthermore, underground power supply monitoring systems differ from general discrete detection systems. Their operation often involves multiple processing stages, including event access, status resolution, alarm aggregation, result publishing, cache holding, queue scheduling, and lock contention. This involves both continuously flowing real-time business data and cross-thread, cross-stage resource occupancy relationships. While existing technologies can detect and handle external manifestations such as abnormal equipment operating parameters, abnormal power grid status, and network bandwidth bottlenecks, they are typically limited to addressing internal non-productive occupancy phenomena—where resources are continuously occupied but processing stages and result boundaries do not advance—through empirical threshold alarms, manual log inspection, overall service restarts, or switching to backup nodes. These methods struggle to distinguish between accumulation anomalies (where input continues to advance but results do not) and stagnation anomalies (where neither input nor results advance). Furthermore, they fail to identify, without disrupting the operational environment, whether the release condition has been met but not released, or whether the blockage is a self-blocking loop caused by the holding unit itself. Especially in scenarios where underground power supply monitoring platforms need to continuously receive on-site status and maintain uninterrupted main monitoring functions, simply using shutdown analysis or global restart methods can easily lead to problems such as loss of abnormal on-site data, broken responsibility chains, excessively large recovery scope, and recurrence after recovery. These shortcomings are not the focus of existing coal mine power supply monitoring literature, and their published content does not provide solutions for freezing, tracing, responsibility closure, and minimum boundary recovery for abnormal resource instance-level occupancy.

[0004] Furthermore, most existing Chinese patent documents related to fault recovery focus on equipment failure, host failure, or node takeover. For example, Chinese patent document CN115442571A discloses a fault recovery method and apparatus for a video surveillance system. Its core is that when a second machine fails, the first machine takes over the corresponding services based on available resources and load, which belongs to the fault takeover and load adjustment approach based on available host resources. This type of solution can improve service continuity, but it focuses on the transport migration after machine-level failures. It does not perform freeze analysis on the state maintenance, direct holding relationship, single-hop waiting relationship, blocking source, and non-release reason of a single resource instance in the running state, nor does it involve generating a read-only frozen copy of the abnormal scene, performing write deflection on subsequent write operations, reverse verification of resource release prerequisites, determining the responsibility closure endpoint, determining the cut-off boundary through virtual culling verification, and performing link break verification after recovery. Summary of the Invention

[0005] The purpose of this invention is to provide a method for self-diagnosis and online recovery of abnormal resource occupancy in an underground power supply monitoring system, thereby solving some of the drawbacks and shortcomings pointed out in the background art.

[0006] The technical solutions adopted by the present invention to solve the above-mentioned technical problems include:

[0007] Collect abnormal resource usage data; based on the operational data, determine whether a resource instance has entered a non-output resource usage retention state, and freeze the corresponding abnormal situation.

[0008] Based on the abnormal site traceability after freezing, resource instances, holding units, waiting conditions, sources of blockage and reasons for non-release are traced to form an abnormal resource occupation responsibility chain, and responsibility closure judgment is performed to determine the responsibility closure endpoint.

[0009] The cut-off boundary is determined based on the responsibility closure endpoint and online recovery is performed; after recovery, the chain of responsibility is checked for breakage based on the frozen abnormal scene, and the cut-off boundary is re-determined if the check fails.

[0010] Furthermore, an input boundary identifier, a processing phase advancement identifier, and a result boundary identifier are set for the resource instance. During the continuous occupation of the resource instance, if the processing phase advancement identifier remains unchanged and the result boundary identifier does not advance, it is determined to enter the non-output type resource occupation maintenance determination. After the maintenance determination is confirmed, if the input boundary identifier advances but the result boundary identifier does not advance, it is determined to be the accumulation type non-output type resource occupation. If neither the input boundary identifier nor the result boundary identifier advances, it is determined to be the stagnation type non-output type resource occupation.

[0011] Furthermore, taking the resource instance as the center, abnormal situation units are extracted along the direct holding relationship and single-hop waiting relationship to generate read-only frozen copies; write bias is performed on subsequent write operations involving the abnormal situation units so that subsequent write operations act on running data.

[0012] Furthermore, the resource release prerequisites of each holding unit in the frozen abnormal scene are reverse-verified; when the resource release prerequisites have been met but the resources are still held, the corresponding holding unit is determined to be a unit with missing release responsibility; when the source of the blockage that the holding unit is waiting for is generated or maintained by itself, and the removal of the source of the blockage still depends on the holding unit itself, the holding unit is determined to be a self-blocking responsibility unit; the unit with missing release responsibility or the self-blocking responsibility unit is used as the responsibility closure endpoint, and the cut-off boundary is determined.

[0013] Furthermore, a unique association identifier is set for the resource instance, and the input boundary identifier, processing stage advancement identifier, and result boundary identifier are bound to the unique association identifier; wherein, the input boundary identifier is updated when entering the processing link, the processing stage advancement identifier is updated when crossing a preset processing stage, and the result boundary identifier is updated when a submission result is formed or a result is published.

[0014] Furthermore, when the processing stage advancement indicator remains unchanged and the result boundary indicator does not advance for the first time, a retention determination is initiated for the resource instance; if the result boundary indicator advances during the retention determination period, the retention determination for non-output resource occupation is revoked; if the result boundary indicator still does not advance, non-output resource occupation is confirmed; wherein, if the input boundary indicator advances, it is confirmed as accumulation-type non-output resource occupation, and if the input boundary indicator does not advance, it is confirmed as stagnation-type non-output resource occupation.

[0015] Furthermore, starting from the resource retention state in the frozen abnormal scene, the process traces back along the state items on which the resource release depends to the release trigger state; when all state items corresponding to the resource release prerequisites have been satisfied and the resource is still retained, the corresponding holding unit is determined to be the release missing responsibility unit.

[0016] Furthermore, in the frozen abnormal scene, if the source of the blockage is generated or maintained by the holding unit before entering the waiting state, and the release of the source of the blockage still depends on the holding unit itself, then the holding unit is determined to be the self-blocking responsibility unit.

[0017] Furthermore, taking the responsibility closure endpoint as the center, tracing outward along the resource holding relationship and waiting relationship, the associated holding units necessary to maintain the resource holding state corresponding to the responsibility closure endpoint are defined as the cut-off boundary, and other holding units that do not affect the resource holding state are excluded from the cut-off boundary.

[0018] Furthermore, a virtual removal check is performed on each associated holding unit obtained by tracing along the resource holding relationship and the waiting relationship; after any associated holding unit is virtually removed, if the resource holding state corresponding to the responsibility closure endpoint is released, the associated holding unit is retained at the cut-off boundary; if the resource holding state remains unchanged, the associated holding unit is excluded from the cut-off boundary.

[0019] Compared with existing technologies, this invention does not stop at the level of electrical quantity anomaly detection, communication optimization or whole machine takeover, but focuses on the internal operation of the underground power supply monitoring platform. It establishes input boundary markers, processing stage advancement markers and result boundary markers around resource instances, thereby distinguishing between accumulated non-productive resource occupation and stagnant non-productive resource occupation, and improving the pertinence and accuracy of anomaly identification.

[0020] This invention freezes the abnormal scene and generates a read-only frozen copy, while performing write deflection on subsequent write operations. This allows the traceability of responsibility to be preserved without interrupting the main monitoring functions. Furthermore, by combining release prerequisite reverse verification, self-blocking identification, responsibility closure endpoint determination, virtual removal verification, and chain break verification, online recovery under the minimum removal boundary is achieved. This helps to reduce the scope of recovery disturbance, reduce the risk of global restart, and improve the verifiability of recovery results. Attached Figure Description

[0021] Figure 1 This is a flowchart of the online diagnosis and recovery method for abnormal resource usage according to the present invention.

[0022] Figure 2 This is a multi-indicator collaborative evolution diagram of the abnormal resource occupancy formation stage in Example 1.

[0023] Figure 3 This is a comparison diagram showing the progress of the input boundary and the result boundary within the decision window in Example 1.

[0024] Figure 4 This is a comparison chart of key indicators during the freeze, write deflection, and online recovery phases in Example 1.

[0025] Figure 5 This is a timing diagram of key events and resource status during the process of abnormal resource occupation from freezing to recovery in Example 2.

[0026] Figure 6 This is a diagram showing the relationship between the necessity index of the responsibility unit boundary and the determination of the minimum cut-off boundary in Embodiment 2.

[0027] Figure 7 This is a comparison chart of key operational indicators before and after online recovery in Example 2.

[0028] Figure 8 This is a matrix diagram showing the relationship between the holding unit and key resources under the frozen snapshot in Embodiment 2. Detailed Implementation

[0029] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0030] Combined with appendix Figure 1As shown in this embodiment, during the operation of the underground power supply monitoring system, there are resource instances such as task threads, communication sessions, cache queues, connection handles, lock objects, and transaction contexts. Each resource instance forms associative operational fact data during the execution of acquisition, transmission, processing, and control. The operational fact data includes information such as resource instance identifier, occupancy status, holding unit, processing stage, result output status, waiting conditions, and blocking sources. The system continuously collects and records data according to a unified time base through a diagnostic agent to achieve time-series alignment and correlation analysis of different resource states.

[0031] After acquiring operational data, the system performs a non-productive resource occupancy determination on resource instances. Specifically, this is achieved by analyzing the progress of processing phases and the output status of resources within a continuous cycle, combined with resource occupancy duration and waiting status changes. When a resource is continuously occupied without progress in the processing phase and without generating valid output results, it is determined to enter a non-productive resource occupancy hold determination. To avoid false positives, the system continues to monitor the result status during the hold determination period. If progress is observed, the determination is revoked; otherwise, the abnormal state is confirmed, and the corresponding abnormal state is immediately frozen. The freeze focuses on the abnormal resource instance, extracting only its relevant holding and waiting relationships to form a read-only snapshot of the state. Subsequent write operations are redirected to running data, thus preserving the true state at the time of the anomaly without affecting system operation.

[0032] After freezing, the system traces the chain of responsibility based on the abnormal situation. Starting with the abnormal resource instance, it identifies its current holding unit and obtains the unit's execution status, trigger source, and waiting conditions. Then, it traces the source of the blockage and its maintaining unit level by level along the waiting relationship. Simultaneously, it performs reverse verification of the reason for the resource not being released, checking whether the relevant status is met item by item according to the resource release conditions. If the release conditions are met but the resource is still not released, the corresponding holding unit is identified as the unit responsible for the missing release. If the source of the blockage is generated or maintained by the holding unit itself and is no longer dependent on itself, it is identified as a self-blocking unit responsible. The system connects the above elements according to causal relationships to form an abnormal resource occupancy responsibility chain, used to characterize the path of abnormal formation and maintenance.

[0033] After the chain of responsibility is formed, the system performs a responsibility closure determination. By continuously tracing back along resource dependencies, the system locates the responsibility convergence node that can explain the abnormal state without further expansion, and identifies this node as the responsibility closure endpoint. Subsequently, a cutoff boundary is determined around this endpoint, that is, the associated units necessary to maintain the abnormal resource occupation in the chain of responsibility are screened. For each candidate unit, the system judges its impact on the abnormal state through virtual elimination. If the abnormality is resolved after removal, it is included in the cutoff boundary; otherwise, it is excluded, thereby obtaining the cutoff set with the smallest impact range.

[0034] After determining the cutoff boundary, the system performs an online recovery operation. The recovery process involves releasing resources, clearing blockages, dismantling waiting relationships, and cleaning up the state of units within the boundary. Task termination or rescheduling may be performed as needed to restore the abnormal resources to an available state. The recovery operation only applies to units within the cutoff boundary, thus avoiding impact on other normally operating units and achieving localized repair of local faults.

[0035] After recovery, the system performs a chain-of-responsibility check based on the frozen anomaly scenario. By comparing the responsibility chain structure before recovery with the operational status after recovery, it verifies whether the anomaly resource has been freed from its original holding relationship, whether the source of the blockage has been eliminated, and whether the resource release conditions have been restored to validity. If the anomaly holding relationship corresponding to the responsibility chain has been severed, the recovery is considered successful; if there are still unresolved critical relationships or new holding paths have formed, the check is considered a failure. At this point, the system re-analyzes the responsibility relationships based on the original frozen scenario, expands the cutoff boundary, and performs recovery again until the anomaly resource occupation is completely released.

[0036] In this implementation, to achieve refined identification and classification of resource instance occupancy behavior, the system establishes a multi-dimensional advancement identifier system for each resource instance, which serves as the core basis for determining non-productive resource occupancy. Specifically, when a resource instance is created or enters the scheduling system, it is assigned a unique association identifier to uniquely identify the lifecycle of the resource instance and its associated processing links globally. The unique association identifier is bound to the resource instance's input boundary identifier, processing stage advancement identifier, and result boundary identifier, enabling the three types of identifiers to establish a correspondence with the running trajectory of the same resource instance at any given time, thereby supporting unified analysis across stages and modules.

[0037] Input boundary identifiers describe the input progress status of a resource instance as it enters the processing chain. These identifiers are updated when the resource instance is introduced into the business processing flow, such as when data acquisition is completed and the instance enters the processing queue, when a communication message enters the parsing module, or when a control command enters the execution chain, to reflect that new input has been received. Processing stage progress identifiers describe the stage progress of a resource instance within its internal processing flow. The system updates these identifiers according to predefined processing stage divisions when the resource instance crosses stage nodes, such as when it moves from the parsing stage to the calculation stage or from the decision stage to the execution stage, to characterize whether the processing process continues to progress. Result boundary identifiers describe whether a resource instance produces a valid output result. These identifiers are updated when a submission result is formed, a result is published, or a control output is completed, to reflect whether the resource instance has produced an externally visible processing result. Through the coordinated recording of these three types of identifiers, the entire process progress status of a resource instance from input to processing to output can be fully depicted.

[0038] During the continuous occupation of resource instances, the system periodically monitors the three types of identifiers mentioned above. When the processing stage progress identifier remains unchanged and the result boundary identifier does not progress within multiple consecutive sampling periods, it indicates that although the resource instance is still occupied, its internal processing has not continued and no effective output has been generated. Based on this, the system triggers a non-output resource occupation maintenance judgment. To avoid misjudgment caused by short delays or normal long processing processes, only the maintenance judgment process is initiated when this condition is first met, and the changes in the result boundary identifier are continuously tracked in subsequent observation windows.

[0039] During the hold period, if the result boundary marker is detected to advance, it indicates that the resource instance has generated valid output, breaking the previous stagnant state. The system then cancels the non-output resource occupancy hold decision and resumes normal monitoring. If the result boundary marker does not advance throughout the entire hold period, the resource instance is confirmed to have entered a non-output resource occupancy state. After this confirmation, the system further subdivides the anomaly type based on changes in the input boundary marker. If the input boundary marker continues to advance during the confirmation period, but the result boundary marker does not, it indicates that new inputs are continuously entering but not being converted into output results, creating a backlog within the resource instance. The system classifies this as a backlog-type non-output resource occupancy. If neither the input boundary marker nor the result boundary marker advances during the confirmation period, it indicates that neither new inputs are entering nor output results are being generated, and the resource instance is in a stagnant state. The system classifies this as a stagnant non-output resource occupancy.

[0040] In this implementation, after confirming that a resource instance has entered a non-productive resource occupancy state, the system performs a freeze process on the corresponding abnormal situation to maintain a stable analytical foundation without interrupting system operation. The freeze process unfolds centered on the target resource instance. First, it identifies the holding unit that directly holds the resource instance and establishes a holding relationship mapping between the resource instance and the holding unit. Then, it extracts the waiting state starting from the holding unit, determining the waiting conditions it depends on and the corresponding blocking source unit. To control the data scale and ensure the integrity of key relationships, the system only extracts the upper-level related units along the single-hop waiting relationship, thereby forming a set of abnormal situation units containing resource instances, holding units, waiting conditions, and blocking sources.

[0041] Based on the aforementioned set of abnormal situation units, the system constructs a read-only frozen copy. This frozen copy fully maps the relevant states at the time of the anomaly, records the state information, dependencies, and resource usage of each unit, and remains unchanged after its generation, thus forming a stable and consistent snapshot of the abnormal situation. This snapshot can be used for subsequent chain of responsibility tracing and anomaly analysis, preventing the situation information from being overwritten or altered due to continuous system operation.

[0042] While generating the frozen copy, the system controls the data access path and performs write deflection processing on subsequent write operations involving abnormal field units. Specifically, when a write operation is detected to be targeting a frozen unit, the write request is redirected to the running data area for processing, without affecting the frozen copy, thus keeping the frozen copy unchanged. Read operations, on the other hand, selectively access either the frozen copy or the running data as needed, thereby balancing anomaly backtracking analysis with real-time system operation.

[0043] By combining the frozen copy mechanism with the write deflection mechanism, the system achieves logical isolation and data stability in abnormal situations. The frozen copy ensures the traceability of abnormal states, while the write deflection ensures uninterrupted subsequent business processing. Together, they provide a reliable foundation for subsequent chain of responsibility analysis and online recovery, while maintaining the continuous operation capability of the underground power supply monitoring system.

[0044] In this implementation, to obtain stable anomaly analysis data without interrupting system operation, the corresponding anomaly scenario is frozen after confirming that a resource instance has entered a non-productive resource occupancy state. The freezing process unfolds centered on the target resource instance. First, the holding unit that directly holds the resource instance is identified, and a direct holding relationship is established between the resource and the holding unit. Subsequently, the current waiting state of the holding unit is extracted, the corresponding waiting conditions and blocking sources are identified, and the directly related upper-level dependent units are obtained along the single-hop waiting relationship, thereby forming a set of anomaly scenario units containing the resource instance, holding unit, waiting conditions, and blocking sources.

[0045] After obtaining the aforementioned set of abnormal situation units, the system constructs a read-only frozen copy. This copy fully maps the state, dependencies, and resource usage information of each unit at the time of the anomaly. Once generated, this copy remains unchanged, serving as a stable snapshot of the abnormal situation for subsequent analysis. Simultaneously, the system controls access to runtime data. Subsequent write operations involving abnormal situation units are redirected to the runtime data area, thus preventing any impact on the frozen copy.

[0046] In this implementation, after freezing the abnormal scene and obtaining a stable snapshot, the system performs responsibility unit identification and responsibility closure analysis on the frozen abnormal scene. First, it performs reverse verification of the resource release prerequisites for each holding unit. Specifically, starting from the current resource holding state, it traces back each of the state conditions it depends on according to the resource release rules, including execution completion state, dependent event satisfaction state, and control trigger state, and verifies the actual values ​​of these state items in the frozen scene. When all state items corresponding to the resource release prerequisites have been satisfied, but the resource is still in a held state, it indicates that the holding unit has not performed the required release action, and the system determines it as a unit with missing release responsibility.

[0047] While performing release prerequisite verification, the system further analyzes the waiting relationships and blocking sources of each holding unit. For holding units in a waiting state, the system determines whether the blocking source was generated or maintained by the holding unit before entering the waiting state, and further determines whether the removal of the blocking source still depends on the subsequent execution of the holding unit itself. If the above conditions are met, it indicates that the unit has formed a self-sustaining blocking structure, and it is both the generator of the blocking and the dependent object for unblocking. Based on this, the system determines the holding unit as a self-blocking responsible unit.

[0048] After identifying missing or self-blocking responsibility units, the system designates them as responsibility closure endpoints. A responsibility closure endpoint represents a convergence node where the current abnormal resource holding state can be explained without further outward expansion. Based on this endpoint, the system determines the cutoff boundary around it. Specifically, centered on the responsibility closure endpoint, the system traces all associated holding units outward along resource holding and waiting relationships, and selects associated units that have a practical role in maintaining the current resource holding state.

[0049] To minimize the determination of the cutoff boundary, the system performs a virtual removal check for each candidate associated unit. Specifically, by temporarily removing the influence of the associated unit on the resource preservation relationship in the logical model, the system re-evaluates whether the resource preservation state corresponding to the responsibility closure endpoint still holds. If the resource preservation state is lifted after the virtual removal of the unit, it indicates that the unit is necessary for the continued existence of the anomaly and should be included in the cutoff boundary; if the resource preservation state still exists, it indicates that the unit is not a critical dependency and should be excluded from the cutoff boundary. By performing the above check on all associated units, the system ultimately determines a set of minimum necessary units constituting the cutoff boundary.

[0050] Through the aforementioned reverse verification, responsibility unit identification, and boundary screening process, the system can accurately locate the abnormal responsibility convergence point based on the frozen site, and determine the set of objects requiring intervention within the smallest range, providing a precise basis for subsequent online recovery, while avoiding interference with irrelevant units, thereby improving system recovery efficiency and operational stability.

[0051] Example 1:

[0052] In this embodiment, the underground power supply monitoring system is deployed in the surface dispatch room of a coal mine and is connected to the central substation, the East Third Mining Area substation, feeder switches, protection devices, and telemetry acquisition links via an industrial Ethernet network. During the night shift, the system continuously receives underground high-voltage switch status data, protection action data, and telemetry voltage and current data, and processes and publishes them sequentially through the event access thread T_in_EVT, the status parsing thread T_parse_ST, the alarm aggregation thread T_alarm_AG, and the result publishing thread T_pub_RES. To facilitate resource-level monitoring, the system configures a unique association identifier RI-2025-0414-17 for the publishing queue resource instance in the result publishing link, and binds the input boundary identifier, processing stage advancement identifier, and result boundary identifier to this unique association identifier. The input boundary identifier is used to record the latest sequence number of the message entering the processing link, the processing stage advancement identifier is used to record the latest stage number crossing the preset processing stage, and the result boundary identifier is used to record the latest result number when a submission result or a published result is formed.

[0053] Specifically, from 22:10:00 to 22:18:11, the power supply monitoring services of the central substation and the East Third Mining Area substation operated normally. The event access thread T_in_EVT updated the input boundary identifier once for each telemetry message received. The state parsing thread T_parse_ST advanced the processing stage advancement identifier when completing message decoding, state merging, and alarm association. The result publishing thread T_pub_RES advanced the result boundary identifier after forming alarm results or state publishing results. During this period, the publishing queue depth corresponding to RI-2025-0414-17 remained stable between 18 and 33, and the resource occupation time remained between 0.8 and 1.4 seconds. The continuous advancement of the result publishing time indicates that although this resource instance was continuously reused, it could produce effective output and did not constitute abnormal resource occupation.

[0054] At 22:18:12, a 7-second transient voltage fluctuation occurred on the 6kV feeder line in the East Third Mining Area. From 22:18:12 to 22:18:39, the telemetry acquisition device concentrated on sending voltage, current, and protection self-test messages, causing the event access rate to increase from 26 messages per second to 124 messages per second. The resource monitoring module M_res_watch first detected the abnormal resource usage of RI-2025-0414-17 on the result publication link. Its collected operational data included resource usage duration, number of input messages, number of outgoing results, number of stage advancements in the last 10 seconds, thread wait time, queue depth, CPU usage, memory usage, lock holding time, and result publication timestamp. At 22:18:40, M_res_watch found that the processing stage advancement flag remained unchanged for 10 consecutive seconds, and the result boundary flag also did not advance, thus triggering the first hold judgment and entering a 20-second hold judgment window.

[0055] Table 1: Snapshot of Operational Fact Data During the Anomaly Occurrence Phase

[0056]

[0057] Table 1 provides a snapshot of key operational facts during the anomaly formation phase. The coordinated changes in resource consumption duration, queue depth, thread wait time, and phase advancement count can be combined with... Figure 2 understand.

[0058] In this embodiment, the input boundary identifier is denoted as I, the processing stage advancement identifier as P, and the result boundary identifier as O, and these are bound and stored with RI-2025-0414-17. At 22:18:40, the three identifiers corresponding to this resource instance are I-18420, P-346, and O-9120, respectively. After entering the hold determination window, the event access thread T_in_EVT continues to read packets from the downhole acquisition link, so the input boundary identifier is continuously updated; however, the status parsing thread T_parse_ST does not cross the preset processing stage again, so the processing stage advancement identifier remains P-346; the result publishing thread T_pub_RES does not generate new submission results or publishing results, so the result boundary identifier remains O-9120. By 22:19:00, the three identifiers have changed to I-18916, P-346, and O-9120, and the system accordingly cancels the normal occupancy assumption and confirms that RI-2025-0414-17 has entered non-productive resource occupancy. If O advances within the 20-second window, the hold decision should be revoked. However, in this embodiment, O never advances, so the hold decision is valid.

[0059] To ensure the verifiability of the judgment process, this embodiment further performs least-squares linear fitting on the observation window from 22:18:40 to 22:19:00. The selected time sequences are 0 seconds, 5 seconds, 10 seconds, 15 seconds, and 20 seconds, corresponding to input boundary sequences of 18420, 18545, 18668, 18792, and 18916, and corresponding to result boundary sequences of 9120, 9120, 9120, and 9120. The slope is calculated using the following formula:

[0060] in, Indicates the sampling time within the observation window. Indicates the corresponding boundary sequence value. Represents the mean over time. This represents the mean of the boundary sequence.

[0061]

[0062] For the input boundary sequence,

[0063]

[0064] For the result boundary sequence,

[0065]

[0066] This indicates that maintaining the input boundary markers within the decision window at a rate of 24.78 markers per second, while the result boundary markers have a slope of 0, satisfies the characteristic of continuous input advancement and no result advancement. Maintaining the difference in advancement between the input and result boundaries within the decision window can be combined with... Figure 3 understand.

[0067] Preferably, the system further classifies non-output resource occupation into accumulation type and stagnation type. If neither the input boundary identifier nor the result boundary identifier advances, it should be identified as stagnation type non-output resource occupation. In this embodiment, from 22:18:40 to 22:19:00, the input boundary identifier advances from I-18420 to I-18916, and 496 new packets enter the processing link. The result boundary identifier remains unchanged at O-9120, so it should be identified as accumulation type non-output resource occupation. At the same time, the queue depth increases from 267 to 511, and the thread waiting time increases from 382ms to 911ms, further indicating that new inputs continue to flow in without corresponding outputs. The abnormal behavior is that accumulation occurs on the result publishing side, rather than stagnation caused by a stoppage at the source.

[0068] At 22:19:03, the system, centered on RI-2025-0414-17, extracts the abnormal situation units along the direct holding relationship and single-hop wait relationship. These units include the cache holding unit CBUF-07, the publishing queue Q_RES_07, the publishing lock resource L_PUB_04, the result publishing thread T_pub_RES, and the alarm aggregation thread T_alarm_AG which has a single-hop wait relationship with it. A frozen copy with version number SNAP-221903 is then generated. This frozen copy only stores the queue head and tail pointers, slot occupancy bitmap, lock ownership information, thread state, stage number, and result number at the time of 22:19:03. All attributes are set to read-only, disallowing subsequent threads from overwriting or writing to this copy. Starting at 22:19:03, subsequent write operations of the event access thread T_in_EVT and the status resolution thread T_parse_ST uniformly perform write deflection. Newly arriving packets are no longer written to Q_RES_07 and CBUF-07 corresponding to the frozen replicas, but are instead written to Q_RES_07_RUN and CBUF-07_RUN in the running data area. Since the frozen replicas remain read-only, the resource holding state and progress state at the time of the anomaly are fully preserved, thus not compromising the on-site consistency required for subsequent chain of responsibility tracing. Furthermore, since the write deflection only applies to the running data, new packets can still continue to enter the processing chain, thus preventing the interruption of the main downhole monitoring functions.

[0069] Table 2: Effectiveness data during the freeze and recovery phases

[0070]

[0071] Table 2 compares the performance data during the freeze, write deflection, and online recovery phases. The 121 records per second during the write deflection period after freeze correspond to the short-term peak input rate after triggering, which can be compared with… Figure 4 The contrast between the stages shown should be understood in conjunction with the overall picture.

[0072] After completing the frozen replica establishment and write offset, the system performed online recovery on the result publishing link. At 22:28:06, the recovery module only performed partial cutoff and reconstruction on the publishing sub-link containing RI-2025-0414-17. Recovery was completed at 22:28:13, and the result publishing thread T_pub_RES was reconnected. From 22:19:03 to 22:28:06, the running data area received a total of 1236 new messages, all of which were temporarily stored by Q_RES_07_RUN and not written to the frozen replica. After recovery, the result boundary marker continued to advance from O-9120 to O-10384, the average publishing latency decreased from 8.7 seconds to 1.3 seconds, the queue depth decreased to below 39 within 5 minutes, CPU utilization decreased from 73% to 48%, and lock holding time decreased from 1246ms to 72ms. After recovery, the system performs a link break check based on the frozen anomaly scenario, confirming that the persistence relationship on the original anomaly link no longer exists, and that the running link does not exhibit the combined characteristics of processing stage stagnation and result boundary stagnation again. Changes in key indicators during the freeze, write deflection, and online recovery phases can be combined with... Figure 4 understand.

[0073] As can be seen from the above implementation process, this embodiment continuously collects operational factual data on abnormal resource usage based on RI-2025-0414-17, and achieves timely identification of non-output resource usage by binding and updating input boundary markers, processing stage progress markers, and result boundary markers. In particular, within the decision window, the least squares fitting data obtained... and It can directly support the confirmation of the occupation of accumulated non-productive resources, thus distinguishing between accumulated and stagnant resources. Furthermore, by establishing a read-only frozen copy and performing write deflection on subsequent write operations, the system maintains the continued flow of running data while preserving the traceability of abnormal situations, avoiding downhole monitoring gaps caused by full-link write shutdowns.

[0074] Example 2:

[0075] The system has completed the abnormal site freeze based on operational fact data. The next phase involves responsibility chain tracing, responsibility closure determination, boundary clearing, and online recovery. In this embodiment, a coal mine underground power supply monitoring system simultaneously handles high-voltage feeder switch status uploads, protection action playback, load forecast result publication, and alarm linkage services during the day shift handover period. From 08:15:08, two feeder switches in the east wing mining area continuously uploaded status change frames. The protection action playback task overlapped with the load forecast publication task, increasing the load on the result publication link. At 08:15:19, the diagnostic module discovered that the publication token resource R_TK_09 had remained unreleased for 17.6 seconds, the result publication time delay increased to 9.8 seconds, alarm backlog reached 286, the publication queue depth reached 302, and CPU usage rose to 78%. Based on this, the starting point for this continuous period is approximately 08:15:01.4. At 08:15:26, the system froze the abnormal scene, creating snapshot SNAP-2026-0414-081526, and locked the holding units directly related to the target resource: U1 Event Receiving Unit, U2 Status Aggregation Unit, U3 Result Submission Unit, U4 Alarm Issuance Unit, and U5 Recovery Coordination Unit.

[0076] In this embodiment, the key event nodes from anomaly triggering and site freezing to partial recovery completion, and their correspondence with resource status, can be combined with... Figure 5 understand.

[0077] After freezing, the system traces outwards from the holding state of the release token resource R_TK_09. Snapshots show that U4 currently holds R_TK_09 and queue slot resource R_QS_17, and is waiting for U3 to release the mutex lock resource R_MX_02 before performing alarm tail cleanup and returning the release token. U3 currently holds R_MX_02, its thread state is "committed and held," and is waiting for U4 to release R_TK_09 to write back the commit confirmation number. U2 waits for U3 to return a commit confirmation. U1 only continues to receive status frames and write incremental data to the bypass runtime area, no longer participating in this freeze chain. U5 is only used for subsequent local recovery coordination and does not participate in resource holding. This forms a bidirectional waiting chain with U4 and U3 at its core, and the system continues to identify the responsibility type of this chain.

[0078] The correspondence between each holding unit, key resources, and waiting objects under the frozen snapshot can be further combined. Figure 8 understand.

[0079] Table 3: Snapshots of the abnormal scene after freezing

[0080]

[0081] Table 3 lists the abnormal snapshots after freezing, used to support the determination of the relationship between each holding unit and the status of critical resources at the time of freezing. In this embodiment, the system first performs a reverse verification of the resource release prerequisite on U3. The verification starts from the holding state of R_MX_02 of U3, and traces back to the release trigger state along the state items on which the release depends. The system reads the frozen snapshot and knows that at 08:15:24, the submission frame PF-118722 has been written to the publication buffer and the status is sent; at 08:15:24 618ms, the submission sequence number CI-552981 has been written to disk; at 08:15:24 625ms, the result submission completion flag S_done has been set to 1. All three of the above conditions were met. According to the normal procedure, U3 should have released R_MX_02 around 08:15:24 (approximately 630ms). However, the lock was still held until the freeze time at 08:15:26, an extra 1.37 seconds compared to the expected release time. Furthermore, the lock was not released until 08:16:04 before online recovery, accumulating an extra holding time of approximately 39.37 seconds. Therefore, U3's resource release prerequisites were met, yet the resource was still held. Thus, U3 can be identified as the unit responsible for the missing release. This judgment is not based on empirical assumptions about thread timeouts, but rather on a step-by-step verification of the traceable release trigger states in the frozen state, demonstrating clear engineering verifiability.

[0082] Subsequently, the system performed a self-blocking verification on U4. The frozen snapshot shows that before entering the waiting state at 08:15:14, U4 first created and maintained the alarm publication completion flag F_pub_42, and marked queue slot R_QS_17 as occupied. According to the system design, the clearing of F_pub_42 is completed by the alarm tail cleanup process executed by U4 itself, and this cleanup process can only run after U4 regains the conditions to submit the link access. At this time, U4 holds R_TK_09 on one hand, and waits for U3 to release R_MX_02 on the other hand before continuing to perform cleanup, thus keeping F_pub_42 indefinitely. Since this blocking source is generated and maintained by U4 before entering the waiting state, and its release still depends on U4's continued operation, U4 constitutes a self-blocking responsible unit. In other words, U4 is not unidirectionally blocked by external units, but rather first forms internal blocking conditions, and then is constrained in reverse by the conditions it maintains during the waiting process.

[0083] The bidirectional waiting of U3 and U4 in the frozen link, the self-blocking sources of U4, and its responsibility closure trend can be combined. Figure 6 understand.

[0084] To further verify whether the chain of responsibility forms a closed loop at a certain unit, the system establishes a Boolean adjacency matrix of holding units and waiting relationships. If a unit is waiting for another unit, the corresponding position is set to 1; otherwise, it is set to 0. Since U1 only continues to receive status frames and no longer participates in freezing the link, the corresponding row of U1 is set to 0 when constructing the adjacency matrix. Based on the relationships after freezing, we can obtain...

[0085] Considering that U1 will only continue to receive status frames and will no longer participate in freezing the link, the row corresponding to U1 will be set to zero when constructing the adjacency matrix, and only the waiting relationship between U2, U3 and U4 will be retained.

[0086]

[0087] Further Boolean multiplication yields...

[0088]

[0089] in and This indicates that both U3 and U4 have closed-loop paths that return to themselves via two hops. However, in this embodiment, the target resource holding state is the continuous holding of R_TK_09 by U4. Tracing outward along this holding state, it can be seen that the root cause of U4's wait eventually falls back to F_pub_42, which it generates and maintains itself, and the removal of this flag still depends on U4 itself. Therefore, the chain of responsibility is closed at U4, and it is no longer necessary to extend outward to explain why the target resource is continuously held. Therefore, U4 is determined as the end point of the responsibility closure, while U3, as the release missing responsibility unit on the same chain, enters the subsequent cut-off boundary determination.

[0090] After determining U4 as the endpoint of the responsibility closure, the system traces outward along resource holding relationships and waiting relationships, extracting only the associated holding units necessary to maintain the holding state of R_TK_09. The candidate set is U2, U3, and U4. U1 is excluded from the candidate set because it only undertakes bypass receive writes and does not maintain the target resource holding, and U5 is also excluded because it does not participate in the holding chain. Subsequently, the system performs virtual elimination verification on the candidate units. Assume that the total number of critical blocking relationships maintaining the target resource holding state at the time of freezing is 4, corresponding to U4 holding R_TK_09, U4 occupying R_QS_17, U4 maintaining F_pub_42, and U3 maintaining R_MX_02. Define the boundary necessity index as follows:

[0091]

[0092] When the virtual device is removed from U4, R_TK_09, R_QS_17, and F_pub_42 are simultaneously released, leaving 0 blocked devices. When the virtual object is removed from U3, R_MX_02 is deactivated, U4 can continue to clean up and release R_TK_09, leaving only one item at the tail of the queue for short-term recycling. When the virtual resource is moved out of U2, the target resource retains its state. This indicates that U3 and U4 are both necessary units within the cutoff boundary, while U2 is not. Based on this, the system limits the minimum cutoff boundary to U3 and U4, avoiding a global restart of U1, U2, and other normal links.

[0093] The boundary necessity index and minimum cut boundary determination results of each holding unit can be further combined. Figure 6 understand.

[0094] Table 4: Virtual Removal Validation Results

[0095]

[0096] Table 4 presents the results of the virtual removal verification. Figure 6 The quantification relationship of boundary necessity and the decision of removal.

[0097] At 08:16:04, U5 performs online recovery on U3 and U4 based on the aforementioned minimum cutoff boundary, cutting only their local holding and waiting relationships, without restarting U1 and U2, and without interrupting the main link for receiving high-voltage feeder switch status and replaying protection actions. During the recovery process, the system first releases R_MX_02 held by U3, then resets the F_pub_42 and R_QS_17 states of U4, and then reclaims and retransmits the 43 pending results corresponding to R_TK_09. At 08:16:11, the partial recovery is complete, with a total time of 7 seconds. After the recovery is complete, the system performs a link break check based on the frozen state and the recovered operating state, confirming that the waiting edge from U4 to U3 and the reverse waiting edge from U3 to U4 have both disappeared, and the new reachability relationship no longer contains closed loops that return to themselves from U3 or U4 in two hops. The result publishing link re-enters the continuous advancement state.

[0098] The sequence of freeze, restore, and chain break verification times from 08:15:08 to 08:16:11 can be further combined. Figure 5Understood. Within 5 minutes of recovery, the result publication time delay decreased from 9.8 seconds to 1.1 seconds, the backlog of alarms decreased from 286 to 23, the average holding time of the publication token R_TK_09 decreased from 24.8 seconds to 0.9 seconds, the publication queue depth decreased from 302 to 28, CPU usage dropped from 78% to 49%, and the number of missed results decreased from 43 pending before freezing to 0. Because the system employs a combination of responsibility chain tracing, responsibility closure endpoint determination, virtual removal verification, and minimum cut-off recovery, only the two necessary units, U3 and U4, are subjected to online cut-off and reconstruction. Therefore, without affecting other normally holding units, responsibility closure positioning and minimum boundary recovery are achieved, avoiding the disruption of downhole power supply monitoring continuity caused by a global restart, and improving the accuracy and verifiability of online recovery under abnormal resource usage scenarios. The improvements in result publication time delay, alarm backlog, publication queue depth, token holding time, and CPU usage before and after recovery can be seen in the following figures. Figure 7 understand.

Claims

1. A method for self-diagnosis and online recovery of abnormal resource occupancy in an underground power supply monitoring system, characterized in that... include: Collect data on abnormal resource usage during operation; Based on the operational fact data, the resource instance is determined to be in a non-productive resource occupancy state, and the corresponding abnormal situation is frozen. Based on the abnormal site traceability after freezing, resource instances, holding units, waiting conditions, sources of blockage and reasons for non-release are traced to form an abnormal resource occupation responsibility chain, and responsibility closure judgment is performed to determine the responsibility closure endpoint. The cut-off boundary is determined based on the responsibility closure endpoint and online recovery is performed; after recovery, the chain of responsibility is checked for breakage based on the frozen abnormal scene, and the cut-off boundary is re-determined if the check fails.

2. The method for self-diagnosis and online recovery of abnormal resource occupancy in the downhole power supply monitoring system according to claim 1, characterized in that... An input boundary identifier, a processing phase advancement identifier, and a result boundary identifier are set for the resource instance. During the continuous occupation of the resource instance, if the processing phase advancement identifier remains unchanged and the result boundary identifier does not advance, it is determined to enter the non-output type resource occupation maintenance determination. After the maintenance determination is confirmed, if the input boundary identifier advances but the result boundary identifier does not advance, it is determined to be the accumulation type non-output type resource occupation. If neither the input boundary identifier nor the result boundary identifier advances, it is determined to be the stagnation type non-output type resource occupation.

3. The method for self-diagnosis and online recovery of abnormal resource occupancy in the downhole power supply monitoring system according to claim 1, characterized in that... Centered on the resource instance, extract the abnormal situation unit along the direct holding relationship and single-hop waiting relationship, and generate a read-only frozen copy; perform write bias on subsequent write operations involving the abnormal situation unit, so that the subsequent write operations act on the running data.

4. The method for self-diagnosis and online recovery of abnormal resource occupancy in the downhole power supply monitoring system according to claim 1, characterized in that... The resource release prerequisites of each holding unit in the frozen abnormal scene are reverse-verified; when the resource release prerequisites have been met but the resources are still held, the corresponding holding unit is determined to be a unit with missing release responsibility; when the blocking source that the holding unit is waiting for is generated or maintained by itself, and the removal of the blocking source still depends on the holding unit itself, the holding unit is determined to be a self-blocking responsibility unit; the unit with missing release responsibility or the self-blocking responsibility unit is used as the responsibility closure endpoint, and the cut-off boundary is determined.

5. The method for self-diagnosis and online recovery of abnormal resource occupancy in the downhole power supply monitoring system according to claim 2, characterized in that... A unique association identifier is set for the resource instance, and the input boundary identifier, processing stage advancement identifier, and result boundary identifier are bound to the unique association identifier; wherein, the input boundary identifier is updated when entering the processing link, the processing stage advancement identifier is updated when crossing a preset processing stage, and the result boundary identifier is updated when a submission result is formed or a result is published.

6. The method for self-diagnosis and online recovery of abnormal resource occupancy in the downhole power supply monitoring system according to claim 2, characterized in that... When the processing stage advancement indicator remains unchanged and the result boundary indicator does not advance for the first time, a retention determination is initiated for the resource instance; if the result boundary indicator advances during the retention determination period, the retention determination for non-output resource occupation is revoked; if the result boundary indicator still does not advance, non-output resource occupation is confirmed; among them, if the input boundary indicator advances, it is confirmed as accumulation-type non-output resource occupation, and if the input boundary indicator does not advance, it is confirmed as stagnation-type non-output resource occupation.

7. The method for self-diagnosis and online recovery of abnormal resource occupancy in the downhole power supply monitoring system according to claim 4, characterized in that... Starting from the resource retention state in the abnormal situation after freezing, trace back along the state items on which resource release depends to the release trigger state; when all state items corresponding to the resource release prerequisite have been satisfied and the resource is still retained, the corresponding holding unit is determined as the release missing responsibility unit.

8. The method for self-diagnosis and online recovery of abnormal resource occupancy in the downhole power supply monitoring system according to claim 4, characterized in that... In the abnormal situation after freezing, if the source of the blockage is generated or maintained by the holding unit before entering the waiting state, and the release of the source of the blockage still depends on the holding unit itself, then the holding unit is determined to be the self-blocking responsibility unit.

9. The method for self-diagnosis and online recovery of abnormal resource occupancy in an underground power supply monitoring system according to claim 4, characterized in that... Centered on the responsibility closure endpoint, trace outward along the resource holding relationship and waiting relationship, define the associated holding units necessary to maintain the resource holding state corresponding to the responsibility closure endpoint as the cut-off boundary, and exclude other holding units that do not affect the resource holding state from the cut-off boundary.

10. The method for self-diagnosis and online recovery of abnormal resource occupancy in an underground power supply monitoring system according to claim 9, characterized in that... A virtual removal check is performed on each associated holding unit obtained by tracing along the resource holding relationship and the waiting relationship; after any associated holding unit is virtually removed, if the resource holding state corresponding to the responsibility closure endpoint is released, the associated holding unit is retained at the cut-off boundary; if the resource holding state remains unchanged, the associated holding unit is excluded from the cut-off boundary.

Images