A power equipment operation state monitoring method based on a heterogeneous data correlation graph

Abstract

This invention relates to the field of knowledge graph technology, specifically to a method for monitoring the operational status of power equipment based on a heterogeneous data association graph. The method includes receiving multi-source heterogeneous operational data from power equipment, outputting equipment status parameters, computing load parameters, and access control levels, and instantiating these into object nodes, machine nodes, and access control nodes. Connection edges are established, the interaction frequency and latency between nodes are extracted, cross-domain coupling degree is calculated, and values ​​are assigned to the connection edges to generate a weighted graph. An anomaly flag is output when the equipment status parameters exceed a preset threshold. Path traversal is performed in the weighted graph to select a cloud-edge collaborative scheduling path. Target machine nodes and target access control nodes are extracted, access control levels are read, preemption priorities are generated, and computing resources corresponding to the target machine node are allocated. Collaborative control commands are then output and transmitted to the object nodes. This invention reduces the probability of critical commands being preempted or discarded by constructing a global monitoring system based on deep coupling of heterogeneous data.

Description

Technical Field

[0001] This invention relates to the field of knowledge graph technology, specifically to a method for monitoring the operating status of power equipment based on heterogeneous data association graphs. Background Technology

[0002] In complex distributed computing systems oriented towards ubiquitous electrical equipment, monitoring and collaboratively managing the operational status of massive heterogeneous processing nodes and underlying hardware is a crucial technological application area in the Industrial Internet of Things (IIoT) and underlying foundational software fields. Existing technologies generally employ a monitoring method combining single-dimensional resource indicator threshold comparison with a static rule engine. This involves real-time collection of individual operational attributes of underlying electrical and computing devices. When an attribute value exceeds a preset safety boundary, within a pre-defined program execution logic framework, an alarm is automatically triggered for the abnormal node, or a fixed task scheduling instruction is issued. This is currently the mainstream technical approach for achieving underlying hardware status monitoring and anomaly intervention.

[0003] However, existing technical solutions have significant shortcomings in cross-domain data collaborative processing and underlying computing resource allocation. Their monitoring architecture for characterizing the system's data processing status suffers from fragmentation and flattening. Existing monitoring models are mostly built on isolated static hardware resource catalogs or single data bus topologies. Their associated structures cannot interact and evolve dynamically in real time, making it difficult to accurately match the highly complex cross-domain processing of heterogeneous data and process coupling relationships. This leads to a serious bias in the monitoring perspective of the global computing power operation status. The system only issues "open-loop" unidirectional instructions at the moment of data processing anomaly trigger, failing to achieve closed-loop collaboration across system hardware and software layers. It cannot discern whether the underlying computing nodes have sufficient computing power to support fault recovery or process migration tasks when massive computing tasks are concurrently abnormal, nor can it confirm whether the scheduling control logic conforms to the memory and access permission level limits of the current system processes. This easily leads to computing resource deadlock, data bus congestion, or unauthorized execution risks. In addition, collaborative instructions suffer from mechanical output. The mechanically triggered control flow continuously crowds out the scarce execution channels at the underlying level, causing critical fault repair instructions to frequently time out and be discarded.

[0004] To address this, a method for monitoring the operating status of power equipment based on heterogeneous data association graphs is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a method for monitoring the operating status of power equipment based on heterogeneous data association maps, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for monitoring the operating status of power equipment based on heterogeneous data association maps, comprising: It receives multi-source heterogeneous operation data of power equipment in the cloud-edge-device architecture, and outputs equipment status parameters, computing load parameters and access control levels by field decoding. The device status parameters, computing load parameters, and access control levels are instantiated as object nodes, machine nodes, and access control nodes. Connection edges are established between the nodes, the interaction frequency and latency between the nodes are extracted, and aggregation operations are performed to obtain the cross-domain coupling degree. The cross-domain coupling degree is assigned to the connection edges to construct a weighted graph. When the device status parameter corresponding to the object node exceeds the preset boundary, an abnormality flag is output. Path traversal is performed in the weighted graph, and the path with the highest cumulative cross-domain coupling degree is selected as the cloud-edge collaborative scheduling path. The target machine node and target access control node contained in the cloud-edge collaborative scheduling path are extracted. When the computing power load parameter corresponding to the target machine node exceeds the load threshold, the corresponding access control level is read. Based on the access control level and the exception identifier, a preemption priority is generated, a priority preemption instruction is sent to the target machine node, the internal low-priority background processes are frozen and idle computing resources are released, and the cooperative control instruction is output to the object node corresponding to the exception identifier.

[0007] Preferably, the specific generation process of the device status parameters, computing load parameters, and access control levels includes: extracting the underlying communication protocol feature codes carried by the multi-source heterogeneous operation data of the power equipment; comparing the preset protocol parsing rule base and the underlying communication protocol feature codes to output a matching target protocol unpacking strategy; calling the target protocol unpacking strategy to perform byte stripping processing on the multi-source heterogeneous operation data of the power equipment, separating physical layer feature data segments, system layer feature data segments, and interaction layer feature data segments; performing semantic mapping parsing on the physical layer feature data segments to output device status parameters; performing format conversion and reorganization on the system layer feature data segments to output computing load parameters; and performing decryption verification and matching on the interaction layer feature data segments to output access control levels.

[0008] Preferably, the specific generation process of the interaction frequency and delay time includes: dividing the graph data structure into a physical device mapping area, a computing resource mapping area, and an interaction permission mapping area; importing device status parameters to generate object nodes in the physical device mapping area; importing computing power load parameters to generate machine nodes in the computing resource mapping area; importing access control levels to generate access control nodes in the interaction permission mapping area; extracting historical collaborative interaction logs carried by the graph data structure; retrieving cross-domain collaborative request records initiated between object nodes, machine nodes, and access control nodes; performing feature separation on the cross-domain collaborative request records to extract the total number of calls and output the interaction frequency; extracting the time difference between the cross-domain collaborative request initiation point and the receiving response point in the cross-domain collaborative request records and outputting the delay time.

[0009] Preferably, the specific generation process of the cross-domain coupling degree includes: inputting the interaction frequency into the minimax linear scaling model, compressing it through numerical transformation, and outputting the normalized frequency coefficient; collecting the distribution characteristic parameters of all delay times within a preset historical observation window; calculating the mathematical variance value corresponding to the distribution characteristic parameters and outputting the stability coefficient; configuring a positive amplification factor corresponding to the normalized frequency coefficient; configuring a bias correction attenuation factor corresponding to the stability coefficient; performing a product operation on the normalized frequency coefficient and the positive amplification factor to output the frequency gain variable; performing a product operation on the stability coefficient and the bias correction attenuation factor to output the delay stability gain variable; and performing an aggregation summation operation on the frequency gain variable and the delay stability gain variable to calculate and output the cross-domain coupling degree.

[0010] Preferably, the specific generation process of the weighted graph includes: scanning all object nodes, machine nodes, and access control nodes contained within the graph data structure; retrieving the communication handshake logs at the underlying level of all object nodes, machine nodes, and access control nodes; comparing the communication handshake logs to extract node pairs with actual data exchange records; allocating virtual memory pointers between node pairs to establish bidirectional data channel output connection edges; converting the cross-domain coupling degree corresponding to the node pair into hexadecimal weight feature codes; writing the hexadecimal weight feature codes into the attribute storage area of ​​the bidirectional data channel to complete the assignment operation of the cross-domain coupling degree; and encapsulating all object nodes, machine nodes, access control nodes, and the bidirectional data channel with the hexadecimal weight feature codes written in to output the weighted graph.

[0011] Preferably, the specific generation process of the cloud-edge collaborative scheduling path includes: obtaining a preset safe operation benchmark interval; sending the device status parameters contained in the object node and the safe operation benchmark interval into a comparator to perform comparison logic and output difference comparison values; determining that the difference comparison values ​​exceed the boundary limit and generating an anomaly identifier; combining the anomaly identifier to extract the object node that generates the device status parameters as the initial object node; setting the depth constraint condition of the breadth-first search algorithm; advancing the probe ray hop by hop from the initial object node along the connection edge based on the depth constraint condition; performing dynamic accumulation on the cross-domain coupling degree carried by the connection edge through which the probe ray passes and outputting a candidate path set; verifying the network connectivity code traversed by each candidate path in the candidate path set; eliminating candidate paths whose network connectivity code matches the disconnection state and outputting a candidate sequence; comparing the accumulated cross-domain coupling degree contained in the candidate sequence; extracting the candidate path corresponding to the top position of the sort and outputting it as the cloud-edge collaborative scheduling path.

[0012] Preferably, the specific process of reading the access control level corresponding to the target access control node includes: performing a disassembly operation on the cloud-edge collaborative scheduling path to output the target machine node and the target access control node; fusing the processor thread occupancy ratio and the remaining memory capacity to output load parameters; obtaining a preset load threshold; sending the load parameters and load threshold into a comparator to perform a judgment operation; determining that the load parameters exceed the load threshold and outputting an over-limit interrupt signal; responding to the over-limit interrupt signal to establish an access channel to the target access control node; extracting the identity credential code embedded in the target access control node along the access channel; and performing matching logic on the identity credential code to read the access control level.

[0013] Preferably, the specific process of outputting the cooperative control command to the object node corresponding to the anomaly identifier includes: extracting the first feature weight coefficient corresponding to the fault level code contained in the anomaly identifier, and extracting the second feature weight coefficient corresponding to the highest authority code contained in the access control level; performing a product operation on the first feature weight coefficient and the second feature weight coefficient, adding a preset emergency compensation constant, and outputting the preemption priority; sending a priority preemption command matching the preemption priority to the target machine node; parsing the system resource occupancy table and mutex lock request list of the background process, evaluating whether freezing the background process will cause a deadlock through a directed graph loop detection algorithm, and outputting a freeze whitelist; for background processes located in the freeze whitelist, pushing key running data and program counter register status into the system protection stack depth for storage, and forcibly modifying the state machine parameters to a sleep state; releasing the idle computing resources occupied by the background process; calling the idle computing resources to perform logical operations and outputting the cooperative control command payload; encapsulating the cooperative control command payload with a protocol header and outputting the cooperative control command; and transmitting the cooperative control command to the object node corresponding to the anomaly identifier according to the network routing table.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By instantiating device status parameters, computing power load parameters, and access control levels into physical nodes, machine nodes, and access control nodes, and performing aggregation operations to obtain cross-domain coupling degree, a weighted graph is generated. When an anomaly is detected, path traversal is directly performed in the weighted graph, and the path with the highest cumulative cross-domain coupling degree is selected as the cloud-edge collaborative scheduling path. This quantifies the real communication status between nodes, enabling command transmission to dynamically select the link with the optimal carrying capacity based on the actual network load. This reduces the probability of commands entering congested channels at the network level and lowers the risk of critical fault repair commands timeout due to channel congestion.

[0015] 2. A state awareness and permission verification mechanism is introduced during the instruction issuance phase. This mechanism extracts the target machine node and target access control node from the cloud-edge collaborative scheduling path during the execution path scheduling process. It determines whether the computing power load parameters corresponding to the target machine node exceed the load threshold and reads the access control level corresponding to the target access control node. It correlates physical layer anomalies, computing resource status, and management permissions to make an objective judgment. This enables the objective identification of whether the underlying channel is severely occupied by the mechanized control flow. When it is confirmed that computing power resources are tight, the regular instruction distribution is temporarily suspended, and high-level permission evidence is extracted. This provides compliant data support before instruction congestion and reduces the possibility of critical instructions directly entering the timeout queue state.

[0016] 3. By constructing a computing resource preemption mechanism based on multi-dimensional cross-validation, after confirming the occurrence of over-limit anomalies and resource congestion, a preemption priority is generated based on the access control level and anomaly identifier. Then, the computing resources corresponding to the target machine node are allocated and the collaborative control instructions are transmitted to the object node corresponding to the anomaly identifier. In specific operations, the status of low-priority background processes is frozen according to the priority order, and idle computing resources are released. The routine mechanized control flow that is occupying the channel is actively cleared, and the limited computing resources are targeted and matched to high-risk repair tasks. The congested channel is cleared from the execution side, and the success rate of critical repair instructions under complex working conditions is improved.

[0017] 4. By constructing a global monitoring system based on deep coupling of heterogeneous data, device status parameters, computing load parameters, and access control levels are classified and output, and instantiated as corresponding nodes. The cross-domain coupling degree is assigned to the connection edge to construct a weighted graph, breaking the state of separation between physical device computing resources and management permissions. When judging anomalies and allocating computing resources corresponding to target machine nodes, the underlying hardware status edge computing power and top-level permissions are evaluated as a highly coupled whole, avoiding the mechanical distribution of instructions triggered by a single-dimensional threshold. Through the collaborative logic of the global perspective, the instruction output can match the actual carrying capacity, reducing the invalid occupation of the underlying execution channel and reducing the probability of critical instructions being squeezed out and discarded from the architectural level. Attached Figure Description

[0018] Figure 1 The flowchart is a method for monitoring the operating status of power equipment based on heterogeneous data association maps, as proposed in an embodiment of this invention application. Figure 2 This is a flowchart of the weighted graph generation and dynamic cloud-edge collaborative scheduling path optimization process proposed in an embodiment of this invention. Figure 3 This is a flowchart of the computing load determination and resource preemption control proposed in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figures 1-3 The present invention provides a method for monitoring the operating status of power equipment based on heterogeneous data association graphs, the specific steps of which are as follows: It receives multi-source heterogeneous operation data of power equipment in the cloud-edge-device architecture, and outputs equipment status parameters, computing load parameters and access control levels by field decoding. The device status parameters, computing load parameters, and access control levels are instantiated as object nodes, machine nodes, and access control nodes. Connection edges are established between the nodes, the interaction frequency and latency between the nodes are extracted, and aggregation operations are performed to obtain the cross-domain coupling degree. The cross-domain coupling degree is assigned to the connection edges to construct a weighted graph. When the device status parameter corresponding to the object node exceeds the preset boundary, an abnormality flag is output. Path traversal is performed in the weighted graph, and the path with the highest cumulative cross-domain coupling degree is selected as the cloud-edge collaborative scheduling path. The target machine node and target access control node contained in the cloud-edge collaborative scheduling path are extracted. When the computing power load parameter corresponding to the target machine node exceeds the load threshold, the corresponding access control level is read. Based on the access control level and the exception identifier, a preemption priority is generated, a priority preemption instruction is sent to the target machine node, the internal low-priority background processes are frozen and idle computing resources are released, and the cooperative control instruction is output to the object node corresponding to the exception identifier.

[0021] The technical solution of the present invention will be further described in detail below with reference to specific embodiments.

[0022] Example 1 This application discloses a method for monitoring the operating status of power equipment based on heterogeneous data association maps. (See attached document.) Figure 1The specific steps proposed in this invention include: S1, receiving multi-source heterogeneous operation data of power equipment in a cloud-edge-device architecture, and outputting equipment status parameters, computing load parameters, and access control levels through field decoding; S2, instantiating the equipment status parameters, computing load parameters, and access control levels into object nodes, machine nodes, and access control nodes, establishing connection edges between nodes, extracting the interaction frequency and latency between nodes, performing aggregation operations to obtain cross-domain coupling degree; assigning the cross-domain coupling degree to the connection edges to construct a weighted graph; S3, determining that the equipment status parameters corresponding to the object node exceed the pre-defined parameters. When setting boundaries, output an anomaly flag, perform path traversal in the weighted graph, and select the path with the highest cumulative cross-domain coupling value as the cloud-edge collaborative scheduling path; extract the target machine node and target access control node contained in the cloud-edge collaborative scheduling path, and when the load parameter corresponding to the target machine node exceeds the load threshold, read the corresponding access control level; S4, generate a preemption priority based on the access control level and the anomaly flag, send a priority preemption command to the target machine node, freeze the internal low-priority background processes and release idle computing resources, and output the collaborative control command to the object node corresponding to the anomaly flag.

[0023] Furthermore, it receives multi-source heterogeneous operational data from power equipment in the cloud-edge-device architecture, decodes the fields, and outputs classified equipment status parameters, computing load parameters, and access control levels; corresponding to step S1 above; the specific implementation process includes: Extract the underlying communication protocol signatures carried in the multi-source heterogeneous operation data of power equipment; compare the signatures with a pre-set protocol parsing rule base and output a matching target protocol unpacking strategy; invoke the target protocol unpacking strategy to perform byte stripping processing on the multi-source heterogeneous operation data of power equipment, separating physical layer feature data segments, system layer feature data segments, and interaction layer feature data segments; perform semantic mapping parsing on the physical layer feature data segments to output device status parameters; perform format conversion and reorganization on the system layer feature data segments to output computing power load parameters; and perform decryption verification and matching on the interaction layer feature data segments to output access control levels.

[0024] Specifically, the generation process for device status parameters, computing load parameters, and access control levels is as follows: In this embodiment, the overall data input terminal is deployed in a heterogeneous communication network on the substation and industrial control system side. The received data is collectively referred to as multi-source heterogeneous operation data of power equipment. The sources of this data span three technical dimensions: first, physical indicators such as temperature, pressure, and partial discharge periodically collected by the sensor matrix deployed inside the 110 kV oil-immersed transformer at the substation site; second, low-level status data extracted by hardware probes from the edge intelligent gateway deployed in the local control cabinet of the substation when processing concurrent daemon processes; and third, digital authentication and instruction request data sent by operation and maintenance personnel when accessing the network through a personal identity verification smart card terminal.

[0025] The system continuously reads multi-source heterogeneous operating data from power equipment and extracts the underlying communication protocol signatures. A pre-loaded protocol parsing rule base is used. When a data frame is extracted, if the preceding signature of its transmission control protocol load portion is a hexadecimal sequence and the subsequent function code field is 03 or 04, the system compares it against the rule base to determine if it matches the industrial control Ethernet protocol. By comparing the pre-loaded protocol parsing rule base and the underlying communication protocol signatures, the system outputs the target protocol unpacking strategy.

[0026] The target protocol unpacking strategy is invoked to perform byte stripping on the multi-source heterogeneous operating data of power equipment. Following the Open Systems Interconnection Reference Model, a 14-byte header containing the target and source media access control addresses, a 20-byte header containing the source and destination Internet Protocol (ISP) addresses, and a 20-byte header of the transmission control protocol are sequentially stripped. After byte stripping, physical layer characteristic data segments, system layer characteristic data segments, and interaction layer characteristic data segments are separated.

[0027] Semantic mapping parsing is performed on the physical layer feature data segments: This involves calling the internally pre-built system configuration description file and logical node model template. The hexadecimal object reference paths within the data segments are mapped to the system configuration description file, identifying the floating-point values ​​they carry, converting them into temperature or gas concentration values, and outputting the device status parameters. For a 110 kV oil-immersed transformer, the output device status parameters are: top oil temperature 81.5 degrees Celsius, hydrogen content concentration 315 parts per million.

[0028] In this embodiment, a sliding window-based deviation test is performed on the converted floating-point value. The deviation between the current value and the historical mean within the window is calculated. If the deviation exceeds a set tolerance, a Kalman filter algorithm is introduced for smoothing correction. Specifically, a first-in-first-out sliding window with a length of 10 sampling periods is constructed, and the converted floating-point values ​​are sequentially stored in this window. The arithmetic mean of the nine historical floating-point values ​​within the window is calculated as a reference value. The latest floating-point value is subtracted from the reference value, and the absolute value is divided by the reference value to obtain the relative deviation. The sensor tolerance threshold is set to 15%. When the relative deviation is determined to exceed 15%, the one-dimensional Kalman filter model is activated. The state transition matrix is ​​set to 1, the observation matrix is ​​set to 1, the variance of the historical arithmetic mean is used as the initial estimation error covariance (set to 0.5 in this embodiment), the currently fluctuating floating-point value is used as the current observation value, and the measurement noise covariance is set to a constant of 0.2. The posterior state estimate from the previous time step is multiplied by the state transition matrix to obtain the prior state estimate. This prior state estimate is then calculated using the observation matrix and the measurement noise covariance. Finally, the Kalman gain is used to correct the deviation between the current observation and the prior state estimate, resulting in a posterior state estimate. This posterior state estimate replaces the current floating-point value indicating abnormal fluctuations. If the relative deviation does not exceed 15%, the original value is retained. The processed value is then confirmed as the final value.

[0029] The state transition matrix is ​​set based on the following: the top oil temperature of a large oil-filled transformer exhibits strong thermal inertia within a short, second-level sampling period, and its physical state can be approximated as a constant. Setting the matrix parameter to 1 aligns with the steady-state observation requirements of this physical device during transient processes, preventing excessive algorithm divergence. The first-in-first-out sliding window length is selected based on the typical duration of high-frequency fluctuations in the transformer sensor's white noise. Setting it to 10 cycles effectively smooths transient interference without causing significant delays in data response due to an excessively long window. The 15% tolerance threshold is determined based on test data of the maximum permissible transient fluctuation rate of the transformer's top oil temperature under normal operating conditions. The measurement noise covariance of 0.2 is determined based on the white noise variance benchmark calibrated at the factory for this sensor model. The above process eliminates glitches and noise caused by transient drift of the bottom sensor, improving the authenticity of the equipment's state parameters and the accuracy of subsequent anomaly detection.

[0030] The system-level characteristic data segment undergoes format conversion and reorganization: the device management information base object identifier encoded by abstract syntax tags in the data segment is read, and the hexadecimal format bytecode in its data field is converted into a decimal percentage floating-point number. The reorganization operation mathematically integrates the processor time slice allocation rate with the available physical memory capacity, outputting the computing load parameters, which are: CPU utilization 87%, available physical memory 9%.

[0031] After extracting the interaction layer feature data segment, a decryption verification and matching process is executed: The personal identity verification credential feature field conforming to information processing standards is extracted from the message. The asymmetric public key certificate chain deployed within the local secure hardware encryption machine is invoked to perform an asymmetric decryption algorithm on the intercepted signature data segment, recovering the digital certificate serial number. After decryption, the serial number is compared and matched against the local access control list and role permission database, outputting the access control level. The output access control level is represented by hexadecimal encoding 7.

[0032] By matching target unpacking strategies with underlying protocol signatures, byte stripping and layered parsing are performed on heterogeneous data to separate physical layer, system layer, and interaction layer data, thus avoiding the mechanical mis-issuance of instructions caused by data confusion from the source. Furthermore, by extracting and mapping multi-dimensional parameters, a high-precision structured data foundation is provided for graph construction and dynamic scheduling, thereby reducing the amount of invalid data processing and initially alleviating the implicit crowding out of the underlying execution channel by conventional mechanical control flow.

[0033] Furthermore, the device status parameters, computing load parameters, and access control levels are instantiated as object nodes, machine nodes, and access control nodes, and connection edges are established between the nodes. The interaction frequency and latency between nodes are extracted, and aggregation operations are performed to obtain the cross-domain coupling degree. The cross-domain coupling degree is assigned to the connection edges to construct a weighted graph; this corresponds to step S2 above. The specific implementation process includes: Within the graph data structure, divide it into physical device mapping area, computing resource mapping area, and interaction permission mapping area; import device status parameters to generate object nodes into the physical device mapping area; import computing power load parameters to generate machine nodes into the computing resource mapping area; import access control levels to generate access control nodes into the interaction permission mapping area; extract historical collaborative interaction logs carried by the graph data structure; retrieve cross-domain collaborative request records initiated between object nodes, machine nodes, and access control nodes; perform feature separation on the cross-domain collaborative request records to extract the total number of calls and output the interaction frequency; extract the time difference between the cross-domain collaborative request initiation point and the receiving response point in the cross-domain collaborative request records and output the delay time.

[0034] Specifically, the generation process for interaction frequency and latency is as follows: Within the graph data structure, physical device mapping area, computing resource mapping area, and interaction permission mapping area are defined. Device status parameters are imported into the physical device mapping area, and entity records are generated using the unique media access control address of the electrical device and the logical device entity identifier as a composite primary key. The physical node is then instantiated and output (the node code is physical node instance one).

[0035] Import computing load parameters into the computing resource mapping area, combine them with the Internet Protocol address characteristics of the edge computing gateway device to generate the corresponding memory object, and output the machine node (the node code is machine node instance one).

[0036] Import the access control level into the interactive permission mapping area, use the digital certificate serial number obtained from the previous decryption and verification and permission code 7 to generate the permission authentication attribute block, and instantiate and output the access control node (the node code is access control node instance one).

[0037] The background log analysis daemon starts, extracting historical collaborative interaction logs carried by the graph data structure. The logs record remote procedure calls and message creation specification read / write behaviors. By retrieving cross-domain collaborative request records initiated between object nodes, machine nodes, and access control nodes, feature separation is performed on the matched cross-domain collaborative request records. In the frequency domain feature separation operation, the application layer transaction sequence numbers contained in the cross-domain collaborative request records are traversed, and the number of requests with the same source node address and destination node address is counted, outputting the interaction frequency, which is 450 times.

[0038] In the feature separation operation in the time domain, network time protocol timestamps are extracted one by one, and the absolute time difference between the initiation point of the cross-domain collaborative request and the receiving response point is extracted. The arithmetic mean of this time difference is calculated, and the mean output is the delay time, which is 12.5 milliseconds.

[0039] Within the graph data structure, dedicated mapping areas are divided to generate objects, machines, and access control nodes. By retrieving collaborative request records, the number of calls and time differences are extracted to generate interaction frequency and latency, objectively quantifying the historical interaction efficiency between nodes. Furthermore, through rigorous log feature extraction, high-latency regular links are identified, preventing critical instructions from being assigned to congested channels. This provides data support for subsequent weight calculations and avoiding scarce channel congestion areas.

[0040] The interaction frequency is input to the minimax linear scaling model, compressed through numerical transformation, and the normalized frequency coefficient is output. Distribution characteristic parameters of all delay times within a preset historical observation window are collected. The mathematical variance corresponding to the distribution characteristic parameters is calculated, and the stability coefficient is output. A positive amplification factor corresponding to the normalized frequency coefficient is configured. A correction attenuation factor corresponding to the stability coefficient is configured. The normalized frequency coefficient and the positive amplification factor are multiplied to output a frequency gain variable. The stability coefficient and the correction attenuation factor are multiplied to output a delay stability gain variable. The frequency gain variable and the delay stability gain variable are aggregated and summed to calculate the cross-domain coupling degree.

[0041] Specifically, the process for generating cross-domain coupling degree is as follows: The interaction frequency is input into the minimax linear scaling model. The difference between the current input interaction frequency and the preset historical minimum interaction frequency is divided by the difference between the preset historical maximum interaction frequency and the preset historical minimum interaction frequency. This yields a normalized frequency coefficient. Substituting this into the numerical calculation, the normalized frequency coefficient equals (450 minus 0) divided by (1000 minus 0), resulting in an output value of 0.45. In this embodiment, the preset historical maximum interaction frequency is selected based on the theoretical maximum packet forwarding rate of the network switch backplane bandwidth and is set to 1000 times; the preset historical minimum interaction frequency is set to 0 times.

[0042] The distribution characteristic parameters of all delay times within a preset historical observation window are collected, and the corresponding mathematical variance values ​​are calculated. To eliminate the time dimension, the baseline variance value under preset historical normal operating conditions is extracted. The currently calculated mathematical variance value is divided by this baseline variance value to obtain a dimensionless relative variance variable. To prevent division-by-zero overflow when the relative variance variable is 0, the relative variance variable is incremented by 1 as the denominator base, and then the reciprocal of the denominator base is taken to obtain a stability coefficient of 0.82. The duration parameter of the preset historical observation window is selected based on the median of the historical failure occurrence cycle of the system equipment and is set to 60 minutes. The preset historical normal operating conditions refer to the extraction of all delay time series recorded by the equipment under laboratory testing environment and stable operation at rated load for 24 hours during the initial deployment phase, and the overall variance of this series is calculated as the initial baseline variance value.

[0043] The positive amplification factor and the correction attenuation factor are selected based on the weight allocation of the linear regression analysis results of the historical equipment scheduling success rate. The specific determination steps are as follows: using the normalized frequency coefficient in the historical operation data as the independent variable and the collaborative scheduling task success rate as the dependent variable; using the condition that the sum of the values ​​of the positive amplification factor and the correction attenuation factor equals 1 as the equality constraint; using minimizing the sum of squared regression residuals as the optimization objective; constructing a Lagrange multiplier optimization function that includes the above equality constraint and optimization objective; calculating the partial derivatives of the positive amplification factor, the correction attenuation factor, and the introduced Lagrange multiplier, and setting each partial derivative equal to 0 to construct a system of linear equations; solving the system of linear equations simultaneously to obtain the unique values ​​of the positive amplification factor and the correction attenuation factor that satisfy the constraint that the sum equals 1. In this embodiment, after solving and constraining based on massive sample data from one year of history, the positive amplification factor is set to 0.65 and the correction attenuation factor is set to 0.35.

[0044] Performing the aggregation and summation calculation of the cross-domain coupling degree, the frequency gain variable is equal to the normalized frequency coefficient multiplied by the forward amplification factor, resulting in a value of 0.2925. The delay stabilization gain variable is equal to the stability coefficient multiplied by the correction attenuation factor, resulting in a value of 0.2870. The cross-domain coupling degree is equal to the frequency gain variable plus the delay stabilization gain variable, with a final output value of 0.5795.

[0045] By introducing minimax linear scaling and variance operations, combined with amplification and attenuation factors, frequency gain and delay-stable gain variables are generated, and cross-domain coupling is aggregated and output. This transforms static network connections into dynamic weights that reflect the actual network load. When issuing commands, the dynamic weights can intelligently avoid inefficient links that are excessively crowded by mechanical control flows, ensuring that the selection of cloud-edge collaborative scheduling paths is based on objective data and improving congestion resistance.

[0046] The process involves: scanning all object nodes, machine nodes, and access control nodes within the scan graph data structure; retrieving the underlying communication handshake logs of all object nodes, machine nodes, and access control nodes; comparing the communication handshake logs to extract node pairs with actual data exchange records; allocating virtual memory pointers between node pairs to establish bidirectional data channel output connection edges; converting the cross-domain coupling degree corresponding to the node pairs into hexadecimal weight feature codes; writing the hexadecimal weight feature codes into the attribute storage area of ​​the bidirectional data channel to complete the cross-domain coupling degree assignment operation; and encapsulating all object nodes, machine nodes, access control nodes, and the bidirectional data channel with the hexadecimal weight feature codes written in them to output a weighted graph.

[0047] Specifically, the process of generating the weighted graph is as follows: The system scans the graph data structure to obtain a list of point sets for the current active graph. It then retrieves the underlying communication handshake logs of all object nodes, machine nodes, and access control nodes, including records of transmission control protocol synchronization requests, synchronization acknowledgments, and acknowledgment requests. The system compares these handshake logs to extract node pairs with actual data exchange records. The criteria for selection are: successful establishment of the underlying network socket connection and extraction of a message containing valid application data payload (such as a write request or report action in the message specification); valid node pairs are then selected. Virtual memory pointers are allocated between the node pairs, and the memory addresses of the source and target node objects are mutually pointed to, establishing a bidirectional data channel, with the output as a connection edge.

[0048] In this embodiment, the fragmentation level and actual read / write speed of the virtual memory block containing the bidirectional data channel are extracted. The channel attenuation rate is calculated by subtracting the read / write speed from a preset baseline speed. The data channel is then downgraded based on the fragmentation level and channel attenuation rate. The underlying memory management unit interface is invoked to scan the contiguous virtual memory blocks allocated to the bidirectional data channel, and the number of non-contiguous physical memory pages is counted. This number is then divided by the total number of allocated pages to calculate the memory fragmentation rate. The average time of the past 100 cross-domain request interactions is extracted to calculate the current actual read / write speed. The preset local area network baseline read / write speed is 500 megabytes per second, which is set based on the theoretical maximum throughput of the internal high-speed data bus or gigabit Ethernet card configured in the edge computing gateway in full-duplex mode. The channel attenuation rate is obtained by subtracting the current actual read / write speed from the baseline read / write speed and then dividing by the baseline read / write speed. The fragmentation threshold is set to 20%, and the attenuation threshold is set to 10%. The memory fragmentation rate and channel decay rate are fed into a Boolean decision logic: if the fragmentation rate is greater than 20% and the channel decay rate is greater than 10%, a "degradation" flag is written to the attribute descriptor of the bidirectional data channel. In subsequent operations, the hexadecimal weight feature code written to this channel is logically shifted one bit to the right to reduce its weight. If both are below the threshold, the default attribute state is maintained. The 20% and 10% thresholds are derived from the historical extreme critical values ​​of the edge computing gateway operating system before read / write blocking occurs. The above process introduces a low-level memory health awareness mechanism, avoiding the allocation of high-frequency critical scheduling instructions to channels with limited physical read / write capabilities or severe fragmentation.

[0049] The cross-domain coupling degree values ​​corresponding to the node pairs are extracted and sent to the data type format conversion processor. The processor multiplies the cross-domain coupling degree value representing 0.5795 by 10000 and rounds down to obtain the integer value 5795. This integer value is then converted into a hexadecimal weight feature code. The converted hexadecimal weight feature code value is 16A3. The controller triggers the direct memory access mechanism to write the hexadecimal weight feature code 16A3 into the attribute storage area of ​​the bidirectional data channel, completing the assignment operation of the cross-domain coupling degree. All object nodes, machine nodes, access control nodes, and the bidirectional data channel containing the hexadecimal weight feature code are encapsulated, and the weighted graph is output.

[0050] During continuous operation, the weighted graph undergoes a dynamic evolution and update mechanism. Specifically, the background daemon periodically reads the most recent collaborative interaction logs using a set time window (60 seconds in this embodiment); it recalculates the interaction frequency and latency between each node pair according to the aforementioned steps, and updates the cross-domain coupling degree value; the latest cross-domain coupling degree value is then converted back into a hexadecimal weight feature code and overwritten into the attribute storage area of ​​the corresponding bidirectional data channel. Furthermore, when the underlying network interface detects a new device node access or an existing node disconnection, a non-periodic full graph scan is triggered to reconstruct the connection edges and weight feature codes of the affected area.

[0051] By extracting node pairs that exchange data, establishing bidirectional channels using virtual memory pointers, and converting coupling degree into hexadecimal feature codes and writing them into the storage area, the efficiency of reading and calling graph weights is improved. Moreover, by establishing a hardware-level channel directly to the bottom layer and fast weight retrieval, the mechanical flow and queuing of upper-layer software business is effectively skipped, preventing repair instructions from timeouts due to excessive time consumption during the flow, and ensuring the timeliness of the bottom layer response.

[0052] Furthermore, when the device status parameter corresponding to the object node exceeds the preset boundary, an anomaly flag is output. Path traversal is performed in the weighted graph, and the path with the highest cumulative cross-domain coupling degree is selected as the cloud-edge collaborative scheduling path. The target machine node and target access control node contained in the cloud-edge collaborative scheduling path are extracted. When the computing power load parameter corresponding to the target machine node exceeds the load threshold, the corresponding access control level is read. This corresponds to step S3 above. (See also...) Figure 2 The specific implementation process includes: Obtain a preset safe operation benchmark interval; input the device status parameters contained in the object node and the safe operation benchmark interval into the comparator to perform comparison logic and output the difference comparison value; determine if the difference comparison value exceeds the boundary limit and generate an anomaly label; combine the anomaly label to extract the object node that generates the device status parameters as the initial object node; set the depth constraint condition of the breadth-first search algorithm; advance the probe ray hop by hop from the initial object node along the connection edge based on the depth constraint condition; perform dynamic accumulation on the cross-domain coupling degree carried by the connection edge through which the probe ray passes and output the candidate path set; verify the network connectivity code traversed by each candidate path in the candidate path set; eliminate candidate paths whose network connectivity code matches the disconnection state and output the candidate sequence; compare the accumulated cross-domain coupling degree contained in the candidate sequence; extract the candidate path corresponding to the top position of the sort and output it as the cloud-edge collaborative scheduling path.

[0053] Specifically, the process of generating a cloud-edge collaborative scheduling path is as follows: The device status parameters contained in the corresponding node are sent to the comparator to perform comparison logic. The current detected value of the device status parameter is subtracted from the upper limit threshold value of the safe operation benchmark range to obtain the difference comparison value, which is 16.5 degrees Celsius. At the same time, it is determined that the device status parameter (81.5 degrees Celsius) has exceeded the preset boundary limit of 80 degrees Celsius, and an anomaly identifier in hexadecimal value is output, coded as 81. The safe operation benchmark range is selected based on the rated allowable temperature rise of the large oil-filled power equipment's nameplate specification and the local historical highest ambient temperature, and the upper limit is set to 65 degrees Celsius. The boundary limit is selected based on the critical temperature parameter for accelerated aging of transformer insulating oil and is set to 80 degrees Celsius.

[0054] Based on anomaly indicators, the object node generating the device status parameters is extracted as the initial object node. A depth constraint of 3 hops is set, and the termination node of the effective probe ray must be a machine node with computational resource allocation capabilities. The probe ray is advanced hop-by-hop along the connecting edges from the initial object node. The hexadecimal weight feature code recorded in the attribute storage area of ​​each connecting edge traversed by the probe ray is read and converted into decimal cross-domain coupling degree. The total cross-domain coupling degree of the path is equal to the sum of the cross-domain coupling degree values ​​carried by all single-hop connecting edges traversed by the probe ray. After accumulation, a set of candidate paths is output, where each path in the candidate path set terminates at a machine node. The upper limit of the hop count for the depth constraint of the breadth-first search algorithm is selected based on the maximum number of routing levels in the local area network and the maximum allowable communication delay time, and is set to 3 hops.

[0055] Verify the network connectivity code of each node passed through in the candidate path set, extract the echo response code of the Internet Control Message Protocol or check the port connectivity status of the Simple Network Management Protocol, eliminate candidate paths whose network connectivity codes match the disconnection status, and output the candidate sequence.

[0056] The system compares the cumulative cross-domain coupling values ​​contained in the candidate sequences, sorts them in descending order, extracts the candidate path corresponding to the first position of the sorted sequence, and outputs the cloud-edge collaborative scheduling path.

[0057] Based on the anomaly identifier, a breadth-first search is triggered. Alternative paths are explored under depth constraints, and network connectivity is verified to eliminate disconnected links. Finally, the optimal cloud-edge collaborative scheduling path is output based on the cumulative coupling degree. This opens up an anti-interference detour path for critical repair commands, actively avoids network trunks that are severely crowded by mechanical control flow, and reduces the probability of commands being dropped due to timeout during transmission.

[0058] The system performs a disassembly operation on the cloud-edge collaborative scheduling path, outputting the target machine node and the target access control node; it integrates the processor thread occupancy ratio and remaining memory capacity to output load parameters; it obtains a preset load threshold; it sends the computing power load parameters and load threshold to a comparator to perform a judgment operation; it determines that the computing power load parameters exceed the load threshold and outputs an over-limit interrupt signal; it responds to the over-limit interrupt signal to establish an access channel to the target access control node; it extracts the identity credential code embedded in the target access control node along the access channel; and it performs matching logic on the identity credential code to read the access control level.

[0059] Specifically, the process of executing the matching logic to read the access control level is as follows: The cloud-edge collaborative scheduling path is disassembled, and the endpoint node of the path is extracted and locked as the target node. Simultaneously, the weighted graph is searched to extract the access control nodes with direct edges to the target node as the target access control nodes. The processor thread occupancy ratio and memory occupancy rate of the target node are read. The load parameter is obtained by multiplying the processor thread occupancy ratio by the processor weight and adding the memory occupancy rate by the memory weight. Substituting the values, the load parameter is calculated as 87% multiplied by 0.6 plus 91% multiplied by 0.4, resulting in a final output load parameter of 88.6%. The processor weight and memory weight are selected based on the attribution probability distribution of historical device resource exhaustion leading to system crashes. In this embodiment, the processor weight is set to 0.6 and the memory weight to 0.4. The specific mapping logic of the attribution probability distribution is as follows: All cases of system crashes caused by resource exhaustion during the historical operation of the target model device are statistically analyzed, and the number of cases caused by CPU overload and physical memory exhaustion are extracted. Divide the number of each of these two types of cases by the sum of their respective numbers to obtain two normalized percentage values. These two percentage values ​​are then directly assigned to the processor weight and memory weight values, respectively, ensuring that the sum of the two weight values ​​is always equal to 1.

[0060] The load parameters and load threshold are fed into a comparator to perform a judgment operation. If the load parameters exceed the load threshold, an over-limit interrupt signal is output. The load threshold is selected based on a comprehensive evaluation of the full-load performance inflection point value in the hardware stress test report of the edge computing gateway device and the system's reserved buffer margin. In this embodiment, it is set to 85%.

[0061] In response to the over-limit interruption signal, an access channel to the target access control node is established. Following the access channel, the identity credential code embedded in the target access control node is extracted. The identity credential code is matched against the locally deployed security authentication rule base to read the access control level, which is the hexadecimal value 7.

[0062] When the target computing load exceeds the load threshold, an over-limit interrupt signal is output, and the identity credentials are extracted and the access control level is read through the edge-access channel. This mechanism introduces a strong intervention-level permission verification logic when computing resources are on the verge of exhaustion. When the underlying execution channel is severely crowded by a large number of mechanical control flows, the over-limit interrupt can promptly block the blind spread of non-core processes; and by quickly pulling and verifying high-level permissions, it provides a rigorous and legal basis for subsequent forced deprivation and high-priority allocation of computing resources.

[0063] Furthermore, based on the access control level and the anomaly identifier, a preemption priority is generated, a priority preemption command is sent to the target machine node, low-priority background processes are frozen and idle computing resources are released, and a cooperative control command is output to the object node corresponding to the anomaly identifier; this corresponds to step S4 above; see [link to relevant documentation]. Figure 3 The specific implementation process includes: Extract the first feature weight coefficient corresponding to the fault level code contained in the anomaly identifier, and extract the second feature weight coefficient corresponding to the highest privilege code contained in the access control level; multiply the first feature weight coefficient and the second feature weight coefficient, add a preset emergency compensation constant, and output the preemption priority; send a priority preemption command matching the preemption priority to the target machine node; parse the system resource occupancy table and mutex lock request list of the background process, evaluate whether freezing the background process will cause a deadlock through a directed graph loop detection algorithm, and output a freeze whitelist; for background processes located in the freeze whitelist, push the key running data and program counter register state into the system protection stack for deep sealing, and forcibly modify the state machine parameters to sleep state; release the idle computing resources occupied by the background process; call the idle computing resources to perform logical operations and output the cooperative control instruction payload; encapsulate the cooperative control instruction payload with a protocol header and output the cooperative control instruction; and transmit the cooperative control instruction to the object node corresponding to the anomaly identifier according to the network routing table.

[0064] Specifically, the process of transmitting collaborative control commands to the object node corresponding to the anomaly identifier is as follows: Extract the fault level code from the anomaly identifier; the current anomaly identifier corresponds to a level 1 fault. Extract the highest privilege code from the access control level, corresponding to code 7 in the previous output. Input the fault level code and the highest privilege code into the cross-tabulation logic to output the preemption priority.

[0065] A complete hierarchical mapping relationship is pre-set: for level 4 faults such as general alarms that do not affect operation, the first feature weight coefficient is set to 2; for level 3 faults that cause equipment to operate at reduced capacity, it is set to 4; for level 2 faults that pose a risk of power outage, it is set to 7; and for level 1 faults that pose a risk of breakdown or explosion, it is set to 10. For the basic permissions of on-site operators at levels 1 to 3, the second feature weight coefficient is set to 3; for the management permissions of team leaders at levels 4 to 6, it is set to 6; and for the highest permissions of safety specialists or shift supervisors at level 7, it is set to 9.

[0066] Obtain the first feature weight coefficient corresponding to the first-level fault level. This first feature weight coefficient is selected by step mapping based on the power equipment fault severity assessment matrix. In this embodiment, it is set to 10 for the first-level fault with insulation breakdown risk. Obtain the second feature weight coefficient corresponding to permission code 7. This second feature weight coefficient is selected by matching based on the system network security access control hierarchy directory. In this embodiment, it is set to 9. Obtain the system's preset emergency compensation constant. This constant is set based on the basic priority displacement amount reserved by the underlying real-time operating system for core hardware interrupts. In this embodiment, it is set to 9.

[0067] By multiplying the first feature weight coefficient by the second feature weight coefficient and then adding the system's preset emergency compensation constant, the final result is 99, which corresponds to the highest absolute preemption priority flag value of 99 in the real-time operating system scheduling strategy.

[0068] Based on the newly generated highest absolute hierarchical ranking identity, a priority preemption command matching the preemption priority parameter (priority level 99) is sent to the core processor control unit of the target node via the control system bus. The real-time operating kernel scheduler of the target node receives and responds to this sudden, unmasked priority preemption command signal, forcing the runtime stack to immediately perform a context protection switch. The underlying instruction manager performs a full search of the queues currently running within the target node, filtering and matching all currently running background resident system services whose ranking level is far lower than the preemption priority parameter, i.e., background processes that consume a large amount of processor cycles and memory.

[0069] In this embodiment, to prevent the deadlock detection logic from failing due to dynamic changes in the lock request state of background processes during loop detection, the system kernel must call an atomic operation before starting the directed graph loop detection algorithm to perform a snapshot instruction on the mutex lock holding and request states of all background processes within the current target machine node. This state data is then copied to a separate dedicated memory buffer and isolated and locked. Subsequent directed graph construction and depth-first loop traversal operations are strictly based on the static snapshot data in this dedicated memory buffer.

[0070] In this embodiment, the system resource usage table and mutex lock request list of the background process are parsed to determine whether it holds a shared resource lock required by critical system services. A resource allocation graph simplification algorithm is used to assess whether freezing the process would cause a deadlock, and a freeze whitelist is output. Specifically, the process control block structure of each identified background process is read, and its file descriptor table and mutex lock request queue are extracted. The resource waiting queues for high-priority system services that maintain uninterrupted basic network communication and hardware guards are retrieved. The current resource allocation directed graph is constructed. In the resource allocation directed graph, edges requesting resources that the background process to be evaluated has applied for but not yet allocated are simulated and removed, while edges for allocating resource locks already held are forcibly maintained to generate a resource usage snapshot after the process is forcibly frozen and suspended. An attempt is made to perform node simplification on the directed graph of resource allocation in the snapshot state. If a high-priority system service node is found to be unable to complete simplification due to waiting for a resource lock held by the frozen process, thus causing the high-priority system service to enter a permanent blocked waiting state (i.e., a resource deadlock is determined), the process identifier of the background process is removed from the set to be frozen. If all high-priority system service nodes can acquire the required resources and successfully simplify, or if they only hold non-shared exclusive temporary business resources, their process identifiers are added to the safe freeze whitelist. After the evaluation is completed, freezing is only performed on items in the whitelist. The above process prevents resource deadlock and system-level priority inversion caused by brute-force process termination, ensuring the underlying kernel operation security of the system.

[0071] For the identified background process, the kernel scheduling system invokes the spinlock mechanism and its improved sleep mutex mechanism to perform state freezing. The specific micro-mechanism is as follows: the kernel forcibly deprives the background process of its allocated virtual execution control processor time slice, pushes the process's critical runtime data and program counter register state onto the system protection stack for deep storage and protection, and then forcibly modifies the state machine parameters of the business process from active execution and marks it as sleeping / suspended blocked.

[0072] By forcibly freezing and suspending low-priority background processes on a large scale, the previously severely scarce physical register pipeline and memory buffer pool blocks were instantly and massively released to clear the execution stack space.

[0073] Next, the preset control state determination logic is loaded. Its internal structure is as follows: extracting a first state flag indicating whether the top oil temperature exceeds the upper limit of 80 degrees Celsius, extracting a second state flag indicating whether the temperature change rate exceeds a set threshold, and extracting a third state flag indicating whether the hydrogen content exceeds a preset concentration. When the determination conditions are met, the corresponding state flag is assigned the value 1; otherwise, it is assigned the value 0.

[0074] At the logic execution layer, when the first state flag bit and the second state flag bit are both set to 1, the first logic judgment branch is executed, and the instruction bit code for driving the high-power cooling fan is output. When the first state flag bit is set to 1 and the third state flag bit is also set to 1, the second logic judgment branch is executed, and the instruction lockout bit code for cutting off the on-load tap changer to control the motor is output.

[0075] In this embodiment, the input top oil temperature is 81.5 degrees Celsius and the hydrogen content is 750 parts per million. The system determines that both the first state flag bit and the third state flag bit are 1, which satisfies the activation condition of the second logic judgment branch. Based on this, the system encodes and reassembles the corresponding binary machine code instruction string set, and finally generates and outputs the collaborative control instruction payload data block.

[0076] In accordance with the communication protocol specifications of the manufacturing message, the payload of the newly generated cooperative control command was encapsulated and the protocol header was added. A unique transaction processing matching identifier sequence was embedded in the front of the data packet, and a specific application layer function control code representing the intention of forced operation was attached. The detailed tree-structured semantic addressing logic path of the controlled device object was embedded inside the encapsulation structure, and the cooperative control command was output.

[0077] Finally, the network routing packet delivery engine reads and loads the Ethernet network routing table and the LAN physical network card media access control address allocation pool, queries and confirms the static WAN IP routing address information and the corresponding LAN MAC addressing physical hardware interface for the initial source anomaly identifier's originating physical node. It then delivers the coordinated control command, containing the emergency forced cooling start control word and the voltage regulation system cut-off interlocking mechanical action feature code, to the physical node's body protection control loop corresponding to the anomaly identifier at the very front end.

[0078] By cross-calculating fault levels and permission codes to generate preemption priorities, a priority preemption command is sent to freeze low-priority background processes, releasing computing power to generate and transmit collaborative control commands. This allows scarce underlying computing power to be allocated to high-risk repair tasks. This preemption mechanism fundamentally unblocks the execution channel, ensuring that critical repair commands can be executed with priority and certainty.

[0079] This invention provides a method for monitoring the operational status of power equipment based on heterogeneous data association graphs. By instantiating electrical parameters, computing load, and personnel permissions into physical, machine, and access control nodes and constructing a weighted graph, it breaks through the barriers of traditional single-dimensional monitoring. Furthermore, after anomaly detection, it no longer relies on rigid static rules but instead uses the weighted graph for dynamic cloud-edge collaborative scheduling path optimization. It also introduces a cross-domain resource allocation mechanism, generating preemption priorities based on access control levels and anomaly identifiers when the target node's computing power is overloaded. This logically rigorous collaborative design can identify the urgency of instructions and dynamically allocate computing resources, ensuring that high-risk anomaly instructions obtain the highest priority execution channel, effectively avoiding instruction timeout and discarding, and improving the reliability of emergency management.

[0080] Example 2 This embodiment describes a power equipment operation status monitoring method based on heterogeneous data correlation maps, applied to a critical safety scenario where severe partial discharge occurs inside a 110 kV oil-immersed transformer, leading to a rapid increase in characteristic gases and posing a significant risk of equipment breakdown. The specific implementation process is as follows: The system continuously receives multi-source heterogeneous operational data from power equipment, extracts underlying communication protocol signatures, compares them against a pre-set rule base, and invokes the target protocol unpacking strategy to perform byte stripping processing on the data stream. For physical layer feature data segments, semantic mapping parsing is performed to output equipment status parameters from transformer dissolved gas analysis, currently detecting a hydrogen concentration of 750 parts per million. For system layer feature data segments, format conversion and reassembly are performed to output the computing power load parameters of the edge computing gateway, currently showing a central processing unit utilization of 92% and only 5% of available physical memory remaining. For interaction layer feature data segments, through decryption verification and matching, the access control level of the on-duty maintenance security specialist is output, corresponding to the hexadecimal value 7.

[0081] Within the memory space, the system is divided into physical device mapping areas, computing resource mapping areas, and interaction permission mapping areas. The three types of parameters are imported respectively, and corresponding object nodes, machine nodes, and access control nodes are instantiated. Subsequently, historical collaborative interaction logs are extracted, and cross-domain collaborative request records between these nodes are retrieved. After feature separation, the interaction frequency within the historical observation window is calculated to be 600 times. By calculating the difference and averaging the timestamps of data packets sent and received between Internet Protocol addresses, the latency is calculated to be 8.5 milliseconds.

[0082] The interaction frequency (600 times) is input to the minimax linear scaling model. Linear scaling compression is performed using a preset historical maximum of 1000 times and a minimum of 0 times, outputting a normalized frequency coefficient of 0.6. Simultaneously, the mathematical variance of the delayed time series is calculated, and after inverse proportional mapping, a stability coefficient of 0.85 is output. Multiplying the normalized frequency coefficient by a set positive amplification factor of 0.65 yields a frequency gain variable of 0.39. Multiplying the stability coefficient by a bias correction attenuation factor of 0.35 yields a delay stability gain variable of 0.2975. Finally, aggregation and summation are performed on both, outputting a cross-domain coupling degree of 0.6875.

[0083] The system scans the graph data structure, retrieves the communication handshake logs of the Transmission Control Protocol (TCP), and extracts node pairs with actual application layer data exchange records. Virtual memory pointers are allocated between the node pairs to establish bidirectional data channels (i.e., connection edges). The system converts the cross-domain coupling degree value of 0.6875 into a hexadecimal weighted feature code (e.g., 16A3) that is more easily matched by the computer's underlying layer, writes it to the channel attribute storage area to complete the assignment operation, and finally encapsulates and outputs a weighted graph containing physical connection features.

[0084] Obtain the preset hydrogen safety operation baseline range (normal upper limit set at 700 parts per million concentration). The hardware comparator compares the current detection value (750) with the baseline upper limit, determines that the boundary limit has been exceeded, and immediately outputs an anomaly flag of hexadecimal value 82. Based on this flag, locate the initial object node, and advance the breadth-first search for the probe ray in the weighted graph with a depth constraint of 3 hops. Perform dynamic accumulation on the cross-domain coupling degree of the connected edges, and after eliminating disconnected paths with timeout Internet control message protocol responses, select the high-reliability path with the highest accumulated coupling degree value as the final cloud-edge collaborative scheduling path.

[0085] The cloud-edge collaborative scheduling path was disassembled, and the target machine node and target access control node were extracted. The CPU and memory usage of the target machine node were merged and weighted, resulting in a comprehensive computing load parameter of 93.2%. Since this value exceeded the preset 90% load threshold, an over-limit interrupt signal was triggered and output to the bus. In response to this signal, the system reads the identity credential code embedded in the target access control node along the secure access channel. Matching logic confirmed that the access control level met the system's highest-level intervention requirements.

[0086] The fault level code and the highest privilege code within the access control level are extracted from the anomaly identifier. Through matrix cross-operation and the addition of a low-level emergency compensation constant, the highest preemption priority number (99) in the scheduling strategy is output. A priority preemption command is sent to the target node, invoking the kernel sleep mutex mechanism to forcibly freeze the state of low-priority background processes, instantly releasing ample idle computing resources. Finally, the idle computing resources are used to execute protection logic operations at full speed, generating a coordinated control command payload to drive the main circuit breaker to trip urgently to isolate the discharge fault. After encapsulating the protocol header, the command is transmitted to the object node corresponding to the anomaly identifier according to the network routing table, thus completely completing the state monitoring and control closed loop from perception and discovery to cross-domain preemption and physical blocking.

[0087] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring the operating status of power equipment based on heterogeneous data association maps, characterized in that, include: It receives multi-source heterogeneous operation data of power equipment in the cloud-edge-device architecture, and outputs equipment status parameters, computing load parameters and access control levels by field decoding. The device status parameters, computing load parameters, and access control levels are instantiated as object nodes, machine nodes, and access control nodes. Connection edges are established between the nodes, the interaction frequency and latency between the nodes are extracted, and aggregation operations are performed to obtain the cross-domain coupling degree. The cross-domain coupling degree is assigned to the connection edges to construct a weighted graph. When the device status parameter corresponding to the object node exceeds the preset boundary, an abnormality flag is output. Path traversal is performed in the weighted graph, and the path with the highest cumulative cross-domain coupling degree is selected as the cloud-edge collaborative scheduling path. Extract the target machine node and target access control node contained in the cloud-edge collaborative scheduling path, and when the load parameter corresponding to the target machine node exceeds the load threshold, read the corresponding access control level. Based on the access control level and the exception identifier, a preemption priority is generated, a priority preemption instruction is sent to the target machine node, the internal low-priority background processes are frozen and idle computing resources are released, and the cooperative control instruction is output to the object node corresponding to the exception identifier.

2. The method for monitoring the operating status of power equipment based on heterogeneous data association graphs according to claim 1, characterized in that, The specific generation process of the device status parameters, computing load parameters, and access control levels includes: Extract the underlying communication protocol signatures carried in the multi-source heterogeneous operation data of power equipment; compare the signatures with a pre-set protocol parsing rule base and output a matching target protocol unpacking strategy; invoke the target protocol unpacking strategy to perform byte stripping processing on the multi-source heterogeneous operation data of power equipment, separating physical layer feature data segments, system layer feature data segments, and interaction layer feature data segments; perform semantic mapping parsing on the physical layer feature data segments to output device status parameters; perform format conversion and reorganization on the system layer feature data segments to output computing power load parameters; and perform decryption verification and matching on the interaction layer feature data segments to output access control levels.

3. The method for monitoring the operating status of power equipment based on heterogeneous data association graphs according to claim 1, characterized in that, The specific process for generating the interaction frequency and delay time includes: Within the graph data structure, divide it into physical device mapping area, computing resource mapping area, and interaction permission mapping area; import device status parameters to generate object nodes into the physical device mapping area; import computing power load parameters to generate machine nodes into the computing resource mapping area; import access control levels to generate access control nodes into the interaction permission mapping area; extract historical collaborative interaction logs carried by the graph data structure; retrieve cross-domain collaborative request records initiated between object nodes, machine nodes, and access control nodes; perform feature separation on the cross-domain collaborative request records to extract the total number of calls and output the interaction frequency; extract the time difference between the cross-domain collaborative request initiation point and the receiving response point in the cross-domain collaborative request records and output the delay time.

4. The method for monitoring the operating status of power equipment based on heterogeneous data association graphs according to claim 1, characterized in that, The specific process for generating the cross-domain coupling degree includes: The interaction frequency is input to the minimax linear scaling model, compressed through numerical transformation, and the normalized frequency coefficient is output. Distribution characteristic parameters of all delay times within a preset historical observation window are collected. The mathematical variance corresponding to the distribution characteristic parameters is calculated, and the stability coefficient is output. A positive amplification factor corresponding to the normalized frequency coefficient is configured. A correction attenuation factor corresponding to the stability coefficient is configured. The normalized frequency coefficient and the positive amplification factor are multiplied to output a frequency gain variable. The stability coefficient and the correction attenuation factor are multiplied to output a delay stability gain variable. The frequency gain variable and the delay stability gain variable are aggregated and summed to calculate the cross-domain coupling degree.

5. The method for monitoring the operating status of power equipment based on heterogeneous data association graphs according to claim 1, characterized in that, The specific process for generating the weighted graph includes: The process involves: scanning all object nodes, machine nodes, and access control nodes within the scan graph data structure; retrieving the underlying communication handshake logs of all object nodes, machine nodes, and access control nodes; comparing the communication handshake logs to extract node pairs with actual data exchange records; allocating virtual memory pointers between node pairs to establish bidirectional data channel output connection edges; converting the cross-domain coupling degree corresponding to the node pairs into hexadecimal weight feature codes; writing the hexadecimal weight feature codes into the attribute storage area of ​​the bidirectional data channel to complete the cross-domain coupling degree assignment operation; and encapsulating all object nodes, machine nodes, access control nodes, and the bidirectional data channel with the hexadecimal weight feature codes written in them to output a weighted graph.

6. The method for monitoring the operating status of power equipment based on heterogeneous data association graphs according to claim 1, characterized in that, The specific process for generating the cloud-edge collaborative scheduling path includes: Obtain a preset safe operation benchmark interval; input the device status parameters contained in the object node and the safe operation benchmark interval into the comparator to perform comparison logic and output the difference comparison value; determine if the difference comparison value exceeds the boundary limit and generate an anomaly label; combine the anomaly label to extract the object node that generates the device status parameters as the initial object node; set the depth constraint condition of the breadth-first search algorithm; advance the probe ray hop by hop from the initial object node along the connection edge based on the depth constraint condition; perform dynamic accumulation on the cross-domain coupling degree carried by the connection edge through which the probe ray passes and output the candidate path set; verify the network connectivity code traversed by each candidate path in the candidate path set; eliminate candidate paths whose network connectivity code matches the disconnection state and output the candidate sequence; compare the accumulated cross-domain coupling degree contained in the candidate sequence; extract the candidate path corresponding to the top position of the sort and output it as the cloud-edge collaborative scheduling path.

7. The method for monitoring the operating status of power equipment based on heterogeneous data association maps according to claim 1, characterized in that, The specific process of reading the access control level corresponding to the target access control node includes: The system performs a disassembly operation on the cloud-edge collaborative scheduling path, outputting the target machine node and the target access control node; it integrates the processor thread occupancy ratio and remaining memory capacity to output load parameters; it obtains a preset load threshold; it sends the load parameters and load threshold to a comparator to perform a judgment operation; it determines that the load parameters exceed the load threshold and outputs an over-limit interrupt signal; it responds to the over-limit interrupt signal to establish an access channel to the target access control node; it extracts the identity credential code embedded in the target access control node along the access channel; and it performs matching logic on the identity credential code to read the access control level.

8. The method for monitoring the operating status of power equipment based on heterogeneous data association graphs according to claim 1, characterized in that, The specific process of outputting the collaborative control command to the object node corresponding to the anomaly identifier includes: Extract the first feature weight coefficient corresponding to the fault level code contained in the anomaly identifier, and extract the second feature weight coefficient corresponding to the highest privilege code contained in the access control level; multiply the first feature weight coefficient and the second feature weight coefficient, add a preset emergency compensation constant, and output the preemption priority; send a priority preemption command matching the preemption priority to the target machine node; parse the system resource occupancy table and mutex lock request list of the background process, evaluate whether freezing the background process will cause a deadlock through a directed graph loop detection algorithm, and output a freeze whitelist; for background processes located in the freeze whitelist, push the key running data and program counter register state into the system protection stack for deep sealing, and forcibly modify the state machine parameters to sleep state; release the idle computing resources occupied by the background process; call the idle computing resources to perform logical operations and output the cooperative control instruction payload; encapsulate the cooperative control instruction payload with a protocol header and output the cooperative control instruction; and transmit the cooperative control instruction to the object node corresponding to the anomaly identifier according to the network routing table.

Images