A cloud-edge collaborative data processing method and system for remote pressure operation

Abstract

The embodiment of the application provides a cloud-edge collaborative data processing method and system for remote pressure operation, and belongs to the technical field of edge computing. The method comprises the following steps: collecting operation data of the pressure operation equipment, and generating label information for expressing a job state based on the operation data at an edge node; constructing a task graph for representing a job flow based on the label information, and dividing the task graph into an edge subgraph and a cloud subgraph according to control requirements, so as to execute the corresponding subgraphs at the edge node and the cloud platform respectively; generating a data processing result for reflecting a job state evolution trend based on the execution result of the edge subgraph and the execution result of the cloud subgraph; generating a job stage snapshot based on the data processing result, and uploading the snapshot to the cloud for archiving at the edge node. The scheme realizes real-time identification and structured retention of key states in the whole pressure operation process, and improves the response accuracy of remote control and the traceability of the operation process.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of edge computing, in particular to a cloud-edge collaborative data processing method for remote pressure operation and a cloud-edge collaborative data processing system for remote pressure operation. BACKGROUND

[0002] With the development of remote pressure operation technology in oil and gas fields, automated operation equipment has gradually replaced manual operation in high-risk operation areas, and cooperates with remote control terminals to complete operation tasks. In this process, a large amount of operation data generated by the equipment during operation (such as action response, electro-hydraulic parameters, image monitoring information, etc.) becomes a key basis for supporting operation safety assessment and process monitoring. However, existing solutions mostly rely on a single data upload mechanism to synchronize all raw data to a cloud platform for unified processing and analysis, which has problems such as high data transmission delay, large resource consumption, slow feedback response, etc. Especially under the condition of unstable well site network or limited bandwidth, it is more likely to cause processing lag or lack of key state.

[0003] In addition, due to the frequent state transition, nonlinear operation behavior and transient mutation in the execution process of the pressure operation equipment, the traditional rule matching or static label-based method is difficult to accurately express the current operation stage state, resulting in insufficient accuracy of the system in task scheduling, risk warning or snapshot archiving, etc. At the same time, the current data processing framework generally lacks the modeling ability of the dynamic evolution characteristics of the operation process, cannot effectively extract the evolution trend of the operation state in the time dimension, and also lacks a mechanism for fusion analysis of the difference changes between stages.

[0004] Therefore, how to construct a dynamic expression mechanism for the operation process during the pressure operation process, and under the premise of ensuring the response real-time and resource utilization efficiency, to complete the accurate expression and structured processing of the key operation state, has become one of the technical problems to be solved in this field. SUMMARY

[0005] The purpose of the embodiments of the present application is to provide a cloud-edge collaborative data processing method and system for remote pressure operation, to at least solve the problems of high data processing delay, inaccurate state recognition and lack of operation process evolution modeling in the prior art.

[0006] To achieve the above object, the first aspect of the present application provides a cloud-edge collaborative data processing method for remote pressure operation, which is applied to data processing of pressure operation equipment with automatic operation components and remote control terminal, and comprises the following steps: collecting operation data of the pressure operation equipment during operation, and generating label information for expressing operation state based on the operation data at an edge node; constructing a task graph for representing operation flow based on the label information, and dividing the task graph into edge sub-graph and cloud sub-graph according to control requirements, so as to execute corresponding sub-graphs at the edge node and cloud platform respectively; generating a data processing result for reflecting evolution trend of operation state based on execution results of the edge sub-graph and the cloud sub-graph; generating an operation stage snapshot based on the data processing result, and uploading the snapshot to the cloud platform for archiving at the edge node.

[0007] Optionally, the label information comprises an action identification label and a state confidence label extracted based on operation data; the action identification label is used to reflect specific action type of operation components in current operation flow; the state confidence label is generated based on a fuzzy confidence model built in the edge node, and is used to quantify stability degree of operation state.

[0008] Optionally, generating label information for expressing operation state based on the operation data at the edge node comprises the following steps: performing change feature extraction based on the operation data, and constructing an initial feature vector containing action duration, pipe column response amplitude and boundary image definition; calling a lightweight label generation model deployed in the edge node, performing inference on the initial feature vector, and outputting a candidate label result containing candidate action categories and corresponding confidence scores; taking the action categories in the candidate label result as the action identification label, and converting the confidence scores in the candidate label result into state confidence labels by using a phased adaptive mapping function; if the confidence scores are lower than a phased adaptive threshold set in the edge node, marking the current label information as unstable, and uploading the initial feature vector to the cloud platform, so that the cloud platform performs inference and returns complete action identification label and state confidence label as label information output; if the confidence scores are not lower than the phased adaptive threshold, the edge node directly arranges the candidate label result as label information output.

[0009] Optionally, the task graph for representing the job flow is constructed based on the label information, including: generating an initial task node set for each operation stage in the job flow according to the action identification label and the state confidence label contained in the label information, and arranging each initial task node according to the time sequence of the action identification label to obtain a task node sequence; establishing a directed connection relationship between the node sequences based on a preset stage dependency rule and a fluctuation trend of the state confidence label, and constructing a task graph with operation sequence constraints and state continuity; wherein the task graph is a directed acyclic graph, each node carries a corresponding label field and a stage number, and each edge carries a priority weight and a state dependency strength value for graph partitioning.

[0010] Optionally, the preset stage dependency rule is: based on the label information of each stage in the job flow, judging whether the operation types between adjacent labels constitute a sequential dependency relationship, if the action identification label of the subsequent label belongs to a predefined dependent action set, then a directed edge relationship is established between the corresponding nodes in the task graph; judging whether there is state transition uncertainty under a preset state transition criterion in the current stage according to the change trend of the state confidence label, if a decreasing trend of confidence or a fuzzy stage boundary is detected, then an auxiliary edge is connected to the predecessor node to express the reversible dependent path; wherein the dependent action set and the preset state transition criterion are both generated by modeling historical job data before the task graph is constructed, and are embedded into the edge node and the job logic library of the cloud platform in the form of a rule configuration file.

[0011] Optionally, the task graph is divided into edge sub-graphs and cloud sub-graphs according to control requirements, so as to execute the corresponding sub-graphs in the edge node and the cloud platform, including: performing preliminary screening in the edge node, dividing task nodes with a control delay index less than a preset response threshold value and a computing resource index lower than an edge node resource threshold value into edge sub-graphs for execution by the edge node; screening out task nodes whose task dependency paths span multiple job stages or require fusion of multiple sub-label features for judgment from the task nodes remaining after screening and not divided into edge sub-graphs, and dividing them into cloud sub-graphs for execution by the cloud platform; marking task nodes not divided by the edge sub-graphs or the cloud sub-graphs as low adaptation nodes, and performing delayed execution or consolidation processing by the cloud platform according to a resource dynamic scheduling strategy at runtime.

[0012] Optionally, based on the execution result of the edge subgraph and the execution result of the cloud subgraph, a data processing result reflecting the evolution trend of the job state is generated, including: extracting the stage label sequence and the parameter change sequence in the execution result of the edge subgraph, constructing a stage index timeline for identifying the order of the job stages based on the stage label sequence, and constructing a state fluctuation curve for reflecting the dynamic response within the stage based on the parameter change sequence; extracting the global job state transition path and the associated parameter characteristics in the execution result of the cloud subgraph, performing key jump node information identification based on the transition path, and constructing a stage difference vector group for measuring the state offset degree before and after the stage conversion based on the associated parameter characteristics; fusing the stage index timeline, the state fluctuation curve, the key jump node information, and the stage difference vector group to generate a data processing result reflecting the evolution trend of the job state.

[0013] Optionally, the stage index timeline, the state fluctuation curve, the key jump node information, and the stage difference vector group are fused and processed to generate a data processing result reflecting the evolution trend of the job state, including: constructing a unified evolution reference axis based on the stage index timeline, and extracting a corresponding mutation window segment centered on each key jump node on the evolution reference axis; within each mutation window segment, calculating a segment response gradient curve based on the state fluctuation curve and the stage difference vector group, the response gradient curve being a continuous function of state change rate over time; normalizing all response gradient curves by length and performing first derivative transformation, calculating the dynamic time warping distance between each pair of curves; under the condition that the dynamic time warping distance is less than a preset similarity threshold, the corresponding mutation window segment is classified into the same evolution mode cluster, and a trend confidence score is generated based on the sample number and the cross-stage coverage of each mode cluster; mapping the trend confidence score to the evolution reference axis to generate a data processing result expressing the evolution trend of the job state.

[0014] Optionally, based on the data processing result, a job stage snapshot is generated, and the snapshot is uploaded to the cloud archive at the edge node, including: extracting the structural information reflecting the evolution trend of the job in the data processing result; based on the structural information, marking the feature state at the trend inflection point in each evolution stage, and attaching the operation data and task graph node record of the current stage to construct a structured snapshot; performing filtering and sorting on the structured snapshot to obtain a high-sensitivity snapshot set expressing the evolution characteristics; performing lightweight encoding compression processing on the high-sensitivity snapshot set, and uploading to the cloud archive.

[0015] The second aspect of the present application provides a cloud-edge collaborative data processing system for remote pressure operation, which is applied to data processing of a pressure operation device with an automatic operation assembly and a remote control terminal. The system comprises: a collection unit configured to collect operation data of the pressure operation device during operation and generate label information for expressing operation status based on the operation data at an edge node; a division unit configured to construct a task graph for representing an operation flow based on the label information, and divide the task graph into an edge subgraph and a cloud subgraph according to control requirements, so as to execute the corresponding subgraphs at the edge node and the cloud platform respectively; a processing unit configured to generate a data processing result for reflecting an evolution trend of the operation status based on execution results of the edge subgraph and the cloud subgraph; and an output unit configured to generate an operation stage snapshot based on the data processing result and upload the snapshot to the cloud platform for archiving at the edge node.

[0016] Through the above technical solution, the present application scheme generates label information for expressing operation status in real time at an edge node, constructs a task graph to divide an edge subgraph and a cloud subgraph, realizes hierarchical processing of operation data and collaborative execution of tasks, can effectively reduce data processing delay and improve timeliness and accuracy of status recognition. Meanwhile, the execution results of the edge and the cloud are fused to generate a data processing result for reflecting an evolution trend of the operation status, and a key snapshot is extracted for archiving accordingly, which enhances the time sequence modeling capability of the operation process and the availability of subsequent backtracking analysis.

[0017] Other features and advantages of the present application will be described in detail in the following detailed description. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings are included to provide a further understanding of the present application and constitute a part of the specification, which together with the detailed description, serve to explain the present application. In the drawings:

[0019] Figure 1 is a step flow chart of a cloud-edge collaborative data processing method for remote pressure operation provided by an embodiment of the present application;

[0020] Figure 2 is a full-process pressure response curve diagram of remote pressure operation provided by an embodiment of the present application;

[0021] Figure 3 is a system structure diagram of a cloud-edge collaborative data processing system for remote pressure operation provided by an embodiment of the present application. DETAILED DESCRIPTION

[0022] The specific embodiments of the present application are described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are intended to illustrate and explain the present application, and are not intended to limit the present application.

[0023] The pressurized operation equipment with automatic operation assembly and remote control terminal applied in the present application scheme refers to engineering equipment that can carry out operation tasks without pressure stop, and the core feature is that it can execute sealing, hole opening, maintenance, pipe replacement and other operations while maintaining the running state of high-pressure pipelines or pressure vessels. Such equipment is generally equipped with automatic operation assemblies with self-positioning and self-execution capabilities, such as hydraulic arms, mechanical tool heads, intelligent sealing structures, etc., which can accurately complete specific operation actions. At the same time, the equipment also includes a communication module and a control interface that interact with a remote operation control platform, and a remote control terminal can monitor and intervene in the operation process through video feedback, data link, task instructions, etc. Such equipment is mainly used in fields such as petrochemical industry and natural gas transportation, which have very high requirements for safety and continuity in working conditions.

[0024] Figure 1 is a step flow chart of a cloud-edge collaborative data processing method for remote pressurized operation provided by an embodiment of the present application. As shown in Figure 1 The present application embodiment provides a cloud-edge collaborative data processing method for remote pressurized operation, which is applied to data processing of pressurized operation equipment with automatic operation assembly and remote control terminal. The method comprises the following steps:

[0025] Step S10: Collecting operation data of the pressurized operation equipment during the operation process, and generating label information for expressing the operation state based on the operation data at the edge node.

[0026] Specifically, the label information includes: action identification label and state confidence label extracted based on operation data; the action identification label is used to reflect the specific action type of the operation assembly in the current operation process; the state confidence label is generated based on the built-in fuzzy confidence model in the edge node, and is used to quantify the stability degree of the operation state.

[0027] Further, the edge node generates label information for expressing a job state based on the operation data, including: performing change feature extraction based on the operation data, constructing an initial feature vector containing action duration, pipe column response amplitude and boundary image definition; calling a lightweight label generation model deployed in the edge node, performing inference on the initial feature vector, and outputting a candidate label result containing candidate action categories and corresponding confidence scores; taking the action category in the candidate label result as an action identification label, and converting the confidence score in the candidate label result into a state confidence label by a phased adaptive mapping function; if the confidence score is lower than the phased adaptive threshold set in the edge node, marking the current label information as unstable, and uploading the initial feature vector to the cloud platform, performing inference by the cloud platform and returning the complete action identification label and state confidence label as label information output; if the confidence score is not lower than the phased adaptive threshold, the edge node directly arranges the candidate label result as label information output.

[0028] In the embodiment of the application, when the under-pressure operation equipment is in a running state, its operation data is continuously collected, which can include but is not limited to motion parameters (such as displacement, acceleration) of the operation assembly, execution pressure curve, real-time video stream, boundary visual image, vibration response data, etc. After the edge node receives the operation data, it first performs feature extraction processing to construct an initial feature vector. The feature extraction process can use a predefined sliding time window strategy to aggregate and normalize the operation data within a certain time span, thereby reducing noise influence and enhancing time continuity. The extracted features include but are not limited to three aspects: 1) action duration, i.e. operation action duration; 2) pipe column response amplitude, i.e. displacement, stress response or vibration response features generated by the operation pipe column during specified action execution; and 3) boundary image definition, which reflects the structural definition and occlusion degree of the camera image boundary area, and can be obtained by edge gradient analysis or texture intensity calculation.

[0029] Based on the initial feature vector constructed above, the edge node calls the deployed lightweight label generation model to perform model inference. The model usually uses a shallow convolutional network or a small graph structure model, which has the characteristics of high inference efficiency and low resource occupation, and is suitable for embedded edge device running environment. The output of the model inference is a set of candidate label results, containing multiple possible action categories and their corresponding confidence scores. The action categories here correspond one-to-one to the operation nodes in the preset operation process, such as "hole positioning", "pipe column insertion", "tool head recovery", etc., and the confidence score represents the degree of determination of the model on the classification result.

[0030] After obtaining the candidate label result, the action class in it needs to be extracted as an action identification label, and a phased adaptive mapping function is performed on the confidence score to generate a state confidence label. The so-called phased adaptive mapping refers to dynamically adjusting the correspondence between the score and the state confidence according to the stability threshold allowed by the current work phase. For example, in the initial positioning phase, a lower confidence may be tolerated and still judged as a stable state, while in the key operation such as high-pressure interface locking phase, a higher confidence is required to determine that the state is stable. The mapping function can be a piecewise linear function, a Sigmoid function or a nonlinear function based on statistical historical data regression training.

[0031] When the confidence score of the candidate label result is not lower than the set phased adaptive threshold, the edge node is considered to be locally reliable, and the action class and the confidence label are directly output as the label information at the current time. The label information contains two parts of content: the action identification label, which is used for subsequent construction of the work flow task graph; and the state confidence label, which is used to determine the stability of the current label and whether to trigger a remediation mechanism.

[0032] If the confidence score is lower than the phased adaptive threshold, it is considered that the edge judgment result is insufficient in reliability and is not suitable for direct use. At this time, the initial feature vector is reported to the cloud platform, and the current phase label is marked as unstable. The cloud platform has a more complex and larger resource-consuming label inference model, which can complete and correct the label based on historical context, cross-device association information or a more powerful graph neural structure. After the cloud platform is processed, the action identification label and the state confidence label after enhanced inference are returned as the final label information to overwrite the original edge result.

[0033] Step S20: constructing a task graph for representing a work flow based on the label information, and dividing the task graph into edge subgraphs and cloud subgraphs according to control requirements, so as to execute corresponding subgraphs at the edge node and the cloud platform respectively.

[0034] Constructing a task graph for representing a work flow based on the label information includes: generating an initial task node set for operation phases in the work flow according to the action identification label and the state confidence label contained in the label information, and arranging each initial task node according to the time sequence of the action identification label to obtain a task node sequence; establishing a directed connection relationship between the node sequence based on a preset phase dependency rule and a fluctuation trend of the state confidence label, and constructing a task graph with operation sequence constraints and state continuity; wherein the task graph is a directed acyclic graph, each node carries a corresponding label field and a phase number, and each edge carries a priority weight and a state dependency strength value for graph division.

[0035] Specifically, the preset stage dependency rule is: based on the label information of each stage in the job flow, it is judged whether the operation type between adjacent labels constitutes a sequential dependency relationship, if the action identifier label of the next label belongs to a predefined dependent action set, a directed edge relationship is established between the corresponding nodes in the task graph; according to the change trend of the state confidence label, it is judged whether there is state transition uncertainty under the preset state transition criterion in the current stage, if it is detected that the confidence has a downward trend or the stage boundary is fuzzy, an auxiliary edge is introduced to the predecessor node to express the reversible dependent path; wherein, the dependent action set and the preset state transition criterion are both modeled by historical job data before the task graph is constructed, and embedded into the edge node and the job logic library of the cloud platform in the form of a rule configuration file.

[0036] In the embodiment of the application, the task graph construction process is first based on the label information generated by the edge node, wherein the label information is composed of action identifier labels and state confidence labels. The action identifier label is used to identify the specific operation action of the job component in the current time period, such as "tool insertion", "valve opening", "high pressure connection", etc.; the state confidence label is used to quantify the stability degree of the current action state, reflecting the dynamic volatility and execution reliability in the job process.

[0037] Before constructing the task graph, the label information generated on the continuous time axis needs to be arranged in stages. First, the time periods with different action identifier labels in adjacent time periods are regarded as independent operation stages, and an initial task node set is constructed based on the start and end time stamps of each operation stage. Each initial task node carries three core fields: one is the action identifier label of the stage to which the node belongs; two is the time boundary information, which is used to construct the time sequence logic of the task graph; three is the stage number, which identifies the relative position in the complete job flow.

[0038] After obtaining the initial task node set, the task nodes are arranged according to the time sequence of the label information, forming a time-ordered node sequence. Here, time stamp deduplication and merging of short-time interference nodes (such as transition actions less than a preset time) are used to ensure that the stages reflected by the task graph have representativeness and structural stability.

[0039] Next, the core graph structure construction step is entered. This step establishes a directed connection relationship between nodes based on the preset stage dependency rule and the trend change of the state confidence label. The essence of the stage dependency rule is to judge whether there is a control dependency relationship between two stages through the logic of operation action, control requirements and historical experience.

[0040] The specific rules are as follows: if the action identification label of node N(i) is a normal control action, and the action identification label of N(i+1) belongs to a predefined "dependent action set", it is considered that there is a sequential constraint dependent relationship between the two, and a directed edge from N(i) to N(i+1) needs to be added in the task graph. The dependent action set is a set of predefined action types, such as "pipe insertion"→ "valve closure"→ "pressure injection start", which forms an action chain with physical or operational logic irreversible dependence. This set is usually formed by a large number of historical operation processes with fixed time sequence operation pairs.

[0041] Further, to enhance the expression ability of the task graph to the actual state uncertainty, a "state transition uncertainty criterion" is introduced. The judgment principle is: if the state confidence label in the adjacent nodes shows a downward trend, that is, the confidence score is significantly lower in N(i+1) than in N(i), or the confidence label fluctuates sharply in a certain stage (for example, exceeds the pre-set confidence standard deviation threshold), it is considered that there is a problem of state ambiguity or unclear stage division in this stage.

[0042] At this time, the "auxiliary edge" mechanism is introduced, that is, a edge from the current node to the predecessor node is added in the task graph to express a "reversible dependence" relationship. This kind of edge does not participate in the control flow judgment of the main path of the task graph, but can be used as a compensation path or risk propagation link in subsequent operation exception analysis or graph segmentation operation.

[0043] In addition, in order to make the task graph have computability and divisibility, a structural weight needs to be assigned to each graph edge, including: task priority (used for subsequent execution control sorting) and state dependence strength value (used to represent the closeness of state dependence between nodes). The state dependence strength value can be obtained by comprehensive evaluation of the state confidence label difference, historical correlation score and confidence propagation model of adjacent nodes.

[0044] The final task graph is a directed acyclic graph (DAG) with clear stage number, operation label field and state constraint logic, which can accurately express the logical path, operation evolution and state evolution of the entire operation process. This graph structure not only provides clear subgraph boundaries for cloud edge division (such as extracting subgraphs with cut points or low dependence edges of the graph), but also provides a structural basis for subsequent operation evolution trend analysis.

[0045] It should be emphasized that the dependent action set and the state transition criterion are not static presets, but are formed by analyzing common stage sequences, abnormal state transition patterns and operation failure records in historical operation data, and have certain data adaptability. The modeling results are embedded in the operation logic library of the edge node and the cloud platform in the form of rule configuration file, which can be continuously iterated according to the field experience, ensuring that the task graph construction process has both universality and flexibility in adjusting to different operation environments.

[0046] Further, the task graph is divided into edge sub-graphs and cloud sub-graphs according to control requirements, so as to execute the corresponding sub-graphs on the edge node and the cloud platform respectively, including: performing preliminary screening on the edge node, dividing task nodes with a control delay index less than a preset response threshold value and a computing resource index lower than an edge node resource threshold value into an edge sub-graph for execution on the edge node; screening the remaining task nodes that are not divided into the edge sub-graph after screening, screening out task nodes with a task dependency path spanning multiple job stages or requiring fusion of multiple sub-label features for judgment, and dividing the task nodes into a cloud sub-graph for execution on the cloud platform; and marking task nodes not divided into the edge sub-graph or the cloud sub-graph as low-adaptation nodes, and performing delayed execution or combined processing on the low-adaptation nodes by the cloud platform according to a resource dynamic scheduling strategy.

[0047] In the embodiment of the present application, after the task graph is constructed, preliminary screening is performed on the task nodes to determine a task set suitable for local execution on the edge node. The preliminary screening process is based on two core judgment indexes: one is the control delay index of the task, and the other is the computing resource index of the task. Specifically, the control delay index is used to represent whether the control feedback of the task node has a high real-time requirement for the job, and is usually obtained based on the time required for control feedback closed loop in historical execution data. If the index is lower than a preset response threshold value, it indicates that the task has a high requirement for control timeliness and is suitable for responding nearby on the edge side.

[0048] At the same time, the computing resource index of each task node, i.e., the estimated computing load (such as model inference time, memory occupation, thread concurrency overhead, etc.) required for executing the task, needs to be evaluated and compared with the current schedulable resource pool of the edge node. When the computing resource index of the task node is lower than the resource threshold value set by the edge node and at the same time meets the control delay index requirement, the task node is divided into a node in the edge sub-graph. The edge sub-graph represents a task set that can be executed locally on the edge node and has a real-time control priority, ensuring that the edge node can complete the fast response and control closed loop of the key action under low delay conditions.

[0049] For the remaining task nodes that are not divided into the edge sub-graph, further feature structure judgment needs to be performed to determine whether they are suitable for being divided into the cloud sub-graph. The division rules include the following two aspects:

[0050] 1) Determine whether the task dependency path spans multiple job stages. If there is a strong dependency edge (such as confidence state transmission, action delay compensation, etc.) between a task node and multiple non-continuous stages, it indicates that the control logic or state judgment corresponding to the node has global dependency across stages. Such nodes are difficult to perform effective inference and scheduling on the edge side and need to be divided into the cloud sub-graph for centralized processing by the cloud platform based on the global job flow logic.

[0051] 2) If the action judgment of a task node needs to fuse multiple sub-label features (for example, referring to image clarity, pipe column response amplitude, execution stability score and other multi-source data at the same time), it indicates that the task judgment complexity is high, and a fusion model or a historical data driven model is suitable for feature fusion judgment, and such models often require higher computing resources and historical sample support. Therefore, such nodes are also divided into a cloud subgraph, and the inference is performed by a cloud platform.

[0052] The nodes divided into the cloud subgraph can perform graph division sub-tasks concurrent processing or full graph path prediction optimization in the cloud platform, and are suitable for task node processing scenarios that require stronger global modeling capability.

[0053] It should be noted that after the above two subgraph divisions, there may still be some task nodes that do not meet the edge side execution capability, do not have clear cross-stage dependency or fusion feature requirements, and are in a state of resource sensitivity and logic boundary ambiguity. For such task nodes that are not divided into the edge subgraph or the cloud subgraph, special marking is required, and they are defined as "low adaptation nodes". These nodes do not directly participate in the control scheduling of the regular task graph path, but are flexibly processed by the cloud platform according to the dynamic resource scheduling strategy during job running.

[0054] Specifically, the cloud platform monitors various resource indicators (such as the current number of concurrent tasks, model loading load, network bandwidth usage, etc.) during running. When the system load is at a low level, these low adaptation nodes can be scheduled for delayed execution tasks or executed in a resource idle window. If such nodes have shared input data or similar phase numbers with other high priority nodes in the graph, they can also be combined with adjacent nodes to form a group of inference units through combined processing logic, in order to save scheduling resources and improve overall processing efficiency.

[0055] Step S30: Based on the execution results of the edge subgraph and the execution results of the cloud subgraph, a data processing result reflecting the evolution trend of the job state is generated.

[0056] Specifically, based on the execution result of the edge subgraph and the execution result of the cloud subgraph, a data processing result for reflecting the evolution trend of the job state is generated, including: extracting the stage label sequence and the parameter change sequence in the execution result of the edge subgraph, constructing a stage index timeline for identifying the order of job stages based on the stage label sequence, and constructing a state fluctuation curve for reflecting the dynamic response within the stage based on the parameter change sequence; extracting the global job state transition path and the associated parameter characteristics in the execution result of the cloud subgraph, performing key jump node information identification based on the transition path, and constructing a stage difference vector group for measuring the state offset degree before and after the stage conversion based on the associated parameter characteristics; the stage index timeline, the state fluctuation curve, the key jump node information and the stage difference vector group are fused and processed to generate a data processing result for reflecting the evolution trend of the job state.

[0057] Further, the stage index timeline, the state fluctuation curve, the key jump node information and the stage difference vector group are fused and processed to generate a data processing result for reflecting the evolution trend of the job state, including: constructing a unified evolution reference axis based on the stage index timeline, and extracting a corresponding mutation window segment centered on each key jump node on the evolution reference axis; within each mutation window segment, a segment response gradient curve is calculated based on the state fluctuation curve and the stage difference vector group, and the response gradient curve is a continuous function of the state change rate with time; all response gradient curves are normalized by length and subjected to first derivative transformation, and the dynamic time warping distance between the curves is calculated; under the condition that the dynamic time warping distance is less than a preset similarity threshold, the corresponding mutation window segment is classified into the same evolution mode cluster, and a trend confidence score is generated based on the sample number and the cross-stage coverage of each mode cluster; the trend confidence score is mapped to the evolution reference axis to generate a data processing result for expressing the evolution trend of the job state.

[0058] In the embodiment of the application, the dynamic evolution information within the job stage is extracted based on the execution result of the edge subgraph. Since the edge subgraph is usually responsible for processing operation nodes with relatively high real-time requirements and relatively low computational complexity, its execution result contains rich operation labels and continuous state parameters. Specifically, they include:

[0059] 1) Stage label sequence extraction: from the task node results completed by the edge nodes, the action identification label and the state confidence label carried are extracted, and the stage label sequence is obtained by timestamp sorting. Each label corresponds to an explicit operation action, which can be regarded as a boundary marker of stage activity, and is used to construct the main clue of job stage division.

[0060] 2) Parameter variation sequence construction: Synchronously extract the state parameter values recorded by each node (such as pressure, displacement, temperature, voltage, etc.), and form a continuous parameter variation sequence according to the stage label index to express the response process of the physical quantity or sensor data within the stage.

[0061] On the basis of the above, the edge side job dynamic reconstruction is performed, and the stage index timeline and state fluctuation curve are constructed. The stage index timeline is used to represent the stage switching sequence in the time dimension and provide a main reference for subsequent jump analysis; the state fluctuation curve is used to describe the response change trend within each stage, which is usually a multi-variable change curve with time, and can be used to identify the stability and fluctuation intensity within the stage.

[0062] Based on the execution results of the cloud subgraph, the job process transfer structure is extracted from a macro perspective. Since the cloud has stronger computing resources and global information perception ability, its execution tasks usually involve full-process modeling, multi-source information fusion and complex dependence analysis. The key steps include:

[0063] 1) Global job state transfer path extraction: By tracking the key dependence path in the task graph, the state transfer link across stages is constructed, and the state jump nodes and potential nonlinear response sections in the operation process are identified. The transfer path represents the evolution sequence and logical dependence between different stages in the form of a directed chain.

[0064] 2) Associated parameter feature construction: For each stage node in the transfer path, extract its output parameter mean, slope, maximum change rate and other statistical features, and construct a feature vector set reflecting the response difference between different stages. Then form a stage difference vector group for measuring the state change degree before and after the stage conversion, providing a quantitative basis for jump identification and pattern classification.

[0065] In the edge and cloud information fusion stage, a unified evolution analysis framework needs to be established, which includes:

[0066] 1) Construct the evolution reference axis: Based on the stage index timeline constructed on the edge side, a unified coordinate axis (i.e. evolution reference axis) is established in the time dimension, which is used for synchronous reference and alignment of all trend analysis results. This reference axis takes the stage label switching point as the segmentation anchor point, forming a discrete structure with equal time steps.

[0067] 2) Identify the key jump window paragraph: Taking the key jump node identified by the cloud as the center, expand a fixed time window along the evolution reference axis, forming a mutation window paragraph for local analysis. Each paragraph contains a local segment of the state fluctuation curve and a cross-stage difference vector input.

[0068] 3) Calculate response gradient curve: Within each mutation window paragraph, construct a combined input based on state fluctuation curve and inter-phase difference vector set, perform numerical differentiation of parameter value to time, form a response gradient curve describing the rate of state change. This curve is the first derivative representation of state value change, embodying the instantaneous response strength and change acceleration characteristics.

[0069] 4) Normalization and time warping distance calculation: To compare the response trends of different paragraphs horizontally, all response gradient curves need to be normalized by time length to have a unified length dimension. Then, use dynamic time warping (DTW) algorithm to calculate the similarity distance between each curve pair to quantitatively match the change trend. The DTW algorithm can effectively deal with local nonlinear shifts of time series curves, and is suitable for scenarios such as stage overlap and response delay in job response.

[0070] 5) Trend pattern cluster division and confidence score generation: When the DTW distance between any two response gradient curves is less than a pre-set similarity threshold, it is considered to exhibit the same evolution trend and is assigned to the same evolution pattern cluster. Calculate the number of mutation window paragraphs (i.e. sample number) and the number of job stages covered (i.e. cross-stage coverage) for each pattern cluster, and generate the trend confidence score of the pattern based on a weighted function. The confidence score is used to assess whether the current job tends to a certain established evolution pattern, which has important early warning and control reference value.

[0071] 6) Trend score mapping and result output: Map the trend confidence scores of each pattern cluster to the evolution reference axis according to their corresponding time period to form a data processing curve with a confidence interval. The final output of the evolution trend result includes the evolution time axis, the high-risk trend segment, the high-confidence pattern identifier, and the state response rate change curve within the stage.

[0072] Step S40: Generate a job stage snapshot based on the data processing result and upload the snapshot to the cloud archive at the edge node.

[0073] Specifically, generating a job stage snapshot based on the data processing result and uploading the snapshot to the cloud archive at the edge node includes: extracting the structural information reflecting the job evolution trend in the data processing result; based on the structural information, marking the characteristic state at the trend inflection point in each evolution stage, and attaching the operation data and task graph node record of the current stage to construct a structured snapshot; performing filtering and sorting on the structured snapshot to obtain a high-sensitivity snapshot set for expressing evolution characteristics; performing lightweight encoding compression processing on the high-sensitivity snapshot set and uploading it to the cloud archive.

[0074] In the embodiments of the present application, in view of the continuity, abruptness and stage characteristics of the job state evolution process, in order to realize the structured storage and efficient archival management of key evolution information, it is necessary to further construct a job snapshot for expressing the stage evolution characteristics based on the aforementioned data processing results. Specifically, in the embodiments, the structural information in the aforementioned data processing results reflecting the job state evolution trend is first extracted, including: stage division nodes extracted based on the stage index timeline, local extreme points identified based on the state fluctuation curve, mutation positions marked based on key jump node information, and evolution sensitive segments generated based on trend confidence scores. These structural information together constitute the positioning anchor points for constructing the stage snapshot.

[0075] In each job evolution stage determined by the aforementioned anchor points, the characteristic state near the trend inflection point is identified, and the task nodes, control variables and operation instruction data corresponding to the characteristic state are extracted from the historical task graph execution record to form a stage state snapshot with consistent structure and clear fields. The snapshot includes: stage number for identifying the stage, timestamp of the trend inflection point, operation data summary at that moment (such as pressure, temperature, valve opening, etc.), current task graph node and its execution parameters, and state vector fields for subsequent pattern matching and evolution comparison.

[0076] To improve the practicality and query efficiency of subsequent archived data, after completing the preliminary snapshot construction, further sensitivity evaluation and priority sorting of all snapshots are performed based on mutation strength, trend coverage range and state change gradient, and a high-sensitivity snapshot set with high analysis value is selected. Each snapshot in the high-sensitivity snapshot set is attached with an evolution paragraph feature label, which serves as the basis for subsequent job optimization and risk prediction analysis.

[0077] Considering the limited transmission capability of edge nodes in remote jobs, to avoid excessive bandwidth occupation, lightweight encoding and compression processing of the high-sensitivity snapshot set is required. In the encoding process, the structural fields are compressed using field indexing, the state variable sequence is compressed using differential encoding, and the operation data is merged into a vector using a fixed template. The encoded snapshot is packaged into an upload package using a fast decompression algorithm, and the edge node uploads it to the cloud archival platform according to the archival scheduling strategy during network stable or job gap periods.

[0078] Embodiments:

[0079] Taking the "DN300 pressure blocking valve replacement" project implemented by a natural gas pipeline node station as an example, the specific application process of the cloud-edge collaborative data processing method for remote pressure operation is described.

[0080] The job site is equipped with a pressurized operation device with automatic operation components, including a hydraulic propulsion unit, a rotary cutting device, a sealing and positioning mechanism, etc. The device body integrates an edge computing node and is connected to a remote control terminal through a 5G private network. The cloud scheduling platform is deployed in the headquarters operation center and has a task graph scheduling module and an operation status analysis module.

[0081] Before the operation starts, the edge node performs preprocessing on the accessed operation data, collects parameters including hydraulic oil pressure (MPa), cutter head speed (rpm), main control state code (Boolean type), displacement encoder reading (mm), etc., and generates standard tag data after analysis, such as "tag press 14.7" (indicating that the current oil pressure is 14.7 MPa) and "tag state exec1" (indicating that the current state is in the execution of the propulsion action).

[0082] Based on the tag information, the edge node constructs a task graph and divides the subgraph according to the following process: In the edge node side, filter out the task nodes such as "cutter head start", "seal propulsion", "initial pressure compliance judgment", etc., and form the edge subgraph; At the same time, identify nodes such as "seal stability judgment" and "historical data correlation matching" that involve stage jump judgment, and divide them into cloud subgraph. After the division is completed, the task graph structure and initial data are pushed to the cloud platform by the edge node.

[0083] During the execution of the seal propulsion, the edge node continuously updates the state tag and generates a stage tag sequence (propulsion-seal-stable pressure) and a parameter change sequence (oil pressure rise-stable-drop) during the execution of the task graph. The cloud platform identifies two jump nodes according to the uploaded state sequence, namely "seal completion" and "abnormal stable pressure", and compares the historical evolution path of the past 7 times of the same type of operation to judge that the current state has the characteristic of "frequent pressure fluctuation".

[0084] As Figure 2 , combined with the cloud subgraph analysis results and edge subgraph feedback data, the fusion generates data processing results, including:

[0085] 1) Stage index timeline: the seal stage starts at T0, and the stable pressure starts at T0+120s.

[0086] 2) State fluctuation curve: there is a continuous ±0.3 MPa fluctuation from T0+140s to T0+180s.

[0087] 3) Key jump nodes: T0+120s is identified as the end of the main seal, and T0+180s is identified as the unstable stable pressure.

[0088] 4) Inter-stage difference vector: between the seal end and the stable pressure stage, the pressure response curve slope change value ΔP / Δt is -0.012.

[0089] On this basis, the evolution axis is constructed with T0 as the reference, the mutation window paragraph (T0+120s to T0+180s) is extracted, the response gradient curve is generated, and the first derivative DTW comparison is performed with the plugging stable case sample in the database. The results show that the distance between the mutation window and the historical stable sample is 1.38, and the distance between the mutation window and the abnormal pressure relief sample is 0.62. According to this, the system classifies the current operation state into the "high pressure relief risk" mode cluster, and the confidence score is 0.87.

[0090] Subsequently, the edge node generates an operation stage snapshot based on the analysis result, including: the current stage identifier is "stable pressure failure"; the operation data summary contains the oil pressure value 13.9MPa at T time, and the plugging depth 145.7mm; the accompanying task graph executes the path and the unmet judgment node; the evolution mode is "high pressure relief risk".

[0091] The snapshot is structured and compressed by the edge node to generate a data file with a volume of about 3.2KB, which is uploaded to the cloud platform for archiving in real time, and is used for subsequent decision control, such as reducing the pressure compensation rate and delaying the unblocking step.

[0092] Figure 3 It is a system structure diagram of a cloud-edge collaborative data processing system for remote pressure operation provided by an embodiment of the present application. As shown in Figure 3 The embodiment of the present application provides a cloud-edge collaborative data processing system for remote pressure operation, which comprises: a collection unit for collecting operation data of the pressure operation equipment during operation, and generating label information for expressing the operation state based on the operation data in the edge node; a division unit for constructing a task graph for representing the operation flow based on the label information, and dividing the task graph into edge subgraphs and cloud subgraphs according to control requirements, so as to execute the corresponding subgraphs in the edge node and the cloud platform respectively; a processing unit for generating a data processing result for reflecting the evolution trend of the operation state based on the execution results of the edge subgraphs and the execution results of the cloud subgraphs; an output unit for generating an operation stage snapshot based on the data processing result, and uploading the snapshot to the cloud for archiving in the edge node.

[0093] Those skilled in the art can understand that all or part of the steps of the method for implementing the above-mentioned embodiments can be completed by programs instructing relevant hardware, the programs are stored in a storage medium, and the programs include a plurality of instructions for enabling a single-chip microcomputer, a chip or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various storage media capable of storing program codes.

[0094] The optional embodiments of the present application are described in detail above in combination with the drawings, but the embodiments of the present application are not limited to the specific details in the above-described embodiments. Within the technical concept range of the embodiments of the present application, various simple modifications can be made to the technical solutions of the embodiments of the present application, and these simple modifications all belong to the protection scope of the embodiments of the present application. In addition, it should be noted that each specific technical feature described in the above-described specific embodiments can be combined in any appropriate manner without contradiction. In order to avoid unnecessary repetition, the embodiments of the present application will not be described again for various possible combinations.

[0095] In addition, various different embodiments of the present application can also be combined in any manner, as long as it does not deviate from the idea of the embodiments of the present application, and it should also be considered as disclosed by the embodiments of the present application.

Claims

1. A cloud-edge collaborative data processing method for remote pressure work, applied to data processing of a pressure work device with an automatic work assembly and a remote control terminal, characterized in that, The method comprises: Collecting operation data of the pressurized operation equipment during operation, and generating label information for expressing operation state based on the operation data at the edge node; wherein, The label information comprises: Extracting action identification label and state confidence label based on operation data; The action identification label is used to reflect the specific action type of the operation component in the current operation process; The state confidence label is generated based on the built-in fuzzy confidence model in the edge node, and is used to quantify the stability degree of the operation state; Generating label information for expressing operation state based on the operation data at the edge node comprises: Performing change feature extraction based on the operation data, and constructing an initial feature vector containing action duration, pipe column response amplitude and boundary image definition; Calling a lightweight label generation model deployed in the edge node, performing inference on the initial feature vector, and outputting a candidate label result containing candidate action categories and corresponding confidence scores; Taking the action category in the candidate label result as the action identification label, and converting the confidence score in the candidate label result into a state confidence label through a phased adaptive mapping function; If the confidence score is lower than the phased adaptive threshold set in the edge node, mark the current label information as unstable, and upload the initial feature vector to the cloud platform, and perform inference at the cloud platform and return the complete action identification label and state confidence label as the label information output; If the confidence score is not lower than the phased adaptive threshold, the edge node directly arranges the candidate label result as the label information output; Based on the label information, a task graph for representing the operation process is constructed, and the task graph is divided into edge subgraphs and cloud subgraphs according to control requirements, so that the corresponding subgraphs are executed at the edge node and the cloud platform respectively; Based on the execution results of the edge subgraphs and the execution results of the cloud subgraphs, a data processing result for reflecting the evolution trend of the operation state is generated; Based on the data processing result, an operation stage snapshot is generated, and the snapshot is uploaded to the cloud for archiving at the edge node.

2. The cloud-edge collaborative data processing method for remote snubbing operations of claim 1, wherein, Based on the label information, a task graph for representing the operation process is constructed, comprising: According to the action identification label and the state confidence label contained in the label information, an initial task node set is generated for the operation stage in the operation process, and the time sequence of the action identification label is executed to arrange each initial task node, and a task node sequence is obtained; Based on the preset stage dependency rule and the fluctuation trend of the state confidence label, a directed connection relationship is established between the node sequence, and a task graph with operation sequence constraint and state continuity is constructed; wherein, The task graph is a directed acyclic graph, each node carries a corresponding label field and stage number, and each edge carries a priority weight and a state dependency strength value for graph division.

3. The cloud-edge collaborative data processing method for remote snubbing operations of claim 2, wherein, The preset stage dependency rule is: Based on the label information of each stage in the operation process, it is judged whether the operation types between adjacent labels constitute a sequential dependency relationship, and if the action identification label of the next label belongs to a predefined dependent action set, a directed edge relationship is established between the corresponding nodes in the task graph; According to the change trend of the state confidence label, it is judged whether there is state transition uncertainty under the preset state transition criterion in the current stage. If it is detected that the confidence has a downward trend or the stage boundary is ambiguous, an auxiliary edge is introduced to connect to the predecessor node to express the reversible dependent path. The dependent action set and the preset state transition criterion are both modeled and generated from historical job data before the task graph is constructed, and are embedded into the job logic library of the edge node and the cloud platform in the form of a rule configuration file.

4. The cloud-edge collaborative data processing method for remote snubbing operations of claim 1, wherein, The task graph is divided into edge subgraphs and cloud subgraphs according to control requirements, so as to execute the corresponding subgraphs in the edge node and the cloud platform, including: In the edge node, tasks with a control delay index less than a preset response threshold and a computing resource index lower than an edge node resource threshold are divided into edge subgraphs for execution by the edge node. Task nodes that are not divided into edge subgraphs after screening are screened out, and task nodes whose task dependency paths span multiple job stages or require the fusion of multiple sub-label features for judgment are divided into cloud subgraphs for execution by the cloud platform. Task nodes that are not divided into edge subgraphs or cloud subgraphs are marked as low-adaptation nodes and are executed or merged by the cloud platform according to a resource dynamic scheduling strategy at runtime.

5. The cloud-edge collaboration data processing method for remote snubbing operations of claim 1, wherein, Based on the execution results of the edge subgraphs and the execution results of the cloud subgraphs, a data processing result reflecting the evolution trend of the job state is generated, including: The stage label sequence and parameter change sequence in the execution result of the edge subgraph are extracted, a stage index timeline for identifying the order of job stages is constructed based on the stage label sequence, and a state fluctuation curve for reflecting the dynamic response within the stage is constructed based on the parameter change sequence. The global job state transition path and associated parameter features in the execution result of the cloud subgraph are extracted, key jump node information is identified based on the transition path, and a stage difference vector group for measuring the state offset degree before and after stage conversion is constructed based on the associated parameter features. The stage index timeline, state fluctuation curve, key jump node information, and stage difference vector group are fused to generate a data processing result reflecting the evolution trend of the job state.

6. The cloud-edge collaborative data processing method for remote snubbing operations of claim 5, wherein, The stage index timeline, state fluctuation curve, key jump node information, and stage difference vector group are fused to generate a data processing result reflecting the evolution trend of the job state, including: A unified evolution reference axis is constructed based on the stage index timeline, and a corresponding mutation window segment is extracted around each key jump node on the evolution reference axis; In each mutation window segment, a segment response gradient curve is calculated based on the state fluctuation curve and the stage difference vector group, and the response gradient curve is a continuous function of the state change rate with respect to time; All response gradient curves are normalized by length and subjected to first-order derivative transformation, and the dynamic time warping distance between the curves is calculated; Under the condition that the dynamic time warping distance is less than a preset similarity threshold, the corresponding mutation window segment is classified into the same evolution mode cluster, and a trend confidence score is generated based on the number of samples and the cross-stage coverage of each mode cluster. mapping the trend confidence score to an evolution benchmark axis to generate a data processing result for expressing a job state evolution trend.

7. The cloud-edge collaboration data processing method for remote snubbing operations of claim 1, wherein, generating a job phase snapshot based on the data processing result and uploading the snapshot to a cloud archive at an edge node, including: extracting structural information reflecting a job evolution trend in the data processing result; based on the structural information, marking a feature state at a trend inflection point in each evolution phase and attaching operation data and task graph node records of a current phase to construct a structured snapshot; performing screening and sorting on the structured snapshot to obtain a high-sensitivity snapshot set for expressing evolution features; performing lightweight encoding compression processing on the high-sensitivity snapshot set and uploading to a cloud archive.

8. A cloud-edge collaborative data processing system for remote pressure operation, applied to data processing of a pressure operation device with an automatic operation assembly and a remote control terminal, characterized in that, The system is used to perform the cloud-edge collaborative data processing method for remote pressure-containing jobs according to any one of claims 1-7, and the system includes: a collection unit for collecting operation data of the pressure-containing job equipment during the job process and generating label information for expressing a job state based on the operation data at an edge node; a division unit for constructing a task graph for representing a job flow based on the label information and dividing the task graph into edge subgraphs and cloud subgraphs according to control requirements to perform corresponding subgraphs at an edge node and a cloud platform, respectively; a processing unit for generating a data processing result for reflecting a job state evolution trend based on execution results of the edge subgraphs and execution results of the cloud subgraphs; an output unit for generating a job phase snapshot based on the data processing result and uploading the snapshot to a cloud archive at an edge node.

Images