Scheduling method of intelligent scheduling platform for drainage and gas recovery operation based on cloud platform

Abstract

The invention relates to the technical field of engineering scheduling, in particular to a scheduling method of a drainage gas production operation intelligent scheduling platform based on a cloud platform. The method comprises the following steps: acquiring multi-source operation well monitoring data, and carrying out well site state structure modeling to obtain a well site state representation tensor; reconstructing a well site state change diagram based on the well site state representation tensor to obtain a well site event topological diagram; analyzing each operation event trigger condition in the well site event topological graph into an equipment state acquisition instruction template, and uploading the equipment state acquisition instruction template to an operation scheduling cloud platform; real-time multi-source operation well monitoring data are obtained, operation well scheduling path planning is carried out according to the real-time multi-source operation well monitoring data, and a resource allocation graph is obtained; and performing resource allocation iterative optimization on the resource allocation graph through well site on-site monitoring data acquired after the resource allocation task until the well site operation efficiency reaches a preset efficiency threshold value. According to the invention, the overall production efficiency and management efficiency of a well site can be improved.

Description

Technical Field

[0001] This invention relates to the field of engineering scheduling technology, and in particular to a scheduling method for a cloud-based intelligent scheduling platform for drainage and gas extraction operations. Background Technology

[0002] Drainage gas production technology, a key method in the development of unconventional natural gas such as shale gas, tight gas, and coalbed methane, is widely used in processes such as removing liquid from working wells, dynamic control of gas wells, and production capacity enhancement. Traditional drainage gas production scheduling relies heavily on manual experience and static planning. Scheduling personnel manually formulate operational plans based on parameters such as wellhead pressure, liquid level, and gas production trends, coordinating with on-site teams via telephone, SMS, or offline methods. This approach is not only slow to respond and inefficient, but also ill-suited to the real-time and frequently changing demands of the field, easily leading to resource imbalances, operational delays, and frequent safety hazards.

[0003] In addition, during the actual scheduling process, there are problems such as inconsistent data sources, inconsistent collection frequencies, and serious information silos. Different oil and gas field enterprises or gas production blocks often use their own closed well monitoring systems and operation record methods, resulting in a lack of unified data support for drainage and gas production decisions. Summary of the Invention

[0004] Therefore, it is necessary for the present invention to provide a scheduling method for a cloud-based intelligent scheduling platform for drainage and gas extraction operations, in order to solve at least one of the above-mentioned technical problems.

[0005] To achieve the above objectives, a scheduling method for a cloud-based intelligent scheduling platform for drainage and gas extraction operations includes the following steps:

[0006] Step S1: Acquire monitoring data of multi-source operating wells through the sensor group pre-deployed at the well site, and perform well site state structure modeling on the multi-source operating well monitoring data to obtain the well site state representation tensor;

[0007] Step S2: Identify and classify operational events based on the well site state representation tensor, and establish a multi-source event flow; reconstruct the well site state change map based on the multi-source event flow to obtain the well site event topology map;

[0008] Step S3: Parse the triggering conditions of each operation event in the well site event topology diagram into equipment status acquisition instruction templates, and transmit the equipment status acquisition instruction templates to the sensor group at the well site through the operation scheduling cloud platform to execute the acquisition instruction pre-deployment task;

[0009] Step S4: Obtain real-time multi-source well monitoring data, perform well scheduling adaptability scoring based on the real-time multi-source well monitoring data, and obtain well adaptability scoring matrix; use well adaptability scoring matrix and the obtained well site equipment access logs to plan well scheduling paths and obtain resource allocation map;

[0010] Step S5: Upload the resource allocation map to the job scheduling cloud platform and execute the resource allocation task; use the well site monitoring data collected after the resource allocation task to iteratively optimize the resource allocation map until the well site operation efficiency reaches the preset efficiency threshold.

[0011] Optionally, step S1 includes:

[0012] Step S11: Collect monitoring data of multi-source wells through the sensor group pre-deployed at the well site, and clean the monitoring data of multi-source wells to obtain the monitoring data of the wells to be analyzed;

[0013] Step S12: Align the monitoring data of the well to be analyzed in the time domain and spatial domain to obtain the aligned monitoring data;

[0014] Step S13: Normalize and standardize the format of the aligned monitoring data to obtain a structured matrix frame;

[0015] Step S14: Construct a dynamic well site state structure diagram based on the structured matrix frame;

[0016] Step S15: Vectorize the dynamic well field state structure diagram to obtain the well field state representation tensor.

[0017] Optionally, step S14 includes:

[0018] Step S141: Extract the state features of each working well in the structured matrix frame according to the preset time step, and construct a single well state vector set;

[0019] Step S142: Perform similarity analysis on the wellhead state vectors of continuous time steps in the single well state vector set, and calculate the similarity score matrix between all well pairs;

[0020] Step S143: Construct a wellhead node graph structure using the well number of the working well as the node index, the state features of each working well in the single well state vector set as nodes, and the similarity score matrix as edges;

[0021] Step S144: Obtain well site structure design data through the job scheduling cloud platform, and use the well site structure design data to perform structural constraint topology enhancement on the wellhead node diagram structure to obtain a dynamic well site state structure diagram.

[0022] Optionally, identifying and classifying job events in step S2 includes:

[0023] Extract the sequence of physical features at the wellhead from the well site state representation tensor according to the well number;

[0024] A sliding time window with a fixed width of 5 minutes was set to calculate the rate of change of state and the amplitude of characteristic fluctuations in adjacent time periods of the wellhead physical characteristic sequence. Based on the rate of change of state and the amplitude of characteristic fluctuations in adjacent time periods, the operation behavior window was filtered to obtain candidate operation behavior windows.

[0025] The candidate job behavior window is filtered for valid job behaviors, and event pattern recognition is performed on the valid job behaviors to obtain job event blocks;

[0026] Based on the task event blocks, the overlapping event blocks are split and merged in sequence to obtain the task event blocks to be combined;

[0027] The data structure of the event blocks to be combined is standardized, and the sequences are combined according to the well number and event time to obtain a multi-source event stream.

[0028] Optionally, the effective job behavior screening includes:

[0029] When the rate of change of wellhead fluid level in any time window of a candidate operation behavior window has an average slope greater than 6 cm / min over 3 consecutive minutes, and the rate of change of pump frequency fluctuates within ±10% within that time window, while the pressure change range exceeds 0.8 MPa, the time window is determined to be a strong characteristic operation behavior window.

[0030] The duration of strong characteristic operation behavior is extracted based on the strong characteristic operation behavior window. When the duration of strong characteristic operation behavior is greater than 5 minutes and the instantaneous drop rate of liquid level is less than -4cm / min, the time window is marked as an effective strong operation behavior segment.

[0031] When the wellhead fluid level change rate in any time window of the candidate operation behavior window is between 2 and 6 cm / min over 3 consecutive minutes, and the pressure fluctuation amplitude is less than 0.5 MPa, and the pump frequency remains within ±5% of the previous time window state, the time window is determined to be a weak characteristic operation behavior window.

[0032] Extract the duration of weak feature operation behavior within the weak feature operation behavior window. If the duration of weak feature operation behavior exceeds 10 minutes and the percentage of pumps in the ON state during the duration of weak feature operation behavior is >95%, then mark the time window as a valid weak feature operation behavior segment.

[0033] If the wellhead fluid level in any time window of the candidate operation behavior window fluctuates within a range of less than ±8cm within 5 minutes, the difference between the maximum and minimum values ​​of the pump frequency in that time window accounts for no less than 40% of the maximum value, and the minimum frequency is close to the pump stop threshold of 3Hz, and the gas production rate drops instantaneously by more than 15%, then the time window is determined to be an intermittent operation behavior window.

[0034] Extract the continuous duration of the intermittent operation behavior window. If the continuous duration is greater than 7 minutes and the gas production does not show a continuous downward trend, then mark the time window as an effective intermittent operation behavior segment.

[0035] The effective strong operation behavior segment, effective weak operation behavior segment, and effective intermittent operation behavior segment are merged into the effective operation behavior.

[0036] Optionally, event pattern recognition includes:

[0037] When the liquid level in the effective strong operation behavior section shows a continuous downward trend, with an average downward rate greater than 8 cm / min, and the pump frequency is always higher than 30 Hz and the fluctuation range is less than ±5% in this behavior section, while the pressure rise trend is stable and the gas production increase rate is greater than 10%, then this behavior section is determined to be a deep pumping liquid discharge event block.

[0038] When the liquid level in the effective weak operation behavior section fluctuates slightly within ±10cm, the pump frequency remains between 20–30Hz with fluctuations not exceeding ±3% within the behavior section, the pressure is stable, and the rate of change of gas production rate is <5%, then the behavior section is determined to be a steady-state pressure-maintaining operation event block.

[0039] If the pump frequency in an effective intermittent operation segment changes abruptly from a low frequency of less than or equal to 30 Hz to a high frequency of greater than or equal to 25 Hz at least twice within that segment, the liquid level has a second-order peak structure within that segment, and the gas production shows a synchronous upward trend, then that segment is determined to be an intermittent restart-type liquid discharge event block.

[0040] When the slope of the liquid level changes from negative to positive in the effective strong operation behavior section and the switching time is less than 2 minutes, the pump frequency remains at a high frequency of greater than or equal to 25Hz during this period, the pressure fluctuation exceeds 1.5MPa, and the gas production continues to decrease by more than 15%, then the behavior section is determined to be the liquid discharge over-pumping event block.

[0041] The deep-drainage event blocks, steady-state pressure-maintaining operation event blocks, intermittent restart-type drainage event blocks, and drainage over-drainage event blocks are structurally summarized according to their respective well numbers and start and end times to obtain the operation event blocks.

[0042] Optionally, the reconstruction of the well site state change map in step S2 includes:

[0043] The liquid level height sequence, pump frequency change sequence, pressure response sequence, and gas production rate curve for each operation event in the multi-source event flow are extracted by numbering each well, thereby generating the physical response matrix of the operation event;

[0044] Calculate the state response similarity of the eigenvectors between any two events in the physical response matrix of the task events to obtain the event similarity matrix;

[0045] Analyze the physical proximity relationships of each working well in the dynamic well field state structure diagram to obtain the physical proximity relationship matrix, and establish the initial connection diagram of events based on the physical proximity relationship matrix and the similarity matrix between events;

[0046] Extract all edges with a weight greater than 0.85 from the initial event connection graph as strong dependent event connections, and construct a set of state mutation propagation links;

[0047] Reachability assessment is performed on links in the initial connection graph of the event with a path length less than 2 and an edge weight in the range of [0.85, 0.9). Isolated edges or broken links with an edge weight less than 0.8 in the reachability assessment results are structurally removed to obtain the propagation structure path set.

[0048] The operation event nodes in the propagation structure path set are used as the graph node set, and the state mutation propagation link set is used as the graph edge set; the well site event topology graph is constructed by combining the graph node set and the graph edge set.

[0049] Optionally, the well scheduling suitability score in step S4 includes:

[0050] Acquire real-time multi-source well monitoring data, and extract well status features from the pre-processed real-time multi-source well monitoring data to obtain a set of well status feature vectors;

[0051] Based on the operation event blocks, identify the types of operation events that occurred in the monitoring data of the operation wells to be analyzed within the past 24 hours, and extract the behavioral characteristics of the operation wells within the past 24 hours, thereby constructing a behavioral response label vector for each operation well;

[0052] For each working well, the corresponding working status feature vector set and behavior response label vector are subjected to feature correlation analysis. Based on the correlation analysis results, the status-behavior matching score is calculated to obtain the status-behavior adaptation score matrix.

[0053] Obtain the well site resource allocation log, and use the current schedulable resource status in the well site resource allocation log as a constraint to score and correct the state-behavior adaptation scoring matrix to obtain the well operation adaptability scoring matrix.

[0054] Optionally, the well scheduling path planning in step S4 includes:

[0055] Extract the field access log dataset from the well site resource allocation log, and set the scheduling start point set based on the location of each operational resource in the field access log dataset;

[0056] The scheduling and passage difficulty of each well is assessed based on the location of each operational resource in the scheduling starting point set, the distribution of each operational well in the dynamic well site status structure diagram, and the pipeline layout.

[0057] Based on the distribution of each working well and the pipeline layout, the starting point where the dispatching starting point is concentrated is connected to the working well, and the dispatching passage difficulty is used as the edge weight of the connection to obtain the initial path map;

[0058] The adaptability score of each well in the well adaptability score matrix is ​​taken as the scheduling feedback weight factor, and the weight of each edge in the initial path map is adjusted by weighting to obtain the well site path map.

[0059] Perform optimal path search on each starting point in the well site access map to obtain the optimal scheduling path set;

[0060] By integrating the optimal scheduling paths between each starting point and the working well in the optimal scheduling path set, a resource allocation map is obtained.

[0061] Optionally, step S5 includes:

[0062] Upload the resource allocation map to the job scheduling cloud platform to execute the resource allocation task;

[0063] During the execution of resource allocation tasks, on-site monitoring data at the well site is collected through sensor arrays and on-board terminals on the work vehicle;

[0064] Calculate the execution effect score of each allocation task execution path in the resource allocation diagram based on the well site monitoring data, and output the scheduling execution score matrix;

[0065] Write the scheduling execution score matrix back to the resource allocation graph, and extract the paths with execution performance scores below 0.6 from the written-back resource allocation graph to obtain a set of inefficient scheduling paths;

[0066] Re-execute steps S4 to S5 on the inefficient scheduling path set to update the planned scheduling path. After each round of resource allocation is completed, count the number of working wells covered by the current round of scheduling and the number of scheduling paths with an execution effect score greater than or equal to 0.8 to calculate the scheduling efficiency.

[0067] If the scheduling efficiency is greater than or equal to the preset efficiency threshold of 0.85, the resource allocation graph updated in this round is determined to have converged to optimality, and the iteration stops.

[0068] If the scheduling efficiency is less than the preset efficiency threshold of 0.85, then the updated resource allocation map will be re-executed in steps S4 to S5 to trigger the next round of updating and planning the scheduling path, until the scheduling efficiency of the new round is greater than or equal to the preset efficiency threshold of 0.85.

[0069] This application utilizes multi-source sensors deployed at the well site to collect key monitoring data of the operating well in real time. Combined with data cleaning, temporal and spatial alignment, and structuring, it ensures the integrity, consistency, and high quality of the collected data, significantly improving the reliability of the data foundation. This solves problems such as scattered data sources, inconsistent collection frequencies, and non-standard formats in traditional well site monitoring, eliminating information silos and promoting effective data integration and unified management. By constructing a well site state representation tensor, multi-dimensional monitoring information is uniformly mapped into a structured tensor form, accurately reflecting the spatiotemporal characteristics of the well site's operational status, which helps to achieve dynamic expression of status in complex operating environments. Based on this tensor, operational events are further identified and segmented to form a multi-source event stream, enabling the scientific capture of key time nodes and status changes in well site operations, improving the accuracy and real-time performance of event identification, and providing a detailed dynamic profile of the operation for scheduling. The construction of the well site event topology map combines physical proximity relationships and the similarity of state responses between events to form an intuitive event propagation link. Utilizing multi-attribute nodes (including timestamps and monitoring summary information) and state variation propagation links, a systematic model of well site state variations is achieved, providing structured support for operational risk warning and scheduling optimization. By converting event triggering conditions into equipment status acquisition instruction templates through a cloud platform, automated scheduling and precise distribution of acquisition tasks are achieved, reducing the response latency of traditional manual scheduling and improving the execution efficiency and flexibility of monitoring tasks. Real-time multi-source well monitoring data, after feature extraction and behavioral response tag construction, enables in-depth correlation analysis of well status and behavior. Combined with nearly 24 hours of operational behavior characteristics, this significantly improves the scientific validity and practical guidance of the suitability score. The scoring process incorporates well site resource allocation logs as constraints, extracting the current schedulable resource status and generating a resource status matrix to ensure the feasibility of the scheduling scheme under actual resource conditions. In specific operations, by calculating the resource response latency ratio and setting a latency threshold (e.g., introducing a penalty factor when the latency ratio is greater than 1), the risk of scheduling failure due to excessively long resource response times is effectively avoided. Simultaneously, by combining the maximum and remaining resource capacity, a schedulable weight vector is constructed to dynamically correct the score, making the scheduling scheme more reasonable and stable. Based on the well adaptability scoring matrix and on-site equipment access logs, dynamic planning of well scheduling paths and resource allocation maps is achieved. This enables optimal matching of scheduling paths and efficient resource allocation, avoiding resource conflicts and path congestion, and ensuring smooth and efficient operation. The resource allocation map is uploaded to the cloud platform and iteratively optimized using subsequently collected on-site monitoring data, realizing closed-loop scheduling management, continuously improving well site operation efficiency, and ensuring that scheduling decisions can dynamically adapt to changes on-site.The well site event topology map is stored in a multi-attribute graph database, supporting multi-dimensional queries, real-time updates and expansion of nodes and edges, providing strong data support capabilities, meeting the multi-scenario scheduling needs and risk control in complex well site environments, effectively improving the intelligent management level of drainage and gas production operations, promoting the automation, precision and dynamic development of well site scheduling, greatly enhancing resource utilization and operation scheduling response speed, and significantly improving the overall production efficiency and management effectiveness of oil and gas fields. Attached Figure Description

[0070] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0071] Figure 1 This is a flowchart illustrating the scheduling method of the intelligent scheduling platform for drainage and gas extraction operations based on a cloud platform according to the present invention.

[0072] Figure 2 This is a detailed flowchart of step S1 in the present invention;

[0073] Figure 3 This is a schematic diagram illustrating the application scenario of the well scheduling path planning in step S4 of this invention;

[0074] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0075] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0076] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0077] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0078] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a scheduling method for a cloud-based intelligent scheduling platform for drainage and gas extraction operations, the method comprising the following steps:

[0079] Step S1: Acquire monitoring data of multi-source operating wells through the sensor group pre-deployed at the well site, and perform well site state structure modeling on the multi-source operating well monitoring data to obtain the well site state representation tensor;

[0080] In this embodiment, various types of sensor groups, including pressure sensors, temperature sensors, flow sensors, and vibration sensors, are pre-deployed at the well site in the oil and gas field to collect multi-dimensional physical parameters of the operating well in real time. The collected multi-source data is stored in the form of a time-series matrix, with a time window length set to 10 seconds and a sampling frequency of 5Hz. This data is fused according to spatial location and acquisition time, and state modeling is performed using a multi-dimensional tensor structure (e.g., the tensor dimension can be defined as time × sensor type × spatial coordinates). Each element in the state tensor reflects the monitoring value of the corresponding sensor at a specific time and location. The structure includes a mechanism for handling missing data, such as interpolation for completion. This structured tensor provides basic data support for subsequent event identification and scheduling optimization, ensuring the unified expression and efficient retrieval of multi-dimensional information.

[0081] Step S2: Identify and classify operational events based on the well site state representation tensor, and establish a multi-source event flow; reconstruct the well site state change map based on the multi-source event flow to obtain the well site event topology map;

[0082] In this embodiment, based on the obtained well site state representation tensor, different types of operational event blocks are divided according to temporal variation characteristics and spatial distribution patterns using a rule-based discrimination method, such as equipment startup, pressure anomaly, and flow fluctuation events. The divided events are then used to construct an event flow based on chronological order and spatial proximity. The relationships between events are expressed using an adjacency matrix, where matrix elements reflect the temporal dependence and spatial correlation between events. The event flow is then used to further reconstruct the well site state change graph, where nodes represent operational events and edges represent triggering or dependency relationships between events, forming a well site event topology graph. The topology graph supports dynamic updates, reflecting real-time changes in operational status, facilitating subsequent event analysis and scheduling strategy formulation.

[0083] Step S3: Parse the triggering conditions of each operation event in the well site event topology diagram into equipment status acquisition instruction templates, and transmit the equipment status acquisition instruction templates to the sensor group at the well site through the operation scheduling cloud platform to execute the acquisition instruction pre-deployment task;

[0084] In this embodiment, for each operational event node in the well site event topology map, the corresponding equipment status triggering conditions are extracted, such as pressure exceeding a set threshold or abnormal equipment vibration frequency. These conditions are then converted into a unified equipment status acquisition instruction template. The template definition includes the acquisition object (sensor ID), acquisition period (10Hz), acquisition duration (30 seconds), and data format (time series CSV). The operation scheduling cloud platform sends these instruction templates to the well site sensor group through a secure communication channel. The sensor group accurately executes the pre-deployed acquisition tasks according to the instructions and transmits data back in real time to verify the accuracy of the event triggering conditions and the timeliness of the acquisition. This step ensures the targeted acquisition of equipment status data, improving acquisition efficiency and data relevance.

[0085] Step S4: Obtain real-time multi-source well monitoring data, perform well scheduling adaptability scoring based on the real-time multi-source well monitoring data, and obtain well adaptability scoring matrix; use well adaptability scoring matrix and the obtained well site equipment access logs to plan well scheduling paths and obtain resource allocation map;

[0086] In this embodiment, multi-source well monitoring data is collected and processed in real time. The data matrix rows correspond to different wells, and the columns include indicators such as pressure, temperature, flow rate, and equipment operating status. Feature correlation analysis is performed based on the multi-source well monitoring data. A status-behavior matching score is calculated based on the correlation analysis results, and the score is corrected to obtain a well adaptability scoring matrix. In some other embodiments, based on real-time multi-source well monitoring data, a pre-set adaptability scoring standard can also be used to calculate the scheduling adaptability score for each well. The score is calculated using normalized index weighting, with weights of 0.4 for pressure stability, 0.35 for equipment health, and 0.25 for environmental adaptability. The score range is 0-1, and the adaptability threshold is 0.75. The resulting adaptability scoring matrix is ​​combined with well site equipment access log data (format: equipment ID, timestamp, path coordinate sequence) to construct a resource allocation graph. Graph nodes represent wellheads, and edge weights are determined by equipment access frequency and adaptability scores, with edge weight values ​​ranging from 0.1 to 1.0. This resource allocation map reflects resource utilization efficiency and accessibility, guiding the optimization planning of scheduling routes.

[0087] Step S5: Upload the resource allocation map to the job scheduling cloud platform and execute the resource allocation task; use the well site monitoring data collected after the resource allocation task to iteratively optimize the resource allocation map until the well site operation efficiency reaches the preset efficiency threshold.

[0088] In this embodiment, the resource allocation map is uploaded to the job scheduling cloud platform. The platform dynamically adjusts the resource allocation plan based on the map structure, including equipment scheduling, personnel arrangement, and priority adjustment of access routes. After the scheduling task is executed, on-site monitoring data is continuously collected, and equipment utilization, access delay, and task completion time are used as key performance indicators for evaluation. Based on the evaluation results, the node weights and edge weights in the resource allocation map are automatically adjusted. For example, the weight of the path corresponding to the access delay exceeding a preset threshold (5 seconds) is reduced by 10%, and the weight of nodes with insufficient equipment utilization is appropriately increased. The initial adjustment cycle is set to 10 minutes, and the iterative optimization process continues until the overall operation efficiency indicators reach the set threshold (e.g., equipment utilization ≥ 85% and access delay ≤ 3 seconds), ensuring that resource allocation dynamically adapts to changes in the on-site environment and achieves efficient and safe operation of the well site.

[0089] Optionally, step S1 includes:

[0090] Step S11: Collect monitoring data of multi-source wells through the sensor group pre-deployed at the well site, and clean the monitoring data of multi-source wells to obtain the monitoring data of the wells to be analyzed;

[0091] In this embodiment, multiple types of sensors are pre-deployed within the well site, including pressure sensors, temperature sensors, wellhead flow sensors, and vibration sensors. All sensors have timestamp synchronization capabilities, and the data acquisition frequency is set to 10Hz. The collected multi-source well monitoring data includes raw signals, voltage, current, mechanical vibration, and other multi-dimensional information. The collected data is wirelessly transmitted to the edge computing unit. The edge unit performs preliminary data cleaning, removing obvious outliers (such as sensor signal abrupt changes exceeding a set threshold of 20%), data segments with a disconnection rate higher than 15% during the acquisition period, and interpolating missing values ​​using the median value within a nearby time window. This cleaning process ensures the integrity and accuracy of subsequent analysis data. The output well monitoring data structure remains a time-series multi-dimensional array, encompassing spatial sensor nodes and the time dimension.

[0092] Step S12: Align the monitoring data of the well to be analyzed in the time domain and spatial domain to obtain the aligned monitoring data;

[0093] In this embodiment, a dual alignment method is adopted for the monitoring data of the well to be analyzed: time domain alignment achieves time window alignment of multi-sensor sampling data through unified synchronization timestamps, with the time window length set to 100 milliseconds to ensure consistent sampling point timing; spatial domain alignment maps sensor data from different spatial locations to fixed grid cells in the well site grid model based on the geographical coordinates of the well site sensors. The grid size is set to 5 meters × 5 meters according to the actual scale of the well site, and all sensor data are spatially interpolated to the corresponding grid cells. After alignment, the monitoring data forms a three-dimensional tensor, with the dimensions corresponding to the time step, sensor type, and spatial grid position, ensuring that the data has a unified reference frame in time and space, facilitating subsequent standardization and structured processing.

[0094] Step S13: Normalize and standardize the format of the aligned monitoring data to obtain a structured matrix frame;

[0095] In this embodiment, the aligned monitoring data is normalized using a min-max normalization method, mapping the data uniformly to the [0,1] interval to eliminate the impact of differences in the dimensions of different sensors. The normalization process is performed independently for each sensor type, calculating the maximum and minimum values ​​across the entire time window to ensure the stability and comparability of the data conversion. Subsequently, according to the unified data interface specification for the well site, the normalized data format is converted into a structured matrix frame. The matrix rows correspond to different time steps, and the columns represent the collected values ​​of each sensor type in each spatial grid cell. The matrix frame uses a two-dimensional floating-point matrix format, supporting batch storage and fast access. Additional metadata fields describe the acquisition time range, sensor type identifier, and spatial coordinate information, providing a complete context for data retrieval and parsing.

[0096] Step S14: Construct a dynamic well site state structure diagram based on the structured matrix frame;

[0097] In this embodiment, a dynamic well site state structure graph is constructed based on a structured matrix frame. Graph nodes correspond to spatial grid cells within the well site, and each node stores a multi-source sensor data vector at the current time step. Connections between nodes are set according to spatial adjacency rules; connecting edges represent the physical association between adjacent grid cells, and edge weights are dynamically assigned based on the similarity of the state data of two nodes, ranging from 0 to 1, to reflect the degree of synchronization or coupling between regions. The graph structure is stored in the form of an adjacency matrix, with the matrix dimension being the number of grid nodes multiplied by the number of grid nodes. This dynamic state graph can reflect the spatial dynamic changes of the well site over time, laying the foundation for state pattern recognition and event analysis.

[0098] Step S15: Vectorize the dynamic well field state structure diagram to obtain the well field state representation tensor.

[0099] In this embodiment, the node attributes and edge weight information of the dynamic well site state structure graph are vectorized and encoded. Multidimensional sensor data for each node are merged into a single feature vector, with the dimension equal to the number of sensor types. The edge weight matrix serves as the graph adjacency information encoding. The vectorized encoding result is a three-dimensional well site state representation tensor. The tensor dimensions are time step × number of nodes × feature vector dimension, where the feature vector dimension is the number of sensors plus the edge weight information extension. This representation tensor has a unified structure, supports direct calls from depth analysis and pattern mining modules, and possesses efficient storage and online update capabilities, providing an accurate and real-time well site state representation for subsequent event recognition and scheduling control.

[0100] Optionally, step S14 includes:

[0101] Step S141: Extract the state features of each working well in the structured matrix frame according to the preset time step, and construct a single well state vector set;

[0102] In this embodiment, a time step of 1 second is pre-set in the structured matrix frame. Based on this time step, the state features corresponding to each working well are extracted. These state features include four dimensions: pressure, temperature, flow rate, and vibration intensity, forming a single-well state vector. Each state vector has a fixed dimension of 4, representing the real-time reflection of the collected multi-source monitoring data. This feature extraction is performed on all working wells within consecutive time steps, generating a set of single-well state vectors arranged in a time series. The data structure uses timestamps as indexes, and the vector set is stored in a time series database, supporting fast retrieval and batch export. This set can reflect the dynamic state of a single wellhead over time, meeting the needs of subsequent cross-well state comparison analysis.

[0103] Step S142: Perform similarity analysis on the wellhead state vectors of continuous time steps in the single well state vector set, and calculate the similarity score matrix between all well pairs;

[0104] In this embodiment, for continuous time-step data in the single-well state vector set, a similarity measurement method based on vector dot product is adopted to calculate the state feature similarity score between each pair of operating wells. Specifically, after standardizing the state vectors of corresponding time steps of two wellheads, the cosine of the angle between their vectors is calculated as a similarity index. This similarity value ranges from 0 to 1, where 1 indicates a high degree of consistency between states and 0 indicates that the states are unrelated. Multiple pairs of similarity values ​​from continuous time steps are integrated through time-weighted averaging. The weights decay exponentially with the time interval by default, and the decay coefficient is set to 0.8 to highlight the influence of recent states. Finally, a similarity score matrix is ​​formed between all well pairs. The matrix size is N×N, where N is the total number of operating wells. The matrix is ​​symmetric and has a main diagonal of 1, which facilitates subsequent graph structure construction.

[0105] Step S143: Construct a wellhead node graph structure using the well number of the working well as the node index, the state features of each working well in the single well state vector set as nodes, and the similarity score matrix as edges;

[0106] In this embodiment, the well number is used as the unique index identifier for the graph nodes. Based on the state characteristics in the single-well state vector set, a node attribute vector is constructed. The attribute dimension is 4-dimensional, corresponding to pressure, temperature, flow rate, and vibration, respectively. The similarity score matrix is ​​used as the weight of the edges connecting nodes, with edge weights ranging from 0 to 1. Edges below the similarity threshold of 0.6 are filtered out to ensure that the graph structure highlights the inter-well relationships with strong state correlations. This wellhead node graph structure adopts an adjacency list storage format, which facilitates dynamic updates and space utilization optimization. The graph structure not only reflects the state attributes of single wells but also reflects the correlation between working wells in terms of state changes, providing spatial data support for dynamic scheduling.

[0107] Step S144: Obtain well site structure design data through the job scheduling cloud platform, and use the well site structure design data to perform structural constraint topology enhancement on the wellhead node diagram structure to obtain a dynamic well site state structure diagram.

[0108] In this embodiment, well site structure design data is obtained through the security interface of the job scheduling cloud platform. This data includes the physical coordinates of the wellhead, pipeline layout, and work area distribution information. When performing topology enhancement on the wellhead node graph structure, physical proximity and pipeline connectivity are used as structural constraints. Specifically, the spatial distance threshold in the well site structure design is set to 20 meters. Wellhead nodes with connection distances less than this threshold are enhanced by increasing the edge weight by 0.2; simultaneously, wellhead nodes directly connected to pipelines are enhanced by increasing the edge weight by 0.3. Finally, the original similarity score matrix and the structural design constraint weights are integrated to form a weighted comprehensive adjacency matrix, generating a dynamic well site status structure graph. This graph structure takes into account both monitoring status and physical structure, which is beneficial for the rational planning of scheduling strategies and resource allocation.

[0109] Optionally, identifying and classifying job events in step S2 includes:

[0110] Extract the sequence of physical features at the wellhead from the well site state representation tensor according to the well number;

[0111] In this embodiment, the physical feature sequence of each working well is extracted sequentially from the constructed well site state representation tensor, according to the well number. This sequence covers six indicators: fluid level, casing pressure, oil pressure, pump frequency, gas production rate, and throttling pressure difference, forming a two-dimensional time series matrix with a time resolution of 1 minute. The data sources extracted here are all from various types of sensor groups deployed at the well site, such as pressure sensors, frequency monitoring devices, and flow meters, and timestamp alignment and wellhead identification binding are completed in the data center. To avoid mislabeling of the same well, a spatial clustering criterion of a geographic coordinate offset threshold of no more than 3 meters must be set before binding the well number.

[0112] A sliding time window with a fixed width of 5 minutes was set to calculate the rate of change of state and the amplitude of characteristic fluctuations in adjacent time periods of the wellhead physical characteristic sequence. Based on the rate of change of state and the amplitude of characteristic fluctuations in adjacent time periods, the operation behavior window was filtered to obtain candidate operation behavior windows.

[0113] In this embodiment, a fixed sliding time window of 5 minutes is set for the extracted wellhead physical feature sequence, and the sequence is slidable in 1-minute increments. The rate of change of state is calculated for the data within each window; for example, the slope of the liquid level is calculated using the first-order difference quotient, and the average rate of change is used as the assessment of the liquid level change trend within that window. The fluctuation amplitude is defined as the difference between the maximum and minimum values ​​within the window. A screening threshold is set as follows: windows with a liquid level change rate greater than 2 cm / min and a pressure difference change amplitude exceeding 0.3 MPa are marked as potential operational behavior windows. This threshold selection is based on the fitting of statistical characteristics of 50 typical gas production wells to ensure coverage of the sensitive fluctuation range during the operational behavior initiation phase.

[0114] The candidate job behavior window is filtered for valid job behaviors, and event pattern recognition is performed on the valid job behaviors to obtain job event blocks;

[0115] In this embodiment, after the candidate operational behavior window is determined, it is screened for effective operational behaviors. Specifically, this includes three aspects: First, if the liquid level slope exceeds 6 cm / min for 3 consecutive minutes, the pump frequency fluctuation is less than ±10%, and the pressure range exceeds 0.8 MPa, the window is initially judged as a strong characteristic behavior window; further, it is determined whether its duration exceeds 5 minutes, and whether the liquid level has a drop rate greater than -4 cm / min. If these conditions are met, it is marked as an effective strong behavior. Second, for windows where the liquid level change is between 2 and 6 cm / min, the frequency fluctuation is no more than ±5%, and the pressure fluctuation is less than 0.5 MPa, it is checked whether it is continuously maintained for more than 10 minutes, and during this period, the pump is ON for more than 95% of the time. If these conditions are met, it is classified as an effective weak behavior. The third type of intermittent behavior screening requires that the ratio of the maximum to minimum pump frequency difference is greater than 40%, the minimum frequency is close to 3 Hz, and it is maintained for more than 7 minutes without a decreasing trend in gas production. The above parameter settings are all based on the empirical curve fitting results of the on-site pump frequency stability threshold, the liquid level dynamic response model, and the gas production platform value, and have practical engineering basis. After identifying effective operational behaviors, the start and end times, behavior type, duration, and corresponding well number of each behavior segment are extracted to construct an operational event block.

[0116] Based on the task event blocks, the overlapping event blocks are split and merged in sequence to obtain the task event blocks to be combined;

[0117] The data structure of the event blocks to be combined is standardized, and the sequences are combined according to the well number and event time to obtain a multi-source event stream.

[0118] In this embodiment, all operational event blocks are categorized by well number, and event blocks with overlapping start and end times are split or merged based on their sustained behavioral stability. Specifically, if two event blocks have the same fluid level slope direction and continuous pressure trend within the overlapping interval, they are merged into the same event block; if the trends are opposite, they are split using the median point. After processing, all event blocks are uniformly converted into standard structures. Each structure contains fields such as {well number, event type, start and end time, behavior intensity, signal feature vector}. Finally, these are arranged and combined according to well number and time order to generate a multi-source event stream, which serves as the input data structure for subsequent well site state diagram reconstruction and task scheduling.

[0119] Optionally, the effective job behavior screening includes:

[0120] When the rate of change of wellhead fluid level in any time window of a candidate operation behavior window has an average slope greater than 6 cm / min over 3 consecutive minutes, and the rate of change of pump frequency fluctuates within ±10% within that time window, while the pressure change range exceeds 0.8 MPa, the time window is determined to be a strong characteristic operation behavior window.

[0121] In this embodiment, the liquid level height sequence, pump frequency sequence, and wellhead pressure sequence are extracted from any time window within the candidate operational behavior window, and the rate of change of liquid level height within the sliding time window is calculated. The calculation method uses the average first derivative of the liquid level value per minute within the window. If the average slope exceeds 6 cm / min, it indicates a rapid change in the liquid level. Simultaneously, if the difference between the maximum and minimum pump frequency values ​​within the window does not exceed ±10%, it indicates that the pump is in a relatively stable operating state. If a pressure fluctuation greater than 0.8 MPa is detected simultaneously within the time window (derived from the difference between the maximum and minimum values), it is considered that a high-intensity operational disturbance and drive response has occurred during that period, thus marking it as a high-characteristic operational behavior window. The selection of each parameter is based on the numerical statistical results of typical drainage processes in historical operational records, covering more than 90% of manually confirmed operational segments.

[0122] The duration of strong characteristic operation behavior is extracted based on the strong characteristic operation behavior window. When the duration of strong characteristic operation behavior is greater than 5 minutes and the instantaneous drop rate of liquid level is less than -4cm / min, the time window is marked as an effective strong operation behavior segment.

[0123] In this embodiment, to further confirm whether the strong characteristic operation behavior is persistent and highly significant, it is necessary to detect the continuous duration of the aforementioned strong characteristic window. If the time span of the continuous strong characteristic window exceeds 5 minutes, the rate of decrease of the liquid level height within each 30-second interval during that period is calculated. If any one of the decrease rates is less than -4 cm / min, it indicates that the rapid unloading at the wellhead, accompanied by significant liquid pumping effect, is significant. At this point, it is determined that the strong characteristic behavior not only exists stably but also has a technological function, and is thus marked as an effective strong operation behavior segment. The duration threshold is set to 5 minutes and the decrease rate threshold is set to -4 cm / min because short-term disturbances often correspond to instantaneous pump adjustment operations or liquid level fluctuations, while slopes below -4 cm / min usually only appear at the initial stage of the liquid drainage operation, which can effectively eliminate false judgments.

[0124] When the wellhead fluid level change rate in any time window of the candidate operation behavior window is between 2 and 6 cm / min over 3 consecutive minutes, and the pressure fluctuation amplitude is less than 0.5 MPa, and the pump frequency remains within ±5% of the previous time window state, the time window is determined to be a weak characteristic operation behavior window.

[0125] Extract the duration of weak feature operation behavior within the weak feature operation behavior window. If the duration of weak feature operation behavior exceeds 10 minutes and the percentage of pumps in the ON state during the duration of weak feature operation behavior is >95%, then mark the time window as a valid weak feature operation behavior segment.

[0126] In this embodiment, when a window with slow liquid level changes, stable frequency control, and weak pressure disturbances is identified, the weak characteristic operation behavior identification process begins. At this point, the basic condition is that the liquid level slope within the sliding window is between 2 and 6 cm / min, combined with a frequency change not exceeding ±5%, and a pressure fluctuation amplitude below 0.5 MPa. This combined screening can cover long-term light operation processes, such as wellhead liquid level maintenance and pumping without rapid unloading. Subsequently, the continuous time periods forming the weak characteristic window are statistically analyzed, and the percentage of time the pump is in ON state during this period is calculated. If the pump is in ON state for more than 10 minutes and consistently exceeds 95%, it indicates that this segment is a low-intensity but stable operation, and is thus marked as a valid weak operation behavior segment. The purpose of setting a high percentage of pump status here is to eliminate pseudo-operation behaviors caused solely by natural liquid level changes.

[0127] If the wellhead fluid level in any time window of the candidate operation behavior window fluctuates within a range of less than ±8cm within 5 minutes, the difference between the maximum and minimum values ​​of the pump frequency in that time window accounts for no less than 40% of the maximum value, and the minimum frequency is close to the pump stop threshold of 3Hz, and the gas production rate drops instantaneously by more than 15%, then the time window is determined to be an intermittent operation behavior window.

[0128] Extract the continuous duration of the intermittent operation behavior window. If the continuous duration is greater than 7 minutes and the gas production does not show a continuous downward trend, then mark the time window as an effective intermittent operation behavior segment.

[0129] In this embodiment, to identify frequent start-stop cycles and sudden drops in gas production during certain operations, a specific identification process needs to be constructed for intermittent operation behavior. First, the stability of the liquid level fluctuation within the sliding window is checked, requiring a total fluctuation amplitude of less than ±8cm within 5 minutes to prevent drastic liquid level changes from affecting frequency judgment. Next, it is determined whether the difference between the maximum and minimum pump frequency values ​​accounts for more than or equal to 40% of the maximum value, and the minimum frequency value is confirmed to be close to 3Hz (this threshold is the common minimum frequency value before the pump stops on-site). If the gas production rate drops by more than 15% at a single point, it indicates that the wellhead is in a state of interrupted gas production and the pump control is not completely disconnected, exhibiting typical intermittent drainage characteristics. Continuous cycles formed by such windows are extracted. If the cycle lasts for more than 7 minutes and the gas production curve does not show a trend of decline (i.e., no continuous production reduction signal), it is marked as a valid intermittent operation behavior segment. The fluctuation parameters and pump frequency settings used are referenced from the average fluctuation range of the historical sensor curves of 10 common intermittent drainage wells, ensuring practicality.

[0130] The effective strong operation behavior segment, effective weak operation behavior segment, and effective intermittent operation behavior segment are merged into the effective operation behavior.

[0131] In this embodiment, all valid strong, weak, and intermittent operation behavior segments are integrated according to their chronological order on the timeline, well number labels, and event duration attributes. During the integration process, non-overlapping segments are primarily merged, while overlapping segments are merged using a "behavior intensity priority" approach, meaning the behavior type with the highest characteristic intensity is retained first. For example, if both strong and weak behaviors exist within an overlapping time interval, the strong behavior is retained, and its start and end times are merged. The output is a set of structured operation behavior segments, each containing the fields: {well number, behavior type, start time, end time, average pump frequency, liquid level drop slope, gas production trend index}, serving as the basic data unit for subsequent event identification and map construction. This merging method can accurately characterize the dynamic evolution of well site operations without introducing aliasing events.

[0132] Optionally, event pattern recognition includes:

[0133] When the liquid level in the effective strong operation behavior section shows a continuous downward trend, with an average downward rate greater than 8 cm / min, and the pump frequency is always higher than 30 Hz and the fluctuation range is less than ±5% in this behavior section, while the pressure rise trend is stable and the gas production increase rate is greater than 10%, then this behavior section is determined to be a deep pumping liquid discharge event block.

[0134] In this embodiment, when performing pattern recognition on effective high-intensity operational behavior segments, the first-order differential slope is first calculated from the liquid level height sequence, and the average descent rate threshold is set to 8 cm / min. This threshold is extracted based on the process experience of rapid well fluid descent during typical deep pumping operations. If, during the continuous descent, the pump frequency remains stably above 30 Hz with fluctuations not exceeding ±5%, and the upward slope extracted from the pressure monitoring data is greater than 0.15 MPa / min, and the gas production increase rate is greater than 10%, then this behavior segment is marked as a deep pumping fluid discharge event block. The above parameter combination reflects the synergistic response of enhanced pumping, gas breakthrough, and rapid bottomhole fluid replacement processes, thus possessing high recognizability.

[0135] When the liquid level in the effective weak operation behavior section fluctuates slightly within ±10cm, the pump frequency remains between 20–30Hz with fluctuations not exceeding ±3% within the behavior section, the pressure is stable, and the rate of change of gas production rate is <5%, then the behavior section is determined to be a steady-state pressure-maintaining operation event block.

[0136] In this embodiment, when identifying effective weak operational behavior sections, it is necessary to detect whether the fluctuation range of the liquid level height is consistently maintained within ±10cm. This fluctuation range originates from the periodic disturbances caused by the pump effect during steady-state operation. If the pump frequency operates stably between 20–30Hz, the fluctuation is controlled within ±3%, the pressure remains within ±0.2MPa, and the gas production rate change rate is consistently below 5%, then it can be identified as a steady-state pressure-maintaining operation event block. The above conditions indicate that the downhole gas production state is close to equilibrium, there are no new sudden fluid disturbances, and the overall operation is in a sustainable low-interference phase.

[0137] If the pump frequency in an effective intermittent operation segment changes abruptly from a low frequency of less than or equal to 30 Hz to a high frequency of greater than or equal to 25 Hz at least twice within that segment, the liquid level has a second-order peak structure within that segment, and the gas production shows a synchronous upward trend, then that segment is determined to be an intermittent restart-type liquid discharge event block.

[0138] In this embodiment, when processing effective intermittent operation segments, the number of frequency abrupt changes is counted from the frequency sequence. There must be at least two transitions from below 30Hz to above 25Hz. Simultaneously, the fluid level must exhibit a typical "second-order peak" structure, meaning there are two local extrema, with the second peak higher than the first, and the interval between their positions is not less than 3 minutes. This structure is often used to identify well fluid replenishment after restarting. If the gas production curve shows a significant synchronous upward trend during this period (an increase exceeding 12%), it is determined to be an intermittent restart-type fluid discharge event block. This type of event typically corresponds to a stage where pump capacity is reallocated at the well site or a normal operation cycle is restored.

[0139] When the slope of the liquid level changes from negative to positive in the effective strong operation behavior section and the switching time is less than 2 minutes, the pump frequency remains at a high frequency of greater than or equal to 25Hz during this period, the pressure fluctuation exceeds 1.5MPa, and the gas production continues to decrease by more than 15%, then the behavior section is determined to be the liquid discharge over-pumping event block.

[0140] In this embodiment, when identifying over-pumping events, if the slope of the liquid level rapidly changes from negative to positive within 2 minutes, it indicates that the well fluid has rebounded after reaching its limit. Simultaneously, if the pump frequency remains at a high frequency (≥25Hz) without significant fluctuations, the pressure rises by more than 1.5MPa within a short period, and the gas production rate continuously decreases by more than 15% during the event, then this behavioral segment is considered a typical over-pumping event. Identifying such events is crucial for preventing pump damage and gas lock phenomena. Its critical value is derived from comprehensive statistical analysis of field gas production pump operation monitoring data, demonstrating engineering feasibility.

[0141] The deep-drainage event blocks, steady-state pressure-maintaining operation event blocks, intermittent restart-type drainage event blocks, and drainage over-drainage event blocks are structurally summarized according to their respective well numbers and start and end times to obtain the operation event blocks.

[0142] In this embodiment, the identified four types of job event blocks are structurally summarized, using the well number as the primary key field, the start and end times of the event block as the time index, and a unified event tag field is established to construct a standard set of job event blocks, providing basic semantic support for subsequent event flow construction and path decision-making. All data structures are uniformly organized in a nested JSON format to ensure that they can be parsed and called by the subsequent scheduling platform.

[0143] Optionally, the reconstruction of the well site state change map in step S2 includes:

[0144] The liquid level height sequence, pump frequency change sequence, pressure response sequence, and gas production rate curve for each operation event in the multi-source event flow are extracted by numbering each well, thereby generating the physical response matrix of the operation event;

[0145] In this embodiment, for each operational event in the multi-source event stream, four types of physical monitoring data sequences—liquid level, pump frequency, pressure response, and gas production rate—are extracted based on the corresponding time period of the event. The time resolution of data acquisition is set to 0.5 seconds to ensure the capture of dynamic details. The extracted sequences constitute a four-dimensional time series matrix, with rows corresponding to time steps and columns corresponding to the four types of monitoring indicators. To address the differences in the time length of different events, a linear time interpolation method is uniformly used to adjust to a fixed time step, ensuring the consistency of the physical response matrix dimensions for easy subsequent comparison. This matrix is ​​stored using event numbers as indexes for quick retrieval and cross-event comparison.

[0146] Calculate the state response similarity of the eigenvectors between any two events in the physical response matrix of the task events to obtain the event similarity matrix;

[0147] In this embodiment, the four-dimensional time series features of the physical response matrix of each operational event are converted into a fixed-length state response vector. Specifically, a feature vector is formed by calculating 12 statistical features of the time series, including the mean, variance, maximum, and minimum values. Based on this feature vector, the similarity between any two events is calculated pairwise. The response similarity is quantified using a weighted inner product method, with weights of 0.3 for liquid level, 0.3 for pump frequency, 0.2 for pressure response, and 0.2 for gas production rate, reflecting the relative importance of different indicators in contributing to the similarity. The calculation results are stored as an event similarity matrix, with matrix elements ranging from 0 to 1. Elements closer to 1 indicate more similar event state responses.

[0148] Analyze the physical proximity relationships of each working well in the dynamic well field state structure diagram to obtain the physical proximity relationship matrix, and establish the initial connection diagram of events based on the physical proximity relationship matrix and the similarity matrix between events;

[0149] In this embodiment, based on the spatial distribution of each working well in the dynamic well site state structure diagram, the spatial coordinates of the wellheads and pipeline connectivity data from the well site design documents are collected, and the physical proximity relationships between wells are calculated using Euclidean distance. A proximity threshold of 30 meters is set; well pairs with a distance below this threshold are marked as 1 in the physical proximity matrix, indicating physical proximity, while those above are marked as 0. This proximity matrix is ​​a binary symmetric matrix. Combined with the event similarity matrix, an initial event connection graph is initially established through matrix multiplication. The edge weights of the connection graph are the product of physical proximity and state response similarity, reflecting the composite relationship between events that possesses both physical proximity and state correlation.

[0150] Extract all edges with a weight greater than 0.85 from the initial event connection graph as strong dependent event connections, and construct a set of state mutation propagation links;

[0151] In this embodiment, connections with edge weights greater than 0.85 are selected from the initial event connectivity graph and considered strong dependencies. These edges represent event pairs with a high probability of state mutation propagation. The subgraph formed by these edges is extracted as a state mutation propagation link set, which reflects the influence paths of strong correlations between well site operation events. The link set is stored as an edge list, including edge start point, end point, and weight attributes, supporting subsequent propagation path analysis and job scheduling decision support.

[0152] Reachability assessment is performed on links in the initial connection graph of the event with a path length less than 2 and an edge weight in the range of [0.85, 0.9). Isolated edges or broken links with an edge weight less than 0.8 in the reachability assessment results are structurally removed to obtain the propagation structure path set.

[0153] In this embodiment, reachability checks are performed on links in the initial connection graph of the event with weights between 0.85 and 0.9 and path lengths not exceeding 2. The check process is based on the connectivity and weight continuity of the connections between nodes, removing isolated edges with weights below 0.8 and path interruptions to ensure the integrity and effectiveness of the propagation path. The resulting set of structural paths better reflects the physical characteristics of actual well site state propagation, reducing the impact of misjudgments. This path set is stored in the form of an adjacency matrix for easy and rapid querying of paths and link strengths.

[0154] The operation event nodes in the propagation structure path set are used as the graph node set, and the state mutation propagation link set is used as the graph edge set; the well site event topology graph is constructed by combining the graph node set and the graph edge set.

[0155] In this embodiment, all operational event nodes are extracted based on the propagation structure path set. Each node carries a unique identifier for the event, a timestamp of occurrence, and a summary of associated monitoring data (such as statistical characteristics of key parameters like liquid level and pump frequency), ensuring the multidimensional integrity of node attributes. Next, the propagation relationships and edge weights corresponding to each edge in the state mutation propagation link set are loaded into a graph edge set, where edge weights reflect the propagation strength and dependency between events. Then, a graph database management method is used to model the node and edge sets, storing them according to node type (operation event nodes) and edge type structure (state mutation propagation edges). Both nodes and edges are stored in a graph database (such as Neo4j or JanusGraph), with nodes indexed by event IDs and edges indexed by start and end node IDs. Finally, based on this topology graph, the latest state and dynamic edge weights of the nodes are periodically synchronized and updated.

[0156] Optionally, the well scheduling suitability score in step S4 includes:

[0157] Acquire real-time multi-source well monitoring data, and extract well status features from the pre-processed real-time multi-source well monitoring data to obtain a set of well status feature vectors;

[0158] In this embodiment, multiple sensors deployed at the well site collect real-time monitoring data from various wells, including indicators such as fluid level, pump frequency, pressure response, and gas production. The collected data undergoes pre-defined filtering and outlier removal processes to ensure accuracy and continuity. The data's time resolution is set to 1 second, and it is transmitted to the dispatch cloud platform in real time. After data cleaning, key state characteristics for each well are extracted from the real-time data, including average fluid level, peak pump frequency, pressure fluctuation amplitude, and gas production rate change rate, forming an 8-dimensional operational state feature vector. This vector encompasses the statistical characteristics of multiple physical parameters and is used for subsequent state analysis.

[0159] Based on the operation event blocks, identify the types of operation events that occurred in the monitoring data of the operation wells to be analyzed within the past 24 hours, and extract the behavioral characteristics of the operation wells within the past 24 hours, thereby constructing a behavioral response label vector for each operation well;

[0160] In this embodiment, by combining the operational event block identification results within the past 24 hours, the actual event types occurring in each well are filtered out, such as pumping, fracturing, or well shutdown, ensuring that the event identification accuracy reaches over 90%. For the identified events, behavioral characteristics of each well are extracted, including parameters such as event duration, event frequency, and event interval, to construct a behavioral response label vector. This label vector has a fixed 6-dimensional dimension, with each dimension representing a behavioral characteristic indicator. Smoothing and trend capture are achieved through a moving average technique within a time window to ensure that it can reflect the recent operational behavior patterns of the wells.

[0161] For each working well, the corresponding working status feature vector set and behavior response label vector are subjected to feature correlation analysis. Based on the correlation analysis results, the status-behavior matching score is calculated to obtain the status-behavior adaptation score matrix.

[0162] In this embodiment, a correlation analysis is performed on the state feature vector of each working well and its corresponding behavioral response label vector. A weighted correlation coefficient is used to quantify the degree of matching between the two. The weight parameters are set to 70% for state features and 30% for behavioral labels, reflecting the dominant role of the current state in the adaptability of the working behavior. The correlation analysis results are normalized to generate a matrix. The rows of the matrix correspond to the working well number, the columns correspond to the working event type, and the element values ​​are the state-behavior adaptation scores, ranging from 0 to 1. The higher the value, the better the well state matches the behavior.

[0163] Obtain the well site resource allocation log, and use the current schedulable resource status in the well site resource allocation log as a constraint to score and correct the state-behavior adaptation scoring matrix to obtain the well operation adaptability scoring matrix.

[0164] In this embodiment, the current status of available equipment, personnel, and materials is obtained by synchronizing the well site resource allocation log in real time, serving as constraints for resource allocation. Based on the real-time scheduling capacity and the amount of allocable resources in the resource log, the scores in the state-behavior adaptation scoring matrix are dynamically adjusted. For example, for wells with scarce resources, the adaptation score is automatically reduced to reflect the impact of resource constraints on scheduling priority. The correction uses a linear scaling method, with the scaling factor dynamically adjusted according to resource scarcity to ensure that the scoring matrix accurately reflects the scheduling reality of the well site, ultimately forming a working well adaptation scoring matrix, providing reliable input for subsequent scheduling path planning.

[0165] Most importantly, the rating correction is as follows:

[0166] Extract the current schedulable resource status from the well site resource allocation log and generate a resource status matrix;

[0167] In this embodiment, the current schedulable resource status is extracted from the well site resource allocation log. This resource status includes equipment type (such as pumps, fracturing machines, pipeline valves, etc.), total resource capacity, current remaining available capacity, location coordinates, and resource operating time window. The resource status data is stored in matrix form, called the resource status matrix R, with dimensions m×n, where m represents the number of resource types and n represents the number of resource instances. Each element specifically includes the resource ID, remaining capacity, and resource activation time period, ensuring that the dynamic tracking accuracy of resource status reaches the minute level.

[0168] Based on the resource types and time requirements of each scoring unit in the state-behavior adaptation scoring matrix, perform coverable job behavior requirement matching from the resource state matrix to obtain a subset of required resources.

[0169] In this embodiment, based on each scoring unit in the previously calculated state-behavior adaptation scoring matrix, the corresponding job behavior type and its resource requirements are analyzed, including resource type (e.g., requiring 1 liquid pump and 2 pipeline valves), quantity requirements, and time window requirements (e.g., requiring a duration of 2 hours). By matching these requirements with the resource state matrix R, a subset of resources that can meet the behavior requirements is selected. Resource requirement matching employs a rule-based mapping mechanism to ensure that the selected resources fully cover the time period and quantity requirements of the job behavior. This process is dynamically updated to adapt to real-time changes in resource allocation, ensuring scheduling accuracy.

[0170] Based on the time window required for the extraction of a subset of demand resources, the average response time from the starting point of the current resource to the working well is calculated, and the resource response delay ratio is calculated. When the resource response delay ratio is greater than 1, a scoring penalty factor is defined, thereby constructing a scoring penalty factor matrix.

[0171] In this embodiment, resource availability time windows are extracted based on resource subsets. Combining the coordinates of the working well location and the resource starting point location, the average response time from the starting point to the target working well is calculated. The response time is estimated using a preset distance-weighted model, considering the length of the transportation path within the well site and the average transportation speed (default 5 km / h). The resource response delay ratio is calculated based on the ratio of the actual response time to the operation behavior time window. When this ratio is greater than 1, it indicates a risk of time delay in resource response; a scoring penalty factor is defined accordingly, specifically in the form of… ,in As a scoring penalty factor, This represents the resource response latency ratio. This scoring penalty factor is used to reduce scheduling priority. The scoring penalty factor matrix P consists of the penalty factor corresponding to each scoring unit, and the matrix dimension is consistent with the adaptation scoring matrix.

[0172] Extract the maximum capacity value and current remaining capacity of each type of job resource in the demand resource subset, calculate the resource schedulable weight vector, and then construct the scoring schedulable correction matrix;

[0173] In this embodiment, for each type of resource in the demand resource subset, its maximum capacity value and current remaining capacity are extracted to calculate a resource schedulable weight vector. The weight calculation is based on the ratio of remaining capacity to maximum capacity; the higher the ratio, the greater the weight. The weight range is limited to 0.1 to 1.0 to prevent resources from being excluded due to excessively low weights. The weight vector corresponds one-to-one with the resource type, forming a scoring schedulable correction matrix W. This matrix matches the elements of the scoring matrix in the resource demand dimension of the job behavior, and is used to reflect the weight correction of the current schedulable capacity of the resources.

[0174] The state-behavior fit scoring matrix is ​​jointly corrected by the scoring penalty factor matrix and the scoring schedulable correction matrix to obtain the well fit scoring matrix.

[0175] In this embodiment, the scoring penalty factor matrix P and the scoring schedulable correction matrix W are jointly applied to the state-behavior adaptation scoring matrix, and a weighted superposition method is used for comprehensive correction. The correction formula is set as: Corrected score = Original score × (1 - Penalty factor) × Weight correction factor, ensuring that the score reflects both the risk of resource response delay and the schedulable capacity of available resources. After correction, the work well adaptability scoring matrix is ​​output, with all matrix elements in the range of 0 to 1, serving as an important basis for subsequent resource allocation and scheduling path planning, improving the rationality of scheduling and real-time response capability.

[0176] Figure 3 This is a schematic diagram illustrating an application scenario for well scheduling and access planning provided in an embodiment of this application. For example... Figure 3 As shown, multiple functional areas can be set up at the well site. For example, these functional areas may include: a working area 101 containing multiple working wells, a main access path 102, and a storage area for dispatchable resources 103. The area in 103 closest to 101 is called Area A, and the resources in Area A may include, but are not limited to, a liquid pump truck, liquid nitrogen tank group, and pressure control equipment. The area in 103 farther from 101 is called Area B, and the resources in Area B may include, but are not limited to, an electric drive control unit (used to control variable frequency pumps or other drive equipment), emergency maintenance vehicles, personnel dispatch and assembly points, or portable power distribution rooms.

[0177] exist Figure 3In section 102, the main access path to the well site is defined. 102 can be defined based on the number of paths in different well sites, and can also simultaneously define both main and secondary access paths; this embodiment does not limit this. Since in actual application scenarios, the working areas of the operating wells may not necessarily exist in the same area simultaneously; the working area can be one or multiple areas. Therefore, area 101 can be defined based on the distribution of operating wells in different well sites; this embodiment does not limit this. For Figure 1 The distance relationship between region A and region B in region 103 is not as shown. Figure 1 The distance from area 101 is defined in this way. The reason for this definition in this embodiment is for ease of understanding. It can be defined according to the actual well site terrain. This embodiment does not limit it.

[0178] Optionally, the well scheduling path planning in step S4 includes:

[0179] Extract the field access log dataset from the well site resource allocation log, and set the scheduling start point set based on the location of each operational resource in the field access log dataset;

[0180] In this embodiment, the well site resource allocation logs generated within the past 48 hours are first retrieved from the well site scheduling platform. Fields recording equipment departure time, current location, target well, and actual travel path are extracted to generate a field travel log dataset. This dataset structure includes resource number, current coordinates (latitude and longitude), resource status (standby, executing, occupied), resource type, and its functional area. Figure 3 Taking the area shown as an example, in area A of 103, three liquid pump trucks and one liquid nitrogen tank set were identified, all of which are in a dispatchable state. A dispatch starting point set is constructed using their coordinates, and the spatial coordinates of each resource are recorded as a starting node. To facilitate path evaluation, each dispatch starting point record also includes a set of maximum dispatch radius for the resource, such as 2.5km for the pump trucks and 2.0km for the liquid nitrogen tank set. This method forms the initial set of starting nodes for the dispatch path.

[0181] The scheduling and passage difficulty of each well is assessed based on the location of each operational resource in the scheduling starting point set, the distribution of each operational well in the dynamic well site status structure diagram, and the pipeline layout.

[0182] In this embodiment, the scheduling starting point set is spatially paired with the working well locations in the dynamic well site status structure diagram to extract... Figure 3The coordinate data of each working well within area 101 are matched with the path grid in area 102. The accessibility assessment considers path length, path gradient (calculated using topographic maps and elevation data), road type (paved / unpaved), curve density, and obstacle density. For example, in practical application, the shortest path from a pump truck in area 103A to well W-5 in area 101 is 1.6km, with a maximum gradient of 12%, including two sharp bends and a section of unpaved gravel road. Therefore, the basic accessibility value for this path is set to 0.68 (normalized to 0–1). This data is dynamically updated using a GIS base map combined with data from access sensors to form a well site scheduling accessibility matrix, providing a basis for subsequent path weighting.

[0183] Based on the distribution of each working well and the pipeline layout, the starting point where the dispatching starting point is concentrated is connected to the working well, and the dispatching passage difficulty is used as the edge weight of the connection to obtain the initial path map;

[0184] In this embodiment, after the accessibility assessment is completed, each resource in the scheduling starting point set is connected to each working well in area 101 via a path connection. The connection rules are based on... Figure 3 The area shown as 102 serves as the main route, with the shortest path between each resource's actual starting point and the work well as an edge in the graph. Each edge is assigned a travel difficulty value, constructing an initial route graph with multiple starting and ending points. The graph structure is represented by an adjacency matrix, where the matrix elements are the edge weights of the paths between resource nodes and work well nodes. For example, the edge weight from pump truck R1 to W-5 is 0.68, and from R1 to W-3 it is 0.55. If the route includes secondary paths, a penalty factor of 0.1 is added to the weights to form a weighted route structure that more closely reflects the actual travel environment.

[0185] The adaptability score of each well in the well adaptability score matrix is ​​taken as the scheduling feedback weight factor, and the weight of each edge in the initial path map is adjusted by weighting to obtain the well site path map.

[0186] In this embodiment, after constructing the initial pathway map, the edge weights of each path are adjusted based on the well suitability scoring matrix. The well suitability scoring matrix is ​​derived from the previous steps through behavioral state matching and resource state correction. For example, the current score of well W-5 is 0.82, indicating that it is relatively suitable for scheduling the current pump truck resources; therefore, the initial path edge weight of 0.68 is multiplied by the scheduling feedback factor 1 / 0.82 ≈ 1.22, resulting in a weighted edge weight of 0.83. Conversely, if the score of a well is too low, such as W-7 at 0.45, its path edge weight will be amplified to the original value multiplied by 2.22, reducing the priority of this path in subsequent pathway searches. This operation forms the final well site pathway map, where each edge reflects both the physical access difficulty and the degree of suitability for operational needs.

[0187] Perform optimal path search on each starting point in the well site access map to obtain the optimal scheduling path set;

[0188] In this embodiment, based on the constructed weighted well site access map, all travel paths between the starting and ending points are traversed. First, a set of all unique starting-to-working-well paths is extracted from the map. Each path consists of a series of node sequences and corresponding edge weights. For each path, the total path length, the proportion of closed areas (e.g., sections with construction or closure), the proportion of ramps, and the proportion of sections with water-filled pits are extracted. The travel risk score is the weighted average of these four indicators. For example, if a path is 1.4 km long, has a closed area proportion of 10%, a ramp proportion of 25%, and a water-filled pit proportion of 5%, the travel risk R is: ; where 3.0 is the maximum path length limit set (used for normalization). Finally, the path with the lowest travel risk is selected from all feasible paths as the optimal scheduling route from each starting point to the working well.

[0189] By integrating the optimal scheduling paths between each starting point and the working well in the optimal scheduling path set, a resource allocation map is obtained.

[0190] In this embodiment, the paths between each resource origin and the work well in the optimal scheduling path set are integrated. Figure 3 For example, pump truck R1 (located in area 103A) is ultimately allocated to well W-3, with the optimal route including the middle section of the main road 102 and a section of unpaved auxiliary road; liquid nitrogen tank group R2 is allocated to well W-5, its route crossing the intersection of secondary roads and the main road network. Each route is coded in the form of a standardized path vector, with fields including the path node sequence, total length, estimated time, risk coefficient, allocated resource ID, and target working well ID, ultimately forming a complete resource allocation map. This map is uploaded to the work scheduling cloud platform for the execution system to parse and instruct each work vehicle and personnel to carry out the task.

[0191] Most importantly, the optimal path search is specifically as follows:

[0192] Extract all feasible path sets based on the scheduling starting point set and the well site access map;

[0193] In this embodiment, by parsing the resource coordinate information in the scheduling starting point set and the node connection relationship in the well site access map, a system is constructed from each resource location (e.g., Figure 3This is a set of all accessible routes from pump trucks in Zone A and emergency repair vehicles in Zone B to each working well in Zone 101. The well site access map is generated based on the GIS map and road structure data deployed at the well site. Each path consists of a set of continuous coordinate points and structural parameters (road type, width, access permission, etc.) of the corresponding connecting segments. No default paths are preset; instead, the paths are extended step by step according to all connectivity relationships on the map until all reachable destination wells are covered. Taking pump truck R1 as an example, its path starting from Zone A may detour through the middle section of the main access road 102 and then turn back to well W-5. The path segment structure must cover turning nodes, intersection nodes, and bridge and culvert structures, and filter out nodes marked as closed or dangerous within the current scheduling cycle during the path construction process. All feasible paths will be structured into a path record set. Each path contains fields such as number, start ID, end ID, node sequence, number of path segments, and estimated travel time, forming the complete path set in the working well scheduling scenario.

[0194] The risk of passage of a route is assessed based on the total length of the route, the proportion of closed areas, the proportion of slopes, and the proportion of water pits in the set of feasible routes.

[0195] In this embodiment, after the path set is established, a traffic risk assessment is performed on each path. The assessment uses historical traffic records from the traffic log data and attribute parameters attached to each path structure in the GIS map. The total path length is obtained by accumulating the distances of each segment and normalizing it with the maximum path standard (e.g., 3.0km). The proportion of closed areas is calculated based on the ratio of the number of nodes marked "closed" or "restricted" in the GIS layer to the total number of path segments in the current period. The proportion of slopes is obtained by the ratio of the length of segments with a cumulative slope change exceeding 5% to the total path length. The proportion of water pits is determined by the data returned from the water accumulation monitoring nodes in low-lying areas controlled by the well site; any segment in the path that passes through a node with a water accumulation alarm in the past 12 hours is included. Each indicator is assigned a corresponding risk weight, for example, 0.2 for total length, 0.3 for closed areas, 0.3 for slopes, and 0.2 for water pits, and the path traffic risk value is output in a weighted summation form. For example, the total length of the route from R1 to W-5 is 1.6km, accounting for 0.53 of the maximum route. It includes one closed section (10%), 20% of the route is sloping, and one section is flooded (6%). The final traffic risk R is: This passage risk value will be used for subsequent optimal path selection.

[0196] The optimal scheduling path set is obtained by selecting the path with the minimum path risk to different working wells from the set of feasible paths.

[0197] In this embodiment, the path from each resource to each target well is compared based on minimizing the travel risk value. Specifically, among all paths from the scheduling starting point R1 to multiple wells such as W-3, W-4, and W-5, the path with the lowest travel risk value is selected first and recorded as the optimal scheduling path from the current resource to that well. This path considers not only terrain and distance but also the travel safety and efficiency in the current time period. For example, although the path from R1 to W-3 is only 1.2km, it traverses two closed sections with a gradient exceeding 15%, resulting in a risk value as high as 0.33. Conversely, the path from R1 to W-5, although 1.6km long, has a gentle path, very few closed sections, and no water-filled areas, resulting in a risk value of only 0.174. Therefore, W-5 is marked as the optimal operational target for R1 in the current period, and the path sequence is the optimal scheduling path for R1. All optimal paths between resources and wells are recorded in the optimal scheduling path set, and the path structure data is stored in the well site task scheduling database for subsequent generation of resource allocation maps.

[0198] Optionally, step S5 includes:

[0199] Upload the resource allocation map to the job scheduling cloud platform to execute the resource allocation task;

[0200] In this embodiment, after the resource allocation map is generated, it is uploaded to the job scheduling cloud platform in the form of a structured chart. This chart structure includes node fields (resource number, wellhead number), edge fields (allocation path ID, path sequence, estimated travel time), and control parameter fields (resource type restrictions, scheduling start and end times). The scheduling platform retrieves the idle status of well site resources within the current period and triggers the corresponding resource to start the vehicle-mounted terminal based on the edge fields in the chart. Each vehicle-mounted terminal is equipped with a GPS positioning unit and a wireless communication unit, uploading its current location and vehicle status (including whether resources are loaded and whether the operation has been completed) to the cloud every 15 seconds while the vehicle is on the road. The operation start flag is triggered by a near-field communication node deployed next to the wellhead, confirming that the resource has arrived at the work well and initiating the operation. This mechanism can maintain the continuity of scheduling status transmission even in areas with low signal coverage, ensuring closed-loop tracking of the entire allocation process.

[0201] During the execution of resource allocation tasks, on-site monitoring data at the well site is collected through sensor arrays and on-board terminals on the work vehicle;

[0202] Calculate the execution effect score of each allocation task execution path in the resource allocation diagram based on the well site monitoring data, and output the scheduling execution score matrix;

[0203] In this embodiment, during the operation, the sensor group deployed at the well site (pressure gauge, liquid level monitor, current sensor, etc.) and the operation recorder on the vehicle collect on-site execution data in real time. Data types include, but are not limited to, wellhead liquid level drop rate, operation duration, pump frequency stability rate, and pressure release slope. A standard reference range is set: for example, the liquid level drop rate must exceed 6 cm / min, pressure release must be completed within 3 minutes, and the operation duration error must be controlled within ±10%. Each resource allocation path corresponds to a scoring evaluation. The score is calculated based on the deviation of each indicator from the standard, using a multi-factor scoring model. The highest score for each indicator is 0.25, and the total score is controlled within the range [0,1]. For example, if the liquid level drop rate only reaches 5 cm / min and the pump frequency fluctuates frequently, the score will be below 0.6, indicating an inefficient scheduling path. The range of values ​​for this scoring parameter is based on statistical analysis of data from 150 past operation allocations, effectively distinguishing between efficient and abnormal operations.

[0204] Write the scheduling execution score matrix back to the resource allocation graph, and extract the paths with execution performance scores below 0.6 from the written-back resource allocation graph to obtain a set of inefficient scheduling paths;

[0205] In this embodiment, the execution scores of all scheduled paths are summarized into a scheduling execution score matrix. The matrix index fields are the starting resource ID and the target job well ID, with a score value in the range [0,1]. Paths with scores below 0.6 are filtered, and their corresponding pathway structures and historical score trends are recorded. Taking resource R1 to well W-3 as an example, if the score is only 0.52, its pathway structure will be reconstructed in the next round of scheduling, and the original path sequence will be prohibited. The reconstruction step will re-evaluate the available paths and remove closed or high-risk sections from the pathway graph to ensure that the replanning has better travel conditions. The score threshold is set to 0.6, a critical value established based on the relationship between scheduling failure rate and score in the statistical results. When the score is below 0.6, its allocation success rate is below 70%, and therefore it is defined as an inefficient path.

[0206] Re-execute steps S4 to S5 on the inefficient scheduling path set to update the planned scheduling path. After each round of resource allocation is completed, count the number of working wells covered by the current round of scheduling and the number of scheduling paths with an execution effect score greater than or equal to 0.8 to calculate the scheduling efficiency.

[0207] If the scheduling efficiency is greater than or equal to the preset efficiency threshold of 0.85, the resource allocation graph updated in this round is determined to have converged to optimality, and the iteration stops.

[0208] If the scheduling efficiency is less than the preset efficiency threshold of 0.85, then the updated resource allocation map will be re-executed in steps S4 to S5 to trigger the next round of updating and planning the scheduling path, until the scheduling efficiency of the new round is greater than or equal to the preset efficiency threshold of 0.85.

[0209] In this embodiment, after each round of scheduling is completed, the scheduling platform will statistically analyze two core indicators: the number of wells successfully covered in the current scheduling graph (e.g., 17 out of a total of 20 wells completed in this round) and the number of efficient scheduling paths with a score greater than 0.8 (e.g., 18 out of 22 deployed paths have a score ≥ 0.8). The former represents scheduling coverage, and the latter represents scheduling quality. The scheduling efficiency is calculated using the following formula: Scheduling efficiency = 0.5 × (well coverage rate) + 0.5 × (percentage of efficient paths). When the scheduling efficiency ≥ 0.85, the optimal scheduling graph is considered to have converged, and the iteration is terminated. This threshold is set because, in practical experience, when the efficiency reaches 85% or higher, the corresponding average work cycle meets both the standards of work efficiency and work safety. If it is lower than this value, steps S4 to S5 are re-executed on the current scheduling graph to enter the next round of path reconstruction until the preset efficiency threshold of 0.85 is reached.

[0210] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0211] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A scheduling method for a cloud-based intelligent scheduling platform for drainage and gas extraction operations, characterized in that, Includes the following steps: Step S1: Acquire monitoring data of multi-source operating wells through the sensor group pre-deployed at the well site, and perform well site state structure modeling on the multi-source operating well monitoring data to obtain the well site state representation tensor; Step S2: Identify and classify operational events based on the well site state representation tensor, and establish a multi-source event flow; reconstruct the well site state change map based on the multi-source event flow to obtain the well site event topology map; Step S3: Parse the triggering conditions of each operation event in the well site event topology diagram into equipment status acquisition instruction templates, and transmit the equipment status acquisition instruction templates to the sensor group at the well site through the operation scheduling cloud platform to execute the acquisition instruction pre-deployment task; Step S4: Obtain real-time multi-source well monitoring data, perform well scheduling adaptability scoring based on the real-time multi-source well monitoring data, and obtain well adaptability scoring matrix; use well adaptability scoring matrix and the obtained well site equipment access logs to plan well scheduling paths and obtain resource allocation map; Step S5: Upload the resource allocation map to the job scheduling cloud platform and execute the resource allocation task; use the well site monitoring data collected after the resource allocation task to iteratively optimize the resource allocation map until the well site operation efficiency reaches the preset efficiency threshold.

2. The scheduling method of the intelligent scheduling platform for drainage and gas extraction operations based on a cloud platform according to claim 1, characterized in that, Step S1 includes: Step S11: Collect monitoring data of multi-source wells through the sensor group pre-deployed at the well site, and clean the monitoring data of multi-source wells to obtain the monitoring data of the wells to be analyzed; Step S12: Align the monitoring data of the well to be analyzed in the time domain and spatial domain to obtain the aligned monitoring data; Step S13: Normalize and standardize the format of the aligned monitoring data to obtain a structured matrix frame; Step S14: Construct a dynamic well site state structure diagram based on the structured matrix frame; Step S15: Vectorize the dynamic well field state structure diagram to obtain the well field state representation tensor.

3. The scheduling method of the cloud-based intelligent scheduling platform for drainage and gas extraction operations according to claim 2, characterized in that, Step S14 includes: Step S141: Extract the state features of each working well in the structured matrix frame according to the preset time step, and construct a single well state vector set; Step S142: Perform similarity analysis on the wellhead state vectors of continuous time steps in the single well state vector set, and calculate the similarity score matrix between all well pairs; Step S143: Construct a wellhead node graph structure using the well number of the working well as the node index, the state features of each working well in the single well state vector set as nodes, and the similarity score matrix as edges; Step S144: Obtain well site structure design data through the job scheduling cloud platform, and use the well site structure design data to perform structural constraint topology enhancement on the wellhead node diagram structure to obtain a dynamic well site state structure diagram.

4. The scheduling method of the cloud-based intelligent scheduling platform for drainage and gas extraction operations according to claim 1, characterized in that, Step S2 involves identifying and classifying job events, including: Extract the sequence of physical features at the wellhead from the well site state representation tensor according to the well number; A sliding time window with a fixed width of 5 minutes was set to calculate the rate of change of state and the amplitude of characteristic fluctuations in adjacent time periods of the wellhead physical characteristic sequence. Based on the rate of change of state and the amplitude of characteristic fluctuations in adjacent time periods, the operation behavior window was filtered to obtain candidate operation behavior windows. The candidate job behavior window is filtered for valid job behaviors, and event pattern recognition is performed on the valid job behaviors to obtain job event blocks; Based on the task event blocks, the overlapping event blocks are split and merged in sequence to obtain the task event blocks to be combined; The data structure of the event blocks to be combined is standardized, and the sequences are combined according to the well number and event time sequence to obtain a multi-source event stream.

5. The scheduling method of the cloud-based intelligent scheduling platform for drainage and gas extraction operations according to claim 4, characterized in that, Effective work behavior screening includes: When the rate of change of wellhead fluid level in any time window of a candidate operation behavior window has an average slope greater than 6 cm / min over 3 consecutive minutes, and the rate of change of pump frequency fluctuates within ±10% within that time window, while the pressure change range exceeds 0.8 MPa, the time window is determined to be a strong characteristic operation behavior window. The duration of strong characteristic operation behavior is extracted based on the strong characteristic operation behavior window. When the duration of strong characteristic operation behavior is greater than 5 minutes and the instantaneous drop rate of liquid level is less than -4cm / min, the time window is marked as an effective strong operation behavior segment. When the wellhead fluid level change rate in any time window of the candidate operation behavior window is between 2 and 6 cm / min over 3 consecutive minutes, and the pressure fluctuation amplitude is less than 0.5 MPa, and the pump frequency remains within ±5% of the previous time window state, the time window is determined to be a weak characteristic operation behavior window. Extract the duration of weak feature operation behavior within the weak feature operation behavior window. If the duration of weak feature operation behavior exceeds 10 minutes and the percentage of pumps in the ON state during the duration of weak feature operation behavior is >95%, then mark the time window as a valid weak feature operation behavior segment. If the wellhead fluid level in any time window of the candidate operation behavior window fluctuates within a range of less than ±8cm within 5 minutes, the difference between the maximum and minimum values ​​of the pump frequency in that time window accounts for no less than 40% of the maximum value, and the minimum frequency is close to the pump stop threshold of 3Hz, and the gas production rate drops instantaneously by more than 15%, then the time window is determined to be an intermittent operation behavior window. Extract the continuous duration of the intermittent operation behavior window. If the continuous duration is greater than 7 minutes and the gas production does not show a continuous downward trend, then mark the time window as an effective intermittent operation behavior segment. The effective strong operation behavior segment, effective weak operation behavior segment, and effective intermittent operation behavior segment are merged into the effective operation behavior.

6. The scheduling method of the cloud-based intelligent scheduling platform for drainage and gas extraction operations according to claim 4, characterized in that, Event pattern recognition includes: When the liquid level in the effective strong operation behavior section shows a continuous downward trend, with an average downward rate greater than 8 cm / min, and the pump frequency is always higher than 30 Hz and the fluctuation range is less than ±5% in this behavior section, while the pressure rise trend is stable and the gas production increase rate is greater than 10%, then this behavior section is determined to be a deep pumping liquid discharge event block. When the liquid level in the effective weak operation behavior section fluctuates slightly within ±10cm, the pump frequency remains between 20–30Hz with fluctuations not exceeding ±3% within the behavior section, the pressure is stable, and the rate of change of gas production rate is <5%, then the behavior section is determined to be a steady-state pressure-maintaining operation event block. If the pump frequency in an effective intermittent operation segment changes abruptly from a low frequency of less than or equal to 30 Hz to a high frequency of greater than or equal to 25 Hz at least twice within that segment, the liquid level has a second-order peak structure within that segment, and the gas production shows a synchronous upward trend, then that segment is determined to be an intermittent restart-type liquid discharge event block. When the slope of the liquid level changes from negative to positive in the effective strong operation behavior section and the switching time is less than 2 minutes, the pump frequency remains at a high frequency of greater than or equal to 25Hz during this period, the pressure fluctuation exceeds 1.5MPa, and the gas production continues to decrease by more than 15%, then the behavior section is determined to be the liquid discharge over-pumping event block. The deep-drainage event blocks, steady-state pressure-maintaining operation event blocks, intermittent restart-type drainage event blocks, and drainage over-drainage event blocks are structurally summarized according to their respective well numbers and start and end times to obtain the operation event blocks.

7. The scheduling method of the intelligent scheduling platform for drainage and gas extraction operations based on a cloud platform according to claim 1, characterized in that, The reconstruction of the well site state change map in step S2 includes: The liquid level height sequence, pump frequency change sequence, pressure response sequence, and gas production rate curve for each operation event in the multi-source event flow are extracted by numbering each well, thereby generating the physical response matrix of the operation event; Calculate the state response similarity of the eigenvectors between any two events in the physical response matrix of the task events to obtain the event similarity matrix; Analyze the physical proximity relationships of each working well in the dynamic well field state structure diagram to obtain the physical proximity relationship matrix, and establish the initial connection diagram of events based on the physical proximity relationship matrix and the similarity matrix between events; Extract all edges with a weight greater than 0.85 from the initial event connection graph as strong dependent event connections, and construct a set of state mutation propagation links; Reachability assessment is performed on links in the initial connection graph of the event with a path length less than 2 and an edge weight in the range of [0.85, 0.9). Isolated edges or broken links with an edge weight less than 0.8 in the reachability assessment results are structurally removed to obtain the propagation structure path set. The operation event nodes in the propagation structure path set are used as the graph node set, and the state mutation propagation link set is used as the graph edge set; the well site event topology graph is constructed by combining the graph node set and the graph edge set.

8. The scheduling method of the intelligent scheduling platform for drainage and gas extraction operations based on a cloud platform according to claim 1, characterized in that, The well scheduling suitability score in step S4 includes: Acquire real-time multi-source well monitoring data, and extract well status features from the pre-processed real-time multi-source well monitoring data to obtain a set of well status feature vectors; Based on the operation event blocks, identify the types of operation events that occurred in the monitoring data of the operation wells to be analyzed within the past 24 hours, and extract the behavioral characteristics of the operation wells within the past 24 hours, thereby constructing a behavioral response label vector for each operation well; For each working well, the corresponding working status feature vector set and behavior response label vector are subjected to feature correlation analysis. Based on the correlation analysis results, the status-behavior matching score is calculated to obtain the status-behavior adaptation score matrix. Obtain the well site resource allocation log, and use the current schedulable resource status in the well site resource allocation log as a constraint to score and correct the state-behavior adaptation scoring matrix to obtain the well operation adaptability scoring matrix.

9. The scheduling method of the intelligent scheduling platform for drainage and gas extraction operations based on a cloud platform according to claim 1, characterized in that, Step S4, the well scheduling pathway planning, includes: Extract the field access log dataset from the well site resource allocation log, and set the scheduling start point set based on the location of each operational resource in the field access log dataset; The scheduling and passage difficulty of each well is assessed based on the location of each operational resource in the scheduling starting point set, the distribution of each operational well in the dynamic well site status structure diagram, and the pipeline layout. Based on the distribution of each working well and the pipeline layout, the starting point where the dispatching starting point is concentrated is connected to the working well, and the dispatching passage difficulty is used as the edge weight of the connection to obtain the initial path map; The adaptability score of each well in the well adaptability score matrix is ​​taken as the scheduling feedback weight factor, and the weight of each edge in the initial path map is adjusted by weighting to obtain the well site path map. Perform optimal path search on each starting point in the well site access map to obtain the optimal scheduling path set; By integrating the optimal scheduling paths between each starting point and the working well in the optimal scheduling path set, a resource allocation map is obtained.

10. The scheduling method of the cloud-based intelligent scheduling platform for drainage and gas extraction operations according to claim 1, characterized in that, Step S5 includes: Upload the resource allocation map to the job scheduling cloud platform to execute the resource allocation task; During the execution of resource allocation tasks, on-site monitoring data at the well site is collected through sensor arrays and on-board terminals on the work vehicle; Calculate the execution effect score of each allocation task execution path in the resource allocation diagram based on the well site monitoring data, and output the scheduling execution score matrix; Write the scheduling execution score matrix back to the resource allocation graph, and extract the paths with execution performance scores below 0.6 from the written-back resource allocation graph to obtain a set of inefficient scheduling paths; Re-execute steps S4 to S5 on the inefficient scheduling path set to update the planned scheduling path. After each round of resource allocation is completed, count the number of working wells covered by the current round of scheduling and the number of scheduling paths with an execution effect score greater than or equal to 0.8 to calculate the scheduling efficiency. If the scheduling efficiency is greater than or equal to the preset efficiency threshold of 0.85, the resource allocation graph updated in this round is determined to have converged to optimality, and the iteration stops. If the scheduling efficiency is less than the preset efficiency threshold of 0.85, then the updated resource allocation map will be re-executed in steps S4 to S5 to trigger the next round of updating and planning the scheduling path, until the scheduling efficiency of the new round is greater than or equal to the preset efficiency threshold of 0.85.

Images