Equipment control method, device and equipment of beam-pumping unit and medium

Abstract

The invention discloses an equipment control method and device of a beam-pumping unit, equipment and a medium, and relates to the technical field of artificial intelligence. Equipment operation state data of the beam-pumping unit are obtained; inputting the equipment operation state data into a preset large language model, and outputting a structured decision instruction; the preset large language model is a model obtained by optimizing the initial large language model by using a low-rank adaptation method; analyzing and processing the structured decision instruction by using an intelligent agent cluster to obtain an initial equipment control instruction; performing safety and rationality evaluation on the initial equipment control instruction by using an artificial intelligence model; if the safety and rationality evaluation is passed, the initial equipment control instruction is converted into an equipment control instruction for controlling the equipment to execute corresponding actions, and the equipment of the beam-pumping unit is controlled, so that multi-target optimization such as deep fusion of multi-source data, improvement of data fusion utilization rate, cooperative processing energy efficiency, fault and safety is realized; and the equipment control instruction accuracy is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a device control method and device of a beam pumping unit, equipment and medium. BACKGROUND

[0002] At present, the device control of the beam pumping unit adopts a single parameter control method, mechanical structure optimization and a hydraulic drive scheme to realize the device control of the beam pumping unit. The disadvantages are as follows: the key parameters such as the stroke frequency and stroke of the beam pumping unit are adjusted depending on the experience of engineers, which consumes labor cost and time; the current, load and vibration data are currently analyzed independently, which is difficult to correlate and make decisions, and the utilization rate of the existing system for analyzing 2TB / day of well site data is less than 40%; in addition, the fault diagnosis relies on regular inspection, and the average disposal time of sudden failure is more than 4 hours, resulting in non-planned downtime loss accounting for more than 30% of the annual maintenance cost; a single AI (Artificial Intelligence) model only solves local problems, such as fault diagnosis or energy efficiency optimization, and cannot cooperatively handle multi-objective optimization of energy efficiency, fault and safety.

[0003] From the above, how to realize deep fusion of multi-source data, break through the island effect of current, visual, vibration and load data, improve the utilization rate of data fusion, cooperatively handle multi-objective optimization of energy efficiency, fault and safety, improve the accuracy of device control instructions, and automatically control the equipment of the beam pumping unit are problems to be solved in the field. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a device control method and device of a beam pumping unit, which can realize deep fusion of multi-source data, break through the island effect of current, visual, vibration and load data, improve the utilization rate of data fusion, cooperatively handle multi-objective optimization of energy efficiency, fault and safety, improve the accuracy of device control instructions, and automatically control the equipment of the beam pumping unit. The specific scheme is as follows:

[0005] In a first aspect, the present application discloses a device control method of a beam pumping unit, comprising:

[0006] obtaining device running state data of the beam pumping unit; the device running state data comprises motor parameters, beam swing angle dynamic change images, vibration parameters and pumping unit suspension point load;

[0007] inputting the device running state data into a preset large language model to output a structured decision instruction; the preset large language model is a model optimized by a low-rank adaptive method from an initial large language model;

[0008] Analyze and process the structured decision instruction by using an agent cluster to obtain an initial device control instruction; the agent cluster includes a visual agent, a fault diagnosis agent, an optimization decision agent, and a safety supervision agent;

[0009] Evaluate the safety and rationality of the initial device control instruction by using a preset artificial intelligence model;

[0010] If the safety and rationality evaluation passes, convert the initial device control instruction into a device control instruction for controlling the device to perform a corresponding action, and control the device of the beam pumping unit based on the device control instruction.

[0011] Optionally, the device operating state data of the beam pumping unit is obtained, including:

[0012] Obtain motor parameters of the beam pumping unit; the motor parameters include three-phase voltage, current, and power factor of the motor;

[0013] Obtain dynamic change images of the beam swing angle by using a camera that meets a preset frame rate;

[0014] Obtain vibration acceleration or speed spectrum by using a vibration sensor;

[0015] Measure the pumping unit suspension point load by using a load sensor.

[0016] Optionally, the device operating state data is input into a preset large language model to output a structured decision instruction, including:

[0017] Optimize an initial large language model by using oil well engineering professional literature and labeled well condition data, and using a low-rank adaptive method to obtain a preset large language model;

[0018] Input the device operating state data into the preset large language model, so that the preset large language model screens a target template matching the device operating state data from a preset decision template, fills the target template by using the device operating state data, screens historical data matching the filled target template from a historical fault library, and fills the historical data by using the filled target template to generate and output a structured decision instruction.

[0019] Optionally, the initial device control instruction is obtained by analyzing and processing the structured decision instruction by using an agent cluster, including:

[0020] Identify visual features of the structured decision instruction by using a visual agent in the agent cluster to obtain a visual feature identification result;

[0021] The fault diagnosis agent in the agent cluster is used to locate and analyze the faulty components of the structured decision instructions, and the faulty component location and analysis results are obtained.

[0022] The optimization decision-making agent in the agent cluster is used to calculate the optimization parameters of the structured decision instructions to obtain the oil pumping unit optimization parameters;

[0023] The structured decision instructions are evaluated and risk warnings are given in real time by using the security supervision intelligent agent in the intelligent agent cluster, and the real-time evaluation and risk warning results are obtained.

[0024] Initial equipment control commands are generated based on the visual feature recognition results, the fault component location and analysis results, the oil pumping unit optimization parameters, and the real-time assessment and risk warning results.

[0025] Optionally, the step of using a preset artificial intelligence model to evaluate the safety and rationality of the initial device control commands includes:

[0026] The initial device control command is sent to different types of artificial intelligence models so that the different types of artificial intelligence models can respectively evaluate the safety and rationality of the initial device control command;

[0027] Determine the number of cases that pass the safety and reasonableness assessment;

[0028] If the quantity is greater than a preset threshold, it indicates that the safety and rationality assessment has passed;

[0029] If the number is not greater than a preset threshold, it indicates that the security and rationality assessment has failed, and a manual review prompt message will be sent to the client so that the client can perform the manual review process based on the manual review prompt message.

[0030] Optionally, the step of converting the initial equipment control command into equipment control commands for controlling the equipment to perform corresponding actions, and controlling the beam pumping unit based on the equipment control commands, includes:

[0031] The initial equipment control commands are converted into adjustment commands for controlling the equipment to perform adjustment strokes using a frequency converter;

[0032] The alarm device is used to convert the initial equipment control commands into risk warning commands for real-time risk warning of the equipment.

[0033] The initial equipment control commands are converted into balance commands for controlling the balance of the beam pumping unit using a balance block motor.

[0034] Based on the adjustment command, the risk warning command, and the balance command, the equipment of the beam pumping unit is adjusted for stroke, real-time risk warning, and balance control.

[0035] Optionally, after controlling the beam pumping unit based on the equipment control command, the method further includes:

[0036] Acquire new equipment operating status data after controlling the beam pumping unit;

[0037] The reward function is calculated using the operating status data of the new equipment;

[0038] Based on the reward function and historical device operating status data, and using reinforcement learning algorithms, the decision network parameters of the optimization decision agents in the agent cluster are reinforced and updated.

[0039] Secondly, this application discloses a control device for a beam pumping unit, comprising:

[0040] The data acquisition module is used to acquire equipment operating status data of the beam pumping unit; the equipment operating status data includes motor parameters, dynamic change image of beam swing angle, vibration parameters, and suspension point load of the pumping unit;

[0041] The instruction output module is used to input the device operating status data into a preset large language model to output structured decision instructions; the preset large language model is a model optimized by using a low-rank adaptation method to optimize an initial large language model;

[0042] The analysis and processing module is used to analyze and process the structured decision instructions using an intelligent agent cluster to obtain initial equipment control instructions; the intelligent agent cluster includes a visual intelligent agent, a fault diagnosis intelligent agent, an optimization decision intelligent agent, and a safety supervision intelligent agent.

[0043] The evaluation module is used to evaluate the safety and rationality of the initial device control commands using a preset artificial intelligence model;

[0044] The equipment control module is used to convert the initial equipment control command into an equipment control command for controlling the equipment to perform corresponding actions if the safety and rationality assessment is passed, and to control the equipment of the beam pumping unit based on the equipment control command.

[0045] Thirdly, this application discloses an electronic device, including:

[0046] Memory, used to store computer programs;

[0047] A processor is used to execute the computer program to implement the aforementioned equipment control method for a beam pumping unit.

[0048] Fourthly, this application discloses a computer storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned equipment control method for a beam pumping unit.

[0049] As can be seen, this application provides a method for controlling a beam pumping unit, including acquiring equipment operating status data of the beam pumping unit; the equipment operating status data includes motor parameters, dynamic change images of the beam swing angle, vibration parameters, and suspension point load of the pumping unit; inputting the equipment operating status data into a preset large-scale language model to output structured decision commands; the preset large-scale language model is a model optimized from an initial large-scale language model using a low-rank adaptation method; analyzing and processing the structured decision commands using an intelligent agent cluster to obtain initial equipment control commands; the intelligent agent cluster includes a visual intelligent agent, a fault diagnosis intelligent agent, an optimization decision intelligent agent, and a safety monitoring intelligent agent; using a preset artificial intelligence model to evaluate the safety and rationality of the initial equipment control commands; if the safety and rationality evaluation is passed, the initial equipment control commands are converted into equipment control commands for controlling the equipment to perform corresponding actions, and the equipment of the beam pumping unit is controlled based on the equipment control commands. This application acquires the equipment operating status data of a beam pumping unit, including motor parameters, dynamic changes in the beam swing angle, vibration parameters, and suspension point loads of the pumping unit. It overcomes the data silos of current, visual, vibration, and load data, achieving deep fusion of multi-source data and improving data fusion utilization. The equipment operating status data is input into a pre-set large-scale language model to output structured decision commands, enabling cross-modal correlation analysis. An intelligent agent cluster is used to analyze and process the structured decision commands to obtain initial equipment control commands, eliminating potential safety hazards in industrial control and resolving control command errors caused by the "illusion" of the large-scale language model, thus avoiding unplanned downtime. A pre-set artificial intelligence model is used to evaluate the safety and rationality of the initial equipment control commands. If the safety and rationality evaluation is passed, the initial equipment control commands are converted into equipment control commands to control the equipment to perform corresponding actions. Based on these equipment control commands, the beam pumping unit is controlled, collaboratively handling multi-objective optimization such as energy efficiency, fault tolerance, and safety, improving the accuracy of equipment control commands, and automating the control of the beam pumping unit. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0051] Figure 1 This application discloses a flowchart of a device control method for a beam pumping unit.

[0052] Figure 2 This application discloses a structural diagram of the equipment control system for a beam pumping unit.

[0053] Figure 3 This application discloses a specific flowchart for the equipment control of a beam pumping unit.

[0054] Figure 4 This application discloses a data example diagram illustrating the disadvantages and risks of a prior art.

[0055] Figure 5 This is a schematic diagram of the equipment control device of a beam pumping unit disclosed in this application;

[0056] Figure 6 This application provides a structural diagram of an electronic device. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Currently, beam pumping units are controlled using single-parameter control methods, mechanical structure optimization, and hydraulic drive schemes. However, this approach has several drawbacks: Adjusting key parameters such as stroke rate and stroke relies heavily on engineer experience, resulting in high labor costs and time consumption; current data on current, load, and vibration are analyzed independently, making it difficult to correlate them for decision-making; existing systems utilize less than 40% of the 2TB / day wellfield data; furthermore, fault diagnosis relies on regular inspections, with average response time for sudden faults exceeding 4 hours, leading to unplanned downtime losses accounting for over 30% of annual maintenance costs; and single AI models only address localized problems, such as fault diagnosis or energy efficiency optimization, failing to address multi-objective optimization involving energy efficiency, faults, and safety. Therefore, achieving deep fusion of multi-source data, overcoming the silo effect of current, visual, vibration, and load data, improving data fusion utilization, collaboratively addressing multi-objective optimization involving energy efficiency, faults, and safety, and improving the accuracy of equipment control commands are all pressing issues to be addressed in this field.

[0059] See Figure 1 As shown in the figure, an embodiment of the present invention discloses a device control method for a beam pumping unit, which may specifically include:

[0060] Step S11: Obtain the equipment operating status data of the walking beam pumping unit; the equipment operating status data includes motor parameters, dynamic change image of the walking beam swing angle, vibration parameters, and suspension point load of the pumping unit.

[0061] In this embodiment, the motor parameters of the walking beam pumping unit are obtained; the motor parameters include the three-phase voltage, current, and power factor of the motor; the dynamic change image of the walking beam swing angle is obtained using a camera with a preset frame rate; the vibration acceleration or velocity spectrum is monitored using a vibration sensor; and the suspension load of the pumping unit is measured using a load sensor.

[0062] In this embodiment, a sensor network is used to provide system sensing capabilities and collect real-time equipment operating status data of the beam pumping unit. The sensor network includes the following four modules:

[0063] Electrical parameter module: Real-time acquisition of motor parameters such as three-phase voltage, current, and power factor with high precision (±0.5%).

[0064] High frame rate camera (200fps): Deployed in key locations to capture images of the dynamic changes in the swing angle of the walking beam.

[0065] Vibration sensors: Deployed in critical components such as bearings to monitor vibration acceleration / velocity spectrum (coverage range: 5-5000Hz) for fault diagnosis.

[0066] Load sensor: Real-time measurement of the load at the suspension point of the pumping unit is the core basis for calculating power, judging the balance status, and warning of overload risk.

[0067] Step S12: Input the device operating status data into a preset large language model to output structured decision instructions; the preset large language model is a model optimized by using a low-rank adaptation method to optimize the initial large language model.

[0068] In this embodiment, oil well engineering literature and labeled well condition data are used, and a low-rank adaptation method is employed to optimize the initial large language model to obtain a preset large language model. The equipment operation status data is input into the preset large language model so that the preset large language model can select target templates that match the equipment operation status data from preset decision templates, fill the target templates with the equipment operation status data, select historical data that match the filled target templates from the historical fault database, and fill the fields of the historical data with the filled target templates to generate and output structured decision instructions.

[0069] This application uses a large-scale language model fine-tuned using LoRA (Low-Rank Adaptation) as the core decision engine, such as DeepSeek-R1 / V3 and Tongyi Qianwen. The LoRA fine-tuning process uses 200,000 professional literature articles on oil well engineering and 20,000 sets of labeled well condition data, focusing on optimizing the understanding of physical laws, such as the application of Hooke's Law in the calculation of suspension point loads, to ensure that the model has deep domain knowledge.

[0070] In this embodiment, the central processing unit continuously receives device operating status data from sensors. This device operating status is then input as a state vector into the Prompt (decision-making) engine within the model to generate context-specific instruction prompts. The LLM (Large Language Model) analyzes the prompt content in conjunction with the real-time state vector. Based on the analysis results, a decision is made regarding whether to query the fault history database for historical fault references, in order to generate and output structured decision-making instructions.

[0071] The Prompt engine comprises the following components:

[0072] Input layer: Receives real-time state vector: S={s1, s2, ..., sn}, such as current fluctuation rate, swing angle deviation, vibration spectrum energy, etc.

[0073] Template matching layer: predefined multiple sets of Prompt templates (preset decision templates), such as fault diagnosis templates, parameter optimization templates, and emergency braking templates. Each template contains domain knowledge constraints, such as: "You are an oilfield control expert and need to make a decision within the [safety threshold]..." and task instruction slots;

[0074] The template type is selected based on the risk level identifier in the state vector, which is generated by SRAgent (a tool for automatically querying the SRA database);

[0075] Dynamic fill layer: Inject key parameters of the state vector into the template slot, such as: "Current impulse is {impulse value}, vibration spectrum in {frequency band} Hz abnormal energy percentage {percentage}...";

[0076] Retrieve summaries of similar cases from the fault history database, such as "Historical records show that similar vibration characteristics correspond to a bearing failure probability of 82%";

[0077] Output layer: Generates structured Prompt instructions in JSON (JavaScript Object Notation, a lightweight data interchange format) format, including: decision objectives, constraints, and reference historical data fields.

[0078] Step S13: Analyze and process the structured decision instructions using an intelligent agent cluster to obtain initial equipment control instructions; the intelligent agent cluster includes a visual intelligent agent, a fault diagnosis intelligent agent, an optimization decision intelligent agent, and a safety monitoring intelligent agent.

[0079] In this embodiment, a visual agent within the agent cluster performs visual feature recognition on the structured decision command to obtain visual feature recognition results; a fault diagnosis agent within the agent cluster locates and analyzes faulty components in the structured decision command to obtain faulty component location and analysis results; an optimization decision agent within the agent cluster calculates optimization parameters for the structured decision command to obtain pumping unit optimization parameters; a safety monitoring agent within the agent cluster performs real-time evaluation and risk warning on the structured decision command to obtain real-time evaluation and risk warning results; and initial equipment control commands are generated based on the visual feature recognition results, the faulty component location and analysis results, the pumping unit optimization parameters, and the real-time evaluation and risk warning results.

[0080] The AgentCluster in this application consists of the following specialized intelligent agents that work together to complete complex tasks:

[0081] ViAgent (Visual Agent): Processes image streams from high frame rate cameras to identify visual anomalies such as beam sway and angle in real time.

[0082] FDAgent (Fault Diagnosis Agent): Analyzes current timing data from the electrical parameter module and spectral data from the vibration sensor to locate potentially faulty components (such as bearings) and predict their remaining lifespan.

[0083] ODAgent (Optimization Decision Agent): Based on real-time operating conditions, historical data, and objectives (energy saving, production increase, and failure reduction), it calculates the optimal combination of stroke / stroke parameters for the pumping unit.

[0084] SRAgent (Safety Regulation Agent): Continuously monitors load sensor data, assesses and warns of load over-limit risks in real time, and triggers real-time risk warnings.

[0085] Step S14: Use a preset artificial intelligence model to evaluate the safety and rationality of the initial device control commands.

[0086] In this embodiment, the initial device control command is sent to different types of artificial intelligence models, so that the different types of artificial intelligence models can respectively evaluate the safety and rationality of the initial device control command; determine the number of times the safety and rationality evaluation passes; if the number is greater than a preset threshold, it indicates that the safety and rationality evaluation has passed; if the number is not greater than the preset threshold, it indicates that the safety and rationality evaluation has failed, and send a manual review prompt message to the client, so that the client can execute the manual review process based on the manual review prompt message.

[0087] To ensure the absolute security and reliability of the commands, this application conducts a security and rationality assessment and verification of the initial device control commands. The verification mechanism employs a voting mechanism using five artificial intelligence models. The initial device control commands are simultaneously submitted to five independent models for security and rationality assessment, such as DeepSeek, Tongyi Qianwen, Tencent Hunyuan, Doubao, and Zhipu.

[0088] Execution conditions: If the command receives approval votes from ≥4 AI models, it is deemed safe and can be issued for execution.

[0089] Review Mechanism: If there are fewer than 4 votes in favor, a manual review process will be initiated immediately, with engineers involved in the review and decision-making to ensure system security. If the LLM or agent cluster returns a high-risk rejection instruction (such as a suggestion to shut down), a manual review will be triggered directly, without requiring verification through voting by the five AI models.

[0090] Step S15: If the safety and rationality assessment is passed, the initial equipment control command is converted into an equipment control command for controlling the equipment to perform corresponding actions, and the equipment of the beam pumping unit is controlled based on the equipment control command.

[0091] In this embodiment, if the safety and rationality assessment is passed, the initial equipment control command is converted into an adjustment command for controlling the equipment to perform adjustment strokes using a frequency converter; the initial equipment control command is converted into a risk warning command for controlling the equipment to perform real-time risk warning using an alarm device; the initial equipment control command is converted into a balance command for controlling the balance of the beam pumping unit using a balance block motor; based on the adjustment command, the risk warning command, and the balance command, the beam pumping unit is subjected to stroke adjustment, real-time risk warning, and balance control.

[0092] This application translates initial device control commands that have passed security verification into specific device actions. It includes a key execution unit that translates optimization decision commands into physical actions.

[0093] Logic controller: Receives safety instructions and coordinates the actions of each actuator.

[0094] Alarm device: Issues an audible and visual alarm when SRAgent triggers a risk warning.

[0095] Balance block motor: Receives commands and precisely controls the position of the balance block to optimize the balance of the pumping unit.

[0096] Variable frequency controller: An actuator used to control the stroke rate.

[0097] In this embodiment, the beam pumping unit is controlled based on equipment control commands to obtain new equipment operating status data after the control of the beam pumping unit; a reward function is calculated using the new equipment operating status data; and based on the reward function and historical equipment operating status data, a reinforcement learning algorithm is used to perform reinforcement learning and update the decision network parameters in the optimization decision agent of the agent cluster.

[0098] The reinforcement learning iterations in this application aim to achieve adaptive optimization of control parameters (such as stroke rate and stroke length) to continuously improve the system's energy efficiency, output, and reliability. The formula for the reward function R is as follows:

[0099] ;

[0100] in, , , These are adjustable weighting coefficients, used to quantify the relative importance of energy saving, increased production, and reduced failure rate in the optimization objective. Weighting coefficients , , The settings are dynamically configured by maintenance personnel based on the oilfield development stage, such as: production increase period. =0.2, =0.5, =0.3, stable production period =0.3, =0.2, =0.5.

[0101] Energy saving: The actual reduction in energy consumption compared to the baseline operating conditions (initial parameters or standard operating conditions).

[0102] Production increase: The actual increase in liquid / oil production compared to the baseline operating conditions (initial parameters or standard operating conditions).

[0103] Failure rate: The frequency of downtime or reduced production caused by system-related failures per unit of time.

[0104] The learning algorithm used in this application employs reinforcement learning algorithms such as PPO (Proximal Policy Optimization). The update mechanism involves updating the parameters of the decision network (optimizing the policy part of the decision agent) weekly based on historical operating data and the reward function, enabling the system to automatically learn and approximate the optimal control policy.

[0105] Furthermore, the LLM decision-making hub in this application can also achieve the same purpose using a knowledge graph inference engine, replacing LLM generation with a predefined rule chain; the five-model voting mechanism in this application can also be replaced with three-model voting + hardware lock, triggering a physical security lock when all three votes are in agreement; the reinforcement learning dynamic weights can also be replaced with expert rule base weight adjustments, manually set according to the development stage. , , .

[0106] The equipment control system structure of the beam pumping unit in this application is as follows: Figure 2 As shown, it includes a master control LLM decision-making center, an agent cluster, security control, an execution toolkit, a sensor network, and reinforcement learning iteration. This application constructs a multi-agent cooperative control architecture, through ViAgent / FDAgent / ODAge

[0107] Specialized division of labor and LLM scheduling logic of nt / SRAgent resolve multi-objective optimization conflicts; dynamic template selection based on risk identifiers and injection of fault history database summaries make LLM instructions conform to the physical laws of oilfields, such as Hooke's Law constraints; heterogeneous model combination (≥3 architectures) + manual review triggering rules (≤3 votes pass) eliminate industrial control safety hazards; mapping oilfield development stages (production increase period / stable production period) to reward function weight coefficient adjustment rules achieves adaptive optimization.

[0108] The specific process for implementing the equipment control of the beam pumping unit in this application is as follows: Figure 3 As shown, taking parameter optimization under normal operating conditions as an example, the process is as follows:

[0109] Initial state. Pumping unit stroke rate: 5 strokes / minute; stroke: 3.2m; counterweight position: midpoint.

[0110] Real-time sensor data: current fluctuation rate 8.7% (electrical parameter module), walking beam swing angle deviation ±1.5° (ViAgent), bearing vibration spectrum abnormal energy percentage 0.6% (FDAgent).

[0111] Decision-making process. The main control LLM receives the state vector, triggers the Prompt engine to generate optimization instructions; calls the optimization decision agent (ODAgent), and calculates the optimal parameter combination based on historical energy efficiency data; outputs instructions: increase the stroke rate to 5.8 times / minute, and move the balance block 12cm towards the motor side.

[0112] Safety verification and execution. Five-model voting results: all five votes passed unanimously; the logic controller drives the balance block motor to shift, and the frequency converter adjusts the stroke rate; the effect: energy consumption is reduced by 18%, liquid production is increased by 10%, and abnormal vibration disappears.

[0113] Taking real-time intervention for sudden failures as an example, the specific process is as follows:

[0114] Risk scenario: The load sensor detects a sudden increase of 120% in the load at the suspension point (exceeding the safety threshold), triggering a level 3 risk warning from SRAgent.

[0115] Collaborative decision-making. The LLM central system synchronously calls FDAgent to analyze the vibration spectrum and confirm the risk of gearbox bearing breakage; it generates an instruction: immediately reduce the stroke rate to a safe value of 2 strokes / minute and activate the audible and visual alarm.

[0116] The security mechanism is in effect. In the five-model vote, 3 votes supported implementation, and 2 votes suggested further verification → triggering manual review; after confirming the fault, the engineer manually authorized frequency reduction to prevent serious equipment damage.

[0117] Current technologies are unable to achieve multi-source data fusion analysis, multi-objective collaborative optimization, and secure real-time decision-making. Specific limitations and risks of existing technologies are detailed below. Figure 4 As shown, traditional systems independently analyze current, visual, vibration, and load data (data resolution rate <40%). This invention requires real-time fusion of multimodal sensing networks (target resolution rate ≥90%). Existing solutions can only optimize single objectives (such as energy efficiency or fault diagnosis), while this invention requires simultaneous optimization of three major objectives: energy efficiency, fault prevention, and safety monitoring. This application constructs a multimodal sensing → decision-making → execution closed loop to achieve a data fusion utilization rate ≥90%. It designs a multi-agent collaborative architecture to simultaneously optimize energy saving / output improvement / fault reduction. It develops a five-model voting verification mechanism to reduce the instruction error rate to <1%.

[0118] This application overcomes the data silo effect of current, visual, vibration, and load data by constructing a multimodal sensing network, increasing the well site data analysis utilization rate from less than 40% to over 92% (based on actual measurements of 2TB / day data). Through an LLM decision-making center, cross-modal correlation analysis is achieved, enabling decision-making for motor faults, balance block position adjustments, and stroke adjustments. Under the same operating conditions, energy savings are increased by 10-20%, production is increased by 8-12%, and the failure rate is reduced by 30-40%. The five-model voting mechanism completely eliminates the LLM "illusion" risk, with the probability of critical command errors approaching zero, avoiding unplanned downtime. The entire process from data acquisition to command execution has a latency of ≤2 seconds, supporting second-level risk intervention.

[0119] In this embodiment, the operating status data of the beam pumping unit is acquired. This data includes motor parameters, dynamic changes in the beam swing angle, vibration parameters, and suspension point loads of the pumping unit. The operating status data is input into a preset large-scale language model to output structured decision commands. This preset large-scale language model is an optimized version of an initial large-scale language model using a low-rank adaptation method. An intelligent agent cluster analyzes and processes the structured decision commands to obtain initial equipment control commands. This cluster includes a visual intelligent agent, a fault diagnosis intelligent agent, an optimization decision intelligent agent, and a safety monitoring intelligent agent. A preset artificial intelligence model is used to evaluate the safety and rationality of the initial equipment control commands. If the safety and rationality evaluation is successful, the initial equipment control commands are converted into equipment control commands to control the equipment to perform corresponding actions. The beam pumping unit is then controlled based on these equipment control commands. This application acquires the equipment operating status data of a beam pumping unit, including motor parameters, dynamic changes in the beam swing angle, vibration parameters, and suspension point loads of the pumping unit. It overcomes the data silos of current, visual, vibration, and load data, achieving deep fusion of multi-source data and improving data fusion utilization. The equipment operating status data is input into a pre-set large-scale language model to output structured decision commands, enabling cross-modal correlation analysis. An intelligent agent cluster is used to analyze and process the structured decision commands to obtain initial equipment control commands, eliminating potential safety hazards in industrial control and resolving control command errors caused by the "illusion" of the large-scale language model, thus avoiding unplanned downtime. A pre-set artificial intelligence model is used to evaluate the safety and rationality of the initial equipment control commands. If the safety and rationality evaluation is passed, the initial equipment control commands are converted into equipment control commands to control the equipment to perform corresponding actions. Based on these equipment control commands, the beam pumping unit is controlled, collaboratively handling multi-objective optimization such as energy efficiency, fault tolerance, and safety, improving the accuracy of equipment control commands, and automating the control of the beam pumping unit.

[0120] See Figure 5 As shown, this embodiment of the invention discloses a device control system for a beam pumping unit, which may specifically include:

[0121] Data acquisition module 11 is used to acquire equipment operating status data of the beam pumping unit; the equipment operating status data includes motor parameters, dynamic change image of beam swing angle, vibration parameters, and suspension point load of the pumping unit;

[0122] The instruction output module 12 is used to input the device operating status data into a preset large language model to output structured decision instructions; the preset large language model is a model optimized by using a low-rank adaptation method to optimize an initial large language model.

[0123] Analysis and processing module 13 is used to analyze and process the structured decision instructions using an intelligent agent cluster to obtain initial equipment control instructions; the intelligent agent cluster includes a visual intelligent agent, a fault diagnosis intelligent agent, an optimization decision intelligent agent, and a safety supervision intelligent agent.

[0124] Evaluation module 14 is used to evaluate the safety and rationality of the initial device control commands using a preset artificial intelligence model;

[0125] The equipment control module 15 is used to convert the initial equipment control command into an equipment control command for controlling the equipment to perform corresponding actions if the safety and rationality assessment is passed, and to control the equipment of the beam pumping unit based on the equipment control command.

[0126] In some specific embodiments, the data acquisition module 11 may specifically include:

[0127] The motor parameter acquisition module is used to acquire the motor parameters of the beam pumping unit; the motor parameters include the three-phase voltage, current, and power factor of the motor.

[0128] The image acquisition module is used to acquire images of the dynamic changes in the swing angle of the walking beam using a camera that meets a preset frame rate;

[0129] An acceleration or velocity spectrum monitoring module is used to monitor vibration acceleration or velocity spectrum using vibration sensors.

[0130] The suspension point load measurement module is used to measure the suspension point load of the oil pumping unit using a load sensor.

[0131] In some specific embodiments, the instruction output module 12 may specifically include:

[0132] The model optimization module is used to optimize the initial large language model by using professional literature on oil well engineering and labeled well condition data, and adopting the low-rank adaptation method to obtain the preset large language model.

[0133] The data processing module is used to input the device operating status data into the preset large language model, so that the preset large language model can filter out the target template that matches the device operating status data from the preset decision template, fill the target template with the device operating status data, filter out historical data that matches the filled target template from the historical fault database, and fill the fields of the historical data with the filled target template to generate and output structured decision instructions.

[0134] In some specific embodiments, the analysis and processing module 13 may specifically include:

[0135] The visual feature recognition module is used to perform visual feature recognition on the structured decision instructions using visual agents in the agent cluster, and obtain visual feature recognition results.

[0136] The fault component localization and analysis module is used to locate and analyze the fault components using the fault diagnosis intelligent agent in the intelligent agent cluster, and obtain the fault component localization and analysis results.

[0137] The optimization parameter calculation module is used to calculate the optimization parameters of the structured decision instructions by utilizing the optimization decision intelligence in the intelligence agent cluster to obtain the optimization parameters of the oil pumping unit.

[0138] The real-time assessment and risk warning module is used to perform real-time assessment and risk warning on the structured decision instructions by utilizing the security supervision intelligent agent in the intelligent agent cluster, and obtain the real-time assessment and risk warning results.

[0139] The initial equipment control command generation module is used to generate initial equipment control commands based on the visual feature recognition results, the fault component location and analysis results, the oil pumping unit optimization parameters, and the real-time assessment and risk warning results.

[0140] In some specific embodiments, the evaluation module 14 may specifically include:

[0141] A safety and rationality assessment module is used to send the initial device control command to different types of artificial intelligence models, so that the different types of artificial intelligence models can respectively assess the safety and rationality of the initial device control command;

[0142] The quantity determination module is used to determine the quantity that has passed the safety and reasonableness assessment.

[0143] The safety and rationality assessment module is used to indicate that the safety and rationality assessment has passed if the number is greater than a preset threshold.

[0144] The module for failing the safety and reasonableness assessment is used to indicate that the safety and reasonableness assessment has failed if the number of failures is not greater than a preset threshold. The module then sends a manual review prompt to the client so that the client can perform a manual review process based on the manual review prompt.

[0145] In some specific embodiments, the device control module 15 may specifically include:

[0146] The adjustment command conversion module is used to convert the initial equipment control command into an adjustment command for controlling the equipment to perform the adjustment stroke using the frequency converter;

[0147] The risk warning command conversion module is used to convert initial equipment control commands into risk warning commands for real-time risk warning of the equipment using an alarm device.

[0148] The balance command conversion module is used to convert initial equipment control commands into balance commands for controlling the balance of the beam pumping unit using the balance block motor.

[0149] The stroke adjustment, real-time risk warning, and balance control module is used to adjust the stroke, provide real-time risk warning, and control the balance of the beam pumping unit based on the adjustment command, the risk warning command, and the balance command.

[0150] In some specific embodiments, the device control module 15 may specifically include:

[0151] The new equipment operation status data acquisition module is used to acquire the new equipment operation status data after the beam pumping unit is controlled.

[0152] The reward function calculation module is used to calculate the reward function using the operating status data of the new device.

[0153] The reinforcement learning and update module is used to perform reinforcement learning and update the decision network parameters of the optimization decision agents in the agent cluster based on the reward function and historical device operating status data, and using reinforcement learning algorithms.

[0154] Figure 6 This is a schematic diagram of an electronic device provided in an embodiment of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the device control method for a beam pumping unit executed by the electronic device disclosed in any of the foregoing embodiments.

[0155] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0156] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored on it include operating system 221, computer program 222 and data 223, etc., and the storage method can be temporary storage or permanent storage.

[0157] The operating system 221 manages and controls the various hardware devices on the electronic device 20 and the computer program 222 to enable the processor 21 to perform calculations and processing on the data 223 in the memory 22. It can be Windows, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the device control method of the beam pumping unit executed by the electronic device 20 as disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the device control equipment of the beam pumping unit from external devices, and may also include data collected by its own input / output interface 25.

[0158] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0159] Furthermore, this application also discloses a computer-readable storage medium storing a computer program. When the computer program is loaded and executed by a processor, it implements the equipment control method steps of the beam pumping unit disclosed in any of the foregoing embodiments.

[0160] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0161] The above provides a detailed description of the equipment control method, apparatus, equipment, and storage medium for a beam pumping unit provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for controlling a beam pumping unit, characterized in that, include: Acquire equipment operating status data for beam pumping units; The equipment operating status data includes motor parameters, dynamic change images of the walking beam swing angle, vibration parameters, and pumping unit suspension point load; The device operating status data is input into a preset large language model to output structured decision instructions; the preset large language model is a model optimized from an initial large language model using a low-rank adaptation method. The structured decision instructions are analyzed and processed using an intelligent agent cluster to obtain initial equipment control instructions; the intelligent agent cluster includes a visual intelligent agent, a fault diagnosis intelligent agent, an optimization decision intelligent agent, and a safety monitoring intelligent agent. The initial device control commands are evaluated for safety and rationality using a pre-defined artificial intelligence model. If the safety and rationality assessment is passed, the initial equipment control command is converted into equipment control command for controlling the equipment to perform corresponding actions, and the equipment of the beam pumping unit is controlled based on the equipment control command.

2. The equipment control method for a beam pumping unit according to claim 1, characterized in that, The acquisition of equipment operating status data for the beam pumping unit includes: Obtain the motor parameters of the beam pumping unit; the motor parameters include the three-phase voltage, current, and power factor of the motor. The dynamic changes in the swing angle of the walking beam are captured using a camera that meets the preset frame rate; Vibration sensors are used to monitor vibration acceleration or velocity spectrum; The load at the suspension point of the oil pumping unit is measured using a load sensor.

3. The equipment control method for a beam pumping unit according to claim 1, characterized in that, The step of inputting the device operating status data into a preset large-scale language model to output structured decision instructions includes: Using professional literature on oil well engineering and labeled well condition data, and employing a low-rank adaptation method to optimize the initial large language model, a pre-defined large language model is obtained. The device operating status data is input into the preset large language model, so that the preset large language model can filter out target templates that match the device operating status data from preset decision templates, fill the target templates with the device operating status data, filter out historical data that matches the filled target templates from the historical fault database, and fill the fields of the historical data with the filled target templates to generate and output structured decision instructions.

4. The equipment control method for a beam pumping unit according to claim 1, characterized in that, The process of analyzing and processing the structured decision instructions using an intelligent agent cluster to obtain initial device control instructions includes: The structured decision instructions are visually identified using a visual agent in the agent cluster to obtain visual feature recognition results. The fault diagnosis agent in the agent cluster is used to locate and analyze the faulty components of the structured decision instructions, and the faulty component location and analysis results are obtained. The optimization decision-making agent in the agent cluster is used to calculate the optimization parameters of the structured decision instructions to obtain the oil pumping unit optimization parameters; The structured decision instructions are evaluated and risk warnings are given in real time by using the security supervision intelligent agent in the intelligent agent cluster, and the real-time evaluation and risk warning results are obtained. Initial equipment control commands are generated based on the visual feature recognition results, the fault component location and analysis results, the oil pumping unit optimization parameters, and the real-time assessment and risk warning results.

5. The equipment control method for a beam pumping unit according to claim 1, characterized in that, The process of using a pre-set artificial intelligence model to assess the safety and rationality of the initial device control commands includes: The initial device control command is sent to different types of artificial intelligence models so that the different types of artificial intelligence models can respectively evaluate the safety and rationality of the initial device control command; Determine the number of cases that pass the safety and reasonableness assessment; If the quantity is greater than a preset threshold, it indicates that the safety and rationality assessment has passed; If the number is not greater than a preset threshold, it indicates that the security and rationality assessment has failed, and a manual review prompt message will be sent to the client so that the client can perform the manual review process based on the manual review prompt message.

6. The equipment control method for a beam pumping unit according to claim 1, characterized in that, The step of converting the initial equipment control command into equipment control commands for controlling the equipment to perform corresponding actions, and controlling the beam pumping unit based on the equipment control commands, includes: The initial equipment control commands are converted into adjustment commands for controlling the equipment to perform adjustment strokes using a frequency converter; The alarm device is used to convert the initial equipment control commands into risk warning commands for real-time risk warning of the equipment. The initial equipment control commands are converted into balance commands for controlling the balance of the beam pumping unit using a balance block motor. Based on the adjustment command, the risk warning command, and the balance command, the equipment of the beam pumping unit is adjusted for stroke, real-time risk warning, and balance control.

7. The equipment control method for a beam pumping unit according to any one of claims 1 to 6, characterized in that, After controlling the beam pumping unit based on the equipment control commands, the method further includes: Acquire new equipment operating status data after controlling the beam pumping unit; The reward function is calculated using the operating status data of the new equipment; Based on the reward function and historical device operating status data, and using reinforcement learning algorithms, the decision network parameters of the optimization decision agents in the agent cluster are reinforced and updated.

8. A control device for a beam pumping unit, characterized in that, include: The data acquisition module is used to acquire equipment operating status data of the beam pumping unit; The equipment operating status data includes motor parameters, dynamic change images of the walking beam swing angle, vibration parameters, and pumping unit suspension point load; The instruction output module is used to input the device operating status data into a preset large language model to output structured decision instructions; the preset large language model is a model optimized by using a low-rank adaptation method to optimize an initial large language model; The analysis and processing module is used to analyze and process the structured decision instructions using an intelligent agent cluster to obtain initial equipment control instructions; the intelligent agent cluster includes a visual intelligent agent, a fault diagnosis intelligent agent, an optimization decision intelligent agent, and a safety supervision intelligent agent. The evaluation module is used to evaluate the safety and rationality of the initial device control commands using a preset artificial intelligence model; The equipment control module is used to convert the initial equipment control command into an equipment control command for controlling the equipment to perform corresponding actions if the safety and rationality assessment is passed, and to control the equipment of the beam pumping unit based on the equipment control command.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the equipment control method for a beam pumping unit as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer programs; wherein, when the computer programs are executed by a processor, they implement the equipment control method for a beam pumping unit as described in any one of claims 1 to 7.

Images