Efficient compressed air control method and system based on native agent

Abstract

This invention discloses a highly efficient compressed air control method and system based on native intelligent agents. The invention constructs a three-layer architecture consisting of native intelligent agents, a collaborative control layer, and a system scheduling layer. Each native intelligent agent is deployed adjacent to the physical device, and incorporates a device physics knowledge model and a local optimization objective function to achieve local perception and decision-making. The collaborative control layer handles information sharing and conflict resolution. The system scheduling layer performs global optimization with the goal of minimizing total system energy consumption, generating coordinated commands. Through the synergy of local decision-making and global optimization, this invention achieves decentralized intelligent control of the compressed air system, effectively improving energy efficiency, response speed, and system scalability.

Description

Technical Field

[0001] This invention relates to the fields of industrial automation and energy management technology, and in particular to a highly efficient compressed air control method and system based on native intelligent agents. Background Technology

[0002] Compressed air is a crucial power source for industrial production, typically accounting for 10% to 30% of a factory's total electricity consumption. Traditional compressed air systems often employ centralized PLC control or stand-alone local control, which suffers from the following drawbacks: 1) Insufficient global energy efficiency optimization capabilities; independent operation of subsystems such as air compressors, dryers, and pipelines leads to redundant work due to a lack of coordination; 2) Lagging dynamic response; slow adjustment when facing sudden load changes, resulting in pressure fluctuations or insufficient air supply; 3) High safety risks; lack of linkage early warning and rapid intervention mechanisms for abnormal states such as dew point exceeding standards and pipeline leaks; 4) Poor scalability; adding new equipment requires reprogramming the central controller.

[0003] In recent years, multi-agent systems have shown promising applications in industrial control. However, existing agent architectures mostly rely on cloud or central servers for decision-making, resulting in problems such as large communication latency and high risk of single points of failure. In addition, existing solutions do not fully integrate the physical model of the equipment with real-time operating conditions, making it difficult to achieve "native intelligence" where the equipment itself possesses closed-loop capabilities of perception, reasoning, and execution. Summary of the Invention

[0004] To address the problems of poor global energy efficiency, slow dynamic response, high safety risks, and poor scalability in existing compressed air control systems, this invention proposes a highly efficient compressed air control method and system based on native intelligent agents.

[0005] The specific technical solution is as follows: A highly efficient compressed air control method based on native intelligent agents, applied to a system including native intelligent agents, a cooperative control layer, and a system scheduling layer, wherein the cooperative control layer is communicatively connected to each native intelligent agent, and the system scheduling layer is communicatively connected to the cooperative control layer, the method comprising:

[0006] Perception step: Each native intelligent agent collects the operating parameters of the corresponding physical device via local sensors;

[0007] Local decision-making step: Each native intelligent agent generates a local preliminary control strategy based on the operating parameters, the built-in device physical knowledge model, and the local optimization objective function;

[0008] Information interaction and collaboration steps: Each native intelligent agent uploads its own state information and intention strategy to the collaborative control layer through a preset communication mechanism. The collaborative control layer then performs information sharing and conflict resolution to generate collaborative decision information.

[0009] Global optimization and execution steps: The system scheduling layer receives the gas supply requirements and safety constraints, and combines them with the collaborative decision information to perform global optimization with the goal of minimizing the total energy consumption of the system, generating a coordination instruction; each native intelligent agent integrates the coordination instruction with the local preliminary control strategy to generate the final execution signal.

[0010] Furthermore, the native intelligent agent includes at least: an air compressor intelligent agent deployed on the air compressor, a drying and post-processing system intelligent agent deployed on the drying and post-processing system, a cooling system intelligent agent deployed on the cooling system, a pipeline intelligent agent deployed on the compressed air pipeline network, and a smart valve intelligent agent deployed on the smart valve.

[0011] Furthermore, in the global optimization and execution steps, the global optimization employs model predictive control to solve a preset optimization problem, the objective function of which is:

[0012] ;

[0013] Where i is the device index, and the summation iterates through all devices; ω1, ω2, and ω3 are the energy consumption weighting coefficient, pressure weighting coefficient, and dew point weighting coefficient, respectively; T h For the prediction time domain; u(t) is the control variable; P w,i P(t) represents the electrical power of the i-th device; P(t) represents the system pressure; P target For target pressure; T d (t) represents the dew point temperature; T d,set The target dew point temperature.

[0014] Furthermore, it also includes an adaptive adjustment step for the weighting coefficients:

[0015] The energy consumption weighting coefficient ω1 is dynamically adjusted based on the real-time electricity price, grid demand response command, and the deviation between the system's unit electricity consumption and the historical best unit electricity consumption.

[0016] The pressure weighting coefficient ω2 is dynamically adjusted in steady-state mode, transient mode, and safety boundary mode based on pressure deviation, pressure fluctuation rate, load change rate, and production stage identifier.

[0017] The dew point weighting coefficient ω3 is dynamically adjusted based on production stage identification, product quality feedback, dew point deviation, ambient temperature and humidity, and desiccant health.

[0018] Specifically, when the safety boundary conditions are triggered, the weight values ​​of the pressure weight coefficient ω2 or the dew point weight coefficient ω3 are increased so that the optimization objective prioritizes meeting the safety constraints.

[0019] Furthermore, in the information interaction and collaboration steps, the conflict resolution specifically includes: in response to the conflict between the intention strategies of different native intelligent agents, the collaborative control layer resolves the conflict according to the preset security priority rules, generates a resolution result, and inputs the resolution result as a constraint condition to the system scheduling layer.

[0020] Furthermore, it also includes anomaly handling and fault tolerance procedures:

[0021] When a communication interruption occurs, the native intelligent agent automatically switches to local self-consistent mode and executes built-in control logic;

[0022] When a device failure occurs, the collaborative control layer marks the failed device as unavailable and triggers the device reconfiguration algorithm;

[0023] When a system-level risk occurs, the system scheduling layer switches to a safe mode, bypassing global optimization and directly executing preset emergency instructions.

[0024] A high-efficiency compressed air control system based on native intelligent agents, comprising:

[0025] Multiple native intelligent agents, each of which has a built-in device physical knowledge model and a local optimization objective function, and is configured to: collect the operating parameters of the corresponding physical device via local sensors, generate a local preliminary control strategy based on the operating parameters, the built-in device physical knowledge model and the local optimization objective function, and, in response to a received coordination instruction, fuse the coordination instruction with the local preliminary control strategy to generate a final execution signal;

[0026] The collaborative control layer communicates with each of the native intelligent agents and is configured to: receive the self-state information and intention strategies uploaded by each of the native intelligent agents, perform information sharing and conflict resolution, and generate collaborative decision information;

[0027] The system scheduling layer is communicatively connected to the collaborative control layer and each of the native intelligent agents, and is configured to: receive gas supply requirements and safety constraints, combine the collaborative decision information, perform global optimization with the goal of minimizing the total energy consumption of the system, generate the coordination instructions, and send them to each of the native intelligent agents.

[0028] Furthermore, the plurality of native intelligent agents include:

[0029] The air compressor intelligent agent is a hardware carrier that is an embedded controller integrated in the air compressor control cabinet or an external edge computing gateway, which is deployed in or next to the air compressor control cabinet.

[0030] The intelligent agent of the drying post-processing system is a hardware carrier that is an embedded computer or programmable logic controller integrated in the control cabinet of the dryer, and is deployed in or next to the control cabinet of the dryer.

[0031] The cooling system intelligent agent, wherein the hardware carrier of the cooling system intelligent agent is a communication coprocessor or an open programmable logic controller integrated in the cooling system control cabinet, and is deployed in or next to the cooling system control cabinet;

[0032] The pipeline intelligent agent is a hardware carrier that is an industrial Internet of Things switch deployed in the pipeline network area and installed in the field junction box or control box in the pipeline network area.

[0033] The intelligent valve agent is a dedicated control module integrated into the valve actuator, which is installed on the valve actuator.

[0034] Furthermore, the system scheduling layer is specifically configured to run model predictive control at a preset cycle, and use the optimization results of the previous cycle as the initial guess for solving the current cycle.

[0035] Furthermore, the system also includes a clock synchronization module, which is used to control the clock deviation of each module in the system within the millisecond range through the IEEE 1588 precision clock protocol, so that the system can operate synchronously with a fixed master control cycle.

[0036] The above technical solution has the following advantages or technical effects:

[0037] 1. This invention achieves the organic integration of local device decision-making and global optimization by constructing a three-layer distributed architecture of "native intelligent agent - collaborative control layer - system scheduling layer", which solves the problems of slow response and poor global energy efficiency of traditional centralized control, and significantly improves the dynamic response speed and overall energy efficiency of compressed air system.

[0038] 2. This invention incorporates energy consumption, pressure stability, and air quality into a unified optimization framework through model predictive control and adaptive adjustment of weight coefficients. It can dynamically respond to changes in electricity prices, fluctuations in production conditions, and safety risks, thereby achieving multi-objective dynamic balance and maximizing energy-saving benefits.

[0039] 3. This invention ensures the robustness and operational continuity of the system under communication interruptions, equipment failures, and system-level risks through hierarchical anomaly handling, safety fault tolerance mechanisms, and high-precision clock synchronization, thereby achieving high reliability and collaborative consistency of the distributed intelligent control system.

[0040] 4. This invention, through modular and standardized intelligent agent hardware design, is compatible with the transformation needs of new intelligent devices and traditional devices, and realizes edge deployment of "data processing nearby and decision generation locally". At the same time, the connection of new devices only requires the deployment of the corresponding intelligent agent without modifying the central program, and has good scalability and engineering adaptability. Attached Figure Description

[0041] Figure 1 This is a flowchart of the method of the present invention;

[0042] Figure 2 This is a flowchart of the decision logic of the ω1 adaptive mechanism of the present invention;

[0043] Figure 3 This is a flowchart of the decision logic of the ω2 adaptive mechanism of the present invention;

[0044] Figure 4 This is a flowchart of the decision logic of the ω3 adaptive mechanism of the present invention;

[0045] Figure 5 This is a system structure block diagram of the present invention. Detailed Implementation

[0046] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0047] Example 1

[0048] like Figure 1 As shown, a highly efficient compressed air control method based on native intelligent agents is proposed and applied to a distributed intelligent control system. The system includes multiple native intelligent agents, a collaborative control layer communicating with the native intelligent agents, and a system scheduling layer communicating with the collaborative control layer. Each native intelligent agent is deployed adjacent to its corresponding physical device, possesses a built-in device physical knowledge model and local optimization objective function, and has autonomous capabilities for local perception, local decision-making, and local execution. The collaborative control layer, acting as an information interaction platform, is responsible for receiving status information and intended strategies uploaded by each native intelligent agent, and for information sharing and conflict resolution. The system scheduling layer, acting as the global optimization hub, is responsible for receiving air supply requirements and safety constraints, combining collaborative decision information to perform global optimization, generating coordination instructions, and issuing them to each native intelligent agent.

[0049] Specifically, the method includes the following steps:

[0050] Step S1: System initialization and parameter configuration.

[0051] First, system initialization is performed. Each native agent pre-configures its built-in model parameters based on factory test calibration data and performs model correction through self-learning during the initial stage of system operation. During model verification, each agent compares its model output with actual sensor readings to ensure model reliability. Simultaneously, the system completes communication networking and topology identification according to preset configurations. Pre-defined communication protocols, such as OPC UA or MQTT, are used to connect the native agents and between the agents and the collaborative control layer and system scheduling layer. If OPC UA is used, a unified information model is defined, creating standard node structures for object types such as "air compressor" and "smart valve" to ensure interoperability. If MQTT is used, the naming conventions for topics and the data format of messages are specified. Furthermore, all safety constraints are pre-configured in the system scheduling module and each agent.

[0052] Step S2: Perception Phase.

[0053] Each native intelligent agent continuously collects operating parameters of its corresponding physical equipment through local sensors. Specifically, the sensor network is deployed on key physical equipment such as air compressors, dryers, compressed air pipelines, smart valves, and cooling systems to collect operating parameters in real time, including pressure (bar), flow rate (m³ / min), temperature (°C), relative humidity (%), electrical power (kW), rotational speed (rpm), dew point temperature (°C), current (A), and vibration. The data sampling frequency is no less than 1Hz to ensure that the system has sufficient dynamic response capabilities.

[0054] Step S3: Local decision-making stage.

[0055] Each native agent generates a preliminary local control strategy based on the collected operating parameters, the built-in device physical knowledge model, and the local optimization objective function. Specifically:

[0056] The air compressor intelligent agent has its own efficiency curve, loading and unloading logic, and maintenance history. Its local optimization goal is to operate at the highest efficiency while meeting air demand. The air compressor intelligent agent calls the efficiency map and, under the required pressure, aims to maximize specific power, outputting suggested speed or loading and unloading commands. For a fixed-frequency screw air compressor, it outputs control signals for loading or unloading status; for a variable-frequency screw air compressor, it outputs the inverter's operating frequency or speed command; for a centrifugal air compressor, it outputs the guide vane opening command.

[0057] The intelligent agent of the post-drying treatment system is configured with the relationship between the desiccant regeneration cycle, energy consumption, and dew point. Its local optimization objective is to minimize its own energy consumption while ensuring the quality of compressed air (i.e., meeting the dew point temperature requirements). Based on the adsorbent saturation model and regeneration cycle strategy, the intelligent agent of the post-drying treatment system determines whether to switch the tower or start the heater, and adjusts the relevant actuators accordingly.

[0058] The cooling system intelligent agent is configured with the relationship between heat exchange efficiency and ambient temperature and humidity. Its local optimization goal is to maintain the optimal operating temperature of the air compressor and drying post-treatment system with minimal energy consumption. Based on the heat balance equation, the cooling system intelligent agent adjusts the water pump flow rate or fan speed to ensure that the exhaust temperature does not exceed the safe upper limit.

[0059] The pipeline agent is equipped with the pipeline network topology and pressure loss model, enabling it to assess system pressure distribution in real time, calculate pressure loss, and diagnose potential leaks. Its local optimization objective is to reduce the overall pressure loss of the pipeline network and maintain pressure stability. Based on the Darcy-Weisbach formula or a measured pressure loss model, the pipeline agent evaluates whether the current pressure loss is reasonable and proposes flow rate adjustment suggestions.

[0060] The intelligent valve agent possesses precise flow regulation capabilities and understands the gas consumption characteristics of its subordinate branches. Its local optimization goal is to accurately meet the end-point demand and assist in balancing pipeline pressure. Based on the flow characteristic curve, the intelligent valve agent calculates the optimal valve opening required to maintain the target flow rate and outputs a control signal.

[0061] Step S4: Information Interaction and Collaboration Stage.

[0062] Each native agent uploads its own state information and intended strategy to the collaborative control layer through a pre-defined communication mechanism. Specifically, each native agent generates a standardized "information element," which is a structured data object containing the following: current state (e.g., "Air compressor A is in a loaded state with an efficiency of 82%"), resource constraints (e.g., "Maximum output pressure is 10 bar"), intended strategy (e.g., "It is recommended to increase the speed by 5%), and abnormal alarms (e.g., "Abnormal pressure drop detected in pipe section B, suspected leak").

[0063] The collaborative control layer employs a publish-subscribe mechanism, allowing native agents to subscribe to relevant topics on demand. When an agent publishes event information, other agents subscribed to that topic can automatically respond. For example, when the pipeline agent publishes a "high-pressure demand" event, the air compressor agent automatically responds and prepares to increase its output.

[0064] The collaborative control layer also performs collaborative conflict detection and resolution. When there is a conflict between the intentional strategies of different native agents, the collaborative control layer first makes a preliminary judgment based on preset rules. For example, when the pipeline agent requests to increase the main pipe pressure, but the air compressor agent reports that its optimal efficiency point is not in this range, which would lead to a decrease in efficiency, the collaborative control layer makes a judgment based on the "safety first" rule and inputs the judgment result as a constraint condition to the system scheduling layer to prevent the optimization solver from receiving excessively contradictory inputs and failing to converge. Through the above information sharing and conflict resolution, the collaborative control layer generates collaborative decision information.

[0065] Step S5: Global optimization and execution phase.

[0066] The system scheduling layer receives gas supply requirements and safety constraints from the upper system, and combines them with the collaborative decision information generated by the collaborative control layer to perform global optimization with the goal of minimizing the total energy consumption of the system, generating coordination instructions for each native intelligent agent.

[0067] The system scheduling layer uses model predictive control to solve a predefined optimization problem, the objective function of which is:

[0068] ;

[0069] Where i is the device index, and the summation iterates through all devices; ω1, ω2, and ω3 are the energy consumption weighting coefficient, pressure weighting coefficient, and dew point weighting coefficient, respectively; T h For the prediction time domain; u(t) is the control variable; P w,i P(t) represents the electrical power of the i-th device; P(t) represents the system pressure; P target For target pressure; T d (t) represents the dew point temperature; T d,set The target dew point temperature.

[0070] The model predicts and controls the operation periodically, with a solution cycle of 30 seconds. It uses the optimization results from the previous cycle as the initial guess for the current cycle to accelerate convergence. The optimization results are decomposed into task instruction packages for each native agent, including the target value, tolerance range, and priority.

[0071] In addition to model predictive control, the system scheduling layer is also equipped with alternative optimizers to ensure the real-time performance and reliability of the system under different operating conditions. When model predictive control timeouts or fails, the system scheduling layer automatically switches to the alternative optimizer, such as a rule engine or heuristic algorithm, to execute a rule-based coordination strategy, including but not limited to distributing the load evenly or according to priority. The setting of alternative optimizers ensures that the system can maintain stable control output even under complex operating conditions or abnormal situations.

[0072] During the global optimization process, the system scheduling layer also dynamically adjusts the three weight coefficients in the objective function to adapt to the multi-objective and dynamically changing environment. The specific adaptive adjustment mechanism is as follows:

[0073] like Figure 2 As shown, the energy consumption weighting coefficient ω1 is dynamically adjusted based on real-time electricity prices, grid demand response commands, the deviation between system unit electricity consumption and historical best unit electricity consumption, and the deviation between the operating equipment load rate and the best load rate. When real-time electricity prices are high or the grid issues demand response commands, ω1 is increased to strengthen energy-saving targets; when system unit electricity consumption is high, ω1 is also increased to trigger more aggressive energy-saving optimization.

[0074] like Figure 3 As shown, the pressure weighting coefficient ω2 is dynamically adjusted based on pressure deviation, pressure fluctuation rate, load change rate, and production stage indicators in steady-state, transient, and safety boundary modes. In steady-state mode, the system operates smoothly, and ω2 is appropriately reduced to prioritize energy efficiency. In transient mode, the system experiences significant disturbances, and ω2 is rapidly increased to prioritize pressure stabilization. In safety boundary mode, the pressure approaches the safety limit, and ω2 is assigned its maximum value to force pressure recovery.

[0075] like Figure 4 As shown, the dew point weighting coefficient ω3 is dynamically adjusted based on production stage indicators, product quality feedback, dew point deviation, ambient temperature and humidity, and desiccant health. In production stages with stringent air quality requirements, such as spraying and food packaging, ω3 is set to its maximum value; when the dew point deviation is large or the ambient humidity is high, ω3 is actively increased to enhance the robustness of drying control.

[0076] The adaptive adjustment processes of the three weighting coefficients mentioned above are performed in parallel. The system has a weight arbitrator that follows the principle of absolute safety priority: when safety boundary conditions such as severe pressure deviation or dew point exceeding the standard are triggered, causing the pressure weighting coefficient or dew point weighting coefficient to enter the safety boundary mode and reach its maximum value, the arbitrator temporarily reduces the weight of the energy consumption weighting coefficient to ensure that the system optimization objective focuses on safety and stability, and temporarily suspends energy efficiency optimization. When there is no safety risk, the arbitrator performs weighted normalization on the three weighting coefficients to ensure that the sum of the three weighting coefficients is 1, and then sends the final weight set to the model predictive control optimizer. Through the above weight arbitration and normalization mechanism, the system achieves intelligent coordination between safety priority and energy efficiency balance under multi-objective optimization.

[0077] When safety boundary conditions are triggered (such as severe pressure deviation or dew point exceeding the standard), causing ω2 or ω3 to enter the safety boundary mode and reach its maximum value, the system scheduling layer temporarily reduces the weight of ω1 to ensure that the optimization objective focuses on safety and stability, and temporarily suspends energy efficiency optimization.

[0078] After receiving coordination instructions from the system scheduling layer, each native agent merges them with its own locally generated preliminary control strategy. The merging method can employ weighted averaging or rule-based arbitration. Finally, each native agent generates a final execution signal to control the corresponding physical equipment, including controlling the air compressor frequency converter, adjusting the opening of electric valves, starting and stopping the cooling water pump, and switching the drying tower.

[0079] Step S6: Feedback and closed loop.

[0080] Each native agent continuously provides feedback on its actual operating status, forming a complete closed loop of "perception-decision-execution-feedback," ensuring the system stably converges to the optimal operating point. The entire system operates on a fixed master control cycle. At the beginning of each cycle, each native agent synchronously executes the perception step; local decision-making and information exchange are completed in the early part of the cycle; the system scheduling module completes global optimization calculations in the middle of the cycle; and each native agent executes instructions at the end of the cycle. The system adopts the IEEE 1588 precision clock protocol, controlling the overall system clock deviation within milliseconds to ensure clock synchronization among all agents and guarantee the consistency of coordinated actions.

[0081] Step S7: Exception handling and fault tolerance.

[0082] The system also features anomaly handling and fault tolerance capabilities to ensure robustness under abnormal conditions. Based on the severity of the anomaly, the system employs a tiered processing strategy:

[0083] Level 1 anomalies are minor anomalies, such as data from a single sensor exceeding a reasonable range but with redundant and reliable data. In this case, the system uses Kalman filtering or multi-sensor data fusion algorithms to repair the data, marks the data quality bits, and reports a minor anomaly log to the system.

[0084] Level 2 anomalies are localized faults, such as device communication interruptions or persistent exceedances of critical parameters. In this case, the collaborative control layer marks the device as "unavailable" within one control cycle and triggers the device reorganization algorithm to recalculate the task allocation for other agents.

[0085] Level 3 anomalies represent system-level risks, such as a drop in main pipe pressure exceeding 15% within 3 seconds. In this case, the system switches to "safe mode" within 500 milliseconds, bypassing the optimizer to directly execute preset emergency commands, such as fully activating backup units, shutting down secondary branches, and triggering the highest-level audible and visual alarm.

[0086] When communication is interrupted, the agent loses contact with the cooperative control layer. The agent automatically switches to local self-consistent mode, executes built-in PID control logic to maintain the current setpoint, and continuously attempts to reconnect. When a timeout or failure of model predictive control solution is detected, the cooperative control layer automatically degrades to a rule-based coordination strategy, such as "average load distribution" or "priority allocation".

[0087] Through the above steps, the method provided by the embodiments of the present invention realizes distributed intelligent control of the compressed air system. Each native intelligent agent has local perception and decision-making capabilities, the collaborative control layer realizes information sharing and conflict resolution, the system scheduling layer completes global optimization, and at the same time, the system's stable operation is ensured through anomaly handling and safety fault tolerance mechanisms.

[0088] Example 2

[0089] Corresponding to Example 1, a high-efficiency compressed air control system based on native intelligent agents is also provided, such as Figure 5 As shown, the method applicable to Example 1 is described. The system includes multiple native intelligent agents, a collaborative control layer, and a system scheduling layer. The system adopts a distributed intelligent control architecture of bottom-up perception, top-down optimization, and intermediate collaborative decision-making.

[0090] Native intelligent agent layer

[0091] The native intelligent agent layer is the core connecting physical devices and upper-level optimization. Multiple native intelligent agents correspond to key physical devices deployed in the compressed air system, including air compressor intelligent agents, drying and after-treatment system intelligent agents, cooling system intelligent agents, pipeline intelligent agents, and smart valve intelligent agents. Each native intelligent agent has a built-in device physics knowledge model and a local optimization objective function, and is configured to perform the following functions: collect operating parameters of the corresponding physical device via local sensors; generate a local preliminary control strategy based on the collected operating parameters, the built-in device physics knowledge model, and the local optimization objective function; and, in response to a received coordination command, fuse the coordination command with the local preliminary control strategy to generate a final execution signal to control the corresponding physical device.

[0092] The hardware carriers of each native intelligent agent are deployed close to their corresponding physical devices, following the principles of edge computing to achieve local data processing and local decision generation. The specific hardware configuration and deployment method are as follows:

[0093] The hardware carrier of the air compressor intelligent agent is an embedded controller integrated into the air compressor control cabinet or an external edge computing gateway, deployed inside or next to the control cabinet. Its core processor can use an ARM Cortex-A53 or higher-performance industrial-grade system-on-a-chip, with a main frequency of no less than 1.2GHz, no less than 1GB of memory, and no less than 4GB of storage for storing operating data, historical maintenance records, and optimization models. Key interfaces include analog and digital input interfaces for acquiring sensor signals from the equipment itself, such as pressure, temperature, current, and power, as well as key signals characterizing the real-time operating status of the air compressor; the communication interface integrates an Ethernet port for uplink communication and retains an RS-485 or Modbus RTU interface for data interaction and control with the air compressor's own frequency converter or programmable logic controller.

[0094] The hardware carrier of the intelligent agent in the post-drying treatment system is an embedded computer or programmable logic controller integrated into the dryer control cabinet, deployed inside or next to the control cabinet. Its key interfaces include analog and digital input interfaces for collecting parameters such as system inlet and outlet pressure and temperature, dew point temperature, tower switching status, fan operating status, compressor operating status, and refrigerant pressure; analog and digital output interfaces for controlling the start / stop, speed, or opening degree of heaters, solenoid valves, compressors, fans, frequency converters, or regulating valves; and communication interfaces for connecting high-precision dew point meters, flow meters, and intelligent drying equipment with communication interfaces. This intelligent agent achieves precise adaptation by automatically identifying or manually configuring the type of dryer connected, loading the corresponding control model and strategy library.

[0095] The hardware carrier of the cooling system intelligent agent is a communication coprocessor or open programmable logic controller integrated within the cooling system control cabinet, deployed inside or next to the control cabinet. Its key functions include an analog input interface for acquiring physical quantities such as water temperature, water flow rate, and ambient temperature and humidity; a communication interface for receiving temperature data and coordination commands from the air compressor and dryer intelligent agents; a digital output interface for controlling the start and stop of the water pump and fan; and an analog output or communication interface for adjusting the operating frequency of the water pump and fan frequency converters to achieve stepless speed regulation.

[0096] The hardware carrier of the pipeline intelligent agent is an industrial IoT gateway deployed in the pipeline network area, installed in the field junction box or control box within the pipeline network area. Its key interfaces include multiple RS-485 interfaces for connecting pressure transmitters and flow meters installed at key nodes in the pipeline network; and an Ethernet interface for data exchange with the intelligent valve agent and the system collaboration layer. This gateway is responsible for polling or receiving data from all affiliated pressure and flow sensors and has a built-in pressure loss calculation and leakage diagnosis model.

[0097] The hardware carrier of the intelligent valve agent is a dedicated control module integrated into the valve actuator. Its key functions include a drive output interface for outputting 4-20mA or 0-10V signals to control the valve actuator and achieve precise opening adjustment; a feedback acquisition interface for acquiring valve opening feedback signals; and a communication interface for network access.

[0098] All the aforementioned native intelligent agents are connected to the same local area network via industrial switches or industrial wireless access points, forming a distributed edge computing cluster. The agents connect to each other through standardized industrial interfaces, possessing industrial-grade characteristics such as wide-temperature operation, high protection levels, and resistance to electromagnetic interference.

[0099] Collaborative Control Layer

[0100] The collaborative control layer communicates with each native agent and is deployed on an industrial control computer or lightweight server within the air compressor station. The collaborative control layer is configured to receive the state information and intention policies uploaded by each native agent, perform information sharing and conflict resolution, and generate collaborative decision information. The collaborative control layer runs an MQTT Broker or OPC UA Server and carries the collaborative decision algorithm. When conflicts exist between the intention policies of different native agents, the collaborative control layer resolves the conflicts according to preset safety priority rules, generates a resolution result, and inputs this resolution result as a constraint to the system scheduling layer.

[0101] System scheduling layer

[0102] The system scheduling layer communicates with the collaborative control layer and each native intelligent agent. It is configured to receive gas supply requirements and safety constraints, combine collaborative decision information, perform global optimization with the goal of minimizing the total energy consumption of the system, generate coordination instructions and send them to each native intelligent agent.

[0103] The system scheduling layer uses model predictive control to solve a predefined optimization problem, the objective function of which is:

[0104] ;

[0105] Where i is the device index, and the summation iterates through all devices; ω1, ω2, and ω3 are the energy consumption weighting coefficient, pressure weighting coefficient, and dew point weighting coefficient, respectively; T h For the prediction time domain; u(t) is the control variable; P w,i P(t) represents the electrical power of the i-th device; P(t) represents the system pressure; P target For target pressure; T d (t) represents the dew point temperature; T d,set The target dew point temperature.

[0106] The system scheduling layer runs model predictive control at a preset cycle, such as solving once every 30 seconds, and uses the optimization results of the previous cycle as the initial guess for the current cycle to accelerate convergence.

[0107] In addition, the system scheduling layer is also equipped with an alternative optimizer. When the model predictive control solution times out or fails, it automatically switches to the rule engine or heuristic algorithm to execute rule-based coordination strategies, such as evenly distributing the load or distributing the load according to priority, to ensure that the system can still maintain stable control output under abnormal conditions.

[0108] The system scheduling layer is also configured to dynamically adjust the three weighting coefficients in the above objective function. The energy consumption weighting coefficient ω1 is dynamically adjusted based on real-time electricity price, grid demand response instructions, and the deviation between the system's unit electricity consumption and the historical best unit electricity consumption; the pressure weighting coefficient ω2 is dynamically adjusted based on pressure deviation, pressure fluctuation rate, load change rate, and production stage indicators, in steady-state mode, transient mode, and safety boundary mode; the dew point weighting coefficient ω3 is dynamically adjusted based on production stage indicators, product quality feedback, dew point deviation, ambient temperature and humidity, and desiccant health.

[0109] The system also has a weighted arbitrator that follows the principle of absolute safety priority: when safety boundary conditions are triggered and cause the pressure weight coefficient or dew point weight coefficient to reach its maximum value, the arbitrator temporarily reduces the weight of the energy consumption weight coefficient; when there is no safety risk, the arbitrator performs weighted normalization on the three weight coefficients to ensure that the sum of the three is 1, and then sends the final weight set to the model predictive control optimizer.

[0110] When safety boundary conditions are triggered, the system scheduling layer increases the weight value of either the pressure weight coefficient ω2 or the dew point weight coefficient ω3 to ensure that the optimization objective prioritizes meeting safety constraints.

[0111] Clock synchronization module

[0112] The system also includes a clock synchronization module, which uses the IEEE 1588 precision clock protocol to control the clock deviation of each module within the system to within milliseconds, enabling the system to operate synchronously with a fixed master control cycle. The switches in the network must support the IEEE 1588 precision clock protocol to ensure that the clocks of all agents are synchronized at the microsecond level, providing a foundation for synchronized perception and coordinated action.

[0113] Example 3

[0114] As a specific application, a certain project includes five air compressor stations with a total of 22 centrifugal air compressors, providing compressed air for the main processes such as steelmaking, continuous casting, and steel rolling. Under traditional control methods, the system consumes approximately 220 million kilowatt-hours of electricity annually, and suffers from problems such as large pressure fluctuations, uneven load distribution, and delayed response.

[0115] In the renovation project, the system adopts a four-layer architecture of edge perception and collaborative optimization: The data acquisition layer deploys 318 sensors on 22 air compressors, drying equipment, and key pipeline nodes to collect parameters such as pressure, flow rate, temperature, and dew point in real time, with a sampling frequency of 1Hz; The native intelligent agent layer adds an intelligent control unit to each centrifuge, based on the ARM Cortex-A53 processor, with a built-in efficiency MAP adaptive learning algorithm to achieve precise load rate control; The drying post-treatment system dynamically adjusts the amount of cold blowing air and circulating water according to the production stage; The cooling system achieves efficient pump station control; The pipeline intelligent agent diagnoses leak points in real time through a pressure loss model; The smart valves of 36 branches are equipped with microcontrollers to support adaptive flow adjustment; The collaborative optimization layer's central server uses a model predictive control algorithm for rolling optimization, issuing instructions to each intelligent agent with a control cycle of 30 seconds.

[0116] Regarding the dynamic adjustment of weighting coefficients, the system adopts a multi-objective optimization algorithm: the energy consumption weight ω1 is linked to the real-time electricity price and automatically increases to 0.6 during peak hours; the pressure weight ω2 is dynamically adjusted based on the pressure deviation change rate, and pressure is maintained first when the deviation exceeds 0.1 MPa; the dew point weight ω3 is switched according to the production stage, and the highest level is used during the spraying period, requiring the dew point temperature to be no higher than -40℃.

[0117] After 12 months of operation, the system achieved significant results: annual electricity savings of 11.34 million kWh, an energy saving rate of 5.2%, and power consumption per unit gas volume reduced from 0.105 kWh / m³ to 0.100 kWh / m³; pressure fluctuations decreased by 85%, and fault response time was shortened from 45 minutes to 8 minutes; predictive maintenance reduced sudden failures by 70% and maintenance costs by 32%; annual electricity savings amounted to approximately 6.46 million yuan, with an investment payback period of less than 2 years. This application example verifies the applicability and effectiveness of the technical solution proposed in this invention in complex industrial scenarios.

[0118] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A highly efficient compressed air control method based on native intelligent agents, characterized in that, The method, applicable to a system comprising native agents, a collaborative control layer, and a system scheduling layer, wherein the collaborative control layer is communicatively connected to each native agent, and the system scheduling layer is communicatively connected to the collaborative control layer, includes: Perception step: Each native intelligent agent collects the operating parameters of the corresponding physical device via local sensors; Local decision-making step: Each native intelligent agent generates a local preliminary control strategy based on the operating parameters, the built-in device physical knowledge model, and the local optimization objective function; Information interaction and collaboration steps: Each native intelligent agent uploads its own state information and intention strategy to the collaborative control layer through a preset communication mechanism. The collaborative control layer then performs information sharing and conflict resolution to generate collaborative decision information. Global optimization and execution steps: The system scheduling layer receives the gas supply requirements and safety constraints, and combines them with the collaborative decision information to perform global optimization with the goal of minimizing the total energy consumption of the system, generating a coordination instruction; each native intelligent agent integrates the coordination instruction with the local preliminary control strategy to generate the final execution signal.

2. The efficient compressed air control method based on native intelligent agents according to claim 1, characterized in that, The native intelligent agents include at least: an air compressor intelligent agent deployed on the air compressor, a drying and post-processing system intelligent agent deployed on the drying and post-processing system, a cooling system intelligent agent deployed on the cooling system, a pipeline intelligent agent deployed on the compressed air pipeline network, and a smart valve intelligent agent deployed on the smart valve.

3. The efficient compressed air control method based on native intelligent agents according to claim 1, characterized in that, In the global optimization and execution steps, the global optimization employs model predictive control to solve a preset optimization problem, the objective function of which is: ； Where i is the device index, and the summation iterates through all devices; ω1, ω2, and ω3 are the energy consumption weighting coefficient, pressure weighting coefficient, and dew point weighting coefficient, respectively; T h For the prediction time domain; u(t) is the control variable; P w,i P(t) represents the electrical power of the i-th device; P(t) represents the system pressure; P target For target pressure; T d (t) represents the dew point temperature; T d,set The target dew point temperature.

4. The efficient compressed air control method based on native intelligent agents according to claim 3, characterized in that, It also includes a step for adaptive adjustment of weighting coefficients: The energy consumption weighting coefficient ω1 is dynamically adjusted based on the real-time electricity price, grid demand response command, and the deviation between the system's unit electricity consumption and the historical best unit electricity consumption. The pressure weighting coefficient ω2 is dynamically adjusted in steady-state mode, transient mode, and safety boundary mode based on pressure deviation, pressure fluctuation rate, load change rate, and production stage identifier. The dew point weighting coefficient ω3 is dynamically adjusted based on production stage identification, product quality feedback, dew point deviation, ambient temperature and humidity, and desiccant health. Specifically, when the safety boundary conditions are triggered, the weight values ​​of the pressure weight coefficient ω2 or the dew point weight coefficient ω3 are increased so that the optimization objective prioritizes meeting the safety constraints.

5. The efficient compressed air control method based on native intelligent agents according to claim 1, characterized in that, In the information interaction and collaboration steps, the conflict resolution specifically includes: in response to the conflict between the intention strategies of different native intelligent agents, the collaborative control layer resolves the conflict according to the preset security priority rules, generates a resolution result, and inputs the resolution result as a constraint condition to the system scheduling layer.

6. The efficient compressed air control method based on native intelligent agents according to claim 1, characterized in that, It also includes exception handling and safety fault tolerance procedures: When a communication interruption occurs, the native intelligent agent automatically switches to local self-consistent mode and executes built-in control logic; When a device failure occurs, the collaborative control layer marks the failed device as unavailable and triggers the device reconfiguration algorithm; When a system-level risk occurs, the system scheduling layer switches to a safe mode, bypassing global optimization and directly executing preset emergency instructions.

7. A high-efficiency compressed air control system based on native intelligent agents, characterized in that, include: Multiple native intelligent agents, each of which has a built-in device physical knowledge model and a local optimization objective function, and is configured to: collect the operating parameters of the corresponding physical device via local sensors, generate a local preliminary control strategy based on the operating parameters, the built-in device physical knowledge model and the local optimization objective function, and, in response to a received coordination instruction, fuse the coordination instruction with the local preliminary control strategy to generate a final execution signal; The collaborative control layer communicates with each of the native intelligent agents and is configured to: receive the self-state information and intention strategies uploaded by each of the native intelligent agents, perform information sharing and conflict resolution, and generate collaborative decision information; The system scheduling layer is communicatively connected to the collaborative control layer and each of the native intelligent agents, and is configured to: receive gas supply requirements and safety constraints, combine the collaborative decision information, perform global optimization with the goal of minimizing the total energy consumption of the system, generate the coordination instructions, and send them to each of the native intelligent agents.

8. A high-efficiency compressed air control system based on a native intelligent agent according to claim 7, characterized in that, The plurality of native intelligent agents include: The air compressor intelligent agent is a hardware carrier that is an embedded controller integrated in the air compressor control cabinet or an external edge computing gateway, which is deployed in or next to the air compressor control cabinet. The intelligent agent of the drying post-processing system is a hardware carrier that is an embedded computer or programmable logic controller integrated in the control cabinet of the dryer, and is deployed in or next to the control cabinet of the dryer. The cooling system intelligent agent, wherein the hardware carrier of the cooling system intelligent agent is a communication coprocessor or an open programmable logic controller integrated in the cooling system control cabinet, and is deployed in or next to the cooling system control cabinet; The pipeline intelligent agent is a hardware carrier that is an industrial Internet of Things switch deployed in the pipeline network area and installed in the field junction box or control box in the pipeline network area. The intelligent valve agent is a dedicated control module integrated into the valve actuator, which is installed on the valve actuator.

9. A high-efficiency compressed air control system based on a native intelligent agent according to claim 7, characterized in that, The system scheduling layer is specifically configured to run model predictive control at a preset cycle and use the optimization results of the previous cycle as the initial guess for solving the current cycle.

10. A high-efficiency compressed air control system based on a native intelligent agent according to claim 7, characterized in that, The system also includes a clock synchronization module, which is used to control the clock deviation of each module in the system within the millisecond range through the IEEE 1588 precision clock protocol, so that the system can operate synchronously with a fixed master control cycle.

Images