Gas compressor control system based on AI intelligent analysis management control
By using a deep reinforcement learning AI system, the problems of poor adaptability and limited energy-saving effect of traditional gas compressor control systems have been solved. It achieves highly adaptable, low-energy-consumption and autonomously optimized control effects, and is applicable to multiple industries such as petroleum and chemical industries.
Patent Information
- Application Number
- CN202511858818.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-10
- Publication Date
- 2026-01-13
AI Technical Summary
Traditional gas compressor control systems have poor adaptability, limited energy-saving effects, and lack self-learning capabilities, making it difficult to adapt to complex and ever-changing operating environments and conditions, resulting in low control accuracy, energy waste, and high maintenance costs.
An AI-based intelligent analysis, management, and control system based on deep reinforcement learning is adopted. Through signal acquisition, a deep reinforcement learning control core, and an output execution module, it can autonomously learn and optimize the operating status of the gas compressor, combined with fault diagnosis and remote monitoring functions.
It significantly improves the adaptability and accuracy of the control system, has a significant energy-saving effect, reduces energy consumption by 10%-25%, has self-learning ability, reduces downtime due to failure, lowers maintenance costs, and is adaptable to different types of compressors.
Smart Images
Figure CN121322339A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of gas compression and gas compressor control technology, specifically a gas compressor control system based on AI intelligent analysis and management control. Background Technology
[0002] Gas compressors, as fluid machines that elevate low-pressure gases to high-pressure gases, are widely used in various industrial fields such as petroleum, chemical, metallurgy, mining, power, and refrigeration, and are one of the key pieces of equipment in industrial production. The operating efficiency and energy consumption level of gas compressors directly affect the production costs and economic benefits of enterprises, and are also closely related to social needs such as energy conservation and environmental protection.
[0003] Currently, traditional gas compressor control systems mostly employ preset logic or PID (proportional-integral-derivative) control methods. These control systems typically adjust the compressor's operating state based on pre-set parameter thresholds or fixed control algorithms, such as adjusting motor speed, intake valve opening, and exhaust pressure. However, traditional control systems have the following significant shortcomings:
[0004] 1. Poor adaptability: Gas compressors operate in complex and variable environments, with parameters such as inlet pressure, inlet temperature, gas composition, and load demand frequently fluctuating. Traditional control systems rely on fixed control logic, making it difficult to respond quickly to these dynamic changes, resulting in low control accuracy and an inability to consistently keep the compressor in its optimal operating state.
[0005] 2. Limited energy-saving effect: The control strategies of traditional control systems are based on experience or simplified models, which fail to fully consider the complex characteristics of multi-parameter coupling and nonlinearity during compressor operation, making it difficult to achieve global optimization of energy consumption, and energy waste is relatively common.
[0006] 3. Lack of self-learning ability: Traditional control systems cannot accumulate experience and optimize control strategies based on long-term operating data. When the compressor ages, wears out, or the operating conditions change significantly, parameters need to be manually readjusted, resulting in high maintenance costs and difficulty in adapting to the continuous optimization needs of the production process.
[0007] With the rapid development of artificial intelligence (AI) technology, deep reinforcement learning and other AI technologies have demonstrated enormous potential in the field of complex system control. Deep reinforcement learning, through trial and error between the agent and the environment, can autonomously learn the optimal control strategy, making it suitable for controlling complex systems with high dimensionality, nonlinearity, and dynamic characteristics. Therefore, introducing AI technology into gas compressor control systems to achieve intelligent analysis, management, and control has become an important development direction for addressing the shortcomings of traditional control systems and improving compressor performance. Summary of the Invention
[0008] The purpose of this invention is to provide a gas compressor control system based on AI intelligent analysis and management control to solve the problems mentioned in the background art.
[0009] To achieve the above objectives, the present invention provides the following technical solution: a gas compressor control system based on AI intelligent analysis, management and control, comprising a signal acquisition module, a deep reinforcement learning control core, and an output execution module;
[0010] The signal acquisition module is used to acquire state parameter signals and environmental parameter signals during the operation of the gas compressor, and transmits the acquired signals to the deep reinforcement learning control core after preprocessing.
[0011] The deep reinforcement learning control core is used to receive parameter signals transmitted by the signal acquisition module, perform data analysis and autonomous learning on the parameter signals based on the deep reinforcement learning algorithm, generate an optimized energy-saving control scheme, and convert the energy-saving control scheme into corresponding control commands and send them to the output execution module.
[0012] The output execution module is used to receive control commands sent by the deep reinforcement learning control core, and to control the execution components of the gas compressor according to the control commands, thereby adjusting the operating state of the gas compressor.
[0013] The signal acquisition module includes a sensor unit and a signal preprocessing unit;
[0014] The sensor unit includes a pressure sensor, a temperature sensor, a flow sensor, a speed sensor, a current sensor, and a vibration sensor. The pressure sensor is used to collect the intake pressure, exhaust pressure, and cylinder pressure of the gas compressor. The temperature sensor is used to collect the intake temperature, exhaust temperature, cylinder temperature, and lubricating oil temperature. The flow sensor is used to collect the intake flow rate and exhaust flow rate. The speed sensor is used to collect the speed of the compressor motor. The current sensor is used to collect the operating current of the motor. The vibration sensor is used to collect the vibration signal of the compressor body.
[0015] The signal preprocessing unit is electrically connected to each sensor unit and is used to filter, amplify, convert analog to digital and reduce noise in the acquired raw signal, remove interference components in the signal, and transmit the analog signal to the deep reinforcement learning control core after conversion into a digital signal.
[0016] The deep reinforcement learning control core includes a data receiving unit, a state feature extraction unit, a reinforcement learning agent unit, a control scheme generation unit, and an instruction conversion unit.
[0017] The data receiving unit receives the preprocessed digital signal transmitted by the signal acquisition module. The state feature extraction unit, electrically connected to the data receiving unit, extracts features from the received digital signal, identifying key feature parameters that characterize the operating state of the gas compressor and constructing a compressor operating state space. The reinforcement learning agent unit, electrically connected to the state feature extraction unit, includes a policy network, a value network, and an experience replay pool. The policy network outputs control actions based on the current operating state feature parameters. The value network evaluates the value function of the current control action, providing a basis for policy optimization. The experience replay pool stores... The state-action-reward-next-state sample data generated during the interaction between the agent and the environment are used for training and optimization of the policy network and value network. The reinforcement learning agent unit uses minimizing compressor energy consumption and maximizing operating efficiency as the reward function, and achieves autonomous learning of the optimal control strategy by continuously iterating and training the policy network. The control scheme generation unit is electrically connected to the reinforcement learning agent unit and is used to generate a specific energy-saving control scheme based on the control actions output by the optimized policy network and the operating characteristics of the compressor. The instruction conversion unit is electrically connected to the control scheme generation unit and is used to convert the energy-saving control scheme into digital control instructions that conform to the output execution module interface protocol.
[0018] The output execution module includes a drive circuit and an execution component;
[0019] The drive circuit is used to receive control commands sent by the deep reinforcement learning control core and convert the control commands into drive signals; the execution components include a motor speed controller, an intake valve actuator, an exhaust valve actuator, and a lubricating oil pump controller; the motor speed controller is used to adjust the speed of the compressor motor according to the drive signal; the intake valve actuator is used to adjust the opening degree of the intake valve; the exhaust valve actuator is used to adjust the opening degree of the exhaust valve; and the lubricating oil pump controller is used to adjust the oil supply of the lubricating oil pump.
[0020] The gas compressor control system also includes a data storage module and a communication module;
[0021] The data storage module is electrically connected to the deep reinforcement learning control core and is used to store the raw data, preprocessed data, training data during the deep reinforcement learning process, and control command data acquired by the signal acquisition module. The communication module adopts wired Ethernet, 4G / 5G, or LoRa communication and is electrically connected to the deep reinforcement learning control core to realize data transmission between the control system and the remote monitoring platform, supporting remote data viewing, control parameter configuration, and fault diagnosis.
[0022] The gas compressor control system also includes a fault diagnosis module, which is electrically connected to the deep reinforcement learning control core and the signal acquisition module, respectively. The fault diagnosis module is used to determine whether the compressor has a fault and the type of fault based on the vibration signal, temperature signal and current signal acquired by the signal acquisition module and the abnormal operation mode identified by the deep reinforcement learning model, and to issue a fault warning signal.
[0023] Compared with the prior art, the beneficial effects of the present invention are:
[0024] 1) High level of intelligence and strong adaptability: The present invention adopts a deep reinforcement learning control core, which can adapt to the dynamic changes of multiple parameters during the operation of the gas compressor through autonomous learning, without the need for frequent manual parameter adjustment, which significantly improves the adaptability and control accuracy of the control system to complex working conditions.
[0025] 2) Significant energy-saving effect: The reinforcement learning agent optimizes strategies with the goal of minimizing energy consumption and maximizing operating efficiency. It can fully consider the nonlinear and multi-parameter coupling characteristics of the compressor and formulate a globally optimal energy-saving control scheme. Actual tests show that it can reduce energy consumption by 10%-25% compared with traditional control systems.
[0026] 3) Possesses autonomous learning and continuous optimization capabilities: By storing and utilizing operational data through an experience replay pool, the reinforcement learning model can continuously iterate and optimize control strategies. As the operating time increases, the control effect continues to improve, adapting to long-term operational changes such as compressor aging and wear.
[0027] 4) Integrated fault diagnosis and remote monitoring functions: The system is equipped with a fault diagnosis module, which can promptly identify compressor faults and issue early warnings, reducing downtime due to faults; the communication module enables interaction with the remote monitoring platform, facilitating centralized management and remote operation and maintenance for enterprises, and reducing maintenance costs;
[0028] 5) Strong compatibility and easy to promote and apply: The signal acquisition module and output execution module of this system can be adapted to different types and specifications of gas compressors, with low modification and upgrade costs, and are easy to promote and apply in multiple industries such as petroleum, chemical, and metallurgy. Attached Figure Description
[0029] Figure 1 This is a block diagram of the gas compressor control system.
[0030] Figure 2 This is a block diagram of the signal acquisition module;
[0031] Figure 3 A structural diagram of the control core for deep reinforcement learning;
[0032] Figure 4 This is a block diagram of the output execution module. Detailed Implementation
[0033] 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.
[0034] Example:
[0035] Please see Figure 1-4 The present invention provides a technical solution:
[0036] A gas compressor control system based on AI intelligent analysis, management and control includes a signal acquisition module, a deep reinforcement learning control core, and an output execution module;
[0037] The signal acquisition module is used to collect state parameter signals and environmental parameter signals during the operation of the gas compressor, and transmits the preprocessed signals to the deep reinforcement learning control core. The signal acquisition module uses a pressure sensor (measurement range 0-2MPa, accuracy ±0.5% FS (full scale), a temperature sensor (-40℃-200℃, resolution 0.1℃), a flow sensor (range 0-100m³ / h, accuracy ±1% FS), a speed sensor (measurement range 0-3000rpm, accuracy ±1rpm), and a current sensor (range 0-500A, accuracy ±0.2%). The system includes: a vibration sensor (measuring axial backward ±50g, frequency response 1-1000Hz); a signal preprocessing unit with at least 8 12-bit or higher precision analog-to-digital converter (ADC) channels (adapting to multi-sensor signal acquisition), supporting sampling frequencies above 10kHz (ensuring real-time performance), and integrating digital filtering algorithm modules (such as moving average filtering and Kalman filtering); in this embodiment, an STM32F103 microcontroller is used as the core processor (it has a 12-bit ADC, a maximum sampling frequency of 1MHz, and can implement filtering, amplification, and analog-to-digital conversion functions through software programming). Those skilled in the art can also choose other embedded processors with equivalent functions (such as STM32F407, ATmega328P, etc.), as long as they meet the above signal preprocessing functional requirements. The microcontroller, as the core processor, implements signal filtering, amplification, and analog-to-digital conversion.
[0038] The deep reinforcement learning control core receives parameter signals transmitted by the signal acquisition module, performs data analysis and autonomous learning on the parameter signals based on the deep reinforcement learning algorithm, generates an optimized energy-saving control scheme, and converts the energy-saving control scheme into corresponding control commands to be sent to the output execution module. The deep reinforcement learning control core must have the computing power to run the deep reinforcement learning algorithm in real time, support the synchronous processing of 12-dimensional feature parameters, have a single policy generation time ≤ 0.5s, and be protocol compatible with the signal acquisition module and output execution module (e.g., UART, Ethernet communication). Both the policy network and value network adopt a three-layer fully connected neural network structure. The input layer consists of 12-dimensional feature parameters (including intake pressure, exhaust pressure, intake temperature, exhaust temperature, intake flow rate, exhaust flow rate, motor speed, motor current, cylinder temperature, lubricating oil temperature, engine vibration acceleration, and load demand). The output layer consists of 4-dimensional control actions (including motor speed adjustment, intake valve opening adjustment, exhaust valve opening adjustment, and lubricating oil pump supply adjustment). The reinforcement learning algorithm uses DQN (Deep QN)... The Q-Network algorithm has a reward function set as: R = α × (1 - energy consumption / rated energy consumption) + β × (control accuracy error), where α and β are weight coefficients, set to 0.7 and 0.3 respectively.
[0039] The output execution module receives control commands from the deep reinforcement learning control core and controls the execution components of the gas compressor according to the control commands, adjusting the operating state of the gas compressor. The drive circuit of the output execution module uses an L298N motor drive module (this model is only an example; hardware meeting the functional requirements is protected). The motor speed controller uses a vector frequency converter, the intake and exhaust valve actuators use electric actuators, and the lubricating oil pump controller uses a PWM speed controller.
[0040] The signal acquisition module includes a sensor unit and a signal preprocessing unit;
[0041] The sensor unit includes a pressure sensor, a temperature sensor, a flow sensor, a speed sensor, a current sensor, and a vibration sensor. The pressure sensor is used to collect the intake pressure, exhaust pressure, and cylinder pressure of the gas compressor. The temperature sensor is used to collect the intake temperature, exhaust temperature, cylinder temperature, and lubricating oil temperature. The flow sensor is used to collect the intake flow rate and exhaust flow rate. The speed sensor is used to collect the speed of the compressor motor. The current sensor is used to collect the operating current of the motor. The vibration sensor is used to collect the vibration signal of the compressor body.
[0042] The signal preprocessing unit is electrically connected to each sensor unit and is used to filter, amplify, convert analog to digital and reduce noise in the acquired raw signal, remove interference components in the signal, and transmit the analog signal to the deep reinforcement learning control core after conversion into a digital signal.
[0043] The deep reinforcement learning control core includes a data receiving unit, a state feature extraction unit, a reinforcement learning agent unit, a control scheme generation unit, and an instruction conversion unit.
[0044] The data receiving unit receives the preprocessed digital signal transmitted by the signal acquisition module. The state feature extraction unit, electrically connected to the data receiving unit, extracts features from the received digital signal, identifying key feature parameters that characterize the operating state of the gas compressor and constructing a compressor operating state space. The reinforcement learning agent unit, electrically connected to the state feature extraction unit, includes a policy network, a value network, and an experience replay pool. The policy network outputs control actions based on the current operating state feature parameters. The value network evaluates the value function of the current control action, providing a basis for policy optimization. The experience replay pool stores... The state-action-reward-next-state sample data generated during the interaction between the agent and the environment are used for training and optimization of the policy network and value network. The reinforcement learning agent unit uses minimizing compressor energy consumption and maximizing operating efficiency as the reward function, and achieves autonomous learning of the optimal control strategy by continuously iterating and training the policy network. The control scheme generation unit is electrically connected to the reinforcement learning agent unit and is used to generate a specific energy-saving control scheme based on the control actions output by the optimized policy network and the operating characteristics of the compressor. The instruction conversion unit is electrically connected to the control scheme generation unit and is used to convert the energy-saving control scheme into digital control instructions that conform to the output execution module interface protocol.
[0045] The output execution module includes a drive circuit and an execution component;
[0046] The drive circuit is used to receive control commands sent by the deep reinforcement learning control core and convert the control commands into drive signals; the execution components include a motor speed controller, an intake valve actuator, an exhaust valve actuator, and a lubricating oil pump controller; the motor speed controller is used to adjust the speed of the compressor motor according to the drive signal; the intake valve actuator is used to adjust the opening degree of the intake valve; the exhaust valve actuator is used to adjust the opening degree of the exhaust valve; and the lubricating oil pump controller is used to adjust the oil supply of the lubricating oil pump.
[0047] The gas compressor control system also includes a data storage module and a communication module;
[0048] The data storage module is electrically connected to the deep reinforcement learning control core and is used to store the raw data, preprocessed data, training data during the deep reinforcement learning process, and control command data acquired by the signal acquisition module. The communication module adopts wired Ethernet, 4G / 5G, or LoRa communication and is electrically connected to the deep reinforcement learning control core to realize data transmission between the control system and the remote monitoring platform, supporting remote data viewing, control parameter configuration, and fault diagnosis.
[0049] The data storage module uses a 128GB SD card (this module needs to be adapted to the compressor's operating scenario for storage capacity: small and medium-sized compressors (single unit, short-term operation) can use small-capacity storage media; large compressors (multiple units working together, long-term operation) need to use large-capacity storage media, and support off-site data backup to avoid data loss due to local storage failure;), and the communication module uses a 4G module EC20 (the latency of receiving data on the remote monitoring platform is ≤300ms, the time from the generation of the fault warning signal to the push to the platform is ≤1s, and the test results show that the control system response time is <0.5s and the fault warning accuracy rate is >92%).
[0050] The fault diagnosis module, based on the vibration signal spectrum characteristics and temperature anomaly patterns identified by a deep reinforcement learning model, can diagnose common faults such as motor overload, bearing wear, and valve jamming.
[0051] The control system of this embodiment was applied to a screw air compressor in a chemical plant. The test results showed that under the conditions of intake pressure fluctuation of ±10% and load demand change of 20%-100%, the control system response time was less than 0.5s and the control accuracy error was less than 3%. Compared with the traditional PID control system, the energy consumption per unit displacement of the compressor was reduced by 18%, saving about 52,000 kWh of electricity per year. The fault warning accuracy rate reached more than 92%, which significantly improved the operating performance and economic benefits of the compressor.
[0052] The gas compressor control system also includes a fault diagnosis module, which is electrically connected to the deep reinforcement learning control core and the signal acquisition module, respectively. The fault diagnosis module is used to determine whether the compressor has a fault and the type of fault based on the vibration signal, temperature signal and current signal acquired by the signal acquisition module and the abnormal operation mode identified by the deep reinforcement learning model, and to issue a fault warning signal.
[0053] The workflow is as follows:
[0054] 1. Signal Acquisition and Preprocessing: Each sensor unit in the signal acquisition module acquires the status parameter signals and environmental parameter signals of the gas compressor in real time. The signal preprocessing unit performs filtering, amplification, analog-to-digital conversion and other processing on the raw signals before transmitting them to the deep reinforcement learning control core.
[0055] 2. State Feature Extraction and Analysis: The data receiving unit of the deep reinforcement learning control core receives the preprocessed signal, and the state feature extraction unit extracts key feature parameters from it to construct the compressor's operating state space;
[0056] 3. Reinforcement Learning and Control Scheme Generation: The reinforcement learning agent unit outputs control actions through the policy network based on the feature parameters of the current operating state, and the value network evaluates the value of the control actions. The model is trained and optimized by combining the sample data in the experience replay pool. The control scheme generation unit generates energy-saving control schemes based on the optimized control actions and converts them into control commands through the instruction conversion unit.
[0057] 4. Execution control and status adjustment: The drive circuit of the output execution module converts control commands into drive signals to control the operation of execution components such as the motor speed controller and intake valve actuator, thereby adjusting the operating status of the compressor;
[0058] 5. Data storage and remote interaction: The data storage module stores various types of operational and training data, the communication module enables data interaction with the remote monitoring platform, and the fault diagnosis module monitors the compressor's operating status in real time and issues timely fault warnings.
[0059] This system consists of a physical layer (sensor array), an edge computing layer (real-time analysis module), and an AI decision-making layer (cloud training platform), forming a closed-loop control system. The core technical architecture is as follows:
[0060] The latest DeepSeek big data model is injected into the controller core to analyze parameters such as speed, temperature, energy consumption, pressure, and output in real time and autonomously optimize operating strategies. Through historical work order big data, AI automatically matches the optimal combination of process parameters; provides early warning of potential quality problems; accurately analyzes the energy consumption cost curve of each piece of equipment; analyzes efficiency changes through energy consumption data, monitors equipment operating wear trends in real time, reminds maintenance plans, provides data for accurate cost accounting, and can generate reports to be pushed to design, user, and management personnel.
[0061] DRL-based controllers learn through methodical and continuous practice (what we call machine teaching). Therefore, these controllers can uncover subtleties and anomalies that are difficult to capture in expert systems; they can learn from hundreds of simulations rather than relying on single control data from a single plant. Each simulation executes faster than what is seen in normal real-time, allowing the controller to learn to identify suboptimal behaviors in the short term, thereby optimizing benefits and evolving control in the long term. When taking control actions, it also considers categorized machine alarms and warnings. It can even use audible signals and various sensor inputs to determine how to make process decisions, somewhat similar to how operators respond to audible information, continuously learning and evolving to provide developers with better control strategies and training models.
[0062] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or basic characteristics. Therefore, the embodiments should be considered exemplary and non-limiting in all respects. The scope of the invention is defined by the appended claims rather than the foregoing description. Therefore, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention, and no reference numerals in the claims should be construed as limiting the scope of the claims.
[0063] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A gas compressor control system based on AI intelligent analysis, management, and control, characterized in that, It includes a signal acquisition module, a deep reinforcement learning control core, and an output execution module; The signal acquisition module is used to acquire state parameter signals and environmental parameter signals during the operation of the gas compressor, and transmits the acquired signals to the deep reinforcement learning control core after preprocessing. The deep reinforcement learning control core is used to receive parameter signals transmitted by the signal acquisition module, perform data analysis and autonomous learning on the parameter signals based on the deep reinforcement learning algorithm, generate an optimized energy-saving control scheme, and convert the energy-saving control scheme into corresponding control commands and send them to the output execution module. The output execution module is used to receive control commands sent by the deep reinforcement learning control core, and to control the execution components of the gas compressor according to the control commands, thereby adjusting the operating state of the gas compressor.
2. The gas compressor control system based on AI intelligent analysis and management control according to claim 1, characterized in that: The signal acquisition module includes a sensor unit and a signal preprocessing unit; The sensor unit includes a pressure sensor, a temperature sensor, a flow sensor, a speed sensor, a current sensor, and a vibration sensor. The pressure sensor is used to collect the intake pressure, exhaust pressure, and cylinder pressure of the gas compressor. The temperature sensor is used to collect the intake temperature, exhaust temperature, cylinder temperature, and lubricating oil temperature. The flow sensor is used to collect the intake flow rate and exhaust flow rate. The speed sensor is used to collect the speed of the compressor motor. The current sensor is used to collect the operating current of the motor. The vibration sensor is used to collect the vibration signal of the compressor body. The signal preprocessing unit is electrically connected to each sensor unit and is used to filter, amplify, convert analog to digital and reduce noise in the acquired raw signal, remove interference components in the signal, and transmit the analog signal to the deep reinforcement learning control core after conversion into a digital signal.
3. The gas compressor control system based on AI intelligent analysis and management control according to claim 1, characterized in that: The deep reinforcement learning control core includes a data receiving unit, a state feature extraction unit, a reinforcement learning agent unit, a control scheme generation unit, and an instruction conversion unit. The data receiving unit receives the preprocessed digital signal transmitted by the signal acquisition module. The state feature extraction unit is electrically connected to the data receiving unit and is used to extract features from the received digital signal, extracting key feature parameters that characterize the operating state of the gas compressor, and constructing the compressor operating state space. The reinforcement learning agent unit is electrically connected to the state feature extraction unit. The reinforcement learning agent unit includes a policy network, a value network, and an experience replay pool. The policy network outputs control actions based on the current operating state feature parameters. The value network evaluates the value function of the current control action, providing a basis for policy optimization. The experience replay pool stores state-action-reward-next state sample data generated during the agent's interaction with the environment, for training and optimization of the policy network and value network. The reinforcement learning agent unit uses minimizing compressor energy consumption and maximizing operating efficiency as its reward function. It achieves autonomous learning of the optimal control strategy by continuously iterating and training the optimization policy network. The control scheme generation unit is electrically connected to the reinforcement learning agent unit and is used to generate a specific energy-saving control scheme based on the control actions output by the optimized policy network and the operating characteristics of the compressor. The instruction conversion unit is electrically connected to the control scheme generation unit and is used to convert the energy-saving control scheme into digital control instructions that conform to the output execution module interface protocol.
4. The gas compressor control system based on AI intelligent analysis and management control according to claim 1, characterized in that: The output execution module includes a drive circuit and an execution component; The drive circuit is used to receive control commands sent by the deep reinforcement learning control core and convert the control commands into drive signals; the execution components include a motor speed controller, an intake valve actuator, an exhaust valve actuator, and a lubricating oil pump controller; the motor speed controller is used to adjust the speed of the compressor motor according to the drive signal; the intake valve actuator is used to adjust the opening degree of the intake valve; the exhaust valve actuator is used to adjust the opening degree of the exhaust valve; The lubricating oil pump controller is used to regulate the oil supply of the lubricating oil pump.
5. A gas compressor control system based on AI intelligent analysis and management control according to claim 1, characterized in that: The gas compressor control system also includes a data storage module and a communication module; The data storage module is electrically connected to the deep reinforcement learning control core and is used to store the raw data, preprocessed data, training data during the deep reinforcement learning process, and control command data acquired by the signal acquisition module. The communication module adopts wired Ethernet, 4G / 5G, or LoRa communication and is electrically connected to the deep reinforcement learning control core to realize data transmission between the control system and the remote monitoring platform, supporting remote data viewing, control parameter configuration, and fault diagnosis.
6. A gas compressor control system based on AI intelligent analysis, management, and control according to claim 1, characterized in that: The gas compressor control system also includes a fault diagnosis module, which is electrically connected to the deep reinforcement learning control core and the signal acquisition module, respectively. The fault diagnosis module is used to determine whether the compressor has a fault and the type of fault based on the vibration signal, temperature signal and current signal acquired by the signal acquisition module and the abnormal operation mode identified by the deep reinforcement learning model, and to issue a fault warning signal.