Intelligent regulation and control system and method for variable-component flue gas drive coalbed methane extraction and CO2 storage

Abstract

The invention discloses an intelligent regulation and control system and method for coal bed gas exploitation and CO2 storage through variable-component flue gas drive, and belongs to the technical field of coal bed gas exploitation and CO2 storage. The system comprises a variable component injection module, a multi-parameter distributed sensing network, an extraction and utilization module, a digital twinborn model, an AI intelligent regulation and control engine and a safety interlocking and emergency response module. The component-variable injection module is used for realizing adjustment of CO2 / N2 proportion, pressure, flow, temperature and humidity; the sensing network realizes high-frequency panoramic monitoring of the whole injection-extraction process through a plurality of sensors; the digital twinborn model is coupled with permeability evolution, adsorption and fracture diversion models to realize real-time inversion of key parameters; the AI engine outputs an optimal injection strategy based on LSTM prediction and a PPO reinforcement learning algorithm; and the safety interlocking and emergency response module triggers emergency response at a millisecond level during overrun. According to the system, collaborative optimization of the coalbed methane recovery rate, the CO2 storage rate and the operation energy consumption is achieved, and normal form transformation from blind injection to intelligent regulation and control is completed.

Description

Technical Field

[0001] This invention belongs to the field of coalbed methane extraction and CO2 storage technology, specifically relating to an intelligent control system and method for coalbed methane extraction and CO2 storage using variable composition flue gas. Background Technology

[0002] Direct emissions of industrial flue gas (mainly composed of CO2 and N2) cause significant environmental pollution. Against this backdrop, injecting CO2-rich industrial flue gas into low-permeability coal seams offers a promising "three-in-one" technological approach. CO2, with its stronger adsorption capacity, can effectively seal and replace CH4, while N2, through its partial pressure effect, can more efficiently promote the desorption and migration of CH4, thus driving coalbed methane (CH4) extraction. This approach is considered highly promising, potentially achieving simultaneous reductions in industrial carbon emissions, increased unconventional natural gas production, and energy security. However, the large-scale and commercial application of this technology remains limited by its inherent complexity and the lag of traditional control methods.

[0003] This complexity stems primarily from the strong nonlinear coupling of multiple physicochemical processes. After industrial flue gas is injected into the coal seam, it triggers a series of complex reactions, including competitive adsorption, matrix expansion / contraction, dynamic permeability evolution, and fracture network reconstruction. Traditional fixed-component injection models neglect the dynamic complementarity of CO2 and N2, resulting in suboptimal displacement efficiency and storage potential. Furthermore, current technologies heavily rely on wellbore point measurements and periodic geophysical exploration for understanding subsurface conditions, leaving the entire system operating in a near-black box state. This makes it difficult to capture the CO2 plume front in real time, accurately assess storage safety, and provide early warnings of potential leakage risks. Current process control strategies largely depend on static models and human experience, lacking an "intelligent brain" capable of integrating real-time data, understanding system status, and performing forward-looking optimizations. This prevents the system from adaptively responding to rapidly changing reservoir conditions and makes it difficult to dynamically balance multiple objectives such as maximizing CH4 recovery, maximizing CO2 storage, and minimizing operational energy consumption, thus limiting the overall effectiveness of the technology.

[0004] To overcome the aforementioned bottlenecks, a new method is proposed. The core innovation lies in the deep integration of dynamic component allocation, multi-parameter distributed sensor networks, digital twin models, and artificial intelligence decision engines. This aims to achieve a paradigm shift from "blind injection" to "intuitive analysis" and from "experience-driven" to "intelligent optimization," providing a new research approach for the development of the coalbed methane industry. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides an intelligent control system and method for coalbed methane extraction and CO2 sequestration using variable-component flue gas.

[0006] To achieve the above technology, the following are included: A smart control system for coalbed methane extraction and CO2 storage using variable-component flue gas includes: a variable-component injection module, a multi-parameter distributed sensor network module, an extraction and utilization module, a digital twin model module, an AI intelligent control engine module, and a safety interlock and emergency response module. The variable component injection module injects flue gas into the well according to a preset initial value. The multi-parameter distributed sensor network module collects the well data and transmits it to the digital twin model module for simulation. The simulation results are input into the AI ​​intelligent control engine module to issue the best control command. The extraction and utilization module is used for the separation, purification, and storage of CH4 gas after extraction; The safety interlock and emergency response module, along with the variable component injection module, multi-parameter distributed sensor network module, digital twin model module, and AI intelligent control engine module, operate in parallel and independently. They are activated immediately when any indicator exceeds a preset safety threshold. The indicators are generated by the AI ​​intelligent control engine module and include: a) Pressure drop rate > 1 MPa / 10 min; b) Daily increase in CO2 concentration in monitoring wells > 2%vol; c) Single-event microseismic energy > 10 4 J; Emergency actions include: a) Close the injection valve slightly to reduce the flow rate below the safe threshold; b) Switch the gas source to high N2 ratio mode, where CO2:N2 ≤ 3:7 or ≤ 1:9; c) Force the injection pressure to ≤5 MPa; d) Activate the groundwater chemical monitoring unit to collect data on pH, conductivity, and dissolved CO2; e) Encrypt the operation logs and upload them to the blockchain audit platform, and simultaneously send risk alerts to the regulatory platform.

[0007] Furthermore, the variable component injection module includes: hot flue gas storage tank 3, CO2 gas storage tank 4, N2 gas storage tank 5, intelligent proportional valve group 6, variable frequency gas injection pump 7, humidity conditioning and dust removal system 8, heat exchanger group 9, and gas injection pipeline 11. Specifically, the hot flue gas storage tank 3 is connected to the CO2 gas storage tank 4 through the intelligent proportional valve group 6. The CO2 gas storage tank 4 is connected to the N2 gas storage tank 5 through the intelligent proportional valve group 6. The N2 gas storage tank 5 is connected to the variable frequency gas injection pump 7 through the intelligent proportional valve group 6. The variable frequency gas injection pump 7 sequentially inputs the gas into the humidification and dust removal system 8 and the heat exchanger group 9, and inputs it into the injection well 12 through the gas injection pipeline 11. The gas injection pipeline 11 and the injection well 12 sequentially pass through the rock stratum 1 and the coal seam 2. The gas injection pipeline 11 is placed in the injection well 12, which is L-shaped. Considering the gas injection efficiency, a gas injection plug 13 is installed between the rock stratum 1 and the coal seam 2 in the injection well 12.

[0008] Furthermore, the multi-parameter distributed sensor network module includes: an injection well integrated multi-parameter monitoring device 14, an injection well data transmission line 10, a ground control center 25, a pumping-monitoring combined well integrated multi-parameter monitoring device 16, and a pumping-monitoring combined well data transmission line 15; Specifically, the injection well integrated multi-parameter monitoring device 14 is installed in the injection well 12. The injection well integrated multi-parameter monitoring device 14 synchronously collects injection parameters such as pressure P, temperature T, flow rate Q, CO2 / N2 / CH4 concentration, AE sound wave, micro-vibration, and geothermal gradient at a high frequency of 1Hz through the injection well data transmission line 10 and transmits them to the ground control center 25. The extraction-monitoring combined well integrated multi-parameter monitoring device 16 is installed in the extraction-monitoring combined well 19. The extraction-monitoring combined well integrated multi-parameter monitoring device 16 transmits permeability, temperature, pressure, adsorption / desorption parameters, etc. during the extraction process to the ground control center 25 through the extraction-monitoring combined well data transmission line 15. The extraction-monitoring combined well 19 passes through rock stratum 1 and coal seam 2, and an extraction plug 17 is installed between rock stratum 1 and coal seam 2.

[0009] Furthermore, the extraction and utilization module includes: extraction pipeline 18, coalbed methane extraction pump 20, heat exchanger group 21, gas filtration device 22, gas separation device 23 and CH4 gas storage tank 24. Specifically, considering the impact of leakage during the extraction process, a gas injection plug 17 is placed in the extraction-monitoring combined well 19. The extracted gas is pumped from the extraction pipeline 18 by the coalbed methane extraction pump 20, cooled by the heat exchanger group 21, impurity gas is removed by the gas filtration device 22, and CH4 gas is separated by the gas separation device 23 before entering the CH4 gas storage tank 24 for subsequent use.

[0010] Furthermore, the digital twin model module includes: a permeability evolution model, an adsorption saturation model, and a fracture conductivity model to invert key parameters of the coal seam in real time, outputting permeability field, pressure field, CO2 plume distribution, and CH4 desorption parameters; the AI ​​intelligent control engine module includes an LSTM prediction algorithm and a PPO reinforcement learning algorithm, which predicts the average pressure, CH4 production, and CO2 breakthrough time for the next few hours through AI algorithms, and combines reinforcement learning with "CH4 increment + CO2 sequestration rate - energy consumption" as the reward function to output optimal injection parameters (flue gas injection pressure, flow rate, composition), updates the injection strategy every 1 hour, and issues instructions to the variable composition module; the digital twin model module and the AI ​​intelligent control engine module are integrated on the ground control center 25; in: The permeability evolution model comprehensively considers stress-sensitive effects, matrix shrinkage / expansion effects, shear slip, and chemical dissolution. Its relationship is as follows:

[0011] In the formula, m is the initial permeability. 2 ; The effective stress function; The matrix strain function; It is a chemical solubility correction factor; The adsorption saturation model, based on the extended Langmuir competitive adsorption model, describes the ternary competitive adsorption behavior of CO2-CH4-N2, and its relationship is as follows:

[0012] In the formula, Components Adsorption saturation; Components Langmuir constant, MPa -1 ; Components The partial pressure, MPa; Components Langmuir constant, MPa -1 ; Components The partial pressure, MPa; This indicates summing the three gases; The fracture conductivity model combines the cubic law and shear displacement effect, considering the changes in conductivity caused by fracture surface roughness and chemical dissolution. The relationship is as follows:

[0013] In the formula, The current crack porosity; Initial crack porosity; The current crack width is in meters (m). Let the initial crack width be m; This is the shear displacement correction factor; It is a function of chemical dissolution time; The LSTM prediction algorithm takes at least 30 days of historical downhole pressure, temperature, CH4 / CO2 concentration and gas production time series as input, and outputs the average pressure distribution, CH4 production and CO2 breakthrough time prediction for the next 6 hours. The PPO reinforcement learning algorithm has a state space consisting of average pressure, CH4 concentration, CO2 plume leading edge position, and microseismic energy. Its action space consists of flue gas injection pressure (0-25 MPa), flue gas flow rate (0.1-10 t / h), and CO2 / N2 volume ratio (30%-70%). Its reward function is:

[0014] in, The rate of increase in production compared to the previous cycle, m 3 / h; CO2 sequestration rate within a single cycle, % Energy consumption of the injection pump and heating unit, MWh; These are the first, second, and third weighting coefficients, which can be adjusted.

[0015] Furthermore, after receiving the action command output by the AI ​​intelligent control engine, the variable component injection module adjusts the wellhead flue gas injection pressure to the target value ±0.1 MPa within 1 minute through the servo valve group, and controls the CO2 / N2 ratio within the set value ±2%.

[0016] A smart control method for coalbed methane extraction and CO2 sequestration using variable-component flue gas includes the following steps: S1. System startup, self-test and initialization; Start the system and power on the variable component injection module, multi-parameter distributed sensor network, digital twin model, extraction and utilization module, AI intelligent control engine and safety interlock module; verify the integrity of intelligent proportional valve group 6, variable frequency gas injection pump 7, humidity and dust removal system 8, heat exchanger group 9, injection well integrated multi-parameter monitoring device 14 and extraction-monitoring combined well integrated multi-parameter monitoring device 16, injection well data transmission line 10 and extraction-monitoring combined well data transmission line 15, edge computing node and blockchain audit interface. If abnormal, alarm and stop the system; otherwise, proceed to step 2. S2. Real-time data acquisition and data preprocessing; Synchronous acquisition at 1Hz high frequency: pressure, temperature, flow rate, CO2 / N2 / CH4 concentration, AE sound waves, micro-vibration, and geothermal gradient; The edge cache stores a 5-minute sliding window of data. The ground control center performs data cleaning and outlier removal on the received real-time data, generating a cleaned data packet D. t Data cleaning and outlier removal specifically involve: using statistical methods to identify and filter out abnormal data points caused by instantaneous sensor failures or signal transmission interference, and using interpolation to fill in missing values ​​to ensure the quality and reliability of the data input to the downstream model.

[0017] S3, the digital twin model module is updated on a rolling basis; The cleaned data packet D obtained in S2 t The digital twin model is input with historical drilling, logging and fracturing data, and the permeability evolution model, ternary competitive adsorption saturation model and fracture conductivity model are invoked to invert and obtain the current full-field permeability field, pressure field, CO2 plume front and CH4 desorption degree; the twin state is updated on a rolling basis with a period of 1 hour as the initial field for subsequent prediction and optimization.

[0018] S4, AI intelligent control engine closed-loop decision-making, including: S4.1 State Prediction (LSTM Model): Input the past ≥30 days of clean data and the latest twin state, output the average pressure distribution, CH4 production sequence and CO2 breakthrough time for the next 6 hours; S4.2, PPO reinforcement learning optimization: The key indicators of the twin at time t (mean pressure, CH4 concentration, CO2 plume location, and accumulated microseismic energy) are encoded into a state vector S. t Construct the state space; The action space is defined by the injection pressure P_{inj}, injection flow rate Q_{inj}, and CO2 / N2 ratio R_{CO2:N2}, denoted as action space A. t ={P_{inj},Q_{inj},R_{CO2:N2}}, where P_{inj}∈[0,25]MPa, Q_{inj}∈[0.1,10]t / h, R_{CO2:N2}∈[30%,70%]; Measure ΔCH4, CO2 sequestration rate, and energy consumption E, calculate the immediate reward according to the reward function, and output the optimal action vector; S4.3, Safety Constraint Verification: If the following occurs: The rate of pressure drop at the bottom of the well, P_{drop}, is greater than 1 MPa / 10 min. CO2 increase > 2% / day; The energy released by a single microseismic event is E_{seis}>10 4 J; Then proceed to S5; otherwise, use PPO shearing importance sampling and gradient ascent to update network parameters, complete policy update, and then issue instructions to parse the optimal action vector into specific set values ​​and issue them within <500ms. The objects to be issued include: intelligent proportional valve group → adjust CO2 / N2 ratio; variable frequency injection pump → adjust P_{inj}, Q_{inj}; temperature control unit → adjust flue gas temperature 150-400℃; humidifier → adjust moisture content 0-20%; finally, wait for a 1-hour rolling cycle and return to step 2 to form a closed loop.

[0019] S5, Safety Interlocks and Emergency Response; When any security constraint trigger condition of S4.3 security constraint verification is met, the security instruction is immediately initiated: ① Close the injection valve slightly to reduce the flow rate to ≤20% of the maximum flow rate; ② Switch the gas source to high N2 mode (CO2:N2≤3:7 or≤1:9); ③ Force the injection pressure to be reduced to ≤5MPa; ④ Activate the groundwater chemical monitoring unit to collect data on pH, conductivity, and dissolved CO2; ⑤ Encrypt the operation logs and upload them to the blockchain audit platform, and simultaneously send risk alerts to the regulatory platform; Simultaneously, the temperature control and humidification units are suspended, and then the groundwater monitoring subunit is activated to collect pH, conductivity, and dissolved CO2. The emergency action sequence, timestamps, and original sensor values ​​are packaged and written into the blockchain audit block. Risk alarms and earthquake impact assessment requests are pushed to the regulatory platform through the 5G link to continuously monitor risk indicators. If the data is below the threshold for 30 consecutive minutes and the monitoring platform confirms it remotely, exit the safety mode and return to S4; otherwise, remain in safety mode until manually reset. All raw data throughout the lifecycle, twin inversion results, AI strategy logs, and emergency records are encrypted and stored in a distributed database. Before issuing the command, the system performs a final security check, including: S5.1 Determine whether the optimal strategy parameters obtained by the PPO algorithm meet all preset safety constraints (e.g., whether the bottom hole pressure is lower than the caprock fracture pressure threshold, and whether the predicted microseismic energy is within the safe range). S5.2 Safe Path: If safety constraints are not met, the system will not execute dangerous commands, but will immediately activate the safety interlock and emergency response module, switching to a degraded operation mode. This mode includes: automatically reducing the injection pressure, increasing the proportion of N2 in the flue gas to reduce risk, and sending an alarm to the superior monitoring system; S5.3 Normal Path: If the strategy meets all safety constraints, the system will send the generated optimization instructions (including variable component instructions, pressure setpoints, temperature setpoints, and flow setpoints) to the variable component intelligent injection module.

[0020] S6, Instruction Execution and Closed-Loop Feedback; After receiving the command, the variable-component intelligent injection module drives the intelligent proportional valve group to precisely adjust the ratio of CO2 to N2; the temperature control unit and humidification unit adjust the flue gas to the target temperature and humidity; the variable-frequency injection pump executes the injection operation according to the set flow rate and pressure. Afterwards, the system returns to S1 to read the next round of downhole data after injection, thus forming a continuously iterative, self-optimizing closed-loop intelligent control cycle. This cycle typically operates on a 1-hour control period, adapting in real-time to the dynamic changes in the reservoir.

[0021] Beneficial effects of the present invention 1. This invention realizes a paradigm shift from "blind injection" to "intelligent driving," significantly improving overall efficiency. It breaks through the traditional extensive injection mode of fixed components and experience-driven approach. Through an AI intelligent control engine, it dynamically optimizes the composition, temperature, and pressure of the injected flue gas, giving full play to the synergistic effect of strong CO2 adsorption and storage and N2 partial pressure-assisted desorption. Thus, it simultaneously maximizes the synergistic effect of CH4 recovery rate and CO2 storage rate in the same process, and effectively reduces the system's operating energy consumption.

[0022] 2. This invention constructs a "transparent" underground sensing and decision-making system, solving the problem of the "black box" of the process. By integrating a multi-parameter distributed sensor network and a high-fidelity digital twin model, the system can capture downhole parameter changes in real time and accurately invert key reservoir parameters (such as permeability field, pressure field, and CO2 plume distribution). This breaks through the limitations of traditional technology in understanding underground processes, providing an unprecedentedly accurate data foundation for intelligent decision-making and realizing the visibility, knowability, and predictability of the entire process.

[0023] 3. This invention possesses forward-looking decision-making and adaptive optimization capabilities, enabling a shift from passive response to proactive intervention. Through the introduction of an LSTM prediction model and a PPO reinforcement learning algorithm, the system can not only describe the current state but also predict production dynamics and potential risks in the coming hours, thereby formulating optimal control strategies in advance. This forward-looking, self-learning, and adaptive capability ensures that the system can proactively adapt to the dynamic changes in the reservoir, always maintaining optimal operating conditions.

[0024] 4. This invention establishes a multi-level active safety barrier to ensure long-term geological safety during the storage process. By linking real-time monitoring data, digital twin inversion results, and preset safety thresholds, a complete safety chain is constructed, from early risk identification and intelligent warning to automatic emergency response. Once risks such as sudden pressure changes, sudden increases in concentration, or micro-vibration anomalies are detected, the system can activate graded interlocking protection (such as pressure reduction and gas source switching) within milliseconds, achieving a leap from passive protection to active safety assurance, and greatly reducing leakage and geological risks in the storage environment. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the overall structure of the present invention; Figure 2 This is the logic flowchart of the intelligent control system.

[0026] In the diagram: 1-Rock strata; 2-Coal seam; 3-Hot flue gas storage tank; 4-CO2 gas storage tank; 5-N2 gas storage tank; 6-Intelligent proportional valve group; 7-Variable frequency gas injection pump; 8-Humidity conditioning and dust removal system; 9-Heat exchanger group; 10-Injection well data transmission line; 11-Gas injection pipeline; 12-Injection well; 13-Gas injection plug; 14-Injection well integrated multi-parameter monitoring device; 15-Drainage-monitoring combined well data transmission line; 16-Drainage-monitoring combined well integrated multi-parameter monitoring device; 17-Drainage plug; 18-Drainage pipeline; 19-Drainage-monitoring combined well; 20-Coalbed methane extraction pump; 21-Heat exchanger group; 22-Gas filtration device; 23-Gas separation device; 24-CH4 gas storage tank; 25-Ground control center. Detailed Implementation

[0027] The present invention will be further described in detail below with reference to specific embodiments.

[0028] This invention integrates intelligent proportional valve groups, digital twin models, and AI algorithms, enabling dynamic perception of downhole parameters, real-time forecasting of production and risk, and automatic execution of control commands. This drives the system to upgrade from a "passive response" to a "proactive optimization" mode, achieving optimal synergy between CH4 production enhancement, CO2 storage, and operating energy consumption, and ensuring proactive safety management of the storage process throughout its entire lifecycle.

[0029] A smart control system for coalbed methane extraction and CO2 storage using variable-component flue gas includes: a variable-component injection module, a multi-parameter distributed sensor network module, an extraction and utilization module, a digital twin model module, an AI intelligent control engine module, and a safety interlock and emergency response module. Upon system startup, the variable-component injection module begins injecting industrial flue gas according to preset initial parameters (e.g., CO2 / N2 = 7:3, medium temperature and pressure). The multi-parameter distributed sensor network module collects initial data from injection well 12 and the combined extraction-monitoring well 19 in real time, transmitting comprehensive data such as pressure, temperature, concentration, and microseismic activity to the digital twin model module. The digital twin model module performs real-time simulation based on the mechanistic model, outputting the current permeability field, pressure field, CO2 plume field, and CH4 saturation state to the AI ​​intelligent control engine module. The AI ​​intelligent control engine module uses LSTM to predict the dynamics over the next few hours, calculates the optimal injection parameters using the PPO algorithm, and issues optimal control commands, including composition, pressure, flow rate, temperature, and humidity, to the variable-component injection module. The variable-component injection module, through the coordinated action of intelligent proportional valve groups, temperature control units, humidifiers, and variable-frequency injection pumps, precisely adjusts the flue gas to the target state and injects it underground. The new injection strategy alters the underground flow and reaction processes, generating new sensor data. The system repeats the above steps in a 1-hour cycle, dynamically adapting to reservoir changes and always pursuing the optimal goal. The extraction and utilization module is used to separate, purify, and store CH4 gas after extraction to prevent the effects of leakage during the extraction process. The safety interlock and emergency response module, along with the variable component injection module, multi-parameter distributed sensor network module, digital twin model module, AI intelligent control engine module, and safety interlock and emergency response module, operate in parallel and independently throughout the entire process. It scans the key risk indicators of the sensor network 24 / 7, and the module is activated immediately if and only if any indicator is detected to exceed the preset safety threshold.

[0030] Furthermore, the safety interlock and emergency response module reads downhole pressure, monitors well CO2 concentration and microseismic energy in real time, and immediately triggers an emergency action when any of the following conditions are met: a) Pressure drop rate > 1 MPa / 10 min; b) Daily increase in CO2 concentration in monitoring wells > 2%vol; c) Single-event microseismic energy > 10 4 J; Emergency actions include: a) Close the injection valve slightly to reduce the flow rate below the safe threshold; b) Switch the gas source to high N2 ratio mode, where CO2:N2 ≤ 3:7 or ≤ 1:9; c) Force the injection pressure to ≤5 MPa; d) Activate the groundwater chemical monitoring unit to collect data on pH, conductivity, and dissolved CO2; e) Encrypt the operation logs and upload them to the blockchain audit platform, and simultaneously send risk alerts to the regulatory platform.

[0031] Furthermore, the variable component injection module includes: hot flue gas storage tank 3, CO2 gas storage tank 4, N2 gas storage tank 5, intelligent proportional valve group 6, variable frequency gas injection pump 7, humidity conditioning and dust removal system 8, heat exchanger group 9, and gas injection pipeline 11. Specifically, the hot flue gas storage tank 3 is connected to the CO2 gas storage tank 4 through the intelligent proportional valve group 6. The CO2 gas storage tank 4 is connected to the N2 gas storage tank 5 through the intelligent proportional valve group 6. The N2 gas storage tank 5 is connected to the variable frequency gas injection pump 7 through the intelligent proportional valve group 6. The variable frequency gas injection pump 7 sequentially inputs the gas into the humidification and dust removal system 8 and the heat exchanger group 9, and inputs it into the injection well 12 through the gas injection pipeline 11. The gas injection pipeline 11 and the injection well 12 sequentially pass through the rock stratum 1 and the coal seam 2. The gas injection pipeline 11 is placed in the injection well 12, which is L-shaped. Considering the gas injection efficiency, a gas injection plug 13 is installed between the rock stratum 1 and the coal seam 2 in the injection well 12.

[0032] Furthermore, the multi-parameter distributed sensor network module includes: an injection well integrated multi-parameter monitoring device 14, an injection well data transmission line 10, a ground control center 25, a pumping-monitoring combined well integrated multi-parameter monitoring device 16, and a pumping-monitoring combined well data transmission line 15; wherein, the injection well data transmission line 10 and the pumping-monitoring combined well data transmission line 15 are 5G / LoRa links; the injection well integrated multi-parameter monitoring device 14 and the pumping-monitoring combined well integrated multi-parameter monitoring device 16 are FBG / TDLAS / microseismic / AE sensors; Specifically, the injection well integrated multi-parameter monitoring device 14 is installed in the injection well 12. The injection well integrated multi-parameter monitoring device 14 synchronously collects injection parameters such as pressure P, temperature T, flow rate Q, CO2 / N2 / CH4 concentration, AE sound wave, micro-vibration, and geothermal gradient at a high frequency of 1Hz through the injection well data transmission line 10 and transmits them to the ground control center 25. The extraction-monitoring combined well integrated multi-parameter monitoring device 16 is installed in the extraction-monitoring combined well 19. The extraction-monitoring combined well integrated multi-parameter monitoring device 16 transmits permeability, temperature, pressure, adsorption / desorption parameters, etc. during the extraction process to the ground control center 25 through the extraction-monitoring combined well data transmission line 15. The extraction-monitoring combined well 19 passes through rock stratum 1 and coal seam 2, and an extraction plug 17 is installed between rock stratum 1 and coal seam 2.

[0033] Furthermore, the extraction and utilization module includes: extraction pipeline 18, coalbed methane extraction pump 20, heat exchanger group 21, gas filtration device 22, gas separation device 23 and CH4 gas storage tank 24. Specifically, considering the impact of leakage during the extraction process, a gas injection plug 17 is placed in the extraction-monitoring combined well 19. The extracted gas is pumped from the extraction pipeline 18 by the coalbed methane extraction pump 20, cooled by the heat exchanger group 21, impurity gas is removed by the gas filtration device 22, and CH4 gas is separated by the gas separation device 23 before entering the CH4 gas storage tank 24 for subsequent use.

[0034] Furthermore, the digital twin model module includes: a permeability evolution model, an adsorption saturation model, and a fracture conductivity model to invert key parameters of the coal seam in real time, outputting permeability field, pressure field, CO2 plume distribution, and CH4 desorption parameters; the AI ​​intelligent control engine module includes an LSTM prediction algorithm and a PPO reinforcement learning algorithm, which predicts the average pressure, CH4 production, and CO2 breakthrough time for the next few hours through AI algorithms, and combines reinforcement learning with "CH4 increment + CO2 sequestration rate - energy consumption" as the reward function to output optimal injection parameters (flue gas injection pressure, flow rate, composition), updates the injection strategy every 1 hour, and issues instructions to the variable composition module; the digital twin model module and the AI ​​intelligent control engine module are integrated on the ground control center 25; in: The permeability evolution model comprehensively considers stress-sensitive effects, matrix shrinkage / expansion effects, shear slip, and chemical dissolution. Its relationship is as follows:

[0035] In the formula, m is the initial permeability. 2 ; The effective stress function; The matrix strain function; It is a chemical solubility correction factor; The adsorption saturation model, based on the extended Langmuir competitive adsorption model, describes the ternary competitive adsorption behavior of CO2-CH4-N2, and its relationship is as follows:

[0036] In the formula, Components Adsorption saturation; Components Langmuir constant, MPa -1 ; Components The partial pressure, MPa; Components Langmuir constant, MPa -1 ; Components The partial pressure, MPa; This indicates summing the three gases; The fracture conductivity model combines the cubic law and shear displacement effect, considering the changes in conductivity caused by fracture surface roughness and chemical dissolution. The relationship is as follows:

[0037] In the formula, The current crack porosity; Initial crack porosity; The current crack width is in meters (m). Let the initial crack width be m; This is the shear displacement correction factor; It is a function of chemical dissolution time; The LSTM prediction algorithm takes at least 30 days of historical downhole pressure, temperature, CH4 / CO2 concentration and gas production time series as input, and outputs the average pressure distribution, CH4 production and CO2 breakthrough time prediction for the next 6 hours. The PPO reinforcement learning algorithm has a state space consisting of average pressure, CH4 concentration, CO2 plume leading edge position, and microseismic energy. Its action space consists of flue gas injection pressure (0-25 MPa), flue gas flow rate (0.1-10 t / h), and CO2 / N2 volume ratio (30%-70%). Its reward function is:

[0038] in, The rate of increase in production compared to the previous cycle, m 3 / h; CO2 sequestration rate within a single cycle, % Energy consumption of the injection pump and heating unit, MWh; These are the first, second, and third weighting coefficients, which can be adjusted.

[0039] Specifically, based on pressure / temperature field data, gas concentration data, microseismic / acoustic data, and injection / extraction data collected by the multi-parameter distributed sensor network module, the digital twin model is input. The model calculates the permeability evolution model, adsorption saturation model, and fracture conductivity model, and outputs key state fields (average pressure distribution field, temperature distribution field), component distribution fields (CO2 saturation field, CH4 desorption front location, residual CH4 saturation field), and property evolution fields (dynamic permeability field, fracture network development degree). Based on the state characterization output by the digital twin model, the data is input into the LSTM spatiotemporal prediction model to predict the reservoir state in the next few hours. The algorithm uses pressure field, plume morphology, and key performance indicators (CH4 instantaneous production, wellhead CO2 concentration, etc.) to output the predicted state for future periods. The output data is then used as input data for PPO reinforcement learning optimization. The PPO algorithm continuously updates the parameters of the policy network and value network. Its core is to maximize the reward function and output the predicted CH4 production curve, CO2 sequestration rate curve, breakthrough time warning, and optimal dynamic injection strategy to adjust the strategy in advance to prevent premature CO2 breakthrough or pressure exceedance. It balances multiple sometimes conflicting goals of "increased production, sequestration, energy saving, and safety" and finds the optimal dynamic injection strategy to achieve precise control.

[0040] Furthermore, after receiving the action command output by the AI ​​intelligent control engine, the variable component injection module adjusts the wellhead flue gas injection pressure to the target value ±0.1 MPa within 1 minute through the servo valve group, and controls the CO2 / N2 ratio within the set value ±2%.

[0041] Furthermore, the multi-parameter distributed sensor network module transmits data via downhole FBG optical cables and integrates at least the following: a) FBG sensor, used to acquire downhole pressure and temperature at a fixed point; b) Laser TDLAS analyzer, used for real-time measurement of CH4 and CO2 concentrations in fluids; c) Microseismic detector, frequency response 0.1 Hz-10 kHz, used to record microseismic events induced by fracturing or production; d) Electromagnetic flowmeter, used for online measurement of the volume of injected / output fluid; e) Distributed acoustic wave DAS system, which uses downhole FBG optical cable as sensor array to realize continuous acquisition of acoustic wave signals throughout the well section; f) Distributed fiber optic temperature measurement (DTS) system, which completes distributed temperature field monitoring in the same FBG optical cable; g) Gas chromatography microcompartment, with an analytical accuracy of ±1%, used for timed and quantitative analysis of fluid components; h) Groundwater chemical sensors, which at least detect pH and CO2 fugacity; All the above sensor data are transmitted back to the ground control center 25 via FBG optical cable, realizing multi-parameter joint monitoring of the entire injection-extraction process.

[0042] A smart control method for coalbed methane extraction and CO2 sequestration using variable-component flue gas includes the following steps: S1. System startup, self-test and initialization; Start the system and power on the variable component injection module, multi-parameter distributed sensor network, digital twin model, extraction and utilization module, AI intelligent control engine and safety interlock module; verify the integrity of the intelligent proportional valve group 6, variable frequency gas injection pump 7, humidity and dust removal system 8, heat exchanger group 9, injection well integrated multi-parameter monitoring device 14 and extraction-monitoring joint well integrated multi-parameter monitoring device 16 (FBG / TDLAS / micro-vibration / AE sensor), injection well data transmission line 10 and extraction-monitoring joint well data transmission line 15 (5G / LoRa link), edge computing node and blockchain audit interface. If there is an abnormality, an alarm will be triggered and the system will be shut down; otherwise, proceed to step 2. The system reads real-time downhole data through a multi-parameter distributed sensor network deployed in injection wells, extraction wells, and monitoring wells. The real-time data includes, but is not limited to, bottom hole pressure, temperature, gas component concentration (CO2, CH4, and N2), microseismic signals, and injection flow rate. The data is transmitted to the central processing platform via a 5G / LoRa network. S2. Real-time data acquisition and data preprocessing; Synchronous acquisition at 1Hz high frequency: pressure, temperature, flow rate, CO2 / N2 / CH4 concentration, AE sound waves, micro-vibration, and geothermal gradient; The edge cache stores a 5-minute sliding window of data. The ground control center performs data cleaning and outlier removal on the received real-time data, generating a cleaned data packet D. t Data cleaning and outlier removal specifically involve: using statistical methods (such as the 3σ criterion) to identify and filter out abnormal data points caused by instantaneous sensor failures or signal transmission interference, and using interpolation to fill in missing values ​​to ensure the quality and reliability of the data input to the downstream model.

[0043] S3, the digital twin model module is updated on a rolling basis; The cleaned data packet D obtained in S2 t The digital twin model is input with historical drilling, logging, and fracturing data. The permeability evolution model, ternary competitive adsorption saturation model, and fracture conductivity model are invoked to invert and obtain the current full-field permeability field, pressure field, CO2 plume front, and CH4 desorption degree. The twin state is updated on a rolling basis with a 1-hour cycle as the initial field for subsequent prediction and optimization. This step uses real-time data to assimilate the model, dynamically updates and reflects the most realistic state of the current reservoir, and outputs key parameters including permeability field, pressure field, spatial distribution of CO2 plume and CH4 adsorption saturation.

[0044] S4, AI intelligent control engine closed-loop decision-making, including: S4.1 State Prediction (LSTM Model): Input the past ≥30 days of clean data and the latest twin state, output the average pressure distribution, CH4 production sequence and CO2 breakthrough time for the next 6 hours; This LSTM model, trained on historical data, can accurately predict the evolution of average pressure, the trend of CH4 production, and the breakthrough time of CO2 in the extraction well over the next 6 hours. S4.2, PPO reinforcement learning optimization: The key indicators of the twin at time t (mean pressure, CH4 concentration, CO2 plume location, and accumulated microseismic energy) are encoded into a state vector S. t Construct the state space; The action space is defined by the injection pressure P_{inj}, injection flow rate Q_{inj}, and CO2 / N2 ratio R_{CO2:N2}, denoted as action space A. t ={P_{inj},Q_{inj},R_{CO2:N2}}, where P_{inj}∈[0,25]MPa, Q_{inj}∈[0.1,10]t / h, R_{CO2:N2}∈[30%,70%]; Measure ΔCH4, CO2 sequestration rate, and energy consumption E, calculate the immediate reward according to the reward function, and output the optimal action vector; The reward function of the algorithm is constructed with the core objective of "CH4 production increase benefit + CO2 storage environmental benefit - system injection energy consumption". The PPO algorithm interacts with the environment (i.e., digital twin model) to solve the optimal control strategy that maximizes the cumulative reward. This strategy is specified as the set values ​​of the injected flue gas composition (CO2 / N2 ratio), injection pressure, injection temperature and injection flow rate for the next cycle. S4.3, Safety Constraint Verification: If the following occurs: The rate of pressure drop at the bottom of the well, P_{drop}, is greater than 1 MPa / 10 min. CO2 increase > 2% / day; The energy released by a single microseismic event is E_{seis}>10 4 J; Then proceed to S5; otherwise, use PPO shearing importance sampling and gradient ascent to update network parameters, complete policy update, and then issue instructions to parse the optimal action vector into specific set values ​​and issue them within <500ms. The objects to be issued include: intelligent proportional valve group → adjust CO2 / N2 ratio; variable frequency injection pump → adjust P_{inj}, Q_{inj}; temperature control unit → adjust flue gas temperature 150-400℃; humidifier → adjust moisture content 0-20%; finally, wait for a 1-hour rolling cycle and return to step 2 to form a closed loop.

[0045] S5, Safety Interlocks and Emergency Response; When any security constraint trigger condition of S4.3 security constraint verification is met, the security instruction is immediately initiated: ① Close the injection valve slightly to reduce the flow rate to ≤20% of the maximum flow rate; ② Switch the gas source to high N2 mode (CO2:N2≤3:7 or≤1:9); ③ Force the injection pressure to be reduced to ≤5MPa; ④ Activate the groundwater chemical monitoring unit to collect data on pH, conductivity, and dissolved CO2; ⑤ Encrypt the operation logs and upload them to the blockchain audit platform, and simultaneously send risk alerts to the regulatory platform; Simultaneously, the temperature control and humidification units are suspended, and then the groundwater monitoring subunit is activated to collect pH, conductivity, and dissolved CO2. The emergency action sequence, timestamps, and original sensor values ​​are packaged and written into the blockchain audit block. Risk alarms and earthquake impact assessment requests are pushed to the regulatory platform through the 5G link to continuously monitor risk indicators. If the data is below the threshold for 30 consecutive minutes and the monitoring platform confirms it remotely, exit the safety mode and return to S4; otherwise, remain in safety mode until manually reset. All raw data throughout the lifecycle, twin inversion results, AI strategy logs, and emergency records are encrypted and stored in a distributed database. Before issuing the command, the system performs a final security check, including: S5.1 Determine whether the optimal strategy parameters obtained by the PPO algorithm meet all preset safety constraints (e.g., whether the bottom hole pressure is lower than the caprock fracture pressure threshold, and whether the predicted microseismic energy is within the safe range). S5.2 Safe Path: If safety constraints are not met, the system will not execute dangerous commands, but will immediately activate the safety interlock and emergency response module, switching to a degraded operation mode. This mode includes: automatically reducing the injection pressure, increasing the proportion of N2 in the flue gas to reduce risk, and sending an alarm to the superior monitoring system; S5.3 Normal Path: If the strategy meets all safety constraints, the system will send the generated optimization instructions (including variable component instructions, pressure setpoints, temperature setpoints, and flow setpoints) to the variable component intelligent injection module.

[0046] S6, Instruction Execution and Closed-Loop Feedback; After receiving the command, the variable-component intelligent injection module drives the intelligent proportional valve group to precisely adjust the ratio of CO2 to N2; the temperature control unit and humidification unit adjust the flue gas to the target temperature and humidity; the variable-frequency injection pump executes the injection operation according to the set flow rate and pressure. Afterwards, the system returns to S1 to read the next round of downhole data after injection, thus forming a continuously iterative, self-optimizing closed-loop intelligent control cycle. This cycle typically operates on a 1-hour control period, adapting in real-time to the dynamic changes in the reservoir.

[0047] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A smart control system for coalbed methane extraction and CO2 sequestration using variable-component flue gas, characterized in that, include: Variable component injection module, multi-parameter distributed sensor network module, extraction and utilization module, digital twin model module, AI intelligent control engine module, safety interlock and emergency response module; The variable component injection module injects flue gas into the well according to a preset initial value. The multi-parameter distributed sensor network module collects the well data and transmits it to the digital twin model module for calculation. The calculation results are input into the AI ​​intelligent control engine module to issue the best control command. The extraction and utilization module is used for the separation, purification, and storage of CH4 gas after extraction; The safety interlock and emergency response module, along with the variable component injection module, multi-parameter distributed sensor network module, digital twin model module, and AI intelligent control engine module, operate in parallel and independently throughout the entire process. The module is activated immediately when any indicator is detected to exceed the preset safety threshold. The indicator is generated by the AI ​​intelligent control engine module.

2. The intelligent control system for coalbed methane extraction and CO2 sequestration using variable-component flue gas according to claim 1, characterized in that: The variable component injection module includes: a hot flue gas storage tank (3), a CO2 gas storage tank (4), an N2 gas storage tank (5), an intelligent proportional valve group (6), a variable frequency gas injection pump (7), a humidity control and dust removal system (8), a heat exchanger group (9), and a gas injection pipeline (11); wherein, the hot flue gas storage tank (3) is connected to the CO2 gas storage tank (4), the N2 gas storage tank (5), and the variable frequency gas injection pump (7) in sequence through the intelligent proportional valve group (6); the variable frequency gas injection pump (7) inputs the gas into the humidity control and dust removal system (8) and the heat exchanger group (9) in sequence, and inputs it into the injection well (12) through the gas injection pipeline (11); the injection well (12) passes through the rock strata (1) and the coal seam (2) in sequence, and the gas injection pipeline (11) is placed in the injection well (12), and the injection well (12) is L-shaped; a gas injection plug (13) is provided between the rock strata (1) and the coal seam (2).

3. The intelligent control system for coalbed methane extraction and CO2 sequestration using variable-component flue gas according to claim 1, characterized in that: The multi-parameter distributed sensor network module includes: an injection well integrated multi-parameter monitoring device (14), an injection well data transmission line (10), a ground control center (25), a extraction-monitoring combined well integrated multi-parameter monitoring device (16), and an extraction-monitoring combined well data transmission line (15); wherein, the injection well integrated multi-parameter monitoring device (14) is installed in the injection well (12), and the injection well integrated multi-parameter monitoring device (14) collects and transmits data at a high frequency of 1Hz to the ground control center (25) through the injection well data transmission line (10); the extraction-monitoring combined well integrated multi-parameter monitoring device (16) is installed in the extraction-monitoring combined well (19), and transmits the data during the extraction process to the ground control center (25) through the extraction-monitoring combined well data transmission line (15); the extraction-monitoring combined well (19) passes through the rock strata (1) and the coal seam (2), and an extraction plug (17) is provided between the rock strata (1) and the coal seam (2).

4. The intelligent control system for coalbed methane extraction and CO2 sequestration using variable-component flue gas according to claim 1, characterized in that: The extraction and utilization module includes: an extraction pipeline (18), a coalbed methane extraction pump (20), a heat exchanger group (21), a gas filtration device (22), a gas separation device (23), and a CH4 gas storage tank (24); wherein, the extracted gas is pumped from the extraction pipeline (18) by the coalbed methane extraction pump (20), cooled by the heat exchanger group (21), impurities are removed by the gas filtration device (22), and CH4 gas is separated by the gas separation device (23) before entering the CH4 gas storage tank (24) for subsequent use.

5. The intelligent control system for coalbed methane extraction and CO2 sequestration using variable-component flue gas according to claim 3, characterized in that: The ground control center (25) integrates a digital twin model module and an AI intelligent control engine module.

6. The intelligent control system for coalbed methane extraction and CO2 sequestration using variable-component flue gas according to claim 1 or 5, characterized in that: The digital twin model module includes: a permeability evolution model, an adsorption saturation model, and a fracture conduction model.

7. The intelligent control system for coalbed methane extraction and CO2 sequestration using variable-component flue gas according to claim 1 or 5, characterized in that: The AI ​​intelligent control engine module includes: LSTM prediction algorithm and PPO reinforcement learning algorithm.

8. The intelligent control system for coalbed methane extraction and CO2 sequestration using variable-component flue gas according to claim 6, characterized in that: The expression for the permeability evolution model is: In the formula, Initial penetration rate; The effective stress function; The matrix strain function; It is a chemical solubility correction factor; The expression for the adsorption saturation model is: In the formula, Components Adsorption saturation; Components Langmuir constant; Components The partial voltage; Components Langmuir constant; Components The partial voltage; This indicates summing the three gases; The expression for the fracture conduction model is: In the formula, The current crack porosity; Initial crack porosity; This represents the current crack width; This represents the initial crack width; This is the shear displacement correction factor; This is a function of chemical dissolution time.

9. The intelligent control system for coalbed methane extraction and CO2 sequestration using variable-component flue gas according to claim 7, characterized in that: The LSTM prediction algorithm takes a historical downhole pressure, temperature, CH4 / CO2 concentration and gas production time series of no less than 30 days as input, and outputs the average pressure distribution, CH4 production and CO2 breakthrough time prediction for the next 6 hours. The PPO reinforcement learning algorithm has a state space consisting of average pressure, CH4 concentration, CO2 plume location, and accumulated microseismic energy, and an action space consisting of flue gas injection pressure (0-25 MPa), flue gas flow rate (0.1-10 t / h), and CO2 / N2 volume ratio (30%-70%). Its reward function is: in, The rate of increase in production compared to the previous cycle; CO2 sequestration rate within a single cycle; Energy consumption for injection pump and heating unit; These are the first weighting coefficient, the second weighting coefficient, and the third weighting coefficient.

10. A method for intelligent control of coalbed methane extraction and CO2 sequestration using variable-component flue gas, used to implement the intelligent control system for coalbed methane extraction and CO2 sequestration using variable-component flue gas as described in claims 1-9, characterized in that, Includes the following steps: S1. System startup, self-test and initialization: Start the system and power on the variable component injection module, multi-parameter distributed sensor network, digital twin model, extraction and utilization module, AI intelligent control engine and safety interlock module; verify the integrity of the intelligent proportional valve group (6), variable frequency gas injection pump (7), humidity and dust removal system (8), heat exchanger group (9), injection well integrated multi-parameter monitoring device (14), extraction-monitoring joint well integrated multi-parameter monitoring device (16), injection well data transmission line (10), extraction-monitoring joint well data transmission line (15), edge computing node and blockchain audit interface. If abnormal, alarm and stop the system; otherwise, enter S2. S2. Real-time data acquisition and data preprocessing; Synchronous acquisition at 1Hz high frequency: pressure, temperature, flow rate, CO2 / N2 / CH4 concentration, AE sound waves, micro-vibration, and geothermal gradient; The edge cache is used to process 5-minute sliding window data. The ground control center (25) performs data cleaning and outlier removal on the received real-time data to generate a cleaned data packet D. t ; S3, the digital twin model module is updated on a rolling basis; The cleaned data packet D obtained in S2 t By inputting historical drilling, logging, and fracturing data into the digital twin model, and calling the permeability evolution model, adsorption saturation model, and fracture conductivity model, the current full-field permeability field, pressure field, CO2 plume distribution, and CH4 desorption degree are obtained through inversion. The twin state is updated on a rolling basis with a period of 1 hour, serving as the initial field for subsequent prediction and optimization; S4, AI intelligent control engine closed-loop decision-making, including: S4.1 LSTM Model State Prediction: Input cleaning data for the past ≥30 days and the latest twin state, output the average pressure distribution, CH4 production and CO2 breakthrough time for the next 6 hours; S4.2, PPO reinforcement learning optimization: The key indicators of the twin at time t—mean pressure, CH4 concentration, CO2 plume location, and accumulated microseismic energy—are encoded into a state vector S. t Construct the state space; The action space is defined by the injection pressure P_{inj}, injection flow rate Q_{inj}, and CO2 / N2 ratio R_{CO2:N2}, denoted as action space A. t ={P_{inj},Q_{inj},R_{CO2:N2}}, where P_{inj}∈[0,25]MPa, Q_{inj}∈[0.1,10]t / h, R_{CO2:N2}∈[30%,70%]; Calculate the immediate reward according to the reward function and output the optimal action vector; S4.3, Safety Constraint Verification: If the following occurs: The rate of pressure drop at the bottom of the well, P_{drop}, is greater than 1 MPa / 10 min. CO2 increase > 2% / day; The energy released by a single microseismic event is E_{seis}>10 4 J; Then proceed to S5; otherwise, use PPO shearing importance sampling and gradient ascent to update network parameters, complete strategy update, and then issue instructions to parse the optimal action vector into specific set values ​​and issue them within <500ms. The objects to be issued include: intelligent proportional valve group (6) to adjust CO2 / N2 ratio; variable frequency gas injection pump (7) to adjust P_{inj} and Q_{inj}; humidity and dust removal system (8) to adjust moisture content 0-20%; heat exchanger group (9) to adjust flue gas temperature 150-400℃; finally, wait for 1 hour of rolling cycle and return to S2 to form a closed loop. S5, Safety Interlocks and Emergency Response; When any security constraint trigger condition of S4.3 security constraint verification is met, the security instruction is immediately initiated: ① Close the injection valve slightly to reduce the flow rate to ≤20% of the maximum flow rate; ② Switch the gas source to high N2 mode. The switching conditions include: CO2:N2 ≤ 3:7 or ≤ 1:9; ③ Force the injection pressure to be reduced to ≤5MPa; ④ Activate the groundwater chemical monitoring unit to collect data on pH, conductivity, and dissolved CO2; ⑤ Encrypt the operation logs and upload them to the blockchain audit platform, and simultaneously send risk alerts to the regulatory platform; Before issuing the command, the system performs a final security check, including: S5.1 Determine whether the optimal policy parameters obtained by the PPO algorithm satisfy all preset security constraints; S5.2 If the safety constraints are not met, the system will not execute the dangerous command, but will immediately activate the safety interlock and emergency response module and switch to the degraded operation mode; S5.3 If the strategy satisfies all security constraints, the system will send the generated optimization instructions to the variable component intelligent injection module. S6, Instruction Execution and Closed-Loop Feedback; After receiving the instruction, the variable component intelligent injection module drives the intelligent proportional valve group (6) to adjust the ratio of CO2 and N2, the humidity and dust removal system (8), and the heat exchanger group (9) to adjust the flue gas to the target temperature and humidity. The variable frequency gas injection pump (7) performs the injection operation according to the set flow rate and pressure. After that, it returns to S1 to read the new round of downhole data after injection, thus forming a continuously iterative and self-optimizing closed-loop intelligent control cycle. This cycle has a control cycle of 1 hour and adapts to the dynamic changes of the reservoir in real time.

Images