A quaternary phosphonium amino acid ester phase change ionic liquid and AI4AI delamination agent cooperative carbon capture integrated system

Abstract

The application discloses a quaternary phosphonium amino acid ester phase change ionic liquid and an AI4AI layered agent cooperative carbon capture integrated system, and belongs to the cross field of CCUS wet carbon capture and intelligent control. The application adopts a low-viscosity thermal phase change quaternary phosphonium amino acid ester absorbent, and only regenerates a CO2-rich liquid phase to reduce energy consumption; a central AI4AI four-layer unsupervised cooperative base and a three-level agent cluster are matched, and without labeling working conditions, the system can be self-adaptive to fluctuating flue gas, can diagnose solvent aging on line, and cancels sampling during shutdown. The system has a regeneration energy consumption of 1.45-1.60 GJ / tCO2, and the service life of the solvent is increased by 30%, can be adapted to intelligent transformation of an existing amine method device, and effectively solves the industry technical pain points of high energy consumption, large solvent loss and poor working condition adaptability of the traditional wet capture.

Description

Technical Field

[0001] This invention belongs to the cross-disciplinary technical fields of CCUS (Carbon Capture, Utilization and Storage) industrial flue gas carbon capture, novel absorption solvents, and multi-agent collaborative control using artificial intelligence. It is specifically applicable to wet carbon capture of low-concentration dust-containing industrial flue gas (8%~15%) from steel sintering, cement kilns, and coal-fired power plants. It innovatively integrates original quaternary phosphonium-based phase change ionic liquid materials with AI4AI (Artificial Intelligence for Agent Interaction) centralized four-layer peer-to-peer unsupervised collaborative base and three-level hierarchical Agent (artificial intelligence agent) global control architecture. It belongs to a complete set of wet CCUS technologies for medium- and long-term intelligent large-scale application and has no technical overlap with membrane separation and solid-state electrochemical pump technologies. Background Technology

[0002] CCUS (Carbon Capture, Utilization, and Storage) is a key supporting technology for achieving carbon neutrality in energy-intensive industries such as steel, thermal power, and cement. Currently, the mainstream industrialization process for carbon capture of low-concentration industrial flue gas is mainly based on organic amine chemical absorption, which suffers from high solvent evaporation losses, severe equipment corrosion, and persistently high regeneration energy consumption.

[0003] To overcome the shortcomings of the amine method, academia and industry have been continuously developing novel ionic liquid absorption solvents. Current research focuses on imidazole and pyridine cationic systems. Meanwhile, some studies have introduced shallow machine learning algorithms to optimize single-point parameters for carbon capture devices.

[0004] Most currently available technical solutions are limited to solvent molecule modification or independent adjustment of single equipment parameters, and have not yet formed an integrated solution encompassing material systems, complete processes, and comprehensive intelligent control. Traditional fixed-parameter control methods have significant bottlenecks: ① Conventional ionic liquids have a viscosity of over 80 mPa·s (mPa·s, unit of dynamic viscosity of liquid) at 25℃. When combined with CO2 (carbon dioxide), the viscosity of the system increases further, resulting in high mass transfer resistance between the gas and liquid phases. The energy consumption for system regeneration is generally no less than 2.0 GJ / tCO2 (gigajoules per ton of carbon dioxide, unit of energy consumption). ② The existing ionic liquid absorption process lacks a thermally induced phase separation mechanism. Both the CO2-rich absorbent and the lean absorbent enter the regeneration tower for heating and desorption. A large amount of heat is used to heat the unloaded lean absorbent, and the ineffective energy consumption accounts for nearly 40%. ③ The components of industrial flue gas exhibit strong time-varying characteristics. The concentration of CO2, SO2 (sulfur dioxide) impurities, and dust load fluctuate continuously from day to night. Relying on artificial static parameters for operation, the system's collection efficiency can fluctuate and decline by 10% to 18%. ④ Traditional AI control models require large-scale labeled operating condition datasets for training. The debugging and deployment cycle of newly built industrial CCUS projects can be as long as 3 to 6 months. There is a lack of unsupervised intelligent control architectures that do not require labeled samples and can iterate autonomously after going online.

[0005] It is evident that existing technologies cannot simultaneously achieve closed-loop operation of three aspects: online prediction of solvent aging status, unified scheduling of low-grade waste heat resources throughout the plant, and intelligent collaborative optimization of the entire carbon capture process. This limits the large-scale, long-term, and stable industrialization of ionic liquid carbon capture technology. Summary of the Invention

[0006] 3.1 Purpose of the Invention An original low-viscosity quaternary phosphonium amino acid ester bifunctional phase change ionic liquid, which solves the material shortcomings of traditional ionic liquids such as high viscosity, easy poisoning, and high regeneration energy consumption; A top-down, three-tiered Agent (artificial intelligence agent) global control cluster is built, coupled with a self-developed AI4AI (intelligent agent mutual collaboration framework) centralized four-layer peer-to-peer unsupervised collaborative base, to achieve zero-sample deployment without the need for manual labeling of working conditions; By leveraging AI to predict flue gas fluctuations in real time and quantify solvent degradation online, manual sampling and testing during shutdowns can be eliminated, reducing production capacity losses. The AI4AI base coordinates the waste heat management of the entire unit, including the absorption tower, phase separation tank, and regeneration tower, reducing the system's regeneration energy consumption to 1.45~1.60 GJ / tCO2; It adapts to the full life cycle conditions of factory peak shaving, cross-seasonal temperature fluctuations, and production line expansion, and its continuous operation control accuracy has been improved year by year. The complete system can retrofit existing traditional MEA (monoethanolamine, monoethanolamine absorbent) / MDEA (methyldiethanolamine, alkanolamine absorbent) amine capture devices, reducing hardware retrofit costs by more than 35%.

[0007] 3.2 Complete Integrated Technical Solution Module 1: Original Quaternary Phosphorus Amino Acid Ester Thermoinduced Phase Change Ionic Liquid Material Cationic: Tetrabutylphosphonium [P4444]⁺; Anionic: Amino acid ethyl ester [AA-EA]⁻; One-step esterification-metathesis mild synthesis, yield ≥95%, purity ≥99%; Key physicochemical properties (25℃): viscosity ≤45mPa・s, CO2 saturated adsorption capacity ≥1.8mol / mol (moles per mole, adsorption capacity unit); absorption at 40~60℃, thermally induced phase separation temperature 42℃, CO2-rich phase volume stability 28%~32%, only the liquid-rich phase is sent to the regeneration tower; withstands ≤200ppm (parts per million, gaseous impurity concentration unit) SO2 flue gas, annual solvent loss ≤0.32%, service life ≥5.2 years.

[0008] Module 2: Integrated Hardware Process Unit for Wet Carbon Capture Flue gas pretreatment dust removal and desulfurization tower → countercurrent packed absorption tower → constant temperature phase change phase separation tank → negative pressure regeneration and analysis tower → lean and rich liquid heat exchanger → plant-wide low-grade waste heat heat exchange pipeline network → solvent circulation pump group; 15 types of online sensor arrays are deployed throughout the process, uploading flue gas concentration, solvent viscosity, temperature, and energy consumption time series data to the AI4AI base every second.

[0009] Module 3: AI4AI Centralized Four-Layer Peer-to-Peer Unsupervised Collaborative Base + Three-Level Hierarchical Agent Cluster Agent hierarchy (from top to bottom): Level 1: Global coordination Agent; Middle layer: Three specialized agents: flue gas prediction, solvent diagnosis, and waste heat management; Bottom layer: Single device execution Agent; AI4AI's core architecture features a four-layer intelligent agent bidirectional peer-to-peer interactive clustering framework that can be started without manually labeled samples; it automatically clusters and classifies high / low / transitional load conditions every 24 hours, dynamically adjusting the decision weights of each agent; and it embeds hard constraints on temperature, pressure, and traffic at the bottom layer to prevent AI from exceeding thresholds and going out of control. Complete closed-loop operation steps: S1 sensor array collects real-time data from the entire process and uploads it to the AI4AI base. The S2 bottom layer agent reports back the device status, and the middle layer agents respectively complete the prediction of flue gas trends, the quantification of solvent aging, and the calculation of waste heat distribution; The S3 top-level global agent outputs optimal energy consumption control commands for the entire plant; The S4 command is sent to the underlying execution unit to adjust the circulation flow rate, regeneration temperature, and heat tracing power; The S5 execution results are sent back to the base, and the unsupervised framework automatically iterates the model weights in a closed loop.

[0010] 3.3 Core Innovation Protection Points Original molecular structure and one-step mild synthesis process of tetrabutylphosphonic amino acid ester bifunctional thermoinduced phase change ionic liquid; The first-ever phase change phase separation and rich liquid regeneration complete set of hardware processes significantly reduces ineffective regeneration heat consumption; A three-tiered, top-down agent global control cluster architecture adapted for wet CCUS scenarios; We have developed a self-developed AI4AI centralized four-layer intelligent agent peer-to-peer interaction unsupervised collaborative base to achieve zero-sample deployment of CCUS in industry; A 15-minute advance prediction algorithm for flue gas components offsets the efficiency degradation caused by diurnal load fluctuations. A real-time quantitative diagnostic model for aging and impurity poisoning of multi-parameter coupled ionic liquids; AI4AI base features 24-hour automatic operating condition clustering and a monthly autonomous improvement mechanism for control accuracy iteration. Phase separation tank interface height closed-loop adaptive fine-tuning technology stabilizes the rich phase volume ratio at 30%±2%; Instantaneous SO2 impurity shock buffering and regulation logic, short-term increase in solvent circulation to resist poisoning; The factory's nighttime peak-shaving and low-load AI-powered automatic sleep energy-saving operation mode; Algorithm for hardware and software compatibility modification and adaptation of existing conventional amine-based CCUS devices; AI4AI's underlying device operation safety threshold hard constraint architecture eliminates over-temperature and over-pressure failures; Intelligent control strategy for pipeline heat tracing that adapts to winter and summer ambient temperatures; Monthly carbon reduction target with customizable energy-saving / high-efficiency dual-mode switching module; The production line expansion historical operating conditions pre-trained adaptation framework eliminates the need for system re-debugging during expansion. An online precision dispensing calculation model for ionic liquids eliminates the waste of consumables due to excessive solvent addition.

[0011] 3.4 Beneficial Technical Effects Energy consumption: The system regeneration energy consumption is 1.45~1.60 GJ / tCO2, which is 48% lower than that of traditional MEA and 21% lower than that of static ionic liquid systems; Consumables: Solvent lifespan extended by 30%, annual loss reduced to 0.32%, and maintenance consumable costs decreased by 40%; Production capacity: Monthly shutdown solvent sampling inspections are eliminated, and annual effective operating time increases by more than 220 hours; Implementation: New projects require no data labeling, reducing deployment time by 60%; retrofitting costs for older equipment decrease by 36%. Long-term iteration: After 5 years of continuous operation, the control accuracy has been improved by 35%, adapting to long-term iterations of production capacity and flue gas composition; Applicable to: full coverage of low-concentration industrial flue gas in steel, cement, and thermal power plants; continuous operation reliability of the system is 99.7%. Attached Figure Description

[0012] Figure 1. Overall process flow diagram of the complete wet carbon capture hardware; Figure 2. Schematic diagram of the three-tiered agent hierarchy; Figure 3. Logical block diagram of AI4AI centralized four-layer intelligent agent peer-to-peer interaction; Figure 4. Comparison curves of adaptive control of phase transition and phase separation in ionic liquids; Figure 5. Timing curves of AI energy consumption optimization under multiple operating conditions. Detailed Implementation

[0013] Test platform: Standardized 1000 Nm³ / h (standard cubic meters per hour, flue gas flow rate unit) cement kiln tail pilot plant; constant flue gas CO2=11%, SO2=65ppm; test period of 90 days, ambient temperature -2~33℃ with natural fluctuations, except for the control scheme, all other equipment, solvents and flue gas conditions are uniform.

[0014] Comparative Example 1: Artificially Fixed-Parameter Phase Change Ion System Absorption at 50℃, phase separation at 45℃, regeneration at 95℃, constant circulation flow rate of 12m³ / h (cubic meters per hour, liquid flow rate unit), without intelligent control module; Actual measurements: average energy consumption 1.89 GJ / tCO2, collection efficiency 91.2%; solvent testing was conducted for 12 hours of shutdown on the 68th day of operation, and the cumulative efficiency decline over the year was 187 hours.

[0015] Comparative Example 2: Single-point neural network with back propagation (BP) It can only independently adjust the regeneration temperature, requires 30 days of training on a manually labeled dataset, and lacks a multi-agent collaborative architecture; Actual measurements show that the average energy consumption is 1.72 GJ / tCO2, and the volume fluctuation range of the rich phase is 28%~41%, requiring periodic shutdowns to check the solvent status.

[0016] Implementation 1: The complete centralized AI4AI collaborative system of this invention The model converged within the first 15 days, and the average stable operation from 15 to 90 days was as follows: energy consumption of 1.58 GJ / tCO2, collection efficiency of 94.1%; phase separation volume stability of 30% ± 2%; no downtime for sampling was required throughout the year, and the cumulative efficiency decline was only 21 hours.

[0017] Implementation 2: Gradient Verification under Low Temperature Conditions When the environment remains at a low temperature of 0~8℃, the system automatically activates the heat tracing adaptive strategy, resulting in an energy consumption increase of only 3.7%, compared to an increase of more than 15% in the traditional blank comparison ratio.

[0018] Implementation of 3 Nighttime Low-Load Peak Shaving Verifications When the production line is shut down for 8 hours, it automatically switches to energy-saving sleep logic, reducing solvent loss by 67%. It can reach the rated collection efficiency in 10 minutes after restarting.

Claims

1. A quaternary phosphonium amino acid ester phase change ionic liquid and an AI4AI layered agent synergistic carbon capture integrated system, characterized in that, include: Quaternary phosphorus amino acid ester phase change ionic liquid absorption solvent, wet carbon capture integrated hardware process unit, global online sensing array, three-level hierarchical agent control cluster, AI4AI centralized four-layer peer-to-peer unsupervised collaborative base; The ionic liquid cation is tetrabutylphosphonium [P4444]⁺, and the anion is amino acid ethyl ester [AA-EA]⁻. The viscosity at 25℃ is ≤45mPa・s, and it has thermally induced liquid-liquid phase transition characteristics at 42℃. The three-level agents are divided into a first-level global coordination agent, a mid-level flue gas prediction / solvent diagnosis / waste heat scheduling agent, and a bottom-level equipment execution agent. The AI4AI base has four layers of intelligent agents that interact bidirectionally and peer-to-peer, enabling unsupervised self-iteration without the need to label operating data. The bottom layer is embedded with hard constraints for equipment temperature and pressure safety. The hardware unit realizes a closed loop of flue gas absorption-isothermal phase separation-rich liquid selective regeneration. The sensor array pushes time-series data to the base in real time, and the agents at all levels collaboratively and adaptively regulate the entire carbon capture process.

2. The system according to claim 1, characterized in that, The ionic liquid is synthesized by one-step esterification-metathesis at 50~70℃, with a yield of ≥95%, tolerance to ≤200ppm SO2 flue gas, annual loss of ≤0.32%, and service life of ≥5.2 years.

3. The system according to claim 1, characterized in that, The flue gas prediction agent has a built-in 15-minute load advance prediction algorithm, which dynamically adjusts the constant temperature parameters of the phase separation tank to stabilize the CO2-rich phase volume at 28%~32%.

4. The system according to claim 1, characterized in that, The AI4AI base automatically clusters and classifies operating conditions every 24 hours, improving system control accuracy month by month, and features a nighttime peak-shaving, low-load sleep energy-saving mode for factories.

5. The system according to claim 1, characterized in that, The entire system can be directly connected to existing MEA / MDEA amine carbon capture devices, and the accompanying software and hardware are compatible with the algorithm, eliminating the need for large-scale equipment replacement.

6. An AI-based collaborative carbon capture operation method based on any one of the complete systems described in claims 1 to 5, characterized in that... Includes the following steps: The S1 global sensor array collects real-time data on flue gas, solvents, and equipment energy consumption, and uploads it to the AI4AI centralized base station. The S2 bottom layer agent reports back the hardware operating status, while the middle layer agents respectively predict flue gas fluctuations, quantify solvent aging, and coordinate the distribution of waste heat in the plant area. S3 Level 1 Global Agent outputs plant-wide energy consumption optimization control commands; The S4 command is issued to the underlying unit to adjust the parameters of the entire process of absorption, phase separation, and regeneration; The S5 runtime feedback data is sent back to the base, and the unsupervised clustering framework iterates the Agent decision weights, continuously operating in a closed loop for adaptive operation.

Images