An automated carbon dioxide hydrogenation and intelligent chromatographic analysis method

Abstract

This invention discloses an automated carbon dioxide hydrogenation and intelligent chromatographic analysis method, belonging to the field of chemical production technology. An AI prediction model analyzes input information and outputs optimal experimental conditions, dynamically adjusting parameters based on real-time product data. This invention learns reaction patterns online through the AI ​​prediction model, enabling adaptive adjustment of the experimental scheme. A programmable heater provides a stable and controllable temperature for the reaction during automated carbon dioxide hydrogenation based on a temperature program; a PID control board automatically adjusts the heater's power output according to temperature changes during the actual reaction. This invention's independent programmable heater, combined with PID closed-loop control, achieves a temperature range of 0~800℃ and a range of ≥10℃ for [missing information - likely related to temperature control]. ‑1 Rapid heating and cooling, with a channel temperature difference of ≤±1℃, improves temperature control consistency.

Description

Technical Field

[0001] This invention relates to the field of chemical production technology, and more specifically, to an automated carbon dioxide hydrogenation and intelligent chromatographic analysis method. Background Technology

[0002] Commercial parallel fixed-bed catalytic reaction evaluation devices primarily employ a shared electric furnace or cylindrical furnace, inserting 8–48 microreaction tubes side-by-side into the heating zone for unified temperature control. They are typically equipped with a multi-position switching valve to sequentially introduce the product streams from each channel into a single GC or GC-MS. However, the complete chromatographic cycle for a single channel is often 3–5 minutes, making it difficult to capture sub-second transient kinetics or rapid product changes during temperature-programmed reprocessing (TPR / TPSR), leading to inaccuracies in mechanism studies and steady-state extrapolation. Furthermore, when multiple reaction tubes share a cylindrical or electric furnace, the radial / axial temperature difference within the furnace cavity can reach ±5°C; heating rates exceeding 5°C s⁻¹ are prone to thermal inertia and overshoot. Temperature deviations between reaction tubes are directly amplified into activity differences, resulting in false positive / false negative screening results and reducing experimental reliability.

[0003] Microfluidic chip-based catalytic reaction / analysis platform: Dozens to hundreds of microcavities, with volumes on the order of nL to μL, are etched onto a single chip, integrating thin-film resistance heaters and temperature sensors. Planar imaging product detection is achieved using microscopic infrared thermography or time-of-flight mass spectrometry, capturing spatial distribution information in a single operation. However, due to the packaging materials and microcavity characteristics, it does not support high-temperature / high-pressure / combustible gas systems; fixed-bed devices have slow heating and cooling rates and long holding zones, making it impossible to realistically simulate rapid thermal cycling or temperature pulse conditions.

[0004] At present, preliminary automation results have been achieved in high-throughput catalytic screening, but there are still problems such as sequential analysis, poor temperature control consistency, slow dynamic response, risk of cross-contamination, or narrow applicability of the process. Summary of the Invention

[0005] In view of this, the present invention provides an automated carbon dioxide hydrogenation and intelligent chromatographic analysis method to solve the problems of existing thermocatalytic reaction devices, such as sequential analysis, poor temperature control consistency, slow dynamic response, risk of cross-contamination, or narrow process applicability.

[0006] To achieve the above objectives, the following solution is proposed: An automated carbon dioxide hydrogenation and intelligent chromatographic analysis method includes: Obtain the experimental condition information input by the user; The AI ​​prediction model analyzes the experimental conditions information, uses a Bayesian optimization algorithm to search for the optimal experimental conditions, and outputs the target experimental data, temperature program, feed parameters, catalyst loading parameters and sampling frequency. The reaction temperature control and micro furnace array module, the catalyst automatic loading and activation module, and the feed and flow control module automatically complete temperature control, catalyst loading, in-situ activation, feed ratio and reaction operation according to the temperature program, catalyst loading parameters and feed parameters. The rapid sampling-multi-detection module samples and detects the products of each channel according to the sampling frequency, realizes synchronous online analysis, and transmits the real-time experimental data obtained from the detection back to the cloud database. The data processing module calculates the deviation between real-time experimental data and target experimental data, and dynamically resets experimental parameters based on the deviation value.

[0007] Preferably, the objective function of the training process of the AI ​​prediction model is:

[0008] Where X represents the conversion rate, Starget represents the target product selectivity, Sbyproduct represents the byproduct selectivity, and E represents the energy consumption or cost indicator. ~ As weight.

[0009] Preferably, the process by which the AI ​​prediction model parses the experimental condition information and uses a Bayesian optimization algorithm to search for the optimal experimental conditions includes: Set the search space; Based on the candidate experimental conditions, predict the target experimental data and uncertainty; Scores for each candidate experimental condition are calculated using mathematical expectation, upper confidence, or probability. The experimental conditions with the highest scores from group 1 to group k are selected as the optimal experimental conditions.

[0010] Preferably, the cloud database stores catalyst composition, reaction conditions, product distribution / performance indicators, raw chromatographic signals, and device operating status.

[0011] Preferably, the AI ​​prediction model is built on an XGBoost + LSTM hybrid architecture.

[0012] Preferably, the reaction temperature control and micro furnace array module includes a programmable heating furnace, a vacuum insulation jacket and a reflective screen, and an independent PID control board; The programmable heating furnace provides a stable and controllable temperature for the reaction during the automated carbon dioxide hydrogenation process based on the temperature program. Vacuum insulation jackets are used to reduce heat loss and external environmental interference with reaction temperature; The reflector screen reflects heat and radiation back into the furnace, reducing heat loss. The PID control board automatically adjusts the power output of the heating furnace based on the temperature changes during the actual reaction.

[0013] Preferably, the catalyst automatic loading and activation module includes a pluggable reaction tube bracket, an H2 pretreatment unit, and an automatic feeding device, which can automatically load the catalyst and complete the reduction and activation. The plug-in reaction tube holder facilitates rapid catalyst replacement and loading; The H2 pretreatment unit utilizes hydrogen to reduce the catalyst. The automated feeding device is used to insert and remove the reaction tube, ensuring that the reaction tube is accurately inserted into the reaction temperature control and micro furnace array module.

[0014] Preferably, the rapid sampling-multi-detection module includes a six-way pneumatic high-speed valve, a multi-channel product analysis system, and a constant-temperature transmission line; The six-way pneumatic high-speed valve quickly switches and samples from different gas sources, and transmits the gas samples to the constant temperature transmission line through the pipeline; The constant temperature transmission line ensures that the gas sample maintains a stable temperature during transmission, and sends the gas sample into the multi-channel product analysis system. The multichannel product analysis system includes GC, FID, and TCD, wherein GC is used to separate gas components, and FID and TCD are used to detect organic and inorganic gases, respectively.

[0015] Preferably, the method further includes: the safety monitoring module responding to over-temperature, over-pressure, and leakage anomalies; The safety monitoring module includes a pressure sensor, thermal fuse, H2 / CO probe, electromagnetic pressure relief valve, and flame arrester.

[0016] Preferably, the method further includes: the automatic cleaning and regeneration module flushes the pipeline with high-temperature inert gas and introduces reducing gas to regenerate the catalyst.

[0017] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: (1) The automated carbon dioxide hydrogenation and intelligent chromatographic analysis method provided by this invention uses an AI prediction model to analyze input information and output optimal experimental conditions, dynamically adjusting parameters based on real-time product data. This invention learns reaction patterns online through an AI prediction model, enabling adaptive adjustment of experimental schemes.

[0018] (2) The automated carbon dioxide hydrogenation and intelligent chromatographic analysis method provided by this invention uses a programmable heating furnace to provide a stable and controllable temperature for the reaction during the automated carbon dioxide hydrogenation process based on a temperature program; the PID control board automatically adjusts the power output of the heating furnace according to the temperature changes in the actual reaction. The independent programmable heating furnace of this invention, combined with PID closed-loop control, achieves a temperature range of 0~800℃ and ≥10℃ s. -1Rapid heating and cooling, with a channel temperature difference of ≤±1℃, improves temperature control consistency.

[0019] (3) The six-way pneumatic high-speed valve of the present invention quickly switches and samples from different gas sources, and transmits the gas sample through a pipeline to the constant temperature transmission line; the constant temperature transmission line ensures that the gas sample maintains a stable temperature during transmission, and sends the gas sample into the multi-channel product analysis system; the multi-channel product analysis system detects organic and inorganic gases respectively. The present invention achieves true parallel detection through millisecond-level sampling valves and multi-channel synchronous GC, improving the time resolution by an order of magnitude.

[0020] (4) The "plug-in" reaction tube bracket of this invention facilitates rapid catalyst replacement and loading; the H2 pretreatment unit uses hydrogen to reduce the catalyst; the automated feeding device is used to plug and unplug the reaction tube, allowing the reaction tube to be accurately inserted into the reaction temperature control and micro furnace array module, automatically loading the catalyst and completing the reduction and activation, and enabling rapid catalyst replacement. The feeding and flow control module includes a high-precision mass flow controller and an 8-position electric switching valve. This invention supports safe operation of combustible gases such as H2 / CO at 0~5 MPa. Through the design of a fully isolated flow path, cross-contamination during the detection process is greatly reduced. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0022] Figure 1 This is a flowchart of an automated carbon dioxide hydrogenation and intelligent chromatographic analysis method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the automated carbon dioxide hydrogenation and intelligent chromatographic analysis system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the hardware layer structure of the automated carbon dioxide hydrogenation and intelligent chromatographic analysis system provided in an embodiment of the present invention. Detailed Implementation

[0023] 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.

[0024] This invention provides an automated carbon dioxide hydrogenation and intelligent chromatographic analysis method. It enables programmed temperature control, flow rate adjustment, and real-time sampling analysis of multiple microreactors within a single experimental cycle. This method is used for catalyst screening, kinetic studies, and process parameter optimization, forming a collaborative whole. The software layer can control hardware actions in real time, and the experimental data generated by the hardware is fed back to the software model, achieving integrated hardware and software design.

[0025] Example 1 First, combined Figure 1 This invention provides a multi-channel product analysis method, as described below: Step S01: Obtain the experimental condition information input by the user.

[0026] Specifically, experimental condition information input by the user is obtained through the user interface.

[0027] In step S02, the AI ​​prediction model analyzes the experimental condition information, uses a Bayesian optimization algorithm to search for the optimal experimental conditions, and outputs the target experimental data, temperature program, feed parameters, catalyst loading parameters, and sampling frequency.

[0028] Specifically, the AI ​​model outputs initial experimental scheme target experimental data, temperature program, feed parameters, catalyst loading parameters, pressure and sampling frequency based on historical database (catalyst formulation-reaction conditions-product distribution three data).

[0029] In step S03, the reaction temperature control and micro furnace array module, the catalyst automatic loading and activation module, and the feed and flow control module automatically complete temperature control, catalyst loading, in-situ activation, feed ratio and reaction operation according to the temperature program, catalyst loading parameters and feed parameters.

[0030] Reaction temperature control and micro furnace array module: Temperature control is based on temperature program.

[0031] Automatic loading and activation module: Modular "plug-in" reaction tube holders are ready to use; catalyst is automatically added according to the set mass, and the reduction and activation sequence is triggered.

[0032] Feed and flow control module: Multi-channel MFC (High-Precision Mass Flow Controller) adjusts gas and pressure in real time based on conditions output by AI prediction models or user-defined feed formulations. Electric switching valves can seamlessly switch between different reactants, internal standard gases, or inert gases within milliseconds.

[0033] Step S04: Rapid sampling - The multi-detection module samples and detects the products of each channel according to the sampling frequency, realizes synchronous online analysis, and transmits the real-time experimental data obtained from the detection back to the cloud database.

[0034] Specifically, the rapid sampling-multi-detection module switches a six-way pneumatic high-speed valve according to the sampling frequency, sending the products from each channel to the GC / FID+TCD detector to achieve synchronous online analysis, and transmitting the real-time experimental data obtained from the detection back to the cloud database for storage.

[0035] Step S05: Calculate the deviation between the real-time experimental data and the target experimental data, and reset the dynamic experimental parameters based on the deviation value.

[0036] Specifically, the data processing module calculates the deviation between real-time experimental data and target experimental data, and dynamically resets experimental parameters based on the deviation value. It automatically adjusts the sampling frequency, temperature program, or terminates the operation of a specific channel based on conversion rate, selectivity, or byproduct thresholds.

[0037] Furthermore, the rapid sampling-multi-detection module can quantitatively capture products into pre-calibrated collection tubes or liquid collection bottles via a branch valve, writing the actual conversion rate, selectivity, and kinetic parameters (TOF) data into a cloud database. The cloud database automatically links and archives the corresponding operating conditions with the chromatographic / mass spectrometry data. The data processing module's embedded scripts instantly compare the target-actual deviations and label them as "successful / needs adjustment." New data is used instantly (online) or in batches (offline) for incremental training / transfer learning, updating the AI ​​prediction model weights, and the AI ​​prediction model regenerates more accurate experimental conditions for the next round of experiments.

[0038] After the reaction is completed, the automatic cleaning and regeneration module injects high-temperature inert gas to sequentially flush the reaction tube, valve body, transmission pipeline and detector according to the preset process; if necessary, reducing gas is injected to regenerate the catalyst and test its stability.

[0039] This invention automates five stages: catalyst loading, in-situ pretreatment, reaction operation, product detection, and catalyst regeneration. The fully isolated flow path design significantly reduces cross-contamination.

[0040] Example 2 This invention describes the system structure for implementing the automated carbon dioxide hydrogenation and intelligent chromatographic analysis method described in the foregoing embodiments, such as... Figure 2 As shown, the system comprises a software layer, a hardware layer, and an integration layer. The AI ​​prediction model in the software layer acts as the "intelligent brain," predicting and optimizing experimental conditions; the automated devices in the hardware layer are responsible for executing the separation operations; and the integration layer handles communication scheduling between the two and provides the user interface. These three layers work together to form a complete system.

[0041] The software layer is designed around a five-step closed loop: "data-feature-model-prediction-self-evolution." Its core task is to predict the optimal reaction scheme before running on the high-throughput platform, perform real-time corrections during operation, and iterate automatically after operation. The software layer includes an AI prediction model, a cloud database, and a data processing module. The cloud database stores experimental data. The AI ​​prediction model analyzes the input information and outputs the optimal experimental conditions. The data processing module dynamically adjusts parameters based on real-time product data.

[0042] like Figure 2 , 3 As shown, the hardware layer includes a reaction temperature control and micro-furnace array module, an automatic catalyst loading and activation module, a feed and flow control module, a rapid sampling-multi-detection module, an online safety monitoring module, and an automatic cleaning and regeneration module. The structure of each module is shown in Table 1. Table 1

[0043] (1) Reaction temperature control and micro furnace array module The primary function of a programmable heater is to provide a stable and controllable temperature for the reaction. By precisely controlling the heating temperature, the reaction can be ensured to proceed under set conditions. In automated carbon dioxide hydrogenation processes, programmable heaters can provide the necessary high-temperature environment to promote the reaction.

[0044] The primary function of a vacuum insulation jacket is to reduce heat loss and external environmental interference with the reaction temperature. When used in conjunction with a programmable furnace, it minimizes heat loss, thereby improving the furnace's energy efficiency and enabling precise temperature control during the reaction process. For precision reactions, minimizing temperature fluctuations is crucial for ensuring reproducible results.

[0045] The reflector is used to reflect heat and radiation energy back into the furnace, reducing heat loss. It helps to concentrate heat, thereby further improving the thermal efficiency of the heating furnace. Working together with the vacuum insulation jacket, the reflector reflects heat energy from the external environment back to the reaction zone, reducing the influence of external factors and further ensuring that the reaction proceeds stably at the set temperature.

[0046] The PID control board effectively avoids temperature fluctuations and ensures temperature control accuracy during the reaction process by monitoring the temperature in real time and adjusting the power of the programmable heating furnace.

[0047] (2) Catalyst automatic loading and activation module The reaction tube holder is designed as a "plug-in" type, which facilitates the rapid replacement and loading of catalysts, making the loading and removal of catalysts more efficient and reducing human error and time consumption.

[0048] The H2 pretreatment unit is used for the reduction and activation of the catalyst. This unit uses hydrogen to reduce the catalyst, removing the oxide layer or impurities on the catalyst surface, thereby activating the catalyst's reactivity.

[0049] like Figure 3 As shown, the automated feeding device is mainly completed by a three-wheeled robotic arm, and through computer control and mechanical operation, it ensures that the reaction tube is accurately inserted into the reaction furnace.

[0050] (3) Feeding and flow control module like Figure 3 As shown, the feed and flow control module includes a high-precision mass flow controller (MFC) (0.1–500 mL min⁻¹) and an 8-position electrically operated switching valve.

[0051] MFC (Mass-flow Controller): Employs thermal or differential pressure measurement principles to precisely regulate gas / liquid flow rates via closed-loop valves. In multi-channel systems, MFC ensures consistent and repeatable feed rates across all reaction channels, supporting automated stepped flow experiments and programmed gas mixing.

[0052] (4) Rapid sampling - multiple detection module The six-way pneumatic high-speed valve is responsible for quickly switching and sampling from different gas sources, and transferring the gas samples through pipelines to the constant temperature transmission line.

[0053] The constant temperature transfer line ensures that the gas sample maintains a stable temperature during the transfer process, avoiding temperature fluctuations from affecting the composition and state of the sample.

[0054] The gas sample is fed into a multichannel product analysis system (GC / FID+TCD), where GC is used to separate gas components, while FID and TCD are used to detect organic and inorganic gases, respectively.

[0055] GC (Gas Chromatography): A chromatographic technique that uses an inert carrier gas as the mobile phase and a stationary column as the separation medium. Through step-up or isothermal operations, components with different boiling points or polarities are sequentially separated and introduced into the detector. In the system of this invention, GC mainly undertakes the fractional separation of multi-channel product streams, providing pure peaks for subsequent FID / TCD detection.

[0056] FID (Flame Ionization Detector): Organic matter is burned in a hydrogen-air flame to generate an ion stream; the current signal is proportional to the number of carbon atoms. FID is highly sensitive to carbon-containing compounds such as hydrocarbons and alcohols, and is used in the system described in this invention for the quantitative analysis of hydrocarbons and oxygen-containing organic products.

[0057] TCD (Thermal Conductivity Detector): A universal detector that measures bridge imbalance signals based on differences in thermal conductivity between components. It is suitable for permanent gases such as H2, N2, CO, and CO2. Through calibration curves, it can complement FID to complete carbon budget closure for all components.

[0058] Furthermore, the rapid sampling-multi-detection module can be equipped with GC-MS, FTIR, mass spectrometry, or a micro-sensor array in addition to GC, enabling faster multi-component detection. The number of channels N can be expanded to 8, 16, or more; the programmable furnace can employ an independent micro-furnace or a shared furnace body with a localized compensation heating structure; and the valve assembly can use an array of solenoid valves instead of rotary valves. The automated carbon dioxide hydrogenation and intelligent chromatographic analysis method of this invention integrates key technologies such as microreactor design, precise temperature and flow control, gas chromatography analysis, synchronous data acquisition, and automated control. It aims to improve catalytic experiment throughput, shorten the research and development cycle, and ensure the accuracy, repeatability, and high temporal resolution of experimental data. This invention overcomes the bottlenecks of existing technologies through an integrated design of an independent thermal module, a millisecond-level sampling valve, and a multi-channel synchronous GC, achieving rapid high-temperature rise and fall, isochronous parallel detection, and zero cross-interference across channels. This provides higher resolution and more reliable data support for catalyst development and reaction engineering scale-up.

[0059] Example 3 This invention describes the cloud database. First, an experimental database corresponding to the catalyst's actual performance is established. A ternary cloud database is constructed with "catalyst composition (C) – reaction conditions (R) – product distribution / performance indicators (P)" as its core. The cloud database contains at least the following data tables (or equivalent datasets), as shown in Table 2: Table 2

[0060] The temperature program can be stored either as a time series (e.g., temperature points sampled every 1 second) or as a parameterized description (initial temperature, heating rate, plateau temperature, and dwell time, etc.). For non-steady-state conditions such as programmed heating / step heating, it is recommended to save both the parameterized description and the discrete time series to accommodate both model training and replication experiments.

[0061] Example 4 The software layer comprises AI prediction models, data processing modules, and a cloud database and visualization solution. The data processing module's experimental scheduling algorithm dynamically adjusts the detection plan based on real-time experimental data. For example, when the temperature and pressure of a high-temperature circuit reach a preset level, the scheduling algorithm adjusts the GC detection time to capture critical changes; when the reaction process is relatively stable, the detection frequency is reduced (e.g., 2 hours / time) to avoid resource waste. The hardware layer's rapid sampling-multi-detection module works in conjunction with the experimental scheduling algorithm through a control interface. Based on the scheduling instructions generated by the algorithm, sampling and analysis are initiated only when needed, rather than cyclically detecting at fixed time intervals.

[0062] The AI ​​prediction model incorporates data trained on large-scale historical data and utilizes machine learning algorithms to extract patterns from a three-dimensional data lake of catalyst composition, reaction conditions, and product distribution. Given input experimental conditions (such as catalyst formulation, active component loading, support properties, reactant composition, thermodynamic constraints, and pilot experiment results), the model can intelligently output optimal or alternative experimental conditions, specifically including: Temperature program (initial temperature, heating rate, temperature of each plateau and dwell time); Feed parameters (gas / liquid component concentration, volume hourly space velocity (GHSV), pressure, dilution ratio); Catalyst loading and bed height (to meet stable external and internal diffusion conditions); Sampling frequency and time resolution (matching product formation rate and detector capacity). Estimated conversion rate, selectivity, and byproduct distribution (as experimental monitoring thresholds).

[0063] Table 3 shows the selection and training process of the AI ​​prediction model of this invention: Table 3

[0064] In this embodiment of the invention, the data entering the AI ​​prediction model undergoes unified preprocessing and quality control to ensure comparability and repeatability across different batches and channels.

[0065] Chromatographic data processing: Denoising, baseline correction, peak identification, and peak area integration are performed on the raw chromatographic signal; peak area is converted into molar flow rate / concentration by calibrating the curve using internal or external standard methods.

[0066] Indicator calculation: Calculate conversion rate, selectivity, yield, space-time yield, etc. based on the feed and discharge molar flow rates; calculate carbon balance closure for carbon-containing systems and mark data with closure below the threshold (e.g., 95%) as "needs review / removal".

[0067] Anomaly and drift detection: Set upper and lower limits and drift thresholds for key process quantities such as temperature, pressure, and flow rate; perform regular or model-based detection of chromatographic peak shape, retention time drift, and baseline drift, and automatically label abnormal samples.

[0068] Missing values ​​and unit standardization: Standardize units (°C, kPa / MPa, mL·min) -1 (e.g., GHSV, etc.); missing values ​​are filled using rules or models, and the filling marker features are recorded.

[0069] The feature extraction and processing process of the AI ​​prediction model is as follows: The input feature vector x is composed of the catalyst descriptor x C With operating condition descriptor x R By concatenation, we get x = [x C , x R ].in: Catalyst descriptor x C This includes embedded or specific encoding of the metal / auxiliary type; continuous characteristics such as loading amount and molar ratio; and continuous characteristics related to the support and preparation (BET, pore volume, pore size, calcination temperature, reduction temperature, etc.). Optionally, descriptors from characterization or calculations (e.g., crystal phase, particle size, XPS valence ratio, DFT adsorption energy, etc.) can be added.

[0070] Operating condition descriptor x R The parameters include continuous features such as feed composition, space velocity / flow rate, pressure, and dilution ratio; discrete features such as valve position strategy; and temperature programs input as parameterized vectors (initial temperature, heating rate, plateau temperature, residence time, etc.) or time series vectors (T(t)). When the temperature program is input as a time series, time window features (e.g., temperature, pressure, and flow rate sequences of the most recent 3600 seconds) are further constructed to support the prediction of transient product responses.

[0071] The AI ​​prediction model of this invention is used to learn mapping relationships: Where P includes product distribution and performance indicators, It is composed of catalysts. The reaction conditions are specified. The AI ​​prediction model can employ a tree model, a deep neural network, or an integration of both. Specifically, the AI ​​prediction model in this embodiment can adopt an XGBoost + LSTM hybrid architecture (multi-input, multi-task hybrid architecture), and alternative implementations are also provided. Preferred model: Multi-input fusion network (Tabular Encoder + Sequence Encoder + Multi-head). Its structure includes: 1) Category feature embedding layer: Maps discrete variables such as metal category, carrier category, and reaction type into low-dimensional dense vectors; 2) Continuous Feature Normalization Layer: Standardizes / normalizes continuous variables such as load, BET, flow rate, and pressure; 3) Table Feature Encoder (MLP or XGBoost): Performs non-linear representation learning on non-sequence features to obtain a table representation vector h. tab ; 4) Temperature Program / Process Sequence Encoder (LSTM / TCN): When time series inputs such as T(t), P(t), and F(t) are available, LSTM or temporal convolutional networks are used to extract the dynamic representation h. seq ; 5) Feature fusion layer: This layer integrates h... tab with h seq h is obtained by attention / gating fusion after splicing; 6) Multi-task output head: outputs conversion rate, selectivity / yield of each product, inactivation rate, etc.; for "product distribution", Softmax constraint can be used to make the sum of selectivity 1, or Dirichlet distribution parameterization can be used to model uncertainty.

[0072] In another embodiment, the table encoder can use XGBoost as the reinforcement learner to output intermediate predicted values ​​and importance weights; the sequence encoder outputs dynamic features; the two are weighted and integrated in the fusion layer to balance small sample robustness and non-steady-state fitting ability.

[0073] Alternative implementations include, but are not limited to: LightGBM / CatBoost, support vector regression; or end-to-end Transformer / TCN; or material graph neural networks (such as CGCNN / MEGNet) to directly map catalyst structure maps / crystal structures to performance metrics.

[0074] Furthermore, this embodiment of the invention establishes a closed-loop feedback mechanism for data and control between the software and hardware layers, forming a continuously self-evolving "model-experiment" cycle. The software layer can trigger dynamic parameter resetting based on real-time output data and continuously learn incrementally. The specific process is as follows: The software layer performs real-time deviation monitoring. For example, if the CO selectivity is detected to be less than 5% of the target selectivity online, sampling is immediately performed and data changes are recorded. New data is compressed and uploaded to the cloud database immediately after the experiment ends. Each experiment generates a unique experiment number, and the original signal, processing results, and metadata are uniformly written to the cloud database. The AI ​​prediction model is fine-tuned using idle GPUs at night or in real-time. This closed-loop optimization mechanism enables the invention to self-evolve, make autonomous decisions, and continuously improve reliability.

[0075] The software layer of this invention not only provides high-precision reaction scheme prediction before the experiment, but also enables real-time adaptive control during the experiment and self-iterative upgrades after the experiment. Through multiple iterations, the system can converge to the optimal process window or screen out high-performance catalysts in a short time, achieving "the more it runs, the more accurate it becomes, and the more it is used, the smarter it gets."

[0076] Example 5 This invention not only predicts the product distribution under given conditions, but also uses the AI ​​prediction model as a "surrogate model" to search for optimal experimental conditions. The objective function of the AI ​​prediction model is:

[0077] Where X represents the conversion rate, Starget represents the target product selectivity, Sbyproduct represents the byproduct selectivity, and E represents the energy consumption or cost indicator. ~ As weight.

[0078] The constraints of the AI ​​prediction model include: upper temperature limit, upper pressure limit, upper limit for combustible gas safety, chromatographic cycle / sampling frequency, and carbon balance closure threshold.

[0079] AI prediction models can be optimized using Bayesian optimization algorithms, genetic algorithms, particle swarm optimization, reinforcement learning, or gradient-based constraint optimization. Uncertainty estimation can also be added for active learning of point selection. This invention uses Bayesian optimization as an example to describe the optimization process, as follows: 1) Set the search space: the range of values ​​and distance steps for temperature program parameters, feed composition, air velocity, pressure, etc. 2) Proxy model prediction: Calculate the target experimental data under candidate condition r: and uncertainty σ; 3) Acquisition function: The score of each candidate experimental condition is calculated using mathematical expectation (EI), upper confidence bound (UCB), or probability calculation (PI). 4) Select candidates: Select the 1~k sets of conditions with the highest scores as the "optimal experimental scheme"; 5) Experiment execution and write-back: The device executes the experiment and writes the results back to the data lake; the proxy model is updated and the next iteration begins.

[0080] When multiple objectives exist (e.g., maximizing both conversion rate and selectivity simultaneously), this invention can output several alternative conditions on the Pareto front for users to choose according to their preferences.

[0081] Example 6 The integration layer is responsible for communication scheduling between the software and hardware layers. Communication methods can include OPC UA, Modbus-TCP, serial port (RS485), shared database message queues (MQTT / Kafka), etc., to achieve device linkage. This embodiment of the invention describes the integration layer using a TCP / IP-based implementation.

[0082] In one implementation, the hardware layer triggers commands such as "Start Chromatographic Analysis / Switch Channel / End Analysis" based on the reaction progress. These commands are short messages (e.g., ASCII strings) and include: a command word, experiment number, channel number, timestamp, and a checksum field. For example, START can be defined.<experiment_id> , <channel>,<method_id>, <checksum>;END,<experiment_id> ;ACK,<experiment_id> In addition, to improve reliability, a heartbeat and retransmission mechanism can be introduced.

[0083] When the third-party instrument software at the hardware layer lacks open interfaces, this embodiment of the invention allows for the use of UI automation and visual recognition to obtain the operating status and trigger linkage. The hardware layer connects to the hardware layer control program (using PID control) via pywinauto, bringing the window to the top and clicking a button at preset coordinates to initiate / confirm an operation. Pixel color recognition determines whether the instrument at the hardware layer is in a "running / idle" state: RGB values ​​are read at several fixed coordinates and compared with preset colors to determine the status. OCR identifies numbers / status codes (e.g., serial numbers or method numbers from 0–39) within a fixed area; when the recognition result meets the triggering conditions, a start command is sent to the other end via a TCP socket.

[0084] After detecting the specified state and meeting the numbering condition, the string "1" is sent as a trigger signal to the target IP address and target port. This logic can be generalized to the START command mentioned above, carrying information such as the experiment number.

[0085] After receiving the startup command, the hardware layer host (such as the chromatography system) automatically completes the method loading, file naming, startup and termination determination, and sends the running status back to the software layer host. After the running is completed, it exits the loop process.

[0086] If the online safety monitoring module detects an over-temperature / over-pressure / combustible gas leak alarm, it will prohibit the issuance of the START command or force the sending of the END / STOP command, and simultaneously trigger the device's emergency stop and depressurization. If the chromatogram of the rapid sampling-multi-detection module does not return an ACK within a preset time, the anomaly will be recorded and the command will be resent; if the number of retries is exceeded, it will enter manual confirmation mode. Both commands and feedback include a verification field to avoid cross-channel errors or mismatches.

[0087] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0088] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0089] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.< / checksum> < / channel>

Claims

1. An automated carbon dioxide hydrogenation and intelligent chromatographic analysis method, characterized in that, include: Obtain the experimental condition information input by the user; The AI ​​prediction model analyzes the experimental conditions information, uses a Bayesian optimization algorithm to search for the optimal experimental conditions, and outputs the target experimental data, temperature program, feed parameters, catalyst loading parameters and sampling frequency. The reaction temperature control and micro furnace array module, the catalyst automatic loading and activation module, and the feed and flow control module automatically complete temperature control, catalyst loading, in-situ activation, feed ratio and reaction operation according to the temperature program, catalyst loading parameters and feed parameters. The rapid sampling-multi-detection module samples and detects the products of each channel according to the sampling frequency, realizes synchronous online analysis, and transmits the real-time experimental data obtained from the detection back to the cloud database. The data processing module calculates the deviation between real-time experimental data and target experimental data, and dynamically resets experimental parameters based on the deviation value.

2. The automated carbon dioxide hydrogenation and intelligent chromatographic analysis method according to claim 1, characterized in that, The objective function for training the AI ​​prediction model is: Where X represents the conversion rate, Starget represents the target product selectivity, Sbyproduct represents the byproduct selectivity, and E represents the energy consumption or cost indicator. ~ For weights.

3. The automated carbon dioxide hydrogenation and intelligent chromatographic analysis method according to claim 1, characterized in that, The process by which the AI ​​prediction model analyzes the experimental condition information and uses a Bayesian optimization algorithm to search for the optimal experimental conditions includes: Set the search space; Based on the candidate experimental conditions, predict the target experimental data and uncertainty; Scores for each candidate experimental condition are calculated using mathematical expectation, upper confidence, or probability. The experimental conditions with the highest scores from group 1 to group k are selected as the optimal experimental conditions.

4. The automated carbon dioxide hydrogenation and intelligent chromatographic analysis method according to claim 1, characterized in that, The cloud database stores catalyst composition, reaction conditions, product distribution / performance indicators, raw chromatographic signals, and device operating status.

5. The automated carbon dioxide hydrogenation and intelligent chromatographic analysis method according to claim 1, characterized in that, The AI ​​prediction model is built on a hybrid architecture of XGBoost + LSTM.

6. The automated carbon dioxide hydrogenation and intelligent chromatographic analysis method according to claim 1, characterized in that, The reaction temperature control and micro furnace array module includes a programmable heating furnace, a vacuum insulation jacket and a reflective screen, and an independent PID control board. The programmable heating furnace provides a stable and controllable temperature for the reaction during the automated carbon dioxide hydrogenation process based on the temperature program. Vacuum insulation jackets are used to reduce heat loss and external environmental interference with reaction temperature; The reflector screen reflects heat and radiation back into the furnace, reducing heat loss. The PID control board automatically adjusts the power output of the heating furnace based on the temperature changes during the actual reaction.

7. The automated carbon dioxide hydrogenation and intelligent chromatographic analysis method according to claim 1, characterized in that, The catalyst automatic loading and activation module includes a plug-in reaction tube bracket, an H2 pretreatment unit, and an automatic feeding device, which can automatically load the catalyst and complete the reduction and activation. The plug-in reaction tube holder facilitates rapid catalyst replacement and loading; The H2 pretreatment unit utilizes hydrogen to reduce the catalyst. The automated feeding device is used to insert and remove the reaction tube, ensuring that the reaction tube is accurately inserted into the reaction temperature control and micro furnace array module.

8. The automated carbon dioxide hydrogenation and intelligent chromatographic analysis method according to claim 1, characterized in that, The rapid sampling-multi-detection module includes a six-way pneumatic high-speed valve, a multi-channel product analysis system, and a constant-temperature transmission line. The six-way pneumatic high-speed valve quickly switches and samples from different gas sources, and transmits the gas samples to the constant temperature transmission line through the pipeline; The constant temperature transmission line ensures that the gas sample maintains a stable temperature during transmission, and sends the gas sample into the multi-channel product analysis system. The multichannel product analysis system includes GC, FID, and TCD, wherein GC is used to separate gas components, and FID and TCD are used to detect organic and inorganic gases, respectively.

9. The automated carbon dioxide hydrogenation and intelligent chromatographic analysis method according to claim 1, characterized in that, The method also includes: the safety monitoring module responding to over-temperature, over-pressure, and leakage anomalies; The safety monitoring module includes a pressure sensor, thermal fuse, H2 / CO probe, electromagnetic pressure relief valve, and flame arrester.

10. The automated carbon dioxide hydrogenation and intelligent chromatographic analysis method according to claim 1, characterized in that, The method also includes: the automatic cleaning and regeneration module flushes the pipeline with high-temperature inert gas and introduces reducing gas to regenerate the catalyst.

Images