Intelligent agent for saponification reaction and application thereof

Abstract

The invention discloses an intelligent agent for saponification reaction and application thereof. The intelligent agent comprises a hardware system and a software system. The hardware system comprises a feed liquid storage module, a liquid conveying module, a reaction module, an environment information sensing module, a fraction collection module and a central control module. The software system comprises a saponification reaction condition recommendation module, an experiment operation condition calculation module, a hardware system control code generation module, an experiment design table generation module, an experiment result qualitative analysis and optimization module and an experiment result automatic modeling module. The intelligent agent integrates a large language model and a micro-channel continuous flow technology, can be used for saponification reaction automatic optimization, and has the characteristics of accurate recommendation of process conditions, real-time monitoring and feedback adjustment, full-process automatic control, environmental perception, fault early warning and the like.

Description

Technical Field

[0001] The present invention relates to the field of saponification reactions in chemical production, and in particular to a saponification reaction intelligent agent that integrates a large language model (LLM) and microchannel continuous flow technology, and can be used for automatic optimization of saponification reactions. Background Art

[0002] Traditional saponification processes often use a batch production method. For example, patent specification CN1594350A discloses a method for separating and extracting cholesterol from crude lanolin, comprising the following steps: 1) washing the crude lanolin with an aqueous surfactant solution; 2) subjecting the washed or refined lanolin to a saponification reaction in an alcohol solution under alkaline conditions; 3) extracting the lanolin alcohol from the saponified product using ethyl acetate, acetone, toluene, alkanes, or halogenated alkanes as extractants to obtain a hydrocarbon or halogenated alkanes solution of lanolin alcohol; 4) washing the hydrocarbon or halogenated alkanes solution with an alcohol solution at room temperature, separating the alcohol solution layer after washing, and evaporating the hydrocarbon or halogenated alkanes in the lanolin alcohol solution to obtain a cholesterol-containing lanolin paste; 5) selectively crystallizing the lanolin paste in an alcohol solvent to obtain crude cholesterol, which is then recrystallized to obtain cholesterol crystals.

[0003] The intermittent production method has many disadvantages: the reaction time is long, usually lasting several hours, and the production efficiency is low; to ensure the saponification effect, excessive alkali is often added, resulting in a large amount of solid waste, which is contrary to the atomic economy principle of green chemistry; and during the reaction process, key parameters such as temperature are often uneven throughout the reactor, making it difficult to accurately control product quality.

[0004] With technological advancements, pipelined continuous saponification methods for extracting cholesterol from lanolin have emerged. For example, patent specification CN109851654A discloses a pipeline reactor method for extracting cholesterol from lanolin through saponification. Lanolin, an alcohol as a solvent, and an alkali are mixed as the raw material, with the alcohol to lanolin mass ratio being 2.5-3:1 and the alkali to lanolin mass ratio being 0.15-0.2:1. The raw material enters the pipeline reactor for a saponification reaction at a temperature of 140-160°C and a pressure of 1.0-1.4 MPa. The raw material resides in the pipeline reactor for 24-35 minutes. The saponified liquid discharged from the pipeline reactor outlet undergoes post-treatment to produce a mixed alcohol containing cholesterol. This patented method achieves continuous and stable operation, increasing production capacity.

[0005] Currently, microchannel continuous flow technology has been applied to numerous reactions. While it offers advantages such as efficient mass and heat transfer, strong process controllability, and inherent safety, in practice, complex parameter optimization and equipment control are required to ensure efficient operation and precise control of the reaction process.

[0006] In recent years, large language models have shown great potential in the field of process optimization. They can complete tasks such as literature parsing, experimental operation code generation, and intelligent decision-making. However, there is currently no mature technology that combines them with microchannel continuous flow technology for saponification reactions. Summary of the Invention

[0007] To address the aforementioned technical issues and shortcomings in the field, the present invention provides an intelligent agent for saponification reactions and its applications. The present intelligent agent for saponification reactions, named SapoMind-Prime, integrates a large language model agent with microchannel continuous flow technology to achieve efficient and green manufacturing of saponification reactions, precisely control product quality, and reduce energy and resource consumption.

[0008] The specific technical solutions are as follows: In a first aspect, the present invention provides an intelligent agent for saponification reaction, comprising a hardware system and a software system.

[0009] The hardware system includes: The liquid storage module is used to store raw material liquid, alkali solution and diluent.

[0010] Infusion module, used to transport raw material liquid, alkali solution and diluent.

[0011] The reaction module is used to mix the raw material liquid, diluent and alkali solution and heat and control the temperature for reaction.

[0012] The environmental information sensing module is used to record environmental parameters such as temperature, humidity, atmospheric pressure and sound and provide feedback.

[0013] The fraction collection module is used to collect reaction products.

[0014] The central control module is used to realize the overall automatic control of the hardware system, data transmission, storage and processing, as well as human-computer interaction.

[0015] The software systems include: The saponification reaction condition recommendation module uses vector embedding technology to build a saponification literature knowledge base, and applies retrieval-augmented generation (RAG) technology combined with a large language model to achieve intelligent recommendation of process conditions.

[0016] The experimental operating condition calculation module is based on a large language model and is used to: calculate the hardware system operating conditions according to the process conditions, wherein the process conditions include reaction temperature, reaction time, raw material liquid mass fraction and alkali dosage, and the hardware system operating conditions include alkali liquid flow rate, raw material liquid flow rate, diluent flow rate and residence time, and the calculation formulas are shown in formulas (I)-(IV); write the experimental operating condition calculation component and realize the automatic conversion from process conditions to operating conditions by calling the component.

[0017]

[0018] Where: Represents the volume flow rate of raw liquid; represents the volume of the tubing used for the reaction; stands for reaction time; Represents the density of the raw material liquid; represents the mass fraction of the raw material liquid; Represents the saponification value of raw materials; represents the mass fraction of alkali solution; represents the density of lye; Represents the ratio of actual alkali usage to theoretical alkali usage; Represents the mass fraction of the diluted raw material liquid formed by mixing the raw material liquid and the diluent; represents the diluent density; represents the volume flow rate of alkali solution; represents the diluent volume flow rate; represents the dwell time; It represents the maximum time from the raw material liquid, alkali solution and diluent flowing out from the liquid storage module to the mixing of the three.

[0019] The hardware system control code generation module is used to integrate the large language model to generate the control code of each hardware device in the hardware system, pass the hardware system operating conditions as parameters to the hardware device control code, and realize the accurate generation of hardware device control code under specific conditions.

[0020] The experimental design table generation module is based on a large language model and is used to recommend experimental design types and generate experimental design tables based on the process conditions recommended by the saponification reaction condition recommendation module and / or the process condition range optimized by the experimental result qualitative analysis and optimization module.

[0021] The experimental result qualitative analysis and optimization module, based on the large language model, is used to qualitatively analyze the experimental results corresponding to the process conditions recommended by the saponification reaction condition recommendation module, and optimize the process condition range according to the qualitative analysis results.

[0022] The automatic modeling module of experimental results is used to automatically model the process conditions and results and output the optimal process conditions.

[0023] In some embodiments, the hardware system of the intelligent agent for saponification reaction described in the first aspect further includes an online detection module for at least online detection of the saponification rate of the product. Furthermore, the online detection module can be equipped with ultraviolet spectroscopy and near-infrared spectroscopy equipment.

[0024] In some embodiments, in the intelligent body for saponification reaction described in the first aspect, the infusion module is equipped with multiple horizontal flow pumps to respectively transport the raw material liquid, alkali solution and diluent.

[0025] In some embodiments, the intelligent agent for saponification reaction described in the first aspect comprises a reaction module comprising a first T-shaped tee for mixing a raw material solution and a diluent to form a diluted raw material solution, a second T-shaped tee for mixing an alkali solution with the diluted raw material solution, a stainless steel coil for heating the reaction, an oil bath for controlling the reaction temperature, and a reaction temperature sensor. The reaction temperature sensor can monitor the reaction temperature.

[0026] In some embodiments, the intelligent agent for saponification reaction described in the first aspect, the environmental information sensing module includes an environmental temperature, humidity and atmospheric pressure sensor and a sound sensor. The environmental temperature, humidity and atmospheric pressure sensor is used to record the temperature, humidity and atmospheric pressure of the environmental conditions, and the sound sensor is used to record the sound of the environment and the equipment. The environmental information sensing module can be used to help the intelligent agent to perceive and make decisions. For example, when the recorded sound parameters are abnormal, the intelligent agent determines that the equipment may be faulty.

[0027] In some embodiments, the intelligent agent for saponification reaction described in the first aspect, the large language models used for the saponification reaction condition recommendation module, the experimental operation condition calculation module and the hardware system control code generation module are DeepSeek-V3, ERNIE-3.5-8K and ERNIE-3.5-8K respectively.

[0028] In some embodiments, the intelligent agent for saponification reaction described in the first aspect uses a large language model for the experimental result qualitative analysis and optimization module, which is a fine-tuned QwQ-32B (QwQ-32B-DataAnalysis).

[0029] In some embodiments, the intelligent agent for saponification reaction described in the first aspect uses a large language model for the experiment design table generation module, which is a fine-tuned DeepSeek-R1 (DeepSeek-R1-ExpDesign).

[0030] In some embodiments, the intelligent agent for saponification reaction described in the first aspect, the software system calls the central control module through the SDK, and writes a web page to realize software and hardware integrated control, perception, decision-making and visualization.

[0031] In a second aspect, the present invention provides an application of the intelligent agent described in the first aspect to a saponification reaction. Furthermore, the intelligent agent can be used for automatic optimization of the saponification reaction.

[0032] In a third aspect, the present invention provides a method for automatic optimization of a saponification reaction, comprising: utilizing the intelligent agent described in the first aspect to realize automatic optimization of a saponification reaction.

[0033] In the application of the second aspect and the automatic optimization method of the third aspect, when the intelligent agent performs automatic optimization of the saponification reaction, the execution process includes: The saponification reaction condition recommendation module builds a saponification literature knowledge base and uses search-enhanced generation technology combined with a large language model to achieve intelligent recommendations for saponification process conditions (including reaction temperature, reaction time, lanolin solution mass fraction, and alkali dosage). Based on the experimental design table generated by the experimental design table generation module and / or the optimal process conditions output by the experimental result automatic modeling module, the intelligent agent automatically calls the experimental operation condition calculation module and the hardware system control code generation module to generate the operation condition control code and sends it to the central control module, and then the hardware system automatically runs; The module for qualitative analysis and optimization of experimental results automatically obtains the results of the saponification reaction, analyzes the experimental results through a large language model, optimizes the range of process variables, and outputs the optimized process condition range; The experimental design table generation module generates an experimental design table using a large language model based on the process conditions recommended by the saponification reaction condition recommendation module and / or the optimized process condition range output by the experimental result qualitative analysis and optimization module. The table includes factor level settings and randomizes the experimental sequence. The automatic modeling module of experimental results automatically models the process conditions and results and outputs the optimal process conditions.

[0034] Furthermore, after the hardware system automatically runs, the online detection module reads data in real time, obtains the product saponification rate in real time, and automatically saves it. Furthermore, after the hardware system automatically runs, the online detection module reads spectral data in real time, uses an online model of spectral similarity and saponification rate to obtain the product saponification rate in real time, and automatically saves it.

[0035] Compared with the prior art, the present invention has the following beneficial effects: Accurately recommend process conditions: The saponification reaction condition recommendation module is based on a saponification literature knowledge base constructed based on vector embedding technology and RAG technology combined with a large language model. It can quickly and accurately recommend suitable saponification process conditions based on user input requirements or specific saponification reaction types, including key parameters such as reaction temperature, reaction time, raw material liquid mass fraction and alkali dosage. This avoids the waste of time and resources caused by insufficient human experience or trial-and-error processes in traditional methods, and provides a guarantee for efficient and high-quality saponification reactions from the source.

[0036] Real-time monitoring and feedback adjustment: The online detection module is equipped with ultraviolet spectroscopy and near-infrared spectroscopy equipment, which can detect the saponification rate of the product in real time. The hardware system can timely feedback and adjust the reaction conditions based on the real-time monitoring data, greatly shortening the time and cost of optimizing the reaction.

[0037] Intelligent agent-based full-process automated control: Each module in the software system leverages the powerful knowledge understanding and generation capabilities of large language models to make intelligent decisions based on real-time data and preset goals. For example, the module for qualitative analysis and optimization of experimental results automatically obtains and analyzes saponification reaction results, thereby optimizing the range of process variables. The module for generating experimental design tables generates scientifically sound experimental design tables based on recommended process conditions and optimized condition ranges, including factor level settings and randomized experimental sequence. This provides strong support for efficient experimental execution and reliable results, enabling intelligent control of the entire saponification reaction process, from condition recommendation and operation execution to result analysis and optimization.

[0038] Environmental Perception and Fault Warning: The environmental information sensing module records environmental parameters such as temperature, humidity, atmospheric pressure, and sound in real time. When sound parameters are abnormal, the intelligent agent can determine that the equipment may be faulty and promptly issue a warning signal, prompting operators to inspect and address the problem. This prevents production accidents and product quality issues caused by equipment failure, enhancing production safety. Furthermore, by real-time monitoring of environmental conditions, the saponification reaction is ensured to proceed in a suitable environment, reducing the adverse effects of external environmental factors on the reaction process and further improving production stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a photo of the hardware system of the intelligent entity SapoMind-Prime in a specific implementation method.

[0040] Figure 2 Schematic diagram of the hardware system of the intelligent agent SapoMind-Prime in a specific implementation method.

[0041] Figure 3 This is an example diagram of the software webpage of the intelligent agent SapoMind-Prime in a specific implementation method. DETAILED DESCRIPTION

[0042] The present invention will be further described below with reference to the accompanying drawings and specific examples. It should be understood that these examples are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0043] An intelligent agent SapoMind-Prime for saponification reaction, including a hardware system and a software system.

[0044] like Figure 1 、 Figure 2As shown, the hardware system includes a feed liquid storage module, an infusion module, a reaction module, an environmental information sensing module, a fraction collection module, an online detection module, and a central control module. The feed liquid storage module is used to store and insulate the raw material liquid, alkali solution, and diluent. The infusion module is equipped with multiple horizontal flow pumps, one for transporting the raw material liquid, alkali solution, and diluent. The reaction module is used to mix the raw material liquid, diluent, and alkali solution, heat and control the reaction temperature, and monitor the reaction temperature. The reaction module includes a first T-shaped tee for mixing the raw material liquid and diluent to form a diluted raw material liquid, a second T-shaped tee for mixing the alkali solution and the diluted raw material liquid, a stainless steel coil for heating the reaction, an oil bath for controlling the reaction temperature, and a reaction temperature sensor for monitoring the reaction temperature. The environmental information sensing module is used to collect and provide feedback on environmental parameters such as temperature, humidity, atmospheric pressure, and sound. The environmental information sensing module includes ambient temperature, humidity, and atmospheric pressure sensors, as well as sound sensors. The latter are used to record the temperature, humidity, and atmospheric pressure of environmental conditions, while the sound sensor is used to record the sounds of the environment and equipment. The environmental information sensing module helps the intelligent agent perceive and make decisions. For example, if the recorded sound parameters are abnormal, the intelligent agent can determine that the equipment may have malfunctioned. The fraction collection module is used to collect reaction products. The online detection module is equipped with ultraviolet spectroscopy and near-infrared spectroscopy equipment for online product testing and determination of the product's real-time saponification rate. The central control module is responsible for automatic control and perception of the entire hardware system, data transmission, storage, and processing, as well as human-computer interaction.

[0045] The software system includes a saponification reaction condition recommendation module, an experimental operation condition calculation module, a hardware system control code generation module, an experimental design table generation module, an experimental result qualitative analysis and optimization module, and an experimental result automatic modeling module.

[0046] The saponification reaction condition recommendation module uses vector embedding technology to build a saponification literature knowledge base, and applies retrieval-augmented generation (RAG) technology combined with a large language model to achieve intelligent recommendation of process conditions.

[0047] The experimental operating condition calculation module is based on a large language model and is used to: calculate the hardware system operating conditions according to the process conditions, wherein the process conditions include reaction temperature, reaction time, raw material liquid mass fraction and alkali dosage; the hardware system operating conditions include alkali liquid flow rate, raw material liquid flow rate, diluent flow rate and residence time, and the calculation formulas are shown in formulas (I)-(IV); write the experimental operating condition calculation component and realize the automatic conversion from process conditions to operating conditions by calling the component.

[0048]

[0049] Where: Represents the volume flow rate of raw liquid; represents the volume of the tubing used for the reaction; stands for reaction time; Represents the density of the raw material liquid; represents the mass fraction of the raw material liquid; Represents the saponification value of raw materials; represents the mass fraction of alkali solution; represents the density of lye; Represents the ratio of actual alkali usage to theoretical alkali usage; Represents the mass fraction of the diluted raw material liquid formed by mixing the raw material liquid and the diluent; represents the diluent density; represents the volume flow rate of alkali solution; represents the diluent volume flow rate; represents the dwell time; It represents the maximum time from the raw material liquid, alkali solution and diluent flowing out from the liquid storage module to the mixing of the three.

[0050] The hardware system control code generation module is used to integrate the large language model to generate the control code of each hardware device in the hardware system, pass the hardware system operating conditions as parameters to the hardware device control code, and realize the accurate generation of hardware device control code under specific conditions.

[0051] The experimental design table generation module is based on a large language model and is used to recommend experimental design types and generate experimental design tables based on the process conditions recommended by the saponification reaction condition recommendation module and / or the process condition range optimized by the experimental result qualitative analysis and optimization module.

[0052] The experimental result qualitative analysis and optimization module is based on a large language model and is used to qualitatively analyze the experimental results corresponding to the process conditions recommended by the saponification reaction condition recommendation module, and optimize the process condition range based on the qualitative analysis results.

[0053] The automatic modeling module of experimental results is used to automatically model process conditions and results and output the optimal process conditions.

[0054] The working example of the software and hardware integration of the intelligent agent SapoMind-Prime in this embodiment is introduced as follows: The saponification reaction condition recommendation module builds a saponification literature knowledge base and uses search-enhanced generation technology combined with a large language model to achieve intelligent recommendations for saponification process conditions (including reaction temperature, reaction time, lanolin solution mass fraction, and alkali dosage). According to the experimental design table generated by the experimental design table generation module, the intelligent agent automatically calls the experimental operation condition calculation module and the hardware system control code generation module to generate the operation condition control code and send it to the central control module, and then the hardware system automatically runs; After the hardware system automatically runs, the online detection module reads the spectral data in real time, uses the online model of spectral similarity and saponification rate to obtain the product saponification rate in real time and automatically saves it; The module for qualitative analysis and optimization of experimental results automatically obtains the results of the saponification reaction, analyzes the experimental results through a large language model, optimizes the range of process variables, and outputs the optimized process condition range; The experimental design table generation module generates an experimental design table using a large language model based on the process conditions recommended by the saponification reaction condition recommendation module and / or the optimized process condition range output by the experimental result qualitative analysis and optimization module. The table includes factor level settings and randomizes the experimental sequence. The automatic modeling module of experimental results automatically models the process conditions and results and outputs the optimal process conditions.

[0055] The software system calls the central control module through the SDK and writes a web page to implement SapoMind-Prime's hardware and software integrated control, perception and decision-making. Figure 3 An example diagram of the interface of the intelligent saponification software system based on multi-module collaboration of a large language model for the intelligent agent used for saponification reaction in this embodiment is shown.

[0056] The large language models used for the saponification reaction condition recommendation module, the experimental operation condition calculation module, and the hardware system control code generation module are DeepSeek-V3, ERNIE-3.5-8K, and ERNIE-3.5-8K, respectively.

[0057] The large language model used for qualitative analysis of experimental results and optimization modules is the fine-tuned QwQ-32B (QwQ-32B-DataAnalysis).

[0058] The large language model used in the experimental design table generation module is the fine-tuned DeepSeek-R1 (DeepSeek-R1-ExpDesign).

[0059] The intelligent agent of this embodiment can be applied to saponification reactions for automatic optimization of saponification reactions.

[0060] A method for automatic optimization of a saponification reaction includes: utilizing the intelligent agent of this embodiment to realize automatic optimization of the saponification reaction.

[0061] The following describes specific operational execution cases, including: System Construction: Build the hardware system according to the aforementioned hardware configuration, including the liquid storage module, infusion module, reaction module, fraction collection module, environmental information sensing module, online detection module, and central control module. Simultaneously, build the software system based on the aforementioned software system architecture, and create a web-based platform to integrate the hardware and software, thus constructing SapoMind-Prime, the intelligent agent for saponification reactions.

[0062] Process parameter recommendation and code generation: The user enters a request for recommended lanolin saponification reaction process conditions in the webpage dialog area. The system uses the saponification reaction condition recommendation module to retrieve relevant information from the vector knowledge base and recommend process conditions. After the user confirms the process conditions, SapoMind-Prime calls the experimental operation condition calculation module to automatically convert the process conditions into operating conditions. SapoMind-Prime then calls the hardware system control code generation module to convert the operating conditions into control code.

[0063] Automatic Operation and Environmental Status Monitoring: SapoMind-Prime automatically downloads control code to the hardware system's central control module, which initiates automatic task execution. During this process, the environmental information sensor module monitors parameters such as temperature, humidity, atmospheric pressure, and sound in real time, feeding this data back to SapoMind-Prime. If the temperature, humidity, or atmospheric pressure fluctuate dramatically during the process, SapoMind-Prime will detect this and automatically stop responding. Large fluctuations in sound, which could indicate a pump blockage or other issue, will cause SapoMind-Prime to automatically stop responding and sound an alarm.

[0064] Experimental Result Analysis and Automatic Optimization: The similarity between the fully saponified UV spectrum and the online sample UV spectrum is used as an indicator for evaluating the saponification rate; the higher the similarity, the higher the saponification rate. The SapoMind-Prime hardware system's reaction module begins operation, and the online detection module simultaneously collects the product UV spectrum (every 5 seconds) and saves the similarity value for each process condition. After a round of automated experimentation, SapoMind-Prime automatically calls the Experimental Result Qualitative Analysis and Optimization module to analyze the experimental results, providing optimization suggestions for the process condition range. It then calls the Experimental Design Table Generation module to generate the optimized process condition experimental design table. It then sequentially calls the Experimental Operation Condition Calculation module and the Hardware System Control Code Generation module to conduct a second round of automated optimization experiments.

[0065] Automatic Modeling and Verification: After the second round of automated optimization experiments, SapoMind-Prime invokes the automated modeling module for experimental results. Using stepwise and polynomial regression, it automatically builds a statistical model of process conditions and results. Using a similarity score greater than 0.95 as a benchmark, it selects process conditions that meet the requirements and outputs them in JSON format. Subsequently, SapoMind-Prime sequentially invokes the experimental operating condition calculation module and the hardware system control code generation module to conduct verification experiments on the optimization results.

[0066] Optimization results: SapoMind-Prime's automatic optimization achieved efficient and green saponification of lanolin without any manual operation. The automatic optimization reaction took 275 minutes, and a total of 23 groups of experiments were conducted. The optimized continuous flow process reduced carbon emissions by 64% compared with the traditional batch process, reduced alkali usage by 28%, and shortened the reaction time to 15 minutes.

[0067] The above embodiments are merely illustrative of the application of the present invention and are not intended to limit its scope. In practical applications, the intelligent agent of the present invention can flexibly adjust parameters and configurations according to different saponification reaction requirements and is widely applicable to the optimization and control of various saponification reaction processes.

[0068] In addition, it should be understood that after reading the above description of the present invention, those skilled in the art may make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by the claims attached to this application.

Claims

1. An intelligent agent for saponification reaction, characterized in that: Including hardware system and software system; The hardware system includes: Liquid storage module, used to store raw material liquid, alkali solution and diluent; Infusion module, used to transport raw material liquid, alkali solution and diluent; The reaction module is used to mix the raw material liquid, diluent and alkali solution and heat and control the temperature for reaction; Environmental information sensing module, used to record environmental parameters such as temperature, humidity, atmospheric pressure and sound and provide feedback; a fraction collection module, used for collecting reaction products; Central control module, used to realize the overall automatic control of hardware system, data transmission, storage and processing, as well as human-computer interaction; The software systems include: The saponification reaction condition recommendation module uses vector embedding technology to build a saponification literature knowledge base and applies search-enhanced generation technology combined with a large language model to achieve intelligent recommendation of process conditions. The experimental operating condition calculation module, based on a large language model, is used to: calculate the hardware system operating conditions based on process conditions; write experimental operating condition calculation components and realize automatic conversion from process conditions to operating conditions by calling components; The hardware system control code generation module is used to integrate the large language model to generate the control code of each hardware device in the hardware system. The hardware system operating conditions are passed as parameters to the hardware device control code to achieve accurate generation of hardware device control code under specific conditions. The experimental design table generation module, based on the large language model, is used to recommend experimental design types and generate experimental design tables based on the process conditions recommended by the saponification reaction condition recommendation module and / or the process condition range optimized by the experimental result qualitative analysis and optimization module; The experimental result qualitative analysis and optimization module is based on a large language model and is used to qualitatively analyze the experimental results corresponding to the process conditions recommended by the saponification reaction condition recommendation module, and optimize the process condition range based on the qualitative analysis results; The automatic modeling module of experimental results is used to automatically model the process conditions and results and output the optimal process conditions.

2. The intelligent agent for saponification reaction according to claim 1, characterized in that The hardware system also includes an online detection module, which is at least used for online detection of product saponification rate; The online detection module can be equipped with ultraviolet spectroscopy and near infrared spectroscopy equipment.

3. The intelligent agent for saponification reaction according to claim 1, characterized in that The infusion module is equipped with multiple horizontal flow pumps to transport raw material liquid, alkali solution and diluent respectively; The reaction module includes a first T-shaped tee for mixing the raw material liquid and the diluent to form a diluted raw material liquid, a second T-shaped tee for mixing the alkali solution and the diluted raw material liquid, a stainless steel coil for heating the reaction, and an oil bath pot and a reaction temperature sensor for controlling the reaction temperature.

4. The intelligent agent for saponification reaction according to claim 1, characterized in that The environmental information sensing module includes an environmental temperature, humidity and atmospheric pressure sensor and a sound sensor. The environmental temperature, humidity and atmospheric pressure sensors are used to record the temperature, humidity and atmospheric pressure of the environmental conditions. The sound sensor is used to record the sounds of the environment and equipment. The environmental information sensing module is used to help the intelligent agent perceive and make decisions.

5. The intelligent agent for saponification reaction according to claim 1, characterized in that In the experimental operating condition calculation module, the process conditions include reaction temperature, reaction time, raw material liquid mass fraction and alkali dosage, and the hardware system operating conditions include alkali liquid flow rate, raw material liquid flow rate, diluent flow rate and residence time. The calculation formulas are shown in formulas (I)-(IV): Where: Represents the volume flow rate of raw liquid; represents the volume of the tubing used for the reaction; stands for reaction time; Represents the density of the raw material liquid; represents the mass fraction of the raw material liquid; Represents the saponification value of raw materials; represents the mass fraction of alkali solution; represents the density of lye; Represents the ratio of actual alkali usage to theoretical alkali usage; Represents the mass fraction of the diluted raw material liquid formed by mixing the raw material liquid and the diluent; represents the diluent density; represents the volume flow rate of alkali solution; represents the diluent volume flow rate; represents the dwell time; It represents the maximum time from the raw material liquid, alkali solution and diluent flowing out from the liquid storage module to the mixing of the three.

6. The intelligent agent for saponification reaction according to claim 1, characterized in that The large language models used for the saponification reaction condition recommendation module, the experimental operation condition calculation module, and the hardware system control code generation module are DeepSeek-V3, ERNIE-3.5-8K, and ERNIE-3.5-8K, respectively; The large language model used for qualitative analysis of experimental results and optimization module is the fine-tuned QwQ-32B; The large language model used in the experimental design table generation module is the fine-tuned DeepSeek-R1.

7. The intelligent agent for saponification reaction according to claim 1, characterized in that The software system calls the central control module through the SDK, and writes a web page to realize software and hardware integrated control, perception, decision-making and visualization.

8. Use of the intelligent agent according to any one of claims 1 to 7 in saponification reaction.

9. An automatic optimization method for saponification reaction, characterized in that, include: The intelligent agent according to any one of claims 1 to 7 is used to realize automatic optimization of saponification reaction, and the execution process includes: The saponification reaction condition recommendation module builds a saponification literature knowledge base and uses search-enhanced generation technology combined with a large language model to achieve intelligent recommendation of saponification process conditions. Based on the experimental design table generated by the experimental design table generation module and / or the optimal process conditions output by the experimental result automatic modeling module, the intelligent agent automatically calls the experimental operation condition calculation module and the hardware system control code generation module to generate the operation condition control code and sends it to the central control module, and then the hardware system automatically runs; The module for qualitative analysis and optimization of experimental results automatically obtains the results of the saponification reaction, analyzes the experimental results through a large language model, optimizes the range of process variables, and outputs the optimized process condition range; The experimental design table generation module generates an experimental design table using a large language model based on the process conditions recommended by the saponification reaction condition recommendation module and / or the optimized process condition range output by the experimental result qualitative analysis and optimization module. The table includes factor level settings and implements randomization of the experimental sequence. The automatic modeling module of experimental results automatically models the process conditions and results and outputs the optimal process conditions.

10. The automatic optimization method for saponification reaction according to claim 9, characterized in that: The intelligent agent of claim 2 is used to realize automatic optimization of saponification reaction, and the execution process also includes: after the hardware system automatically runs, the online detection module reads data in real time, obtains the product saponification rate in real time and automatically saves it.

Images