Method for preparing acetic acid by hydrothermal resource of waste liquid crystal panel based on large language model

Abstract

The present application belongs to the technical field of electronic waste treatment, and particularly relates to a method for preparing acetic acid by hydrothermal resourceization of waste liquid crystal panel based on a large language model. The method uses waste liquid crystal panel as raw material, realizes intelligent recommendation and real-time optimization of hydrothermal reaction process parameters by relying on the large language model through raw material component analysis and feature input; combines with the language interaction module to complete human-computer intelligent collaborative control, dynamically regulates and controls the reaction process to efficiently generate acetic acid; the product is purified to obtain high-purity acetic acid product, and the optimization iteration and iterative optimization of process data are realized. The present application combines the intelligent regulation and control of the large language model with the hydrothermal resourceization technology, significantly improves the acetic acid yield and process stability, realizes the intelligentization, self-adaptation and high-value conversion of waste liquid crystal panel treatment, and provides an efficient new technical solution for the resource utilization of electronic waste.

Description

Technical Field

[0001] This invention belongs to the field of electronic waste treatment technology, and in particular relates to a method for preparing acetic acid from waste liquid crystal panels using hydrothermal resource recovery based on a large language model. Background Technology

[0002] With the rapid upgrading and iteration of electronic products, the number of waste liquid crystal displays (LCDs) has surged. LCD panels contain rare metals such as indium, as well as complex components like glass and organic polarizing films. Improper disposal can lead to serious environmental and resource waste. Current technologies for the resource recovery of LCD panels mainly focus on indium recycling, while insufficient attention is paid to the synergistic resource recovery of the glass substrate and organic matter, which constitute the largest proportion of the material. The hydrothermal method is a promising treatment approach that can convert organic matter into small-molecule organic acids (such as acetic acid) under certain temperature and pressure conditions, while simultaneously silicating the glass components. However, this process is greatly affected by fluctuations in raw material composition and reaction conditions (temperature, pressure, time, type and amount of catalyst). Traditional control methods rely on fixed process parameters or manual experience, making it difficult to optimize the acid production rate and yield. Furthermore, it requires highly skilled operators and has poor process adaptability. Hydrothermal methods can achieve directional depolymerization of organic components, but existing processes suffer from the following drawbacks: large fluctuations in raw material composition; process parameters relying on human experience, resulting in poor adaptability; the reaction process is invisible and unadjustable, making it difficult to stably control conversion rate and selectivity; lack of data-driven optimization and iteration, hindering continuous model and process optimization; and human-machine interaction relying on specialized codes / instructions, leading to high operational barriers and poor on-site adaptability. Currently, there is no technical solution that combines large language model semantic understanding, real-time big data decision-making, optimized intelligent control, and hydrothermal preparation of acetic acid from waste LCD panels. Summary of the Invention

[0003] This invention uses waste LCD panels as raw materials. The process involves component analysis and feature input of the waste LCD panels to generate a raw material feature vector, which is then input into a large-scale language model process decision module. This module automatically recommends an initial combination of process parameters for the hydrothermal reaction. Under this initial parameter combination, the hydrothermal reaction unit is initiated. During the reaction, real-time process parameters and intermediate product information are continuously collected, and the system responds to human-machine natural language commands through a language interaction and control module. The large-scale language model dynamically optimizes the process parameter combination based on real-time process parameters and intermediate product information, and controls the hydrothermal reaction unit to execute the optimized parameters. After the reaction, the liquid-phase product containing acetic acid is collected and purified to obtain a high-purity acetic acid product. Data feedback throughout the entire process enables continuous model iteration. This invention overcomes the shortcomings of existing technologies, such as reliance on manual experience, uncontrollable processes, low conversion efficiency, unstable product purity, and inability to iteratively optimize. It provides an intelligent, high-conversion-rate, high-selectivity, and self-optimizing method and system for the hydrothermal resource recovery of acetic acid from waste LCD panels.

[0004] The objective of this invention is achieved through the following technical solution: This invention provides a method for the hydrothermal resource recovery of acetic acid from waste liquid crystal panels based on a large language model, comprising the following steps: Step S1: Perform component analysis and feature input on the waste LCD panels to generate raw material feature vectors; Step S2: Input the raw material feature vector into the large language model process decision module, which will automatically recommend the initial process parameter combination for the hydrothermal reaction; Step S3: Start the hydrothermal reaction unit under the initial process parameter combination to carry out the reaction. During the reaction, continuously collect real-time process parameters and intermediate product information, and respond to human-machine natural language commands through the language interaction and control module. Step S4: The large language model dynamically optimizes the combination of process parameters based on the real-time process parameters and intermediate product information, and controls the hydrothermal reaction unit to execute the optimized parameters; after the reaction is completed, the liquid phase product containing acetic acid is collected and purified to obtain a high-purity acetic acid product; Step S5: Use the raw material feature vector, process parameter combination, process parameters and acetic acid yield data of this batch of reaction as optimization samples, and feed them back to the large language model process decision module for model iteration and update.

[0005] In step S2 of the present invention, the recommendation process for the initial process parameter combination includes: semantically matching the raw material feature vector with similar cases in the historical process database; generating at least three sets of candidate process parameter combinations and their corresponding yield prediction values ​​and safety confidence levels using a large language model; and selecting the combination with the highest safety confidence level and the yield prediction value that meets a preset threshold as the initial process parameter combination.

[0006] In step S3 of the present invention, when the language interaction and control module executes natural language instructions, it adopts a confidence-based parameter change mechanism: the natural language parameter change request input by the operator is converted into a numerical adjustment amount; the large language model module evaluates the expected impact of the adjustment amount on the acetic acid yield and the risk score on the safety of the reaction; the parameter change is executed only when the risk score is lower than the safety threshold and the expected yield impact is not negative.

[0007] In step S4 of the present invention, the dynamic optimization of process parameter combination includes: when the real-time detected rate of increase of acetic acid concentration is lower than the preset minimum threshold, the large language model module automatically determines that the current reaction is in an "inhibited state" and performs at least one of the following optimization actions: increasing the reaction temperature but not exceeding the upper limit temperature of the equipment; extending the heat preservation time; adjusting the pH value to the optimal range for acetic acid production; and adding oxidant.

[0008] In step S5 of the present invention, the optimization iteration adopts reinforcement learning or low-rank adaptive fine-tuning method based on human feedback, and performs domain-specific fine-tuning for the hydrothermal reaction task of discarded liquid crystal panels while retaining the original general knowledge of the large language model.

[0009] The beneficial effects of this invention are as follows: 1. High degree of intelligence: Intelligent decision-making throughout the entire process, from raw material matching and process recommendation to process control, eliminating reliance on human experience; 2. High conversion efficiency: Directed conversion of acetic acid significantly improves selectivity and yield; 3. Process controllability: Real-time data feedback allows for dynamic optimization of the reaction process; 4. Continuous iteration: Data is continuously optimized, and the model becomes more accurate with use; 5. Green and high-value: Waste is recycled and rendered harmless to produce high-value-added acetic acid, resulting in significant economic benefits. Attached Figure Description

[0010] Figure 1 is a schematic diagram of the process for preparing acetic acid from waste liquid crystal panels based on a large language model according to the present invention. Detailed Implementation

[0011] The present invention will now be described in detail with reference to specific embodiments, but these are by no means limitations on the present invention. Example 1

[0012] This embodiment uses waste mobile phone / display LCD panels as raw materials to prepare acetic acid through hydrothermal resource utilization according to the method described in claim 1, and achieves fully automated control of the entire process.

[0013] Step 1: Raw material component detection and raw material feature vector construction Waste LCD panel fragments were collected and mechanically crushed, ground, and subjected to iron and dust removal. The raw material characteristics were then analyzed: Organic content: volatile components such as liquid crystal polymers, polarizers, and binders accounted for 32.6%; Inorganic components: SiO2, ITO conductive layer, glass substrate, and metallic impurities; Impurities: halogens, trace metals, and residual liquid crystal monomers; Particle size distribution: D50 = 0.85 mm; Moisture content: 4.2%. A feature vector of the raw material was constructed based on the above composition, morphology, particle size, moisture content, and impurity content.

[0014] Step 2: Input the raw material feature vector into the large language model process decision module, which will automatically recommend the initial combination of process parameters for the hydrothermal reaction. The model training dataset includes: a historical hydrothermal resource utilization case library; simulation data of LCD panel degradation mechanism; natural language instruction semantic parsing data; the model automatically infers and outputs the optimal hydrothermal process parameter set adapted to this batch of raw materials based on the raw material characteristics: reaction temperature: 245℃; reaction pressure: 4.2MPa; reaction atmosphere: oxygen-rich oxidizing atmosphere; stirring rate: 400r / min; reaction time: 120min; initial system pH: 8.5.

[0015] Step 3: Start the hydrothermal reaction unit according to the initial process parameter combination. During the reaction, continuously collect real-time process parameters and intermediate product information. Real-time data collection includes: real-time temperature, pressure; real-time pH; gas phase components (CO2, small molecule hydrocarbons, residual oxygen). All data is transmitted back to the large language model in real time. The model calculates and dynamically adjusts in real time according to the reaction progress: temperature fine-tuning ±5℃; automatic pressure stabilization; automatic compensation for atmospheric oxygen content; and adaptive adjustment of stirring rate according to viscosity changes. Simultaneously, operators can issue commands through the natural language interaction module, such as "improve acetic acid selectivity," "accelerate reaction rate," and "reduce energy consumption." The language interaction and control module responds to these commands. Step 4: After the reaction is completed, the material is automatically discharged into the intelligent distillation and purification module, and undergoes the following processes in sequence: solid-liquid separation; membrane filtration desalination; vacuum distillation; decolorization and impurity removal; finally, high-purity acetic acid with a purity ≥99.2% is obtained.

[0016] Step 5: Upload all the following data to the big data platform: raw material characteristic data; initial process parameters; real-time reaction process data; acetic acid yield, purity, and selectivity; energy consumption and material conversion rate; the big data model completes incremental learning and iterative updates to make the prediction of the next batch of raw materials more accurate.

[0017] Results of this embodiment: Acetic acid yield: 38.6% (based on organic carbon conversion rate); Acetic acid purity: 99.2%; Resource conversion rate: 91.3%. Compared with the traditional manual adjustment process, acetic acid selectivity is improved by 18%, and energy consumption is reduced by 14%.

[0018] The above description of the embodiments is provided to enable those skilled in the art to understand and use the invention. It will be apparent to those skilled in the art that various modifications can be made to these embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the invention should be within the protection scope of the present invention.

Claims

1. A method for preparing acetic acid from waste liquid crystal panels using hydrothermal resource recovery based on a large language model, characterized in that, Includes the following steps: Step S1: Perform component analysis and feature input on the waste LCD panels to generate raw material feature vectors; Step S2: Input the raw material feature vector into the large language model process decision module, which will automatically recommend the initial process parameter combination for the hydrothermal reaction; Step S3: Start the hydrothermal reaction unit under the initial process parameter combination to carry out the reaction. During the reaction, continuously collect real-time process parameters and intermediate product information, and respond to human-machine natural language commands through the language interaction and control module. Step S4: The large language model dynamically optimizes the combination of process parameters based on the real-time process parameters and intermediate product information, and controls the hydrothermal reaction unit to execute the optimized parameters; after the reaction is completed, the liquid phase product containing acetic acid is collected and purified to obtain a high-purity acetic acid product; Step S5: Use the raw material feature vector, process parameter combination, process parameters and acetic acid yield data of this batch of reaction as optimization samples, and feed them back to the large language model process decision module for model iteration and update.

2. The method according to claim 1, characterized in that, In step S2, the recommendation process for the initial process parameter combination includes: semantically matching the raw material feature vector with similar cases in the historical process database; generating at least three sets of candidate process parameter combinations and their corresponding yield prediction values ​​and safety confidence levels using a large language model; and selecting the combination with the highest safety confidence level and the yield prediction value that meets a preset threshold as the initial process parameter combination.

3. The method according to claim 1, characterized in that, In step S3, when the language interaction and control module executes natural language instructions, it adopts a confidence-based parameter change mechanism: the natural language parameter change request input by the operator is converted into a numerical adjustment amount; the large language model module evaluates the expected impact of the adjustment amount on the acetic acid yield and the risk score on the safety of the reaction; the parameter change is executed only when the risk score is lower than the safety threshold and the expected yield impact is not negative.

4. The method according to claim 1, characterized in that, In step S4, the dynamic optimization of process parameter combination includes: when the real-time detected rate of increase of acetic acid concentration is lower than the preset minimum threshold, the large language model module automatically determines that the current reaction is in an "inhibited state" and performs at least one of the following optimization actions: increasing the reaction temperature but not exceeding the upper limit temperature of the equipment; extending the heat preservation time; adjusting the pH value to the optimal range for acetic acid production; and adding oxidant.

5. The method according to claim 1, characterized in that, In step S5, the optimization iteration adopts reinforcement learning or low-rank adaptive fine-tuning method based on human feedback. While retaining the original general knowledge of the large language model, it performs domain-specific fine-tuning for the hydrothermal reaction task of discarded liquid crystal panels.

Images