Rehmannia processing synergistic intelligent control system based on artificial intelligence large model

Abstract

The invention discloses a rehmannia processing synergistic intelligent control system based on an artificial intelligence large model, and belongs to the technical field of traditional Chinese medicine processing intelligent control. The system comprises a data acquisition module, a data transmission module, a rehmannia processing synergistic large model processing module, an intelligent control module and a man-machine interaction module. The data acquisition module is used for acquiring multi-source data such as temperature, humidity, pressure, moisture content and morphological characteristics in the processing process by using various sensors; the data transmission module realizes the transmission of the acquired data from the sensor to the large model processing module; the large model processing module deeply excavates a component-process-drug effect coupling relationship through data preprocessing, feature extraction, model training and SFT fine tuning control, and intelligently generates recommended process parameters; the intelligent control module accurately regulates and controls the processing equipment according to the process parameters output by the large model; the man-machine interaction module provides an operation and monitoring interface. The operation process of the system comprises the steps of starting, data acquisition, intelligent recommendation of the processing technology, transmission of the processing technology to the processing equipment, execution of the processing technology by the processing equipment and ending of operation. The processing quality stability and the production intelligence level are improved, the core component content of processed rehmannia can be effectively improved, and the problems that a traditional rehmannia processing flow depends on artificial experience, and the quality of processed products is unstable are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control of traditional Chinese medicine processing, and discloses an intelligent control system for processing rehmannia based on an artificial intelligence large model. BACKGROUND

[0002] Rehmannia is a commonly used traditional Chinese medicine with the effects of nourishing yin and blood, and tonifying essence and marrow. The processing technology of rehmannia has an important influence on its effective components and efficacy. The traditional processing technology of rehmannia mainly relies on manual experience, and has problems such as complicated processing steps, long processing time, unstable quality of finished products, low production efficiency, and high labor intensity. With the development of modernization of traditional Chinese medicine, higher requirements are put forward for the quality and efficiency of rehmannia processing, and the traditional processing technology has been difficult to meet the development needs of modern traditional Chinese medicine industry.

[0003] In recent years, the application of artificial intelligence technology in the field of traditional Chinese medicine has gradually increased, providing a new idea for the optimization and intelligent control of the processing technology of rehmannia. However, the current application of artificial intelligence mostly stays in the stage of simple data processing and pattern recognition, lacks deep understanding of the complex mechanism of the processing technology of rehmannia and dynamic optimization capability, fails to excavate the internal coupling relationship among components, technology and efficacy, and is difficult to realize precise control and efficiency improvement of the processing of rehmannia. The traditional deep learning model has limitations such as insufficient generalization ability, low knowledge fusion efficiency, and lagging dynamic optimization when dealing with complex nonlinear problems such as the processing of rehmannia.

[0004] CN116585393B discloses a processing technology of rehmannia, which realizes the effect of reducing the stickiness or yellow center of prepared rehmannia through secondary stewing. CN106421089B discloses a processing method of prepared rehmannia, which is prepared by steaming and drying under specific temperature and pressure conditions using yellow rice wine. CN108653468A discloses a processing method of prepared rehmannia, which is prepared by heating rehmannia added with yellow rice wine and rice flour under water.

[0005] All the above-mentioned patents are completely different from the present application, and do not propose the intelligent control technology and method of using an artificial intelligence large model for efficiency improvement of the processing of rehmannia, which has essential differences from the present application, and no similar application has been proposed. In order to solve the foregoing problems, the present application proposes an intelligent control system for efficiency improvement of the processing of rehmannia based on an artificial intelligence large model, which has important significance for realizing intelligent operation of the intelligent equipment for processing rehmannia and improving the processing efficiency and quality of finished products of rehmannia. SUMMARY

[0006] (I) Technical problems to be solved

[0007] The technical problem to be solved by the present application is that the current processing of prepared rehmannia mainly relies on manual experience, and is still dominated by the traditional "nine steaming and nine drying" process, which has problems such as complicated processing steps, long processing time, unstable processing quality, and low intelligentization, and even if simple artificial intelligence methods such as data processing and pattern recognition are applied, the effective mining of the coupling relationship among rehmannia active ingredients, processing technology and actual efficacy is still lacking. Therefore, the present application proposes a rehmannia processing and efficiency intelligent control system based on a large artificial intelligence model to realize precise control and efficiency of rehmannia processing.

[0008] (II) Technical solution

[0009] To solve the above technical problems, the technical solution of the present application is: a rehmannia processing and efficiency intelligent control system based on a large artificial intelligence model, comprising a data acquisition module, a data transmission module, a rehmannia processing and efficiency large model processing module, an intelligent control module and a man-machine interaction module. The joint deployment of the above modules can effectively improve the processing efficiency and product quality of rehmannia.

[0010] Further, the data acquisition module comprises one or more of a temperature sensor, a humidity sensor, a pressure sensor, a weighing sensor and an industrial camera; wherein the temperature sensor is used to monitor the temperature change inside the processing equipment during the steaming, soaking and drying process; the humidity sensor is used to monitor the humidity inside the processing equipment during the steaming, soaking and drying process; the pressure sensor is used to monitor the pressure inside the processing equipment during the steaming, soaking and drying process; the weighing sensor is used to monitor the water content of rehmannia during the steaming, soaking and drying process; and the industrial camera is used to collect the morphological characteristics of rehmannia such as size, color and texture.

[0011] Further, the data transmission module comprises one or more of Ethernet, Modbus, CAN bus and TCP / IP protocols to transmit the data acquired by the data acquisition module to the rehmannia processing and efficiency large model processing module.

[0012] Further, the rehmannia processing and efficiency large model processing module comprises a data preprocessing unit, a feature extraction unit, a model training unit and an SFT (Supervised Fine-Tuning) fine-tuning control unit.

[0013] Further, the data preprocessing unit preprocesses the collected multi-source data, which includes one or more of temperature, humidity, pressure, water content and morphological characteristics; and the preprocessing includes one or more of cleaning, noise reduction and normalization.

[0014] Further, the feature extraction unit is configured to extract key features related to the processing quality of rehmannia from the pre-processed data, the key features including one or more of temperature change curve, humidity change curve, pressure change curve, moisture content change curve, and rehmannia morphological feature parameters.

[0015] Further, the model training unit is configured to train and optimize the rehmannia processing enhancement model based on historical processing data and expert experience. The historical processing data includes processing parameters of rehmannia and corresponding core ingredient content, the processing parameters being one or more of temperature, humidity, pressure, moisture content, and time length at different processing stages; the processing stages including first steaming, second steaming, third steaming, first dampening, second dampening, third dampening, first drying, second drying, and third drying; the expert experience being the processing experience of old pharmacists, specifically the processing parameters of rehmannia from different producing areas, which are input into the model after digital processing.

[0016] Further, the SFT fine-tuning control unit includes a labeled data generator, an SFT fine-tuning engine, and an expert knowledge embedder.

[0017] Further, the labeled data generator is configured to construct a multi-modal labeled set based on historical processing data, the multi-modal labeled set including processing parameter-core ingredient content pairs and image-processing progress mappings.

[0018] Further, the SFT fine-tuning engine includes a pre-training stage and a fine-tuning stage, the pre-training stage using traditional Chinese medicine processing literature data and texts from old pharmacists' experience to pre-train a language model, preferably using a Transformer encoder with a layer value ranging from 1 to 24.

[0019] Further, the fine-tuning stage optimizes the specificity of the model to the rehmannia processing task through a supervision signal, and the loss function is defined as . The mean square error for process parameter prediction is , where is the model-predicted process parameter (such as temperature, humidity, pressure, moisture content, and time length), is the actual process parameter, and n is the number of samples. The core ingredient content prediction loss is , where is the model-predicted core ingredient content, is the actual content of the core ingredient, and m is the number of ingredient types. The cross-entropy loss based on expert scores of the rehmannia processed product's nourishing yin efficacy and blood tonifying efficacy is , where the model prediction efficacy score, the expert actual score, and The value range of each of the above is 1-10 points; k is the number of scoring samples. α, β, γ are weight coefficients, which satisfy the quantity relationship α+β+γ=1.

[0020] Further, the expert knowledge embedder is used to convert old medicine processing experience into constraint rules, and the SFT fine-tuning control unit is optimized by constructing virtual training samples and establishing a gradient penalty mechanism. The virtual training sample refers to the rule refined based on the old medicine processing experience, and the input-output data pair that is artificially constructed and conforms to the experience logic by simulating the correspondence between the process parameters, material state and expected results in the processing scene. The gradient penalty mechanism refers to imposing a penalty on the model prediction behavior deviating from the experience rule, guiding the model to learn the mapping relationship conforming to the expert knowledge. The loss function is constructed as where f(x) is the model output, r(x) is the rule expected result, λ is the penalty coefficient, and ϵ is the tolerance threshold. The value range of the penalty coefficient is 0.1-1.0, and the value range of the tolerance threshold is 0.01-0.5.

[0021] Further, the intelligent control module is connected with the rehmannia processing and efficiency increasing large model processing module, and control instructions are generated according to the control parameters output by the large model and transmitted to the execution mechanism for precise regulation and control of the rehmannia processing equipment. The intelligent control module is a PLC controller. The PLC controller is used to receive the control parameters output by the large model and convert them into specific control instructions sent to the execution mechanism. The control parameters include temperature, humidity, pressure, and time length at different processing stages. The execution structure includes one or more of a motor, a valve, a vacuum pump, a heater, and a spray head. The motor can control the speed of the processing equipment, the valve is a steam valve that can control the temperature, humidity, and pressure inside the processing equipment, the vacuum pump can control the pressure of the processing equipment, the heater can control the temperature and humidity inside the equipment, and the spray head can be used to spray liquid to control the humidity inside the equipment.

[0022] Further, the human-computer interaction module includes one or both of a touch screen and a monitoring large screen. The touch screen is used for user operation of system functions and can also be used to view real-time monitoring data; the monitoring large screen is used to display monitoring information in real time during the rehmannia processing process, and the monitoring information includes real-time data such as temperature, humidity, and pressure monitored by sensors, alarm information, etc.

[0023] The rehmannia processing and efficiency increasing intelligent control system based on artificial intelligence large model includes the following steps:

[0024] Step 1: Start the system: open the operation interface of the intelligent control system for processing Rehmannia root based on artificial intelligence large model;

[0025] Step 2: Collect data: click the "Start collection" button on the touch screen, the system automatically records the initial data, including temperature, humidity, pressure, etc., and real-time data collection and display on the monitoring screen, forming temperature change curve, humidity change curve, pressure change curve.

[0026] Step 3: Intelligent recommendation of processing technology: manually input the origin and weight of Rehmannia root, click "Start matching", the large model starts to calculate and infer. After the inference is completed, the touch screen displays the recommended process parameters, the predicted core ingredient content and the predicted efficacy score.

[0027] Step 4: Transfer the processing technology to the processing equipment: the intelligent control system automatically standardizes the recommended processing technology parameters, and establishes a safe data channel with the processing equipment control system through the data transmission module. The system first checks the current state of the equipment, if the equipment state is normal, the process parameter package is encrypted and transmitted to the PLC controller of the processing equipment, at the same time, the transmission log is generated in the human-computer interaction module, recording the transmission time, parameter integrity check result, for traceability and troubleshooting.

[0028] Step 5: Processing equipment executes processing technology: after receiving the process parameters, the processing equipment adjusts and controls the processing equipment according to the "steaming-moistening-drying" cycle parameters. In the steaming link, through the linkage of steam generating device and temperature sensor, the steaming temperature fluctuation is real-time feedback and dynamically corrected; In the moistening link, through the cooperation of humidity sensor, spray head and valve, the moistening environment is maintained stable; In the drying link, according to the set humidity and temperature target, the running power of the heater is adjusted. In the execution process, the equipment real-time feedback operation data to the intelligent control system, if the parameter exceeds the pre-set limit value, the early warning is triggered immediately and the automatic correction is tried, if the correction is invalid, the equipment is suspended and the fault prompt is pushed to the human-computer interaction interface.

[0029] Step 6: End of Operation: After the equipment completes three cycles of "steaming-moistening-drying", the system initiates the final judgment stage: An industrial camera is used to acquire images of the processed Rehmannia glutinosa product. Color parameters are analyzed using machine vision algorithms and compared with a preset endpoint color standard. If the color data's match with the preset endpoint is below a threshold, the system pushes a message to the human-machine interface, displaying a pop-up prompt: "The current Rehmannia glutinosa color differs from the standard endpoint. Should we increase the drying / steaming time?" Two operation options are provided: "Increase Time" and "Maintain End". If the system chooses to increase the time, it automatically retrieves the appropriate process, pushes new parameters to the equipment to execute the supplementary process, and triggers color judgment again until the standard is met or manual termination occurs. If the system chooses to "Maintain End" or the second judgment meets the standard, it sends a shutdown command to the equipment, shutting down all subsystems and releasing resources, returning to standby mode.

[0030] (III) Beneficial Effects

[0031] After adopting the above technical solution, the present invention has the following beneficial effects:

[0032] (1) Improve the stability of processing quality: Through in-depth analysis of the internal relationship between processing parameters such as temperature, humidity, pressure, moisture content and time and the changes in the content of core components such as rehmannia glycoside, verbascoside, 5-hydroxymethylfurfural and rehmannia polysaccharide, as well as the coupling relationship between component changes and the effects of nourishing yin and replenishing blood, the transformation from "experience control" to "data-driven" is realized, and the stability of processing quality is significantly improved.

[0033] (2) Improve the level of intelligent production: Integrate multi-dimensional data collected by temperature sensors, humidity sensors, pressure sensors, weighing sensors, industrial cameras, etc., to build a comprehensive large model of the processing process, effectively solve the quality fluctuation problem caused by the single data dimension of traditional methods, and use the ability of the large model to comprehensively process massive data to intelligently match and recommend processing parameters, thereby improving the level of intelligent processing of Rehmannia glutinosa.

[0034] (3) Increase the content of core components of Rehmannia glutinosa after processing: Through the intelligent control system of the present invention, compared with the traditional "nine steaming and nine drying" processing process, the content of rehmannia glycosides is increased from 0.4-0.6% to 0.6-0.8%, verbascoside is increased from 0.08-0.10% to 0.11-0.15%, 5-hydroxymethylfurfural is decreased from 0.4-0.5% to 0.2-0.3%, and Rehmannia glutinosa polysaccharide is increased from 8-9% to 10-15%. Attached Figure Description

[0035] In order to make the technical solutions in the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced. Obviously, the accompanying drawings in the following description only only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of these drawings.

[0036] Figure 1 The system architecture diagram of the present application shows the composition and connection relationship of the data acquisition module, the data transmission module, the rehmannia processing efficiency enhancement model processing module, the intelligent control module and the man-machine interaction module.

[0037] Figure 2 The SFT fine-tuning control unit workflow diagram includes the workflow and mutual relationship of the labeled data generator, the SFT fine-tuning engine and the expert knowledge embedder. DETAILED DESCRIPTION

[0038] In order to make the content of the present application more easily and clearly understood, the present application will be further described in detail according to specific embodiments and in conjunction with the accompanying drawings.

[0039] Embodiment 1

[0040] Take 50 kg of raw rehmannia produced in Wenxian, Henan as raw material, which has an initial form of strip block, yellow, diameter 3-5 cm, moisture content 16%. After starting the system, the temperature sensor collects the initial temperature of the processing container as 25℃, the humidity sensor records the environmental humidity as 60%, and the industrial camera transmits the form feature of rehmannia to the large model processing module. In the pre-training stage of the large model, a 12-layer Transformer encoder is used, and language model training is performed based on 5000 pieces of traditional Chinese processing literature data and 200 groups of old medicine worker experience texts. In the fine-tuning stage, the model is optimized through more than 300 groups of "temperature-humidity-processing stage-time length→core ingredient content" data pairs and image-processing progress mapping. The old medicine worker experience texts adopt natural language structured texts, follow the binary structure of "process parameter description + experience conclusion", and are stored in JSON format text files for structured parsing and model calling; The "temperature-humidity-processing stage-time length→core ingredient content" data pair adopts a five-tuple mapping form, and the processing stage is divided into first steaming, second steaming, third steaming, first dampening, second dampening, third dampening, first drying, second drying and third drying, corresponding to the "three steaming and three drying" process in actual processing.

[0041] In the reasoning process, the large model calculates to generate three cycles of parameters. First steaming: temperature 110°C, duration 6h, after 12h of wetting, dry at 55°C to moisture content 25%; second steaming: temperature 105°C, duration 4h, after 12h of wetting, dry at 60°C to moisture content 22%; third steaming: temperature 105°C, duration 4h, after 10h of wetting, dry at 45°C to moisture content 12%. After receiving the parameters, the PLC controller controls the steaming process through the steam valve opening and the tank body speed. When the temperature overshoots to 112°C, the system triggers the PID compensation by reducing the heater power, and returns to 110±1°C within 30min. At the end of the run, the industrial camera collects the finished product color RGB value (50,30,20), which matches the preset endpoint (45,25,15) with a matching degree of 92%, and the discharge is 61kg. The sample is detected in the laboratory, and the content of rehmannioside is 0.61%, the content of verbascoside is 0.12%, the content of 5-hydroxymethylfurfural is 0.29%, and the content of rehmannia polysaccharide is 10.1%.

[0042] Example 2

[0043] Take 60kg of raw rehmannia produced in Bozhou, Anhui as raw material, the initial form is clump, dark yellow, diameter 2-4cm, moisture content 18%. After starting the system, the temperature sensor collects the initial temperature of the processing container as 28°C, the humidity sensor records the environmental humidity as 55%, and the industrial camera transmits the rehmannia morphological characteristics to the large model processing module. In the training stage of the large model, a 16-layer Transformer encoder is used, and language model training is performed based on 8000 pieces of traditional Chinese medicine processing literature data and 300 sets of old medicine worker experience texts. The old medicine worker experience texts are stored in JSON format. In the fine-tuning stage, the optimized model is used with 400 sets of “temperature-humidity-duration-moisture content→core ingredient content” data, and the loss function is calculated as follows:

[0044] (1) Process parameter prediction mean square error:

[0045] Take the first steaming temperature as an example, the model prediction value is 108°C, the actual process parameter is 110°C, and the sample number n=100:

[0046] (2) Core ingredient content prediction loss:

[0047] Take the content of rehmannioside as an example, the model prediction value is 0.65%, the actual content is 0.68%, and the number of ingredient types m=5:

[0048]

[0049] (3) Efficacy score cross-entropy loss:

[0050] The expert scores the efficacy of nourishing yin as 8 points, the model prediction value is 7.5 points, and the score sample number k=200:

[0051]

[0052] (4) Total loss function (weight coefficients a = 0.4, b = 0.4, g = 0.2):

[0053] After large model calculation and inference, the output process parameters are: first steaming: temperature 108℃, time 5h, after 14h of dampening at 60℃, drying to moisture content 28%; second steaming: temperature 105℃, time 4h, after 12h of dampening at 55℃, drying to moisture content 25%; third steaming: temperature 107℃, time 4h, after 10h of dampening at 48℃, drying to moisture content 14%. After the PLC controller receives the parameters, the steaming process is controlled through the steam valve opening and the tank body rotating speed. When the temperature overshoots to 58℃, the system triggers the PID compensation, and returns to 55±1℃ within 20min. When the operation ends, the industrial camera collects the product color RGB value (48, 32, 22), which matches the preset endpoint (45, 25, 15) with a matching degree of 89%, and the discharge is 70kg. After laboratory detection, the content of rehmannioside is 0.68%, the content of verbascoside is 0.13%, the content of 5-hydroxymethyl furfural is 0.27%, and the content of rehmannia polysaccharide is 11.5%.

Claims

1. An artificial intelligence large model-based processing and synergistic intelligent control system for rehmannia, characterized in that, The system comprises a data acquisition module, a data transmission module, a rehmannia processing and efficiency enhancement large model processing module, an intelligent control module, and a human-computer interaction module. The core component of the rehmannia processing and efficiency enhancement large model processing module is an SFT fine-tuning control unit, which comprises a labeled data generator, an SFT fine-tuning engine, and an expert knowledge embedder. The SFT fine-tuning engine is pre-trained as a language model based on a Transformer encoder, and the specificity of the model to the rehmannia processing task is optimized through a loss function. The operation process of the rehmannia processing and efficiency enhancement intelligent control system comprises starting the system, acquiring data, determining rehmannia processing parameters through large model calculation and intelligent matching, transmitting the processing parameters to the processing equipment, executing the processing parameters by the processing equipment, and ending the operation.

2. The intelligent control system of claim 1, wherein, The data acquisition module comprises one or more of a temperature sensor, a humidity sensor, a pressure sensor, a weighing sensor, and an industrial camera. The temperature sensor is used to monitor the temperature change inside the processing equipment during steaming, soaking, and drying in real time. The humidity sensor is used to monitor the humidity inside the processing equipment during steaming, soaking, and drying in real time. The pressure sensor is used to monitor the pressure inside the processing equipment during steaming, soaking, and drying in real time. The weighing sensor is used to monitor the water content of rehmannia during steaming, soaking, and drying in real time. The industrial camera is used to acquire the size, color, and texture of rehmannia.

3. The intelligent control system of claim 2, wherein, The rehmannia processing and efficiency enhancement large model processing module comprises a data preprocessing unit, a feature extraction unit, a model training unit, and an SFT fine-tuning control unit. The data preprocessing unit performs cleaning, noise reduction, and normalization processing on multi-source data such as temperature, humidity, pressure, water content, and morphological characteristics. The feature extraction unit extracts temperature change curves, humidity change curves, water content change curves, pressure change curves, and rehmannia morphological characteristic parameters related to rehmannia processing quality from the preprocessed data.

4. The intelligent control system of claim 3, wherein, The model training unit trains and optimizes the model based on historical processing data containing different processing parameters such as temperature, humidity, pressure, water content, and duration, as well as the corresponding core ingredient content, and digitized old medicine processing experience.

5. The intelligent control system of claim 4, wherein, The labeled data generator constructs a multi-modal labeled set containing processing parameter-core ingredient content pairs and image-processing progress mappings based on historical processing data.

6. The intelligent control system of claim 5, wherein, The expert knowledge embedder converts the old drug experience into constraint rules, and optimizes the SFT fine-tuning control unit by constructing virtual training samples and establishing a gradient penalty mechanism, where the formula of the gradient penalty mechanism is .

7. The intelligent control system of claim 6, wherein, The intelligent control module is one or more PLC controllers connected to the rehmannia processing and efficiency enhancement large model processing module. The PLC controller receives the control parameters output by the large model and converts them into control instructions sent to the execution mechanism. The execution mechanism comprises one or more of a motor, a valve, a vacuum pump, and a spray head, which are used to control the processing equipment.

8. The intelligent control system of claim 7, wherein, The human-computer interaction module comprises one or both of a touch screen and a monitoring large screen. The touch screen is used to input or modify processing parameters, and view real-time monitoring data. The monitoring large screen is used to display monitoring information such as processing parameters, equipment status, and alarm information in real time.

Images