Preparation method of traditional Chinese medicine compound medicine for treating rapid atrial fibrillation

Abstract

The invention discloses a preparation method of a traditional Chinese medicine compound medicine for treating rapid atrial fibrillation, which comprises the following steps: taking coptis chinensis, and adding a phosphate buffer solution for extraction; collecting optical data of the extracting solution in real time, and inputting the optical data into a multi-channel convolutional neural network model; when the output comprehensive extraction efficiency index reaches a preset threshold value, extraction is stopped, and a coptis chinensis extracting solution is obtained; and acquiring concentration time sequence data of tanshinone IIA, paeoniflorin and glycyrrhizic acid in a decoction in real time, inputting the concentration time sequence data into the long-short-term memory neural network model, stopping decoction when the fluctuation range of interaction effect parameters in three continuous detection periods is less than or equal to + / -5%, and stopping decoction when the fluctuation range of the interaction effect parameters in three continuous detection periods is less than or equal to + / -5%. A mixed decoction is obtained; combining the coptis chinensis extracting solution and the mixed decoction, and concentrating to obtain concentrated extract; preparing the traditional Chinese medicine compound medicine. The preparation efficiency and the quality stability of the traditional Chinese medicine compound medicine can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of traditional Chinese medicine. More particularly, the present application relates to a preparation method of a traditional Chinese medicine compound for treating rapid atrial fibrillation. BACKGROUND

[0002] In the field of traditional Chinese medicine compound preparation, there are long-term technical bottlenecks in the preparation process of drugs for treating rapid atrial fibrillation. Traditional extraction processes use fixed temperature and time parameters, such as the invention patent with application number 2024117757811, which cannot match the differentiated dissolution characteristics of multiple components in medicinal materials. For example, constant temperature extraction can easily lead to degradation of effective components or insufficient dissolution, and single index control at the end of extraction cannot take into account related parameters such as polysaccharide polymerization degree and cell wall breaking rate, resulting in insufficient batch-to-batch efficacy uniformity. The low dissolution efficiency of mineral medicinal materials constitutes another technical difficulty. Mineral medicines such as calcined dragon bone and calcined oyster shell are difficult to break through the physical barriers of mineral lattices and cell walls with traditional decoction methods, even if the process of extending the decoction time or increasing the temperature is used, the dissolution efficiency cannot be effectively improved, and the destruction of heat-sensitive components in plant medicines will be intensified. The dissolution kinetics of liposoluble components such as tanshinones and water-soluble components such as paeoniflorin are significantly different, and traditional processes cannot capture the interaction dynamics between components in real time.

[0003] The essence of the above technical bottlenecks lies in the lack of a multi-dimensional process monitoring system and dynamic control mechanism in traditional processes. Although existing detection technologies attempt to introduce spectral analysis methods, due to the complex correlation of multi-channel data, traditional chemometrics methods cannot effectively analyze them. Therefore, it is necessary to design a technical solution that can overcome the above-mentioned defects. SUMMARY

[0004] An object of the present application is to provide a preparation method of a traditional Chinese medicine compound for treating rapid atrial fibrillation, which can improve the preparation efficiency and quality stability of traditional Chinese medicine compound drugs.

[0005] To achieve these objects and other advantages and in accordance with the purpose of the application, as embodied and broadly described herein, the present application provides a preparation method of traditional Chinese medicine compound medicine for treating rapid atrial fibrillation, comprising: S1: taking Huanglian, adding phosphate buffer for extraction; S2: real-time acquisition of optical data of the extract by ultraviolet-near infrared combined spectrometer, inputting a pre-trained multi-channel convolutional neural network model; wherein in the multi-channel convolutional neural network model, the first convolutional channel analyzes the absorbance data at 280 nm wavelength, the second convolutional channel processes the near-infrared spectrum data of 1050-1100 nm, and the third convolutional channel extracts the spectrum features of 2200-2400 nm; when the comprehensive extraction efficiency index output by the model reaches the predetermined threshold, the extraction is terminated, and the Huanglian extract is obtained; S3: taking Taizishen, Huangjing, Shudihuang, Baishao, Sangshen, Zaoren, Wuyizi, Huaimai, Maidong, Chuanqiong, Danshen, Shengdi, Chishao, Gancao, Foshou, Chuanlonggu, Chuanmushixie and Meigui, and performing decoction; S4: real-time acquisition of the concentration time series data of tanshinone IIA, paeoniflorin and glycyrrhizic acid in the decoction liquid by high performance liquid chromatograph, inputting a long short-term memory neural network model based on attention mechanism; wherein in the long short-term memory neural network model, the bidirectional long short-term memory layer extracts the dissolution rate characteristics of each component, the attention mechanism layer calculates the synergistic weight coefficient of tanshinone IIA and paeoniflorin, and the output layer generates the flavonoid-saponin interaction effect parameter; when the fluctuation range of the interaction effect parameter is ≤±5% for three consecutive detection periods, the decoction is terminated, and the mixed decoction liquid is obtained; S5: combining the Huanglian extract and the mixed decoction liquid, treating by microfiltration membrane, then alcohol precipitation, collecting the alcohol precipitation supernatant, concentrating to obtain the concentrated extract; S6: mixing the concentrated extract with a pharmaceutically acceptable carrier to prepare a traditional Chinese medicine compound medicine.

[0006] Further, in S1, Huanglian is crushed to 40-60 mesh, phosphate buffer with pH 6.8 is added, and the solid-liquid ratio is 1:8-1:12; three-stage temperature-controlled extraction is performed, the first stage is constant temperature extraction at 45°C for 30 minutes, the second stage is pulse ultrasonic extraction by increasing the temperature to 65°C, and the third stage: when the conductivity of the extract increases by ≤1% / 5 minutes for three consecutive detections, switch to 85°C hot reflux extraction.

[0007] Further, in S2, the first convolutional channel is configured to process the 280 nm absorbance time series data using a 3x1 convolution kernel, outputting a first feature map; the second convolutional channel is configured to extract the features of the 1050-1100 nm near-infrared spectrum using a 5x1 dilated convolution kernel, outputting a second feature map, and the dilated rate of the dilated convolution is 2; the third convolutional channel is configured to capture the features of the 2200-2400 nm spectrum using a 7x1 convolution kernel combined with an exponential linear unit activation function, outputting a third feature map; the first feature map, the second feature map, and the third feature map are spliced in the channel dimension to form an initial fusion feature, and a 1x1 convolution kernel operation is applied to the initial fusion feature to generate a channel attention weight matrix; a three-dimensional convolution kernel operation is applied to the initial fusion feature to generate a space-spectrum correlation feature, and an element multiplication operation is performed on the channel attention weight matrix and the space-spectrum correlation feature to output an optimized fusion feature; the optimized fusion feature is input into a fully connected layer to generate a multi-dimensional feature vector containing the berberine dissolution parameter, the polysaccharide degree of polymerization parameter, and the cell wall breaking rate parameter; and a comprehensive extraction efficiency index is calculated based on the multi-dimensional feature vector.

[0008] Further, the maximum value normalization is performed on the berberine dissolution parameter to generate a standard dissolution parameter, the logarithmic conversion processing is applied to the polysaccharide degree of polymerization parameter to generate a standard degree of polymerization parameter, and the Sigmoid function conversion is performed on the cell wall breaking rate parameter to generate a standard wall breaking rate parameter; the standard dissolution parameter, the standard degree of polymerization parameter, and the standard wall breaking rate parameter are input into a fully connected layer to output an initial weight vector; the initial weight vector is spliced with a medicinal material batch characteristic code, and the medicinal material batch characteristic code contains a production place code, a harvesting season code, and a storage time length code; a dynamic weight coefficient is generated through a gated recurrent unit, wherein the berberine dissolution parameter weight coefficient ranges from 0.55 to 0.65, the polysaccharide degree of polymerization parameter weight coefficient ranges from 0.25 to 0.35, and the cell wall breaking rate parameter weight coefficient ranges from 0.08 to 0.12; and the comprehensive extraction efficiency index is calculated according to the following formula: comprehensive extraction efficiency index = dynamic weight coefficient α x standard dissolution parameter + dynamic weight coefficient β x standard degree of polymerization parameter + dynamic weight coefficient γ x standard wall breaking rate parameter, and α, β, and γ are dynamic weight coefficients.

[0009] Further, it also includes: acquiring the comprehensive extraction efficiency index output by the multi-channel convolutional neural network model in real time, calculating the confidence score output by the multi-channel convolutional neural network model, and generating the confidence score by performing Monte Carlo random dropout sampling on the optimized fusion feature based on the inverse variance of 10 consecutive prediction results; when the confidence score is lower than the 0.85 threshold, the extraction is not terminated according to the comprehensive extraction efficiency index, but is terminated when the cumulative dissolution amount of berberine is not less than the maximum dissolution amount.

[0010] Further, in S3, the Oyster Shell Powder and the Sichuan Greenbrier Rhizome Powder are crushed to 100-120 mesh, the first stage uses a 1.0% citric acid solution with a liquid-solid ratio of 3:1 at 50°C for 15 minutes, the second stage uses a 0.5% citric acid solution with a liquid-solid ratio of 5:1 at 70°C for 20 minutes, the filtrates of the first stage and the second stage are combined to obtain a mineral filtrate; the Radix Pseudostellariae, the Polygonatum, the Prepared Rehmannia, the Raw Rehmannia, the Red Peony Root, the Szechuan Lovage Rhizome, and the Salvia miltiorrhiza are added with an acetic acid buffer solution with a pH of 5.8 and a liquid-solid ratio of 8:1, and boiled at 80°C for 25 minutes to obtain a rhizome decoction; the Jujube, the Schisandra, the wheat, the Ophiopogon, the Licorice, the Bergamot, and the Rose are added with the rhizome decoction, the mineral filtrate, and purified water, and the total liquid-solid ratio is controlled to be 12:1, and refluxed and boiled at 95°C.

[0011] Further, in S4, the input layer of the long short-term memory neural network model is configured to synchronously receive five independent time sequence data channels: the first channel inputs the tanshinone IIA concentration time sequence data, the second channel inputs the paeoniflorin concentration time sequence data, the third channel inputs the glycyrrhizic acid concentration time sequence data, the fourth channel inputs the decoction system temperature time sequence data, and the fifth channel inputs the decoction system pressure time sequence data; the five channel data are respectively input into the corresponding processing units in the bidirectional long short-term memory layer, and respectively output the tanshinone IIA dissolution feature vector, the paeoniflorin dissolution feature vector, the glycyrrhizic acid dynamic feature vector, the temperature dynamic feature vector, and the pressure dynamic feature vector; the tanshinone IIA dissolution feature vector, the paeoniflorin dissolution feature vector, the glycyrrhizic acid dynamic feature vector, the temperature dynamic feature vector, and the pressure dynamic feature vector are spliced; the spliced features are input into the attention mechanism layer, processed by the full connection layer and the hyperbolic tangent activation function to generate an intermediate feature matrix, and then normalized by the Softmax function to output a collaborative weight coefficient; the collaborative weight coefficient is fused with the tanshinone IIA dissolution feature vector, the paeoniflorin dissolution feature vector, and the glycyrrhizic acid dynamic feature vector; the fused features are processed by the output layer to generate a flavonoid-saponin interaction effect parameter normalized to the interval of 0 to 1.

[0012] Further, in S5, after the Coptis extract and the mixed decoction liquid are combined, microfiltration is performed on the microfiltration membrane with a pore size of 0.22 μm; the microfiltration operation pressure is controlled in three stages: the first stage is 0.25 MPa for 5 minutes, the second stage is increased to 0.40 MPa for 3 minutes, and the third stage is decreased to 0.18 MPa until the filtration endpoint; the obtained microfiltrate is subjected to gradient alcohol precipitation: first add ethanol to make the alcohol concentration reach 60%±2%, separate and discard the precipitate after standing at 4°C for 20 minutes; then add ethanol to the supernatant to make the alcohol concentration reach 80%±1%, collect the alcohol precipitation supernatant after standing at 25°C for 15 minutes; the alcohol precipitation supernatant is concentrated under reduced pressure at 45°C±1°C and a vacuum degree of-0.08 MPa, and the concentration is terminated when the relative density of the concentrated extract is 1.25-1.3.

[0013] The present application at least includes the following beneficial effects: The present application realizes the coordinated dissolution of alkaloids and polysaccharide components in the extraction of Coptis chinensis by controlling temperature in stages and combining a multi-channel spectrum-neural network monitoring system, which not only guarantees the structural stability of heat-sensitive components, but also improves the cell wall breaking efficiency, breaking through the contradiction between component degradation and insufficient dissolution in traditional constant temperature extraction. The gradient acid treatment process of mineral medicine specifically destroys the mineral lattice structure, and cooperates with the grouping decoction strategy to effectively promote the dissolution of inorganic components and the release balance of active components of plant medicine, solving the technical bottleneck of low dissolution efficiency of mineral medicine in traditional water decoction. During the compound decoction process, the dynamic model based on long short-term memory neural network and attention mechanism can capture the time sequence characteristics and interaction effects of multi-component dissolution in real time, realize the precise regulation of the coordinated release of flavonoids and saponins, and avoid the process lag caused by traditional offline detection.

[0014] Other advantages, objects, and features of the present application will be apparent to those skilled in the art from the following specification. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 Flowchart of an embodiment of the present application. DETAILED DESCRIPTION

[0016] The present application will be further described in detail below, so that those skilled in the art can implement it according to the description.

[0017] It should be understood that the terms such as "have", "contain" and "include" used in the embodiments of the present application do not exclude the presence or addition of one or more other elements or combinations thereof. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present application are only used to explain the relative position relationship, movement condition, etc. between the components in a certain posture, and if the posture changes, the directional indications also change accordingly. When an element is referred to as "fixed to" or "disposed on" another element, it can be directly on another element or there can be a middle element. When an element is referred to as "connected to" another element, it can be directly connected to another element or indirectly connected to another element through a middle element. The embodiments of the present application refer to "first", "second", etc. for the purpose of description only, and cannot be understood as indicating or implying the relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features limited by "first", "second" can explicitly or implicitly include at least one of the features.

[0018] It should be explained that the technical solutions among various embodiments of the present application can be combined with each other, but it must be based on that a person skilled in the art can realize, when the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, and is not within the protection scope required by the present application.

[0019] As Figure 1 shown, the embodiments of the present application provide a preparation method of traditional Chinese medicine compound medicine for treating rapid atrial fibrillation, comprising: S1 taking Coptis chinensis, adding phosphate buffer solution for extraction; S2 real-time collecting optical data of the extraction solution by using ultraviolet-near infrared combined spectrometer, inputting a pre-trained multi-channel convolutional neural network model, wherein the first convolutional channel analyzes the absorbance data at 280 nm wavelength, the second convolutional channel processes the near infrared spectrum data of 1050-1100 nm, and the third convolutional channel extracts the spectrum features of 2200-2400 nm, when the comprehensive extraction efficiency index output by the model reaches the predetermined threshold, the extraction is terminated, and the Coptis chinensis extraction solution is obtained; S3 taking Radix Pseudostellariae, Rhizoma Polygonati and other medicinal materials to perform decoction; S4 real-time obtaining the concentration time series data of tanshinone IIA, paeoniflorin and glycyrrhizic acid in the decoction liquid by using high performance liquid chromatograph, inputting a long short-term memory neural network model based on attention mechanism, when the fluctuation range of the interaction effect parameter is ≤±5% for three consecutive detection periods, the decoction is terminated; S5 combining the extraction solution and the decoction liquid, alcohol precipitation after microfiltration membrane treatment, and collecting the supernatant for concentration; S6 mixing the concentrated extract with a pharmaceutical carrier to prepare the medicine.

[0020] Exemplarily, in S1, the Coptis chinensis can be processed by a crushing device to a conventional medicinal mesh size, and added into a phosphate buffer with a pH value of 6.8±0.2, and the solid-liquid ratio is controlled between 1:8 and 1:12, and specifically, 1:8, 1:10 or 1:12 can be selected. In S2, a commercially available device such as PerkinElmer Lambda series can be selected for the ultraviolet-near infrared spectrometer, and the spectral data is collected in real time with a sampling interval of 1 minute, wherein the absorbance at 280 nm corresponds to the characteristic absorption of berberine, the near-infrared band of 1050-1100 nm reflects the hydroxyl vibration of polysaccharide, and the region of 2200-2400 nm captures the functional group information of alkaloids. When the multi-channel convolutional neural network model is pre-trained, the first convolutional channel uses a 3×1 convolutional kernel to analyze the time series characteristics of the absorbance at 280 nm, the second convolutional channel extracts the near-infrared spectral interval characteristics through a 5×1 hollow convolutional kernel (expansion rate 2), and the third convolutional channel uses a 7×1 convolutional kernel combined with an ELU activation function to process the mid-infrared spectrum. After the three-channel feature fusion, the comprehensive extraction efficiency index is output through the fully connected layer, and the predetermined threshold can be set to 0.85. In S3, the cooking operation can use a stainless steel sandwich pot, and the stirring rate is controlled at 50-80 rpm. In S4, the high-performance liquid chromatograph such as Agilent 1260 series is used to monitor the component concentration online, the time series data is input into the Bi-LSTM model, the attention mechanism layer calculates the component synergistic weight, and the interaction effect parameter fluctuation threshold is set to ±5%. In S5, the microfiltration membrane aperture is selected to be 0.22 μm, and the ethanol concentration is adjusted to 60% and 80% in two steps in the alcohol precipitation process, and the concentration temperature is controlled at 45℃±1℃. In S6, the pharmaceutical carrier can be selected from starch, lactose and other commonly used excipients.

[0021] In the prior art, the traditional preparation process uses fixed temperature and time extraction, and only controls the endpoint by offline detection of a single component (such as berberine), which cannot meet the synergistic dissolution requirements of multiple components, resulting in coexistence of degradation of heat-sensitive components and insufficient dissolution of effective components, and significant batch-to-batch efficacy differences. The present embodiment realizes dynamic regulation of the extraction and cooking process by coupling ultraviolet-near infrared spectrum real-time monitoring with a multi-channel neural network model, accurately controls the process endpoint according to the comprehensive efficiency index and the interaction effect parameter, and effectively improves the synergism of multiple component dissolution and the consistency of preparation quality. In another embodiment, in S1, the Coptis chinensis is crushed to 40-60 mesh, added into a phosphate buffer with a pH of 6.8, and the solid-liquid ratio is 1:8-1:12; three-stage temperature-controlled extraction, the first stage is constant temperature extraction at 45℃ for 30 minutes, the second stage is pulse ultrasonic extraction at 65℃, and the third stage is hot reflux extraction at 85℃ when the conductivity of the extraction liquid increases by ≤1% / 5 minutes for three consecutive times. Exemplarily, the pulverization of Coptis chinensis can adopt a universal pulverizer, and the pulverization mesh number is selected as 40 mesh, 50 mesh or 60 mesh to ensure the uniformity of particles. The phosphate buffer solution with pH 6.8 is prepared by a conventional ratio of potassium dihydrogen phosphate and sodium phosphate dibasic, and the solid-liquid ratio can be selected as 1:8, 1:10 or 1:12. In the three-stage temperature control extraction, the first stage uses a constant temperature water bath to maintain 45°C, and the extraction time is 30 minutes to protect the heat-sensitive components such as berberine; the second stage is heated to 65°C, and the pulse ultrasonic extraction instrument (ultrasonic power 300-500W, pulse frequency 10-20Hz) is used to strengthen the cell wall breaking; the third stage is monitored in real time by a DDS-307 type conductivity meter, and when the increase amplitude is less than or equal to 1% for three consecutive times (such as 0.8% / 5min, 0.9% / 5min), the hot reflux device matched with the rotary evaporator is switched to, and dynamic extraction is carried out at 85°C. In the prior art, the extraction of Coptis chinensis usually adopts single temperature (such as 60°C) long time decoction, without considering the difference in dissolution temperature of alkaloids and polysaccharide components, and the end point judgment depends on experience or offline detection, resulting in high degradation rate of berberine and insufficient dissolution of polysaccharides. In this embodiment, by combining the three-stage temperature control with the dynamic conductivity monitoring, a gradient extraction mechanism of “low temperature protection-middle temperature wall breaking-high temperature extraction” is formed, which not only reduces the degradation of heat-sensitive components, but also strengthens the dissolution of components through pulse ultrasonic and hot reflux, solving the contradiction between component loss and incomplete extraction in the traditional process. In another embodiment, in S2, the first convolution channel is configured to process the 280nm absorbance time series data using a 3x1 convolution kernel, outputting a first feature map; the second convolution channel is configured to extract the features of 1050-1100nm near-infrared spectrum using a 5x1 empty convolution kernel, outputting a second feature map, and the expansion rate of the empty convolution is 2; the third convolution channel is configured to capture the features of 2200-2400nm spectrum using a 7x1 convolution kernel combined with an exponential linear unit activation function, outputting a third feature map; the three feature maps are spliced in the channel dimension, and a 1x1 convolution is performed to generate a channel attention weight matrix, which is fused with the spatial-spectral correlation features extracted by three-dimensional convolution, and input into a fully connected layer to generate a multi-dimensional feature vector, and calculate the comprehensive extraction efficiency index.

[0022] Exemplarily, the convolutional neural network is built based on a TensorFlow framework, and input data is 280 nm absorbance (sampling interval 1 min, a total of 100 time points), 1050-1100 nm near-infrared spectrum (51 wavelength points), and 2200-2400 nm mid-infrared spectrum (101 wavelength points). The first convolutional channel processes the absorbance data through 3 layers of 3x1 convolutional layers (step length 1, padding=same), and outputs a feature map of 100x1x32; the second convolutional channel uses a 5x1 hollow convolutional kernel with an expansion rate of 2, and an equivalent receptive field covers 9 spectral points, and outputs a feature map of 51x1x64; the third convolutional channel uses a 7x1 convolutional kernel combined with an ELU activation function (a=1.0), and outputs a feature map of 101x1x48. After feature splicing, a 1x1 convolutional layer (64 kernels) is used to generate channel attention weights, and a 3x3x3 three-dimensional convolutional kernel is used to extract spectral-time correlation features. A fully connected layer (256 nodes) outputs a feature vector containing parameters such as berberine dissolution, polysaccharide degree of polymerization, and comprehensive efficiency index is calculated by linear weighting (initial weights a=0.6, b=0.3, g=0.1).

[0023] In the prior art, spectral analysis is mostly performed by principal component analysis (PCA) to process single near-infrared channel data, without utilizing the synergistic information of ultraviolet characteristic absorption and near-infrared vibration spectrum, and the feature extraction layer cannot capture the dynamic correlation of spectral-time dimension, resulting in difficulty for the model to accurately reflect the complex relationship of multi-component dissolution. In this embodiment, different spectral scale features are adapted by designing multi-channel convolutional kernels, the receptive field is expanded by using hollow convolution, and the spatio-temporal correlation is captured by using three-dimensional convolution, and the channel attention mechanism strengthens the spectral signal of key components, so that the accuracy of extraction efficiency evaluation is significantly improved compared with traditional methods, and the technical problem of insufficient analysis of multi-spectral data is solved.

[0024] In another embodiment, the berberine dissolution parameter is subjected to maximum value normalization to generate a standard dissolution parameter; the polysaccharide degree of polymerization parameter is subjected to logarithmic conversion processing to generate a standard degree of polymerization parameter; the cell wall breaking rate parameter is subjected to Sigmoid function conversion to generate a standard wall breaking rate parameter; the three types of standard parameters are input into a fully connected layer to output an initial weight vector; the initial weight vector is spliced with a medicinal material batch feature code containing origin, harvesting season, and storage time length to generate a dynamic weight coefficient through a gated recurrent unit, wherein the berberine weight ranges from 0.55 to 0.65, the polysaccharide weight ranges from 0.25 to 0.35, and the wall breaking rate weight ranges from 0.08 to 0.12, and the comprehensive extraction efficiency index is calculated according to the formula.

[0025] Exemplarily, the maximum value normalization maps the berberine dissolution to the interval [0, 1], the logarithmic conversion is log10 (polymerization degree of polysaccharide + 1), and the Sigmoid function is 1 / (1+e^(-x)). The fully connected layer (32 nodes) outputs a 3-dimensional initial weight vector, and the medicinal material batch characteristic code adopts one-hot encoding (for example, the production place Beijing = 100, Henan = 010, the harvesting season spring = 10, autumn = 01, and the storage time ≤ 6 months = 1). The number of nodes of the Gated Recurrent Unit (GRU) hidden layer is 64, the dynamic weight is generated through the update gate and the reset gate, α can be 0.55, 0.60, 0.65, β can be 0.25, 0.30, 0.35, γ can be 0.08, 0.10, 0.12, and the comprehensive index = α × standard dissolution + β × standard polymerization degree + γ × standard broken wall rate.

[0026] In the prior art, the extraction efficiency evaluation is mostly based on fixed weight (such as focusing only on the berberine content), without considering the influence of medicinal material batch difference on the dissolution of ingredients, resulting in poor adaptability of the model to medicinal materials from different production places and harvesting times, and lacking of dynamic adjustment mechanism for process parameters between batches. The embodiment realizes self-adaptive regulation and control of the extraction process on the differences of medicinal materials through the standardized processing (maximum value normalization / logarithmic conversion / Sigmoid conversion) of multiple parameters, combined with the generation of dynamic weight coefficients by the medicinal material batch characteristic code and the GRU, so that the fluctuation range of the extraction efficiency of different batches of medicinal materials is reduced, and the problem of insufficient adaptability of the traditional fixed weight model is solved. In another embodiment, further comprising: obtaining the comprehensive extraction efficiency index output by the multi-channel convolutional neural network model in real time, calculating the confidence score of the model output, which is generated based on the inverse variance of 10 consecutive prediction results by performing Monte Carlo random dropout sampling on the optimized fusion features; when the confidence score is lower than the threshold value of 0.85, no longer determining the endpoint according to the comprehensive index, but terminating the extraction when the cumulative dissolution amount of berberine is not less than the maximum dissolution amount.

[0027] Exemplarily, the Monte Carlo random dropout sampling (dropout rate 0.2) performs 10 times of random inactivation processing on the optimized fusion features, generates a predicted value of the comprehensive efficiency index each time, calculates the variance σ² of the 10 results, and the confidence score is 1 / (1+σ²). The threshold value 0.85 can be adjusted to 0.8 or 0.9, when the score is lower than the threshold value, the cumulative dissolution amount of berberine is detected offline by Waters e2695 HPLC, compared with the maximum dissolution amount determined by the pre-experiment of the batch of medicinal materials, and the extraction is terminated when reaching 85%, 90% or 95%. In the prior art, intelligent model control lacks a reliable verification mechanism, which easily leads to incorrect judgment of the extraction endpoint when the spectral data fluctuates or the model deviates, resulting in over-extraction or insufficient extraction of components. The present embodiment calculates the confidence score through Monte Carlo sampling and constructs a triple safeguard mechanism of "model prediction-reliability verification-offline detection". When the model confidence is insufficient, it automatically switches to hard endpoint control based on the content of the components, solving the risk problem of single intelligent model control and improving the reliability of the extraction process. In another embodiment, in S3, the calcined dragon bone and calcined oyster shell are crushed to 100-120 mesh, the first stage is treated with 1.0% citric acid solution at a liquid-solid ratio of 3:1 at 50°C for 15 minutes, the second stage is treated with 0.5% citric acid solution at a liquid-solid ratio of 5:1 at 70°C for 20 minutes, and the filtrate is combined to obtain a mineral filtrate; add pH 5.8 acetic acid buffer solution (liquid-solid ratio 8:1) to the root and stem medicinal materials such as radix pseudoginseng, and extract at 80°C for 25 minutes to obtain a root and stem decoction; add the root and stem decoction, mineral filtrate and purified water to the medicinal materials such as jujube kernel, and control the total liquid-solid ratio to 12:1, and reflux extract at 95°C.

[0028] Illustratively, the calcined dragon bone and calcined oyster shell are crushed to 100 mesh, 110 mesh or 120 mesh by an air flow crusher, the first stage is treated with 1.0% citric acid solution (liquid-solid ratio 3:1) in a constant temperature water bath at 50°C, stirring rate 200 rpm for 15 minutes; the second stage is treated with 0.5% citric acid solution heated to 70°C, liquid-solid ratio 5:1 for 20 minutes, and the filtrate is combined and filtered through a Buchner funnel. Add pH 5.8 acetic acid buffer solution (prepared with acetic acid and sodium acetate) to the root and stem medicinal materials, liquid-solid ratio 8:1, extract in a double boiler at 80°C for 25 minutes. Mix the flower / seed medicinal materials with the root and stem decoction, mineral filtrate, and add purified water to adjust the total liquid-solid ratio to 12:1, and extract in a reflux extraction device at 95°C, with the condensate water temperature ≤25°C.

[0029] In the prior art, mineral medicines and plant medicines often use unified decoction process. Due to the dense structure of mineral medicines, the dissolution rate of inorganic components such as calcium and magnesium is less than 5%, and long-time high-temperature decoction easily causes degradation of heat-sensitive components in plant medicines. The present embodiment destroys the mineral lattice through gradient acid treatment (1.0%-0.5% citric acid) of mineral medicines, combined with grouping decoction strategy (root and stem- flower / seed), and optimizes the dissolution conditions of different medicinal properties of medicinal materials, which significantly improves the dissolution rate of inorganic components of mineral medicines, while avoiding the degradation of plant medicine components, solving the dual problems of low dissolution efficiency of mineral medicines and loss of plant medicine components in traditional co-decoction process.

[0030] In another embodiment, in S4, the input layer of the long short-term memory neural network model synchronously receives five time channel data: tanshinone IIA concentration, paeonol concentration, glycyrrhizic acid concentration, decoction temperature, and decoction pressure; the data are respectively input into the Bi-LSTM processing unit, and the output is a component dissolution feature vector and a process parameter dynamic feature vector; after the features are spliced, they are input into the attention mechanism layer, an intermediate matrix is generated through the full connection layer and the tanh activation function, the Softmax normalization output is a synergistic weight coefficient, and the feature vector is fused to generate a flavone-saponin interaction effect parameter in the interval of 0-1.

[0031] Exemplarily, the model is constructed based on PyTorch, the input data is a time sequence of 5 channels (sampling interval 5 min, a total of 30 time points), each Bi-LSTM processing unit contains 128 hidden nodes, and the forward and reverse LSTM outputs a 256-dimensional feature vector. The five-channel feature is spliced into a 1280-dimensional vector, which is input into the attention mechanism layer: the full connection layer (512 nodes) is reduced in dimension through the tanh activation to generate an intermediate matrix, and the Softmax function is normalized to obtain a weight coefficient (tanshinone IIA weight 0.4-0.6, paeonol weight 0.3-0.5). The output layer maps the interaction effect parameter to [0, 1] through the Sigmoid function, and the fluctuation threshold is set to ±5%. The temperature (80-95°C) and pressure (0.1-0.3 MPa) data are input into the model after standardization processing. In the prior art, the decoction process control is mostly based on a single component concentration or an empirical temperature curve, and the dissolution kinetics of fat-soluble components (such as tanshinone IIA) and water-soluble components (such as paeonol) are not associated with temperature and pressure parameters, and there is a lack of real-time modeling of the synergistic effect between components. In this embodiment, the Bi-LSTM captures the long-term dependence characteristics of the dissolution of multiple components, the attention mechanism dynamically allocates the weight coefficients of tanshinone IIA and paeonol, realizes the dynamic evaluation of the synergistic release of flavone-saponin components, and solves the technical problem that the component interaction effect cannot be real-time regulated in the traditional process, thereby providing a multi-dimensional scientific basis for the decoction endpoint control. In another embodiment, in S5, the combined liquid is treated by a 0.22 μm microfiltration membrane, and the operating pressure is controlled in three stages: 0.25 MPa (5 min)-0.40 MPa (3 min)-0.18 MPa (to the end point); the microfiltrate is subjected to gradient alcohol precipitation: first adjust the ethanol concentration to 60%±2% (4°C for 20 min), then adjust to 80%±1% (25°C for 15 min), and collect the supernatant at 45°C±1°C and-0.08 MPa under reduced pressure to a relative density of 1.25-1.3. Exemplarily, microfiltration adopts a Millipore plate and frame device, the membrane material is PVDF, the pore size is 0.22 μm, and the three-stage pressure is controlled by a diaphragm pump: the first stage is 0.25 MPa for stable filtration for 5 minutes, the second stage is 0.40 MPa for increasing flux for 3 minutes, and the third stage is 0.18 MPa for maintaining until the flow rate is <50 mL / min. During gradient alcohol precipitation, analytical pure ethanol is added by a peristaltic pump, 60% alcohol precipitation is placed in a 4°C refrigerator for 20 minutes, and the precipitate is separated by centrifugation at 3000 rpm; 80% alcohol precipitation is placed at 25°C for 15 minutes, and the supernatant is taken, concentrated in a BUCHI rotary evaporator at 45°C and -0.08 MPa, and the relative density is determined by a specific gravity bottle (1.25, 1.28, 1.30 at 25°C).

[0032] In the prior art, microfiltration often adopts 0.45 μm membrane constant pressure (0.3 MPa) filtration, which easily leads to membrane pollution, and the flux decay rate is more than 40%; alcohol precipitation often adopts a single 70% ethanol concentration, and the removal of macromolecular impurities is not complete and the effective component loss rate is more than 15%. The present embodiment reduces the formation of filter cake layer on the membrane surface by three-stage pressure regulation (first stable, then increase, and then decrease), so that the flux decay rate is controlled to be within 15%, and gradient alcohol precipitation (60%-80%) grades and precipitates impurities according to the polarity difference of components, and the effective component loss rate is <8%, thereby solving the technical problems of low filtration efficiency and incomplete removal of impurities in the post-processing process, and improving the purity and batch consistency of concentrated extract.

[0033] The following is illustrated by specific examples.

[0034] Example: The Coptis chinensis is ground to 50 mesh, pH 6.8 phosphate buffer (solid-liquid ratio 1:10) is added, and three-stage temperature-controlled extraction is performed: the first stage is constant temperature at 45°C for 30 minutes, the second stage is pulse ultrasonic extraction (power 400 W, pulse frequency 15 Hz), and the third stage is switched to hot reflux extraction at 85°C when the conductivity of the extraction liquid increases by ≤1% / 5 minutes for three consecutive times. During the extraction process, data is collected in real time by a UV-near infrared spectrometer, input into a multi-channel convolutional neural network model (the first convolutional channel uses a 3×1 convolutional kernel to analyze the absorbance at 280 nm, the second convolutional channel uses a 5×1 hollow convolutional kernel to process the near-infrared spectrum at 1050-1100 nm, and the third convolutional channel extracts the features at 2200-2400 nm by a 7×1 convolutional kernel combined with an ELU activation function), and the extraction is terminated when the comprehensive extraction efficiency index output by the model reaches 0.85, and the Coptis chinensis extract is obtained.

[0035] The bone and oyster shell were crushed to 110 mesh, the first stage was treated with 1.0% citric acid solution (liquid-solid ratio 3:1) at 50°C for 15 minutes, the second stage was treated with 0.5% citric acid solution (liquid-solid ratio 5:1) at 70°C for 20 minutes, and the filtrate was combined to obtain a mineral filtrate; The pH 5.8 acetic acid buffer solution (liquid-solid ratio 8:1) was added to the root and stem medicinal materials such as radix pseudoginseng, and the root and stem decoction liquid was obtained by decocting at 80°C for 25 minutes; The root and stem decoction liquid, mineral filtrate and purified water were added to the medicinal materials such as jujube kernel (total liquid-solid ratio 12:1), and the mixture was refluxed and decocted at 95°C. The concentration time series data of tanshinone IIA, paeoniflorin and glycyrrhizic acid were obtained in real time by high performance liquid chromatograph, and input into the long short-term memory neural network model based on attention mechanism (bidirectional LSTM layer extracts dissolution rate characteristics, attention mechanism layer calculates the synergistic weight coefficient of tanshinone IIA and paeoniflorin, and output layer generates flavonoid-saponin interaction effect parameters). When the interaction effect parameters fluctuate by ≤±5% for three consecutive detection cycles, the mixed decoction liquid is obtained.

[0036] The extraction liquid and decoction liquid were combined and treated by 0.22 μm microfiltration membrane (three-stage pressure: 0.25 MPa x 5 min-0.40 MPa x 3 min-0.18 MPa to endpoint). The filtrate was first added with ethanol to a concentration of 60% (4°C for 20 minutes), and the precipitate was discarded. The supernatant was supplemented with ethanol to 80% (25°C for 15 minutes), and the supernatant was collected and concentrated to a relative density of 1.28 at 45°C under a vacuum of-0.08 MPa. The compound drug was prepared by mixing with lactose. Comparative Example 1: Traditional fixed parameter extraction process The powder of Coptis was crushed to 40 mesh, and pH 6.8 phosphate buffer solution (solid-liquid ratio 1:8) was added. The mixture was extracted at 60°C for 2 hours. No conductivity monitoring and spectral data were input. The content of berberine was detected offline to control the end point (sample every 30 minutes). The bone and oyster shell were crushed to 80 mesh and added to purified water (liquid-solid ratio 10:1) together with other medicinal materials. The mixture was decocted at 90°C for 1.5 hours. There was no grouping decoction and gradient acid treatment. The decoction liquid was not monitored in real time for the concentration of ingredients. The decoction time was controlled by experience. The microfiltration was performed by 0.45 μm membrane at constant pressure (0.3 MPa). The ethanol concentration was adjusted to 70% at one time (25°C for 30 minutes). Comparative Example 2: Mineral medicine treatment process without grouping decoction The bone and oyster shell were crushed to 90 mesh and not subjected to gradient acid treatment. They were added to purified water (liquid-solid ratio 12:1) together with other medicinal materials such as radix pseudoginseng. The mixture was decocted at 95°C for 2 hours. The root and stem medicinal materials and flower medicinal materials were not grouped for decoction. They were uniformly subjected to reflux decoction at 95°C. The acetic acid buffer solution and mineral filtrate were not used. The concentrations of tanshinone IIA and paeoniflorin were not monitored in real time during the decoction process. The decoction end point was controlled by fixed time (120 minutes). Comparative Example 3: Decoction control process based on single LSTM model In the decoction stage, only the concentration of tanshinone IIA was detected by high-performance liquid chromatography offline (sampled every 10 minutes), the data was input into the single-direction long short-term memory neural network model, the concentrations of paeoniflorin and glycyrrhizic acid and the timing temperature and pressure data were not synchronized, there was no attention mechanism layer in the model, the component synergistic weight coefficient could not be calculated, the decoction endpoint was controlled by the concentration of tanshinone IIA reaching the preset value (80% of the maximum dissolution), and no flavonoid-saponin interaction effect parameter was generated. Control group: traditional water decoction and alcohol precipitation process Huanglian was crushed to 30 mesh, and water was decocted twice (1 hour each time, 80°C), and the decoction was combined; all medicinal materials were mixed and decocted without grouping, calcined longgu and calcined mufu were not crushed to more than 100 mesh, and there was no gradient acid treatment; the decoction process was not real-time monitored for component concentration, and was decocted for 90 minutes according to traditional experience; microfiltration was performed using a 0.45 μm membrane constant pressure filtration, and ethanol was added to a concentration of 70% at one time during alcohol precipitation, and after standing for 30 minutes, filtration was performed, and the relative density was concentrated to 1.15, and the medicine was prepared by mixing with excipients.

[0037] Key index detection means and experimental method: I. Component dissolution efficiency detection 1. Berberine dissolution detection High-performance liquid chromatography (HPLC) was used, referring to Chinese Pharmacopoeia 2025 Edition Volume IV General Chapter 0512. A C18 chromatographic column (250 mm x 4.6 mm, 5 μm) was used, the mobile phase was acetonitrile-0.1% phosphoric acid solution (55:45, containing 0.05 mol / L sodium dodecyl sulfate), the flow rate was 1.0 mL / min, the column temperature was 30°C, and the detection wavelength was 280 nm. The sample solution was filtered through a 0.45 μm filter membrane and injected 10 μL, and the dissolution was calculated by the external standard method of berberine reference substance, and the formula was: dissolution (%) = (berberine content in sample / theoretical berberine content in medicinal material) x 100%.

[0038] 2. Polysaccharide polymerization degree detection High-performance gel permeation chromatography (HPGPC) was used, a TSK-GEL G4000PWXL chromatographic column (300 mm x 7.8 mm) was used, the mobile phase was 0.1 mol / L sodium chloride solution, the flow rate was 0.6 mL / min, the column temperature was 35°C, and the differential refractive index detector was used. The sample was saturated with water and n-butanol to remove protein, filtered through a 0.22 μm filter membrane, and a standard curve was drawn by standard dextran, and the weight average molecular weight (Mw) of polysaccharide was calculated according to the retention time to represent the polymerization degree.

[0039] 3. Calcium and magnesium ion dissolution rate detection Instrument parameters: RF power 1550 W, atomizing gas flow 0.8 L / min, sampling depth 8 mm. The sample was digested with nitric acid-perchloric acid (4:1), and then diluted with 0.5% nitric acid. The concentrations of Ca²⁺ and Mg²⁺ were determined, and the dissolution rate was calculated based on the total amount of corresponding elements in the medicinal material. The formula is: dissolution rate (%) = (ion content in solution / total amount of elements in medicinal material) x 100%. II. Process control and model precision detection 1. Flavone-saponin interaction effect parameter determination The concentrations of tanshinone IIA and paeoniflorin were monitored in real time by high performance liquid chromatography-mass spectrometry (HPLC-MS). The HPLC conditions were the same as those for berberine detection. The mass spectrometry used electrospray ionization (ESI) in positive ion mode, and monitored m / z. The LSTM-attention mechanism model was inputted with the data of boiling temperature and pressure, and the interaction effect parameter was outputted, which was normalized to [0, 1]. The fluctuation range was calculated by the standard deviation of 3 consecutive detection cycles (interval 5 min).

[0040] 2. Confidence score of multi-channel CNN model Monte Carlo random dropout sampling (dropout rate 0.2) was performed on the optimized fused features. The variance σ² was calculated based on the integrated extraction efficiency index of 10 consecutive predictions. The confidence score = 1 / (1+σ²). When the score < 0.85, the cumulative dissolution of berberine was detected by offline HPLC method, and compared with the maximum dissolution in pre-experiment to determine the endpoint.

[0041] III. Detection of post-processing process indicators 1. Determination of microfiltration flux decay rate The filtration time and filtrate volume of each stage of microfiltration were recorded, and the initial flux (average flux of the first stage) and end-point flux (stable flux of the third stage) were calculated. The decay rate = (initial flux-end-point flux) / initial flux x 100%. The Millipore plate and frame microfiltration device was used to monitor the flow and pressure in real time.

[0042] 2. Detection of loss rate of effective components in alcohol precipitation The contents of berberine, tanshinone IIA and paeoniflorin in the filtrate before alcohol precipitation and supernatant were determined by HPLC external standard method. The loss rate = (content of components before alcohol precipitation-content of components in supernatant) / content of components before alcohol precipitation x 100%. The average value was obtained by parallel determination for 3 times. IV. Pharmacodynamic detection of animal models 1. Establishment of rapid atrial fibrillation rat model SD rats (200±20g) were intraperitoneally injected with sodium pentobarbital (30mg / kg) for anesthesia, and an electrode catheter was inserted into the right atrium through the jugular vein to induce atrial fibrillation at a frequency of 200 times / min for 30 minutes. The electrocardiogram was recorded by the PowerLab system, and the model was determined to be successful when the P wave disappeared, the f wave appeared, and the RR interval was absolutely irregular. 2. Drug effectiveness evaluation The model rats were randomly divided into groups and given intragastrically administered drugs (containing crude drugs 1.5g / kg), comparative drugs, and saline as a blank control. The heart rate was recorded 30 minutes, 1 hour, and 2 hours after administration. A decrease in heart rate of ≥20% compared to the model group and maintenance for more than 1 hour was determined to be effective, and the effective rate was calculated as the number of effective animals divided by the number of animals in each group multiplied by 100%.

[0043] Five, comparison of component dissolution efficiency and preparation efficiency data 1. Berberine dissolution and degradation rate Example: dissolution 92.3%, degradation rate 3.1% Comparative Example 1 (traditional fixed parameters): dissolution 65.2%, degradation rate 12.0% Control group (traditional water decoction): dissolution 54.8%, degradation rate 18.5% The data show that the example, through three-stage temperature control and real-time monitoring of the spectrum, improves the dissolution by 27.1% and reduces the degradation rate by 74.2% compared to Comparative Example 1, solving the contradiction between "insufficient dissolution and component degradation" in traditional constant temperature extraction, and significantly improving the preparation efficiency. 2. Polysaccharide polymerization degree retention rate Example: retention rate 88.0% (weight average molecular weight Mw=12500Da) Control group: retention rate 53.0% (Mw=7800Da) The example, through the synergy of the low-temperature protection stage (constant temperature at 45°C) and pulsed ultrasonic wall breaking, improves the polysaccharide polymerization degree retention rate by 35.0% compared to the control group, avoiding polysaccharide degradation caused by traditional high-temperature decoction and ensuring the structural stability of the active ingredients.

[0044] 3. Mineral drug calcium and magnesium ion dissolution rate Example: Ca²⁺ dissolution rate 18.7%, Mg²⁺ dissolution rate 15.3% Comparative Example 2 (no grouping decoction): Ca²⁺ dissolution rate 4.2%, Mg²⁺ dissolution rate 3.8% Control group: Ca²⁺ dissolution rate 2.7%, Mg²⁺ dissolution rate 2.1% The embodiment improves the calcium and magnesium ion dissolution rate by 4.4 times compared with Comparative Example 2 through gradient acid treatment (1.0%-0.5% citric acid) and 110 mesh ultrafine grinding, breaking through the physical barrier of mineral crystal lattice in the traditional decoction process and significantly improving the dissolution efficiency of inorganic ingredients.

[0045] 4. Synergistic dissolution amount of tanshinone IIA and paeoniflorin Example: The dissolution amount of tanshinone IIA is 2.87 mg / g of medicinal materials, and the dissolution amount of paeoniflorin is 8.54 mg / g of medicinal materials, and the dissolution amount ratio of the two is 1:2.98 (close to the golden synergistic ratio 1:3) Comparative Example 3 (single LSTM control): The dissolution amount of tanshinone IIA is 2.12 mg / g, and the dissolution amount of paeoniflorin is 6.31 mg / g, and the ratio is 1:2.98 (but the interaction effect parameter fluctuates ±10.2%) Control group: The dissolution amount of tanshinone IIA is 1.75 mg / g, and the dissolution amount of paeoniflorin is 5.12 mg / g, and the ratio is 1:2.92 (significant degradation of ingredients) The embodiment realizes the precise matching of ingredient release by using the attention mechanism LSTM model to real-time regulate the warm pressure parameters, so that the synergistic dissolution amount of lipid-soluble and water-soluble ingredients is increased by 64.0% and 66.8% respectively compared with the control group, and the interaction effect parameter fluctuation is ≤±4.2%. Six, process control precision and quality stability data 1. Extraction and decoction endpoint control error Example: The prediction error of comprehensive extraction efficiency index is ±3.0%, and the fluctuation of flavonoid-saponin interaction effect parameter is ±4.2% Comparative Example 2 (single spectrum model): The prediction error of extraction efficiency is ±15.0%, and the misjudgment rate of endpoint is 30.0% Control group: The endpoint is controlled by experience, and the ingredient content fluctuation between batches is ±22.0% The multi-channel CNN and LSTM-attention mechanism model of the embodiment improves the process control precision by 80.0% compared with Comparative Example 2, realizes dynamic monitoring, real-time regulation of closed-loop control, and guarantees the ingredient uniformity between batches. 2. Different batch medicinal materials adaptability data Example: The berberine dissolution degree of Sichuan / Henan origin medicinal materials fluctuates ±5.8%, and the comprehensive extraction efficiency index fluctuates ±5.8% Control group: The difference in origin leads to a dissolution degree fluctuation of ±22.0%, and the efficiency index fluctuation is ±22.0% The embodiment improves the self-adaptive ability of the process to the difference of medicinal materials by 73.6% by using the medicinal material batch characteristic code (origin, harvesting season, storage time) and the generated dynamic weight coefficient of the gating recurrent unit, which significantly reduces the quality difference between batches. 3. Post-processing process efficiency and purity Microfiltration flux decay rate: 15.0% for the example, 40.0% for the control Loss rate of effective components in alcohol precipitation: 8.0% for the example, 15.0% for the control Impurity content of concentrated extract: 5.0% (relative density 1.28) for the example, 12.0% (relative density 1.15) for the control The three-stage pressure regulation and gradient alcohol precipitation process improves the microfiltration flux maintenance efficiency by 62.5%, the effective component retention rate by 46.7%, and the extract purity significantly, laying a foundation for the stability of the preparation quality. III. Product quality and pharmacodynamic stability data 1. Heart rate control efficiency of rat fibrillation model Example: 91.0% (heart rate decreased by 28.5±3.2 beats / min after 2 hours of administration) Control group: 65.0% (heart rate decreased by 15.2±5.8 beats / min) Comparative example 1: 72.0% (heart rate decreased by 18.7±4.5 beats / min) The example drug has sufficient and synergistic component dissolution, and the pharmacodynamic efficiency is improved by 26.0% compared with the control group, and the heart rate control amplitude and stability are significantly better than those of the traditional process group. 2. Pharmacodynamic difference coefficient between batches Example: effective rate fluctuation ±8.0% Control group: effective rate fluctuation ±25.0% The example reduces the pharmacodynamic batch difference by 68.0% compared with the traditional process through intelligent control of the whole process, directly reflecting the improvement in quality stability. Although the embodiments of the present application have been disclosed as above, they are not limited to the use listed in the specification and embodiments, and can be fully applied to various fields suitable for the present application, and other modifications can be easily realized by those skilled in the art, and therefore the present application is not limited to specific details and the examples shown and described herein, without departing from the general concept defined by the claims and equivalent scope.

Claims

1. A method for preparing a traditional Chinese medicine compound for treating rapid atrial fibrillation, characterized in that: include: S1: Take Coptis chinensis and add phosphate buffer for extraction; S2: Collecting optical data of the extract in real time using a UV-NIR spectrometer and inputting it into a pre-trained multi-channel convolutional neural network model; wherein, in the multi-channel convolutional neural network model, the first convolution channel analyzes absorbance data at a wavelength of 280 nm, the second convolution channel processes near-infrared spectral data from 1050-1100 nm, and the third convolution channel extracts spectral features from 2200-2400 nm; when the comprehensive extraction efficiency index output by the model reaches a predetermined threshold, the extraction is terminated to obtain the Coptidis rhizome extract; S3: Take Radix Pseudostellariae, Polygonatum sibiricum, Rehmannia glutinosa, White Peony Root, Mulberry Fruit, Ziziphus jujuba seed, Schisandra chinensis, Triticum aestivum, Radix Ophiopogonis, Rhizoma Chuanxiong, Salvia miltiorrhizae, Radix Rehmanniae, Red Peony Root, Licorice, Citron, Calcined Dragon Bone, Calcined Oyster Shell and Rose Flower and boil them: S4: Real-time concentration time series data of tanshinone IIA, paeoniflorin, and glycyrrhizic acid in the decoction are obtained by high-performance liquid chromatography and input into a long-short-term memory neural network model based on an attention mechanism. In the long-short-term memory neural network model, the bidirectional long-short-term memory layer extracts the dissolution rate characteristics of each component, the attention mechanism layer calculates the synergistic weight coefficient of tanshinone IIA and paeoniflorin, and the output layer generates the flavonoid-saponin interaction effect parameter. When the interaction effect parameter fluctuates within a range of ≤±5% for three consecutive detection cycles, the decoction is terminated to obtain a mixed decoction. S5: combining the Rhizoma Coptidis extract and the mixed decoction, treating with a microfiltration membrane, and then performing alcohol precipitation. The supernatant of the alcohol precipitation is collected and concentrated to obtain a concentrated extract. S6: Mixing the concentrated extract with a pharmaceutically acceptable carrier to prepare a traditional Chinese medicine compound.

2. The method according to claim 1, wherein In S1, the coptis root is crushed into 40-60 mesh, and a phosphate buffer solution of pH 6.8 is added, with a solid-liquid ratio of 1:8-1:12; The extraction was carried out in three stages: the first stage was constant temperature extraction at 45°C for 30 minutes; the second stage was heating to 65°C for pulse ultrasonic extraction; the third stage was switching to hot reflux extraction at 85°C when the conductivity of the extract showed an increase of ≤1% / 5 minutes for three consecutive tests.

3. The method according to claim 1, wherein In S2, the first convolution channel is configured to use a 3×1 convolution kernel to process the 280nm absorbance time series data and output the first feature map; the second convolution channel is configured to use a 5×1 dilated convolution kernel to extract the features of the 1050-1100nm near-infrared spectrum and output the second feature map, with the dilation rate of the dilated convolution being 2; the third convolution channel is configured to use a 7×1 convolution kernel combined with an exponential linear unit activation function to capture the features of the 2200-2400nm spectrum and output the third feature map; The first feature map, the second feature map, and the third feature map are concatenated in the channel dimension to form an initial fusion feature, and a 1×1 convolution kernel operation is applied to the initial fusion feature to generate a channel attention weight matrix; Applying a three-dimensional convolution kernel operation to the initial fusion feature to generate a spatial-spectral correlation feature, performing an element-wise multiplication operation on the channel attention weight matrix and the spatial-spectral correlation feature, and outputting an optimized fusion feature; Inputting the optimized fusion features into a fully connected layer to generate a multidimensional feature vector including berberine solubility parameters, polysaccharide polymerization parameters, and cell wall rupture rate parameters; A comprehensive extraction efficiency index is calculated based on the multidimensional feature vector.

4. The method according to claim 3, wherein The berberine dissolution parameters were normalized to the maximum value to generate the standard dissolution parameters. The polysaccharide polymerization parameters were transformed by logarithm to generate the standard polymerization parameters. The cell wall rupture rate parameters were transformed by Sigmoid function to generate the standard cell wall rupture rate parameters. Input standard dissolution parameters, standard polymerization parameters, and standard wall-breaking rate parameters to the fully connected layer and output the initial weight vector; The initial weight vector is concatenated with the medicinal material batch feature code, which includes the origin code, harvest season code, and storage time code; Dynamic weight coefficients were generated through gated cyclic units, where the weight coefficients of berberine solubility parameters ranged from 0.55 to 0.65, the weight coefficients of polysaccharide polymerization parameters ranged from 0.25 to 0.35, and the weight coefficients of cell wall rupture rate parameters ranged from 0.08 to 0.

12. The comprehensive extraction efficiency index was calculated according to the following formula: comprehensive extraction efficiency index = dynamic weight coefficient α × standard dissolution parameter + dynamic weight coefficient β × standard polymerization parameter + dynamic weight coefficient γ × standard cell wall rupture rate parameter, where α, β, and γ are dynamic weight coefficients.

5. The method according to claim 3, wherein Also includes: The comprehensive extraction efficiency index output by the multi-channel convolutional neural network model is obtained in real time, and the confidence score of the multi-channel convolutional neural network model output is calculated. The confidence score is generated by performing Monte Carlo random dropout sampling on the optimized fusion features and is based on the inverse of the variance of 10 consecutive prediction results. When the confidence score is lower than the threshold of 0.85, the extraction is no longer terminated based on the comprehensive extraction efficiency index. Instead, the extraction is terminated when the cumulative dissolution amount of berberine is not less than the maximum dissolution amount.

6. The method according to claim 3, wherein In S3, the calcined dragon bone and calcined oyster are crushed to 100-120 mesh, treated with a 1.0% mass concentration citric acid solution at a liquid-to-solid ratio of 3:1 at 50°C for 15 minutes in the first stage, and treated with a 0.5% mass concentration citric acid solution at a liquid-to-solid ratio of 5:1 at 70°C for 20 minutes in the second stage, and the filtrates from the first and second stages are combined to obtain a mineral filtrate; Add pH 5.8 acetic acid buffer to Pseudostellariae Radix, Polygonatum sibiricum, Rehmanniae Preparata, Rehmanniae Radix, Paeoniae Rubra, Paeoniae Alba, Ligusticum chuanxiong and Salvia miltiorrhiza, and boil at 80°C for 25 minutes to obtain a rhizome decoction; Add rhizome decoction, mineral filtrate and purified water to jujube seeds, schisandra chinensis, Huai wheat, ophiopogon japonicus, liquorice, bergamot and rose, control the total liquid-solid ratio to 12:1, and reflux and decoct at 95°C.

7. The method according to claim 1, wherein In S4, the input layer of the long short-term memory neural network model is configured to synchronously receive five independent time series data channels: the first channel inputs the time series data of tanshinone IIA concentration, the second channel inputs the time series data of paeoniflorin concentration, the third channel inputs the time series data of glycyrrhizic acid concentration, the fourth channel inputs the time series data of decoction system temperature, and the fifth channel inputs the time series data of decoction system pressure; The five channel data are input into the corresponding processing units in the bidirectional long short-term memory layer, and the outputs are the tanshinone IIA dissolution feature vector, the paeoniflorin dissolution feature vector, the glycyrrhizic acid dynamic feature vector, the temperature dynamic feature vector, and the pressure dynamic feature vector. The tanshinone IIA dissolution feature vector, the paeoniflorin dissolution feature vector, the glycyrrhizic acid dynamic feature vector, the temperature dynamic feature vector, and the pressure dynamic feature vector are feature concatenated; the concatenated features are input into the attention mechanism layer, processed by the fully connected layer and the hyperbolic tangent activation function to generate an intermediate feature matrix, and then normalized by the Softmax function to output a synergistic weight coefficient; the synergistic weight coefficient is feature fused with the tanshinone IIA dissolution feature vector, the paeoniflorin dissolution feature vector, and the glycyrrhizic acid dynamic feature vector; The fused features are processed by the output layer to generate the flavonoid-saponin interaction effect parameter normalized to the range of 0 to 1.

8. The method according to claim 1, wherein In S5, the Rhizoma Coptidis extract and the mixed decoction were combined and then microfiltered through a microfiltration membrane with a pore size of 0.22 μm. The microfiltration operating pressure was controlled in three stages: the first stage was 0.25 MPa for 5 minutes, the second stage was increased to 0.40 MPa for 3 minutes, and the third stage was reduced to 0.18 MPa and maintained until the end of the filtration. The obtained microfiltrate was subjected to gradient alcohol precipitation: ethanol was first added to make the alcohol concentration reach 60%±2%, and the mixture was allowed to stand at 4°C for 20 minutes, after which the precipitate was separated and discarded; ethanol was then added to the supernatant to make the alcohol concentration reach 80%±1%, and the mixture was allowed to stand at 25°C for 15 minutes, after which the alcohol precipitation supernatant was collected; the alcohol precipitation supernatant was concentrated under reduced pressure at 45°C±1°C and a vacuum degree of -0.08 MPa, and the concentration was terminated when the relative density of the concentrated extract reached 1.25-1.3.

Images