A traditional Chinese medicine decoction piece whole life cycle data management method and system and a storage medium

Abstract

The application discloses a traditional Chinese medicine decoction piece whole life cycle data management method and system and a storage medium, relates to the technical field of big data management and artificial intelligence, and collects planting environment and original data, generates a source feature vector after denoising by using a LOF algorithm, and chains up; in the processing stage, multi-modal features such as color, texture and temperature evolution are collected, an initial quality baseline index is output by an evaluation network, in the circulation stage, a dynamic quality index is calculated based on environmental fluctuations, real-time monitoring and early warning of quality decay are realized, in the clinical stage, subjective and objective curative effect data are extracted, a curative effect feedback vector is generated, a whole-link knowledge graph is constructed, reverse attribution is carried out by using a graph neural network, and process feedback optimization strategies are output; by means of the blockchain, deep learning and dynamic decay model, the application realizes credible traceability, dynamic evaluation and closed-loop optimization of traditional Chinese medicine decoction pieces from the field to the clinic, and significantly improves the quality controllability and drug safety of the traditional Chinese medicine decoction pieces.

Description

Technical Field

[0001] This invention belongs to the field of big data management and artificial intelligence technology, and specifically relates to a method, system and storage medium for managing the entire life cycle data of traditional Chinese medicine decoction pieces. Background Technology

[0002] As an important component of TCM clinical medication, the quality of prepared herbal pieces directly affects clinical efficacy and medication safety. With the acceleration of TCM industrialization, prepared herbal pieces have achieved large-scale production and distribution. However, their production chain is long and involves many links, including planting, harvesting, processing, storage, logistics, and end-use. Supervision of a single link can no longer meet the needs of modern TCM quality control.

[0003] Currently, the industry primarily uses RFID or QR code technologies for traceability management, with applications mainly focused on recording planting information and tracking distribution routes. Technically, existing methods largely rely on static database storage, and quality assessment in the processing stage still primarily depends on manual judgment of sensory indicators such as color and texture, lacking in-depth fusion analysis of multimodal data. In the distribution stage, existing management systems typically set fixed shelf lives, ignoring the dynamic catalytic degradation of active ingredients caused by temperature and humidity fluctuations during warehousing and logistics. Furthermore, clinical efficacy data is often isolated and cannot be effectively correlated with upstream production parameters.

[0004] This reveals the following problems with existing technologies: 1. Traceability data is limited to information recording and lacks dynamic quantitative assessment of the evolution of herbal medicine quality; 2. There is a data gap between processing techniques and clinical efficacy, making it difficult to achieve precise closed-loop optimization of production processes through clinical feedback, resulting in a lack of scientific data-driven means to improve the quality of Chinese herbal medicine. Summary of the Invention

[0005] (a) Technical problems to be solved To address the problems in related technologies, this invention provides a method, system, and storage medium for managing the entire lifecycle data of traditional Chinese medicine decoction pieces, thereby overcoming the aforementioned technical problems existing in the prior art.

[0006] (II) Technical Solution To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: S1. Obtain source production environment monitoring data and basic attribute data of the target batch of Chinese medicinal materials, perform time series alignment and outlier removal preprocessing, generate source feature vector, and upload it to the consortium blockchain node after binding it with the identity identifier of the target batch of Chinese medicinal materials. S2. During the processing stage of the target batch of Chinese medicinal materials, the operating time parameters of the processing equipment, the time series data of ambient temperature, and the color gradient characteristics and texture roughness characteristics of the sliced ​​medicinal materials are collected simultaneously. After feature fusion, the data are input into the processing quality benchmark evaluation network to obtain the initial quality baseline index of the target batch of Chinese medicinal materials, and then added to the consortium blockchain. S3. During the storage and logistics of the Chinese herbal medicine pieces, real-time storage environment temperature and humidity data and light intensity data are obtained, and combined with the initial quality baseline index, the cumulative loss value of the effective ingredients is calculated, and then the dynamic quality index is updated and generated in real time. S4. In the clinical application stage of the Chinese herbal medicine, obtain the prescription dispensing record, perform association matching based on the identity identifier, extract the subjective symptom improvement entity and the change range of laboratory test index data from the electronic medical record text data of the patient taking the medicine, and calculate and generate the efficacy feedback vector. S5. Using the identity identifier as a node, and the source feature vector, initial quality baseline index, dynamic quality index, and efficacy feedback vector as associated attribute nodes, construct a knowledge graph of the entire chain of traditional Chinese medicine. By using graph neural networks to perform multi-hop path aggregation and high-dimensional feature mining on the entire knowledge graph of traditional Chinese medicine, the optimal parameter combination of the processing stage corresponding to the optimal interval of the target efficacy feedback vector is found, and the process feedback optimization strategy for the target traditional Chinese medicine is output. By constructing a full lifecycle data management system covering the planting, processing, warehousing, logistics, and clinical application of Chinese medicinal materials, this invention uses identity identification as a link to connect source environmental characteristics, processing quality baselines, dynamic quality indices, and clinical efficacy feedback across stages. Blockchain technology is employed to ensure the immutability and reliable sharing of data at each stage. A multimodal deep learning model is used to achieve objective quantitative evaluation of processing quality. Environmental perception and linear decay rules enable dynamic quality monitoring during the circulation process. Graph neural networks are used to aggregate multi-hop paths in a knowledge graph, uncovering deep correlations between clinical efficacy and production parameters, and automatically outputting process feedback optimization strategies. This invention breaks through the limitations of static and fragmented traditional Chinese medicine (TCM) decoction piece quality management, achieving closed-loop optimization from experience-driven to data-driven approaches, thus improving the overall quality control level and clinical medication safety of TCM decoction pieces.

[0007] Preferably, step S1 includes the following steps: S11. By deploying agricultural IoT sensors in the Chinese medicinal herb planting base, acquire the average daily precipitation data, soil pH data, and sunshine duration data of the target Chinese medicinal herb throughout its entire growth cycle; and obtain the original species category and harvesting season information of the target Chinese medicinal herb through a manual input terminal to obtain the initial source dataset. S12. The local anomaly factor algorithm is used to detect outliers in the daily average precipitation data and soil pH data in the initial source dataset. Data points with local anomaly factor scores greater than the preset outlier threshold are identified as noise data generated by sensor malfunctions. The noise data is then smoothly replaced using the arithmetic mean of normal data within adjacent time windows. S13. The daily average precipitation data, soil pH data and sunshine duration data after smooth replacement are resampled and aligned according to a fixed time step, and then spliced ​​with the original species category and harvest solar term information that have undergone unique thermal coding to generate a source feature vector with fixed dimensions. The source feature vector is processed by a secure hash algorithm and the current timestamp to generate an identity identifier consisting of a fixed-length hexadecimal string. The source feature vector is bound to the identity identifier and then uploaded to a pre-built consortium blockchain node; By combining agricultural IoT with outlier detection algorithms for local anomalies, automated cleaning of source environmental data and accurate digital modeling of origin information are achieved. One-hot coding and time-series resampling alignment technology are used to unify heterogeneous multi-source data into a fixed-dimensional source feature vector. Secure hashing algorithms and consortium blockchain technology are used to give each batch of Chinese medicinal materials an immutable digital identity, laying a data foundation for reliable traceability of data throughout the entire life cycle from the source.

[0008] Preferably, in step S13, binding the source feature vector with the identity identifier and uploading it to the pre-built consortium blockchain node specifically includes: S131. Combine the source feature vector with the identity identifier into a raw data string; S132. Generate a digital signature for the original data string using the private key in asymmetric encryption technology; S133. Package the digital signature and the original data string into a block transaction to be uploaded to the blockchain; the sorting nodes in the consortium blockchain network sort the block transactions to be uploaded to the blockchain by timestamp and request each consensus node to verify them through the Byzantine fault-tolerant consensus algorithm. S134. Once more than two-thirds of the consensus nodes have verified the validity of the digital signature, the transaction to be added to the blockchain is written into the distributed ledger, thus completing the blockchain upload. By combining asymmetric encryption digital signature mechanism with Byzantine fault-tolerant consensus algorithm, the integrity and non-repudiation of on-chain data are ensured; the timestamp service of sorting nodes guarantees the consistency of transaction order, while the mechanism of verification by more than two-thirds of consensus nodes effectively prevents data tampering by malicious nodes; this on-chain process provides a highly secure, decentralized, and trusted storage solution for binding source feature vectors and identity identifiers, enabling the initial quality information of Chinese medicinal materials to have judicial-level evidentiary effect.

[0009] Preferably, step S2 includes the following steps: S21. Deploy a high-temperature resistant image acquisition probe inside the processing container to capture multispectral image data of the sliced ​​medicinal material at preset time intervals; convert the multispectral image data to brightness, red-green hue, and yellow-blue hue spaces, and extract the pixel mean values ​​of the red-green hue channel and the yellow-blue hue channel as color gradient features; use the gray-level co-occurrence matrix to calculate the contrast and energy values ​​of the multispectral image data in different directions, and take the average as the texture roughness feature; S22. Obtain the ambient temperature time series data inside the processing container through a thermocouple sensor; input the ambient temperature time series data into the convolutional layer of a one-dimensional convolutional neural network, extract local temperature fluctuation features, and perform dimensionality reduction processing through a maximum pooling layer to output a temperature evolution feature vector. S23. Linearly concatenate the color gradient features, texture roughness features, running time parameters, and temperature evolution feature vectors to construct a multimodal processing feature matrix; S24. Input the multimodal processing feature matrix into the pre-trained processing quality benchmark evaluation network; the processing quality benchmark evaluation network includes an input layer, two fully connected hidden layers and an output layer; after processing by the nonlinear activation function of the fully connected hidden layer, the output layer outputs a value between zero and one hundred as the initial quality baseline index; By using multispectral image processing technology, the traditional method of identifying medicinal materials by color based on human experience in processing is transformed into quantitative indicators of color gradient features and texture roughness features. One-dimensional convolutional neural networks are used to extract local fluctuation features that are difficult for humans to detect from long-term temperature data. By combining multimodal physical features with a deep learning quality assessment network, standardized scoring of processing quality is achieved, eliminating the subjectivity and non-repeatability of human sensory judgment, and providing traceable initial quality baseline proof for each batch of medicinal slices.

[0010] Preferably, the training process of the processing quality benchmark evaluation network in S24 is as follows: S241. Collect the multimodal processing feature matrix of historical processing batches as training input samples; at the same time, perform destructive chemical component extraction and testing on the slices of the historical processing batches to obtain the absolute percentage of the core effective components, and amplify the absolute percentage of the components to the range of zero to one hundred through linear mapping, and use it as the training output label. S242. Construct the initial processing quality benchmark evaluation network, and randomly initialize the connection weights and bias parameters of the fully connected hidden layers; S243. Input the training input samples into the initial processing quality benchmark evaluation network to obtain the predicted quality index; calculate the mean square error between the predicted quality index and the training output label; S244. Calculate the gradient direction of the mean squared error relative to each connection weight and bias parameter using the backpropagation algorithm; use an adaptive moment estimation optimizer to update and fine-tune the connection weights and bias parameters along the opposite direction of the gradient with a preset learning rate step size; repeat this step until the mean squared error is less than the preset convergence threshold, save the current network parameters, and obtain the fabrication quality benchmark evaluation network. Using the absolute percentage of content obtained from destructive chemical component extraction and testing as the training label, and aligning it with the quality index scale through linear mapping, a supervised learning mapping relationship between multimodal processing characteristics and intrinsic component content was established. Iterative training was carried out using the mean squared error loss function and the adaptive moment estimation optimizer, enabling the network to automatically learn the complex nonlinear relationship between processing parameters and quality results from historical batch data. This ensured that the prediction results of the quality benchmark assessment network had objective basis at the chemical component level.

[0011] Preferably, step S3 includes the following steps: S31. Set standard ideal storage conditions, including standard temperature value, standard relative humidity value and standard light intensity value; extract the actual temperature value, actual relative humidity value and actual light intensity value transmitted back in real time by the IoT environmental sensing device; S32. Calculate the temperature difference between the actual temperature value and the standard temperature value; when the temperature difference is greater than zero, set the temperature attenuation weight to the temperature difference multiplied by the preset temperature sensitivity coefficient; when the temperature difference is less than or equal to zero, set the temperature attenuation weight to zero. S33. Calculate the humidity difference between the actual relative humidity value and the standard relative humidity value; when the humidity difference is greater than zero, set the humidity attenuation weight to the humidity difference multiplied by the preset humidity sensitivity coefficient; when the humidity difference is less than or equal to zero, set the humidity attenuation weight to zero. S34. Multiply the actual light intensity value by the light sensitivity coefficient to obtain the light attenuation intensity; add the temperature attenuation weight, humidity attenuation weight, light attenuation intensity and the preset baseline natural attenuation rate to obtain the instantaneous attenuation rate at the current sampling moment; multiply the duration between two adjacent sampling moments by the instantaneous attenuation rate to obtain the single environmental fluctuation loss value. S35. Subtract the sum of all single environmental fluctuation losses during the entire logistics flow stage from the initial quality baseline index to obtain the real-time updated dynamic quality index; when the dynamic quality index is lower than the preset scrapping interception threshold, an early warning blocking instruction is automatically generated and broadcast to the consortium blockchain. By comparing the difference between standard ideal storage conditions and actual environmental parameters, and combining the temperature, humidity, and light sensitivity coefficients calibrated based on accelerated stability tests, the complex environmental degradation process is transformed into a linearly superimposed instantaneous decay rate calculation model. The dynamic quality index is updated in real time with each environmental fluctuation, breaking through the limitations of traditional static management with fixed shelf life. The linkage mechanism between multi-level early warning thresholds and consortium blockchain smart contracts realizes a fully automated closed-loop management system from quality monitoring and early warning to physical blocking of outbound shipments, effectively preventing the risk of substandard drugs entering clinical settings due to environmental out-of-control issues in the logistics process.

[0012] Preferably, step S4 includes the following steps: S41. Obtain prescription dispensing records from the hospital information system and extract the medication patient number containing the identity identifier in the prescription; based on the medication patient number, retrieve electronic medical record text data and laboratory test indicator data for the previous natural month and the three natural months after medication. S42. Use a pre-trained named entity recognition model to segment and tag the electronic medical record text data, and extract entity words belonging to the symptom description category; by comparing the symptom description category entity words before and after medication, count the frequency of symptom disappearance or relief, and map them to a symptom improvement score between zero and one. S43. Extract the target physiological and biochemical indicators corresponding to the target Chinese medicinal material from the laboratory test index data; calculate the percentage decrease or increase of the target physiological and biochemical indicator value after medication relative to the target physiological and biochemical indicator value before medication; convert the percentage into an objective biochemical score between zero and one according to the preset medical evaluation standard table. S44. Assign a first diagnostic weight to the symptom improvement score and a second diagnostic weight to the objective biochemical score; multiply the symptom improvement score by the first diagnostic weight and add the product of the objective biochemical score multiplied by the second diagnostic weight to obtain the comprehensive efficacy score; combine the symptom improvement score, objective biochemical score and comprehensive efficacy score to generate a structured efficacy feedback vector. By using named entity recognition technology, symptom improvement entities are automatically extracted from unstructured electronic medical record texts, transforming subjective clinical descriptions into quantifiable symptom improvement scores. Simultaneously, the variation range of laboratory test indicators is transformed into objective biochemical scores based on evidence-based medicine standards. An inverse weighting method based on historical medical dispute data is used to determine the initial ratio of subjective and objective weights, and a safety penalty mechanism for toxic side effects is used to redistribute the weights, automatically strengthening the dominant position of objective indicators in efficacy evaluation for traditional Chinese medicines with safety risks. This system establishes a data link from real-world clinical data to feedback on the quality of medicinal materials.

[0013] Preferably, the quantification determination rules for the first diagnostic weight and the second diagnostic weight in S44 include the following steps: S441. Compile all medical disputes and complaints of poor efficacy related to the target Chinese medicinal materials in the historical database. S442. Calculate the proportion of complaints due to unrelieved subjective symptoms to the total number of complaints, and use this proportion as the initial value of the first diagnostic weight; calculate the proportion of complaints due to unmet objective biochemical indicators to the total number of complaints, and use this proportion as the initial value of the second diagnostic weight. S443. If the target Chinese medicinal material is clearly marked as having toxic side effects in the National Pharmacopoeia, the initial value of the second diagnostic weight will be forcibly increased by a preset safety penalty fixed value, and the initial value of the first diagnostic weight will be reduced accordingly to ensure that the objective biochemical score plays a dominant role in the comprehensive evaluation. At the same time, a weight boundary verification mechanism is set: if the initial value of the second diagnostic weight plus the safety penalty fixed value is greater than or equal to 1, the initial value of the second diagnostic weight will be directly set to 1 and the initial value of the first diagnostic weight will be set to 0, which means that the efficacy evaluation of this type of highly toxic medicinal material is based entirely on objective biochemical indicators to avoid weight overflow or abnormal situations of negative weight. S444. Normalize the adjusted initial values ​​of the first diagnostic weight and the second diagnostic weight so that their sum equals one, and finally obtain the first diagnostic weight and the second diagnostic weight. The method employs a reverse quantitative weighting approach based on historical dispute data, using objective statistical data on actual medical disputes and complaints of poor efficacy as the basis for weighting, thus avoiding the arbitrariness of traditional expert subjective weighting methods. By statistically analyzing the proportions of complaints about unrelieved subjective symptoms and unmet objective biochemical indicators, the weights accurately reflect the degree of influence of each dimension on efficacy evaluation in clinical practice. A safety penalty mechanism triggered by pharmacopoeia toxicity labeling is introduced, which automatically biases the evaluation of efficacy towards objective biochemical safety indicators for medicinal materials with safety risks such as hepatotoxicity.

[0014] Preferably, step S5 includes the following steps: S51. Create entity nodes in the graph database. Entity node types include medicinal material batch nodes, source feature nodes, processing parameter nodes, environmental attenuation nodes, and efficacy feedback nodes. S52. Based on the flow path of the identity identifier, establish directed edges between various entity nodes. The types of directed edges include originating from, processed from, experienced environment, and resulting in therapeutic effect. Use the directed edges to connect isolated entity nodes and construct a full-chain knowledge graph of traditional Chinese medicine. S53. The knowledge graph of the entire chain of traditional Chinese medicine is processed using a graph convolutional neural network. In each layer of the graph convolutional neural network, each entity node collects the feature vectors contained in its directly connected neighboring nodes, calculates the arithmetic mean of its own feature vector and the feature vectors of its neighboring nodes as aggregated features, and performs nonlinear mapping through an activation function to update the hidden state vector of the entity node in the next layer. S54. After multiple iterations of graph convolutional layer updates, extract the final hidden state vectors of all medicinal material batch nodes; sort the medicinal material batch nodes in descending order according to the comprehensive efficacy score in the efficacy feedback node, and select the set of high-quality medicinal material batch nodes with the comprehensive efficacy score in the top 10%. S55. Perform a statistical averaging operation on the running time parameters and temperature evolution feature vectors in the processing parameter nodes associated with the set of high-quality medicinal material batch nodes to generate a recommended processing time and recommended processing temperature range for the target Chinese medicinal material, which is then output as a process feedback optimization strategy. By constructing a full-chain knowledge graph with identity identifiers as the core hub nodes, connecting the four stages of source, processing, logistics, and clinical application through a graph database, the originally scattered and isolated data is transformed into a network knowledge structure with clear semantic relationships. Utilizing the multi-hop message passing mechanism of graph convolutional neural networks, each batch node of medicinal materials can aggregate feature information from each stage of its entire life cycle, forming a high-dimensional hidden state vector containing global context. By ranking high-quality batches based on efficacy and tracing back their shared processing parameter patterns, reverse attribution analysis from clinical terminal efficacy to source production processes is achieved. The generated process feedback optimization strategy provides pharmaceutical companies with precise and quantitative parameter adjustment basis.

[0015] Preferably, a data management system for the entire lifecycle of traditional Chinese medicine decoction pieces is provided to implement the aforementioned method for managing the entire lifecycle of traditional Chinese medicine decoction pieces; the system includes: The source tracing and blockchain module is responsible for connecting with agricultural IoT sensors to obtain basic and environmental data, performing resampling and unique thermal coding alignment, generating identity identifiers, and interacting with the consortium blockchain. The multimodal processing evaluation module, which incorporates a one-dimensional convolutional neural network and a multispectral image processing unit, is responsible for extracting color, texture, and temperature features from the processing container and running the processing quality benchmark evaluation network to output the initial quality baseline index. The dynamic environment attenuation calculation module integrates environmental parameter comparison logic and linear attenuation summation calculation unit, and is responsible for receiving warehousing and logistics environment data and updating the dynamic quality index in real time. The clinical efficacy quantification feedback module connects to the hospital's electronic medical record database and includes a named entity recognition unit and a biochemical indicator standardization conversion unit, which is responsible for outputting structured efficacy feedback vectors. The graph feedback optimization decision module is used to construct and store entity nodes and directed edges, run graph convolutional neural networks to mine high-quality batch features, and output specific process feedback optimization strategies.

[0016] (III) Beneficial Effects The present invention has the following beneficial effects: This invention changes the traditional Chinese medicine decoction piece management system that relies solely on static shelf-life management by constructing a dynamic quality index model. By combining linear decay rules with real-time environmental data, it can accurately calculate the loss of effective ingredients in the circulation process, ensuring the actual effectiveness of medication for patients.

[0017] This invention employs multimodal feature fusion technology, combining color and texture features extracted by machine vision with extracted temperature evolution features; replacing subjective human experience judgment, and providing standardized quality certification for each batch of medicinal slices through the initial quality baseline index given by the processing quality benchmark assessment network.

[0018] This invention utilizes graph neural networks to perform multi-hop path aggregation in knowledge graphs, enabling reverse tracing and attribution analysis from clinical efficacy to front-end processing technology; through automatically generated process feedback optimization strategies, it can accurately guide pharmaceutical companies to adjust production parameters, realizing a continuous evolutionary cycle of the quality of traditional Chinese medicine decoction pieces.

[0019] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, the drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating a method for managing the entire lifecycle data of traditional Chinese medicine decoction pieces according to the present invention. Figure 2 This is a schematic diagram illustrating the process of storing source feature vectors and identity identifiers into the blockchain according to the present invention. Figure 3 This is a schematic diagram of a module of a data management system for the entire life cycle of traditional Chinese medicine decoction pieces according to the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the invention, and not all embodiments. Based on the embodiments of the invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the invention.

[0023] Please see Figure 1 This invention discloses a method for managing the entire lifecycle data of traditional Chinese medicine decoction pieces, including the following steps: S1. Obtain source production environment monitoring data and basic attribute data of the target batch of Chinese medicinal materials, perform time series alignment and outlier removal preprocessing, and generate source feature vector; use hash algorithm to generate identity identifier for the target batch of Chinese medicinal materials, and upload the source feature vector and identity identifier to the pre-built consortium blockchain node after binding them. S1 includes the following steps: S11. By deploying agricultural IoT sensors in the Chinese medicinal herb planting base, acquire the average daily precipitation data, soil pH data, and sunshine duration data of the target Chinese medicinal herb throughout its entire growth cycle; and obtain the original species category and harvesting season information of the target Chinese medicinal herb through a manual input terminal to obtain the initial source dataset. Specifically, in Example S11, this example selects Polygonum multiflorum from a large-scale standardized Chinese medicinal herb planting base as the target Chinese medicinal herb; within the planting area of ​​approximately 500 mu, 30 agricultural Internet of Things (IoT) sensor nodes are deployed in a grid pattern; during the three-year growth cycle of Polygonum multiflorum, the sensors transmit data daily at regular intervals; the acquired daily average precipitation data is in millimeters, the soil pH data is in pH value, which usually fluctuates between 5.5 and 7.0, and the sunshine duration data is in hours; at the same time, the base's agricultural technicians use handheld terminals to input the original species category of this batch of Polygonum multiflorum as the dicotyledonous root of Polygonum multiflorum (Polygonaceae family), and the harvesting solar term information as after the autumnal equinox to before the first frost; these continuous numerical data and discrete textual data together constitute the initial source dataset; S12. The local anomaly factor algorithm is used to detect outliers in the daily average precipitation data and soil pH data in the initial source dataset. Data points with local anomaly factor scores greater than the preset outlier threshold are identified as noise data generated by sensor malfunctions. The noise data is then smoothly replaced using the arithmetic mean of normal data within adjacent time windows. Specifically, in embodiment S12, due to the harsh outdoor environment, sensor probes often generate erroneous readings due to soil cover or insect interference. For example, on a certain day, the soil pH sensor suddenly returned extreme data with a pH value of 1.2, which is highly acidic and plants cannot survive. This embodiment uses a local anomaly factor algorithm, which judges anomalies by comparing the local density of a data point with its surrounding neighboring nodes. The calculation shows that the local density of the data point with pH=1.2 is much lower than that of the surrounding normal points with pH=6.2, and its local anomaly factor score is as high as 4.5. This embodiment sets the preset outlier threshold to 2.0. Since 4.5 is greater than 2.0, the system automatically judges the point as noise data. The system extracts normal pH value data for 3 days before and after the anomaly point, for a total of 6 data points, and calculates the arithmetic mean of these 6 data points to replace the original erroneous data. After this step, the underlying dirty data is cleaned up, ensuring the purity of subsequent feature extraction. Furthermore, the calculation process of the local anomaly factor is as follows: Before calculating the Euclidean distance of each physical quantity, the daily average precipitation and soil pH series were Z-scor standardized to unify the dimensions into dimensionless standard deviation distance, thereby ensuring the physical comparability of distance measurements in a multivariable space. Calculate the reachable distance from point p to point o. Where d(p,o) is the Euclidean distance between data points p and o. Let k be the k-th distance from point o; calculate the local reachability density. , representing the neighborhood N of point p k (p) The reciprocal of the average reachable distance; calculate the LOF score. ;in, This represents the local reachability density of the neighboring point o; When the LOF value is close to 1, it means that the density of point p is similar to that of its neighborhood; when the LOF value is greater than 1, such as 4.5 in this embodiment, it means that the density of point p is much lower than that of its neighborhood, and it is determined to be an isolated outlier. S13. The daily average precipitation data, soil pH data and sunshine duration data after smooth replacement are resampled and aligned according to a fixed time step, and then spliced ​​with the original species category and harvest solar term information that have undergone unique thermal coding to generate a source feature vector with fixed dimensions. The source feature vector is processed by a secure hash algorithm and the current timestamp to generate an identity identifier consisting of a fixed-length hexadecimal string. The source feature vector is bound to the identity identifier and then uploaded to a pre-built consortium blockchain node; Please see Figure 2 In step S13, the source feature vector is bound to the identity identifier and then uploaded to the pre-built consortium blockchain node, specifically including: S131. Combine the source feature vector with the identity identifier into a raw data string; S132. Generate a digital signature for the original data string using the private key in asymmetric encryption technology; S133. Package the digital signature and the original data string into a block transaction to be uploaded to the blockchain; the sorting nodes in the consortium blockchain network sort the block transactions to be uploaded to the blockchain by timestamp and request each consensus node to verify them through the Byzantine fault-tolerant consensus algorithm. S134. Once more than two-thirds of the consensus nodes have verified the validity of the digital signature, the transaction to be added to the blockchain is written into the distributed ledger, thus completing the blockchain upload. Specifically, in embodiment S13, the data transmission frequency of the sensors may vary, such as precipitation once a day and pH value once an hour. To unify the dimensions, this embodiment sets a fixed time step of one week. The precipitation within each week is accumulated, and the pH value and sunshine duration are averaged weekly, thereby compressing and aligning the three-year time series into a sequence matrix with 156 time steps. For non-numerical harvest solar terms, such as the autumnal equinox, one-hot encoding is used to convert them into sparse vectors composed of 0s and 1s. For example, the twenty-four solar terms are mapped to a vector of length 24, with the autumnal equinox at position 1 and the rest at position 0. The numerical sequence matrix and the one-hot encoded vector are concatenated end to end to form a source feature vector of length 512. The precise timestamp of the current operation is extracted, such as 20231015093000. This timestamp is combined with a 512-dimensional source feature vector and input into the SHA-256 secure hash algorithm engine. After complex bitwise operations and nonlinear transformations, a 64-bit hexadecimal string is output, such as e3b0c44298fc1c149afbf4c8996fb924... This string is the DID of this batch of Polygonum multiflorum. The DID and the source feature vector are packaged and sent to a consortium blockchain network jointly maintained by the drug regulatory authority, core pharmaceutical companies, and hospitals. After verification by consensus nodes, it is permanently recorded on the distributed ledger, realizing the immutability of the source data. By using agricultural IoT and LOF anomaly detection algorithm, precise cleaning of source environmental data and digital modeling of origin information were achieved. Blockchain technology was used to ensure the authenticity and immutability of the initial data of Chinese medicinal materials, laying a reliable data foundation for subsequent quality assessment. S2. During the processing stage of the target batch of Chinese medicinal materials, the operating time parameters of the processing equipment, the time series data of ambient temperature, and the multispectral image data of the slices during the processing are collected simultaneously. The color gradient features and texture roughness features of the multispectral image data are extracted, and the color gradient features, texture roughness features, running time parameters, and ambient temperature time series data are fused together and input into the processing quality benchmark evaluation network. The initial quality baseline index of the batch of Chinese herbal medicine slices is calculated and output, and the initial quality baseline index is added to the consortium blockchain. S2 includes the following steps: S21. Deploy a high-temperature resistant image acquisition probe inside the processing container to capture multispectral image data of the sliced ​​medicinal material at preset time intervals; convert the multispectral image data to brightness, red-green hue, and yellow-blue hue spaces, and extract the pixel mean values ​​of the red-green hue channel and the yellow-blue hue channel as color gradient features; use the gray-level co-occurrence matrix to calculate the contrast and energy values ​​of the multispectral image data in different directions, and take the average as the texture roughness feature; Specifically, in Example S21, taking Polygonum multiflorum as an example, raw Polygonum multiflorum is hepatotoxic and must be processed using the traditional method of steaming with black bean juice to be transformed into medicinal Polygonum multiflorum. The experience lies in observing the color. In this example, a high-temperature resistant multispectral camera with a sapphire protective cover is deployed in an industrial steaming tank. The cross-sectional image of Polygonum multiflorum is captured every 2 hours. The original image is in RGB format, and the system converts it to CIELAB color space, including L brightness, a red-green saturation, and b yellow-blue saturation. As steaming proceeds, the color of the cross-sectional of Polygonum multiflorum gradually changes from yellowish-white to dark brown. The system extracts the values ​​of all pixels in the a and b channels and calculates the arithmetic mean. For example, the mean value of the a channel is 15 in the early stage of processing and drops to 2 after 32 hours of processing, which clearly quantifies the color gradation characteristics. Furthermore, to quantify the wrinkles on the surface of the medicinal slices caused by moisture loss and tissue changes, the system converts the image into a grayscale image and calculates the gray-level co-occurrence matrix (GLCM). Specifically, the system counts the frequency of grayscale value combinations of adjacent pixels in the horizontal, vertical, and diagonal directions of the image; extracts two statistical quantities from the matrix: contrast, which reflects the local grayscale change rate, and energy, which reflects the uniformity of the texture; and calculates the arithmetic mean of the contrast and energy in the four directions, combining them to form a two-dimensional texture roughness feature vector. S22. Obtain the ambient temperature time series data inside the processing container through a thermocouple sensor; input the ambient temperature time series data into the convolutional layer of a one-dimensional convolutional neural network, extract local temperature fluctuation features, and perform dimensionality reduction processing through a maximum pooling layer to output a temperature evolution feature vector. Specifically, in embodiment S22, since the temperature is not absolutely constant during long-term steaming, the thermocouple sensor records the temperature once per minute, forming a continuous time series of about 2000. Simply calculating the average temperature will mask local temperature drops or overheating phenomena. Therefore, this embodiment constructs a one-dimensional convolutional neural network 1D-CNN. The architecture of the one-dimensional convolutional neural network 1D-CNN includes: 1 input layer; 3 convolutional layers with 32, 64 and 128 convolutional kernels respectively, kernel size of 5 and stride of 1; 3 max pooling layers with pooling size of 2; and 1 fully connected 64-dimensional output layer. Furthermore, the network contains three convolutional layers with a kernel size of 5, meaning it observes temperature changes over a continuous 5-minute period. The kernels slide across the temperature sequence, extracting fluctuation features of local peaks and troughs through multiplication and addition operations. Subsequently, the data enters a max pooling layer with a pooling window size of 2, preserving the most salient local features and halving the data length. After three convolutional and pooling operations, the original 2000-dimensional temperature data is refined and compressed into a 64-dimensional temperature evolution feature vector. This vector captures hidden processing defects such as an unexpected 5-degree temperature drop lasting 20 minutes in the 10th hour. Furthermore, the training process of the 1D-CNN is as follows: using historical batches of temperature time-series data as input, and using the effective component conversion rate obtained from the same period of destructive chemical detection as a supervision label, such as the conversion rate of bound anthraquinone; the loss function adopts the mean squared error function (MSE); the Adam optimizer is used for iterative training, and when the MSE loss value on the validation set decreases by less than 1e-5 for 20 consecutive rounds, or when the preset first maximum number of training iterations is reached, the model is determined to have converged and the training is completed. S23. Linearly concatenate the color gradient features, texture roughness features, running time parameters, and temperature evolution feature vectors to construct a multimodal processing feature matrix; Specifically, in embodiment S23, the system extracts the total running time parameter, such as 32 hours, and quantizes it into the value 32.0; the 2D color feature, 2D GLCM texture feature, 1D total running time, and 64D temperature evolution feature vector obtained in S21, which include the mean values ​​of channels a and b, are concatenated in the data structure. Furthermore, in order to eliminate the huge differences between different units, such as time being 32 while texture energy may be only 0.01, range standardization is performed on all the stitched data, so that the value of each dimension is scaled to between 0 and 1, ultimately forming a multimodal fabrication feature matrix containing 69 feature dimensions. S24. Input the multimodal processing feature matrix into the pre-trained processing quality benchmark evaluation network; the processing quality benchmark evaluation network includes an input layer, two fully connected hidden layers and an output layer; after processing by the nonlinear activation function of the fully connected hidden layer, the output layer outputs a value between zero and one hundred as the initial quality baseline index; The training process of the processing quality benchmark evaluation network in S24 is as follows: S241. Collect the multimodal processing feature matrix of historical processing batches as training input samples; at the same time, perform destructive chemical component extraction and testing on the slices of the historical processing batches to obtain the absolute percentage of the core effective components, and amplify the absolute percentage of the components to the range of zero to one hundred through linear mapping, and use it as the training output label. S242. Construct the initial processing quality benchmark evaluation network, and randomly initialize the connection weights and bias parameters of the fully connected hidden layers; The processing quality benchmark evaluation network adopts a multilayer perceptron structure, including one input layer, two fully connected hidden layers, and one output layer. The dimension of the input layer is consistent with the dimension of the multimodal processing feature matrix. The number of neurons in the first hidden layer is 128, and the number of neurons in the second hidden layer is 64. The activation function of the hidden layer is the ReLU function. The number of neurons in the output layer is 1, and the activation function is the Sigmoid function. The output value is linearly mapped to the 0-100 range as the initial quality baseline index. S243. Input the training input samples into the initial processing quality benchmark evaluation network to obtain the predicted quality index; calculate the mean square error between the predicted quality index and the training output label; S244. Calculate the gradient direction of the mean squared error relative to each connection weight and bias parameter using the backpropagation algorithm; use an adaptive moment estimation optimizer to update and fine-tune the connection weights and bias parameters along the opposite direction of the gradient with a preset learning rate step size; repeat this step until the mean squared error is less than the preset convergence threshold, save the current network parameters, and obtain the fabrication quality benchmark evaluation network. Specifically, in embodiment S24, the aforementioned 69-dimensional multimodal processing feature matrix is ​​fed into the processing quality benchmark evaluation network; the processing quality benchmark evaluation network adopts a multilayer perceptron structure, with an input layer of 69 dimensions, hidden layer 1 of 128 dimensions, hidden layer 2 of 64 dimensions, and an output layer of 1 dimension; the total number of connection weights is (69×128)+(128×64)+(64×1)=17152; this architecture achieves a high degree of fusion between physical features and process features through nonlinear mapping; the physical features include color and texture, and the process features include temperature and time; The optimizer uses the Adam optimizer with adaptive moment estimation, with an initial learning rate of 0.001 and a weight decay coefficient of 5 × 10⁻⁶.-4 The convergence threshold is set to 1×10. -4 The network is considered converged when the mean square error decreases less than the convergence threshold for 10 consecutive training rounds. Furthermore, the first hidden layer of the network contains 128 neurons, and the second layer contains 64 neurons. The input features are multiplied by the connection weights between each neuron and a bias value is added. Then, the ReLU activation function is used to perform a non-linear mapping, i.e., when the value is less than 0, the output is 0, and when it is greater than 0, the output is unchanged. The data is aggregated to the output layer, which contains only 1 neuron. The Sigmoid activation function is used to compress the final result to between 0 and 1, and then multiplied by 100. For example, after complex forward propagation calculations by the network, the output value of this batch of processed Polygonum multiflorum is 94.5; this means that, based on its multimodal performance during the processing, the toxic component bound anthraquinone of this batch of medicinal slices has been fully converted, and the effective component stilbene glycoside has been well preserved, and the initial quality baseline index is rated as 94.5 points; the system appends this score together with the identity identifier DID generated by S1 and writes it into the consortium blockchain as a new record; By using multimodal data fusion technology, the vague empirical indicators in traditional processing are transformed into quantifiable digital features. Through deep learning models, the initial quality of medicinal slices is objectively evaluated, solving the problems of difficulty in standardizing the processing of traditional Chinese medicine and large quality fluctuations.

[0024] S3. During the storage and logistics of the Chinese herbal medicine slices, real-time storage environment temperature and humidity data and light intensity data are obtained through IoT environmental sensing devices. Based on the initial quality baseline index, combined with the real-time storage environment temperature and humidity data and light intensity data, the cumulative loss value of the effective ingredients is calculated through the preset dynamic attenuation quantification rules, and then the dynamic quality index is updated and generated in real time. S3 includes the following steps: S31. Set standard ideal storage conditions, including standard temperature value, standard relative humidity value and standard light intensity value; extract the actual temperature value, actual relative humidity value and actual light intensity value transmitted back in real time by the IoT environmental sensing device; S32. Calculate the temperature difference between the actual temperature value and the standard temperature value; when the temperature difference is greater than zero, set the temperature attenuation weight to the temperature difference multiplied by the preset temperature sensitivity coefficient; when the temperature difference is less than or equal to zero, set the temperature attenuation weight to zero. S33. Calculate the humidity difference between the actual relative humidity value and the standard relative humidity value; when the humidity difference is greater than zero, set the humidity attenuation weight to the humidity difference multiplied by the preset humidity sensitivity coefficient; when the humidity difference is less than or equal to zero, set the humidity attenuation weight to zero. S34. Multiply the actual light intensity value by the light sensitivity coefficient to obtain the light attenuation intensity; add the temperature attenuation weight, humidity attenuation weight, light attenuation intensity and the preset baseline natural attenuation rate to obtain the instantaneous attenuation rate at the current sampling moment; multiply the duration between two adjacent sampling moments by the instantaneous attenuation rate to obtain the single environmental fluctuation loss value. Specifically, in embodiments S31-S34, the environmental changes of Chinese herbal medicine pieces in logistics vehicles and hospital warehouses directly lead to the degradation of effective components; traditional static shelf life cannot reflect this real process; this embodiment abandons the complex calculus Arrhenius equation and adopts a clear and quantified linear decay rule mechanism; the reason for not using the classic nonlinear dynamic model is to adapt to the application scenario constraints of the Internet of Things in the Chinese medicine industry; specifically, Chinese herbal medicine pieces contain dozens to hundreds of chemical components, and the degradation laws of different components are different, and the overall mass decay cannot be described by a single dynamic equation; if each effective component is modeled separately, the parameter calibration cost is extremely high, and it is not feasible for industrial promotion; in addition, the computing power and power consumption of edge sensing devices at the warehousing and logistics end are limited, and the real-time solution of nonlinear differential equations is difficult to deploy on low-power IoT nodes; Existing technologies either rely on static management with fixed shelf life or only conduct single-component kinetic studies at the laboratory level, without proposing a practical dynamic quality calculation method for multi-component systems of traditional Chinese medicine and edge scenarios of industrial IoT. In contrast, this solution pre-calibrates the sensitivity coefficient specific to each variety through accelerated stability testing, decoupling the complex multi-factor degradation process into independent loss terms that can be linearly superimposed. This preserves the scientific basis of kinetic experiments while enabling real-time calculation with low computing power at the edge, demonstrating the non-obvious nature of this invention in the field of whole-chain quality management of traditional Chinese medicine. Furthermore, from the perspective of physicochemical kinetics, the isothermal degradation of the active ingredients in traditional Chinese medicine usually follows first-order kinetics, and the relationship between the rate constant and temperature follows the Arrhenius equation. In conventional warehousing and logistics scenarios, the deviation of environmental parameters from standard conditions is usually small. After performing a first-order Taylor expansion of the Arrhenius equation at the standard temperature point, the temperature deviation and the increment of degradation rate can be approximated as a linear relationship. Similarly, the accelerating effect of humidity and light on degradation can also be described by linear approximation within a small fluctuation range. Traditional Chinese medicine decoction pieces belong to a complex system with multiple components. The degradation activation energy and humidity response coefficient of different effective components vary significantly. The coupling effect between multiple factors cannot be accurately calibrated by a unified nonlinear model. This solution sums the three independent linear terms of temperature, humidity, and light, which is essentially an engineered linear approximation of the multi-factor coupled degradation process within the range of normal storage fluctuations. All sensitivity coefficients are calibrated by accelerated stability tests of the corresponding varieties. They can accurately reflect the cumulative impact of environmental fluctuations on quality within an acceptable error range, while avoiding the practical problem that the parameters of multi-component nonlinear coupling models cannot be measured one by one. Specifically, for processed Polygonum multiflorum, the system sets standard ideal storage conditions: standard temperature of 20℃, standard relative humidity of 55%, and light intensity set to 0 Lux in the dark; the system presets a temperature sensitivity coefficient of 0.05℃. -1 ·h -1 That is, for every 1 degree Celsius above the set temperature, the base attenuation weight increases by 0.05; the humidity sensitivity coefficient is 0.08%. -1 ·h -1 The light sensitivity coefficient is 0.001 Lux. -1 ·h -1 ; Furthermore, the methods for setting the temperature sensitivity coefficient, humidity sensitivity coefficient, and light sensitivity coefficient are as follows: Based on the stability test guidelines of the Chinese Pharmacopoeia, accelerated stability tests are conducted on the target medicinal material; under constant humidity, the rate of change of the content of active ingredients over time at different temperatures, such as 30℃, 40℃, and 50℃, is measured. The content decay slope corresponding to each 1℃ temperature change is obtained through linear fitting of the Arrhenius equation, which is the temperature sensitivity coefficient; similarly, the decay slopes of humidity and light are measured separately using the controlled variable method. These steps ensure that the decay parameters have clear kinetic experimental basis; simultaneously, according to the long-term stability test requirements of the Chinese Pharmacopoeia, the monthly average degradation rate of the active ingredients of the target medicinal material under standard storage conditions is measured, converted into an hourly decay rate, and calibrated as the baseline natural decay rate for this variety; taking processed Polygonum multiflorum as an example, its baseline natural decay rate is set to 0.002h. -1 This characterizes the natural degradation level of the quality of medicinal slices under standard storage conditions; For example, during a long-haul logistics transport, the refrigeration unit of the cold chain truck experienced a brief malfunction; from 2 pm to 3 pm, lasting for 1 hour; the IoT sensor reported the actual temperature as 35°C, the actual relative humidity as 80%, and the actual light intensity as 100 Lux. The calculation process is as follows: Calculate the temperature difference: 35℃ - 20℃ = 15℃, which is greater than zero. Therefore, the temperature decay weight = 15 * 0.05 = 0.75, in hours (h). -1Calculate the humidity difference: 80% - 55% = 25%, which is greater than zero. Therefore, the humidity attenuation weight = 25 * 0.08 = 2.0, in hours (h). -1 ; Calculate the light attenuation weight: 100 * 0.001 = 0.1, unit: h -1 Instantaneous decay rate = 0.75 + 2.0 + 0.1 + 0.002 = 2.852, in hours (h). -1 The single environmental fluctuation loss value = duration (1 hour) * 2.85 = 2.852; if the environment is within the standard range, such as a temperature of 18℃, the temperature difference value is negative, the attenuation weight is directly reduced to zero, and no additional environmental fluctuation loss is generated. S35. Subtract the sum of all single environmental fluctuation losses during the entire logistics flow stage from the initial quality baseline index to obtain the real-time updated dynamic quality index; when the dynamic quality index is lower than the preset scrapping interception threshold, an early warning blocking instruction is automatically generated and broadcast to the consortium blockchain. In specific implementation, the above embodiment S35 is as follows: the initial quality baseline index of this batch of processed Polygonum multiflorum when it leaves the factory is 94.5; after experiencing the above-mentioned 1 hour of high temperature and high humidity fluctuation, the system automatically deducts: dynamic quality index = 94.5 - 2.852 = 91.648; as the logistics cycle is extended and multiple small fluctuations accumulate, the index will continue to decrease dynamically. The system has pre-set control rules: when the dynamic quality index is below 80, a yellow priority allocation warning is triggered, reminding the hospital pharmacy to move the batch to the front of the shelf for priority use; when the dynamic quality index is below the scrap interception threshold of 60, it indicates that the effective ingredients of the batch of medicinal slices have been severely lost or may have become moldy; at this time, the system automatically generates a red warning blocking instruction, which is automatically broadcast to the consortium blockchain through a smart contract; after receiving the blockchain instruction, the hospital pharmacy's HIS system will directly pop up an error message and lock the batch of medicinal slices during the barcode scanning and warehousing process, physically preventing the batch of medicinal slices from being distributed to patients, thereby eliminating potential medication safety hazards; Furthermore, the scrapping interception threshold maps the lower limit of the content in the Chinese Pharmacopoeia to the internal control interception line of the system, and adds 20% as a safety margin; the dynamic quality index threshold is based on the worst environmental conditions recorded in the historical logistics data of the variety, simulating the decay rate of the dynamic quality index; the time required for any index value to decay to the regulatory compliance line is calculated; the required time = the index value corresponding to the longest logistics allocation cycle is set as the early warning interception threshold to ensure that after the early warning is triggered, there is still a sufficient time window to complete the logistics allocation, avoiding the quality falling below the compliance line before delivery to the patient; This step introduces a dynamic quality index, transforming the static shelf life into a dynamic effectiveness evaluation based on real-time environmental fluctuations. This enables precise quality monitoring and automated early warning and interception during the warehousing and logistics process, thereby improving the safety of medication use in the distribution process. S4. In the clinical application stage of the Chinese herbal medicine, obtain the prescription dispensing record, and associate and match the identity identifier with the electronic medical record text data and laboratory test index data of the patient taking the medicine; use a natural language processing model to extract the subjective symptom improvement entity from the electronic medical record text data, and combine it with the change range of the laboratory test index data to calculate and generate a structured efficacy feedback vector. S4 includes the following steps: S41. Obtain prescription dispensing records from the hospital information system and extract the medication patient number containing the identity identifier in the prescription; based on the medication patient number, retrieve electronic medical record text data and laboratory test indicator data for the previous natural month and the three natural months after medication. S42. Use a pre-trained named entity recognition model to segment and tag the electronic medical record text data, and extract entity words belonging to the symptom description category; by comparing the symptom description category entity words before and after medication, count the frequency of symptom disappearance or relief, and map them to a symptom improvement score between zero and one. Specifically, in Example S42, by extracting the set of target symptom items before the patient takes medication, the number of items that completely disappear or are reduced in severity after medication is counted, and the ratio of the number of improved items to the total number of symptom items before medication is used as the symptom improvement score; if all symptoms before medication completely disappear after medication, the symptom improvement score is 1; if all symptoms do not improve or even worsen, the score is 0; partial improvement is linearly mapped to the 0-1 interval proportionally; S43. Extract the target physiological and biochemical indicators corresponding to the target Chinese medicinal material from the laboratory test index data; calculate the percentage decrease or increase of the target physiological and biochemical indicator value after medication relative to the target physiological and biochemical indicator value before medication; convert the percentage into an objective biochemical score between zero and one according to the preset medical evaluation standard table. Specifically, in embodiments S41-S43, when the batch of prepared Polygonum multiflorum with DID e3b0... is allocated to patient Zhang San (patient number 00123) in the hospital for the treatment of hyperlipidemia with hair loss, the system background automatically binds the DID with the patient number; the system retrieves Zhang San's electronic medical records before and after medication; For unstructured patient complaints, such as a patient's self-report of significantly reduced hair loss and improved sleep, but occasional dry mouth, the system employs a named entity recognition model combining a bidirectional long short-term memory network (Bi-LSTM) with a conditional random field (CRF). The model segments the text character by character, identifying entities related to symptom improvement (reduced hair loss) and adverse reactions (dry mouth). The system shows that the frequency of negative symptom entities such as hair loss and dizziness decreased by 80% after medication compared to before medication, and assigns a subjective symptom improvement score of 0.8 according to the linear mapping rule. Furthermore, the named entity recognition model includes a word embedding layer for mapping medical record text into a 128-dimensional vector, a bidirectional LSTM layer with a hidden layer dimension of 256 for capturing the semantic association between the preceding and following symptom descriptions, and a CRF layer for outputting the optimal label sequence. The named entity recognition model is trained using a TCM electronic medical record dataset annotated by medical experts, and negative log-likelihood loss is used as the loss function. Training is considered complete when the model's F1-score on the test set reaches the set test target threshold. Simultaneously, the system extracts structured laboratory test data; the targeted physiological and biochemical indicators for processed Polygonum multiflorum are set as serum total cholesterol (TC), low-density lipoprotein (LDL-C), and liver function indicator alanine aminotransferase (ALT); Zhang San's TC was 6.8 mmol / L before medication, and decreased to 5.1 mmol / L after three months of medication, a decrease of 25%; according to the system's preset standard table, a decrease of more than 20% in lipid-lowering drugs is recorded as a full score of 1.0, and its blood lipid improvement sub-score is 1.0; at the same time, no abnormal increase in ALT was observed, and the safety score is a full score of 1.0; the arithmetic mean of the two scores is finally obtained to obtain an objective biochemical score of 1.0; Furthermore, the standard table is constructed by retrieving the clinical randomized controlled trial (RCT) literature of the target Chinese medicinal material in the past 10 years, extracting the range of changes in biochemical indicators corresponding to significant efficacy, effectiveness, and ineffectiveness in each clinical study, taking the weighted average of the results of each study as the grading boundary, and constructing a structured medical evaluation standard table. S44. Assign a first diagnostic weight to the symptom improvement score and a second diagnostic weight to the objective biochemical score; multiply the symptom improvement score by the first diagnostic weight and add the product of the objective biochemical score multiplied by the second diagnostic weight to obtain the comprehensive efficacy score; combine the symptom improvement score, objective biochemical score and comprehensive efficacy score to generate a structured efficacy feedback vector. The quantification determination rules for the first diagnostic weight and the second diagnostic weight in S44 include the following steps: S441. Compile all medical disputes and complaints of poor efficacy related to the target Chinese medicinal materials in the historical database. S442. Calculate the proportion of complaints due to unrelieved subjective symptoms to the total number of complaints, and use this proportion as the initial value of the first diagnostic weight; calculate the proportion of complaints due to unmet objective biochemical indicators to the total number of complaints, and use this proportion as the initial value of the second diagnostic weight. S443. If the target Chinese medicinal material is clearly marked as having toxic side effects in the National Pharmacopoeia, the initial value of the second diagnostic weight will be forcibly increased by a preset safety penalty fixed value, and the initial value of the first diagnostic weight will be reduced accordingly to ensure that the objective biochemical score plays a dominant role in the comprehensive evaluation. At the same time, a weight boundary verification mechanism is set: if the initial value of the second diagnostic weight plus the safety penalty fixed value is greater than or equal to 1, the initial value of the second diagnostic weight will be directly set to 1 and the initial value of the first diagnostic weight will be set to 0, which means that the efficacy evaluation of this type of highly toxic medicinal material is based entirely on objective biochemical indicators to avoid weight overflow or abnormal situations of negative weight. Specifically, the fixed value of the security penalty is obtained through... Calculate; where N adr N represents the historical number of adverse reaction complaints. total The total sample size is γ, which is the toxicity sensitivity coefficient. Its value is strictly determined according to the legal toxicity levels of medicinal materials as specified in the Pharmacopoeia of the People's Republic of China. It is used to apply differentiated safety weighting amplification effects to medicinal materials with different toxicity risks. For example, the mapping rules between toxicity levels and γ are shown in Table 1 below: Table 1 Toxicity Sensitivity Coefficient Mapping Table

[0025] S444. Normalize the adjusted initial values ​​of the first diagnostic weight and the second diagnostic weight so that their sum equals one, and finally obtain the first diagnostic weight and the second diagnostic weight. Specifically, in embodiment S44, the weight calculation rule abandons the subjective weighting based on expert guesswork and adopts a reverse quantitative weighting method based on historical dispute data; The system statistically analyzed 1,000 medical disputes and complaints about Polygonum multiflorum in the historical database; among them, 300 complaints were due to the failure to relieve subjective symptoms, that is, patients felt that it was ineffective, and 700 complaints were due to the failure to meet objective biochemical standards, that is, abnormal liver function was found after blood tests after taking it. The proportion of subjective complaints is calculated as 300 / 1000=0.3, which is used as the initial value for the first diagnostic weight; the proportion of objective complaints is calculated as 700 / 1000=0.7, which is used as the initial value for the second diagnostic weight. Because Polygonum multiflorum is clearly labeled in the National Pharmacopoeia as having the risk of liver damage, the system triggers a mandatory safety penalty mechanism: the initial value of the second diagnostic weight is forcibly increased by 0.15, becoming a fixed safety penalty value of 0.85; and 0.15 is deducted from the first diagnostic weight, becoming 0.15; since the sum of the two is still 1.0, no additional normalization is required; finally, the first diagnostic weight is determined to be 0.15, and the second diagnostic weight is determined to be 0.85; this reflects that for traditional Chinese medicines with toxic risks, objective biochemical safety must be given primary evaluation. The overall efficacy score was calculated as (0.8*0.15) + (1.0*0.85) = 0.12 + 0.85 = 0.97. Finally, [0.8, 1.0, 0.97] was concatenated to generate the structured efficacy feedback vector for the patient taking this batch of decoction pieces. By collecting data from hundreds or thousands of patients and calculating the average value, the final group efficacy feedback vector for this batch of decoction pieces can be obtained. By mining real-world clinical data using NLP technology, subjective efficacy and objective indicators are deeply integrated to construct a weighted efficacy feedback system, which solves the problems of difficulty in quantifying and evaluating the clinical efficacy of traditional Chinese medicine decoction pieces and the lack of feedback chain. S5. Using the identity identifier as the core node, and the source feature vector, initial quality baseline index, dynamic quality index, and efficacy feedback vector as associated attribute nodes, construct a full-chain knowledge graph of traditional Chinese medicine; use a graph neural network to perform multi-hop path aggregation and high-dimensional feature mining on the full-chain knowledge graph of traditional Chinese medicine, find the optimal parameter combination of the processing stage corresponding to the optimal interval of the target efficacy feedback vector, and output the process feedback optimization strategy for the target traditional Chinese medicine. S5 includes the following steps: S51. Create entity nodes in the graph database. Entity node types include medicinal material batch nodes, source feature nodes, processing parameter nodes, environmental attenuation nodes, and efficacy feedback nodes. S52. Based on the flow path of the identity identifier, establish directed edges between various entity nodes. The types of directed edges include originating from, processed from, experienced environment, and resulting in therapeutic effect. Use the directed edges to connect isolated entity nodes and construct a full-chain knowledge graph of traditional Chinese medicine. Specifically, in embodiments S51-S52, the system constructs a knowledge graph in the Neo4j graph database; using DIDe3b0... as the central hub node, i.e., the medicinal material batch node; establishing the 512-dimensional source features of S1 as a node, with directed edges pointing to the DID node; establishing the feature matrix containing 32 hours, temperature curves, and texture features in S2 as a processing parameter node, with processing parameters pointing to the DID node; establishing the environmental data recording high temperature fluctuations in S3 as an environmental decay node, with experienced environment pointing to the DID node; establishing the efficacy feedback vector with a score of 0.97 in S4 as an efficacy node, with directed edges from the DID node pointing to it; as data is continuously incorporated, a full-chain knowledge graph of traditional Chinese medicine is formed; traditional relational databases and manual statistics are difficult to discover patterns from this network structure, therefore, a graph neural network is further introduced; S53. The knowledge graph of the entire chain of traditional Chinese medicine is processed using a graph convolutional neural network. In each layer of the graph convolutional neural network, each entity node collects the feature vectors contained in its directly connected neighboring nodes, calculates the arithmetic mean of its own feature vector and the feature vectors of its neighboring nodes as aggregated features, and performs nonlinear mapping through an activation function to update the hidden state vector of the entity node in the next layer. Specifically, in embodiment S53, it should be noted that the above aggregation method is mathematically equivalent to the symmetric normalization aggregation mechanism of standard graph convolutional networks. Specifically, standard GCN achieves feature-weighted aggregation of the central node and its neighboring nodes by introducing symmetric normalization of the adjacency matrix and degree matrix of self-loop connections. In the knowledge graph constructed by this scheme, each batch of medicinal materials has a fixed one-to-one association with the central node and the four attribute nodes of source, processing, environment, and efficacy. The semantic association strength between each neighboring node and the central node is consistent, and the edge weights are indistinguishable. Therefore, the arithmetic mean is the simplified form of symmetric normalization aggregation. Using the above aggregation method, there is no need to learn the edge weight parameters, which can significantly reduce the number of model parameters, adapt to the scenario of limited sample size of complete batch data of Chinese medicine throughout its entire life cycle, reduce the risk of overfitting, and at the same time ensure that the contribution of features at each stage to the node representation is balanced. The graph convolutional neural network (GCN) consists of three graph convolutional layers. The first layer maps the initial feature vector of each entity node to a 256-dimensional hidden state vector. The second layer maps the 256-dimensional vector to 128 dimensions. The third layer maps the 128-dimensional vector to 64 dimensions, which serves as the final hidden state vector of the node. Each graph convolutional layer is followed by a ReLU activation function to introduce nonlinear transformation capabilities. The core operation mechanism of the graph convolutional layer is as follows: each node first collects the feature vectors of all its direct neighbors, aggregates its own features with those of its neighbors, calculates the average, performs a linear transformation through the trainable weight matrix of the layer, and finally outputs the updated node representation through the activation function. By stacking three layers of graph convolution, each medicinal batch node can aggregate the information of its multi-hop neighbors layer by layer: the first layer aggregates the information of direct neighbors, such as processing parameter nodes and source feature nodes; the second layer aggregates the information of two-hop neighbors; and the third layer further extends to three-hop neighbors. Ultimately, the hidden state vector of the medicinal batch node integrates the high-dimensional features of the entire chain, including the source environment, processing technology, logistics attenuation, and clinical efficacy. The graph convolutional neural network (GCN) is trained through the following steps: First, batches of traditional Chinese medicine (TCM) decoction pieces with complete lifecycle data records and full efficacy feedback vectors are selected from a historical database. Then, a subgraph is extracted centered on each batch node, and the comprehensive efficacy score corresponding to that batch is used as the training label. The mean squared error (MSE) loss function is used to measure the deviation between the model's predicted efficacy and the actual efficacy. The optimizer is the Adam adaptive moment estimation optimizer, with an initial learning rate set to 0.001 and a weight decay coefficient set to 5 × 10⁻⁶. -4 Each training batch processes 32 sub-image samples; When the mean squared error loss value on the validation set decreases by less than 1% over 20 consecutive iterations If the model has converged, training is terminated early. If the convergence condition is not triggered, training is terminated when the preset second maximum number of iterations is reached. After training, the model weight parameters at the moment with the lowest loss on the validation set are saved as the final deployment weights. S54. After multiple iterations of graph convolutional layer updates, extract the final hidden state vectors of all medicinal material batch nodes; sort the medicinal material batch nodes in descending order according to the comprehensive efficacy score in the efficacy feedback node, and select the set of high-quality medicinal material batch nodes with the comprehensive efficacy score in the top 10%. S55. Perform a statistical averaging operation on the running time parameters and temperature evolution feature vectors in the processing parameter nodes associated with the set of high-quality medicinal material batch nodes to generate a recommended processing time and recommended processing temperature range for the target Chinese medicinal material, which is then output as a process feedback optimization strategy. Specifically, in embodiments S53-S55, the core of the graph convolutional neural network (GCN) lies in the message passing mechanism. In the first layer of network iteration, the node representing a certain batch of Polygonum multiflorum identifies the soil pH value, the 32-hour steaming process, the 35℃ logistics fluctuations experienced, and the comprehensive efficacy score of the cured patients. The node collects all the values ​​of its neighboring nodes, adds them to its own initial value, calculates the arithmetic mean, and then generates a new comprehensive value, i.e., the hidden state vector, through the ReLU activation function. After three layers of such transmission and fusion, the features contained in each batch of medicinal materials are no longer a single attribute, but a global high-dimensional feature that integrates where it comes from, how it is processed, how it is transported, and whether its efficacy is good or not. Using a comprehensive efficacy score greater than 0.90 as a hard indicator, all batches of Polygonum multiflorum in the database were sorted in descending order and truncated to extract the top 10% of high-quality batches. The system traced back the processing parameter nodes connected to these high-quality batches in the atlas. Statistical average calculations showed that, compared with the conventional process of cooking parameters of about 32 hours and constant temperature of 100℃, the batches represented by this set, which had the highest efficacy and the fewest cases of abnormal liver function, had an average actual running time parameter that converged to 36.5 hours. Further temperature evolution feature vector analysis revealed that the process curve of this set had a stable cooling phase at 20 hours, which cooled down to 85℃ and was maintained for 2 hours. The system automatically extracted this data and output a process optimization strategy instruction sheet for processing Polygonum multiflorum: it is recommended to extend the total processing time to 36.5 hours and reduce the steam temperature to 85°C for 2 hours at the 20th hour; this parameter combination suggests that appropriately extending the cooking time and combining it with mid-term staged cooling may be more conducive to promoting the conversion and retention balance of effective components such as stilbene glycosides and anthraquinones, and reducing the residue of hepatotoxic components. Therefore, the associated clinical lipid-lowering efficacy is expected to increase by about 15%, and the liver damage risk index is reduced to the lowest range in the sample. The aforementioned instruction sheets are directly pushed to the production control brain of the front-end pharmaceutical companies, thereby realizing a closed loop where clinical efficacy data guides the optimization of source industrial production parameters, changing the problem that traditional Chinese medicine relies solely on experience and lacks quantitative iteration; by constructing a full life cycle knowledge graph and utilizing the feature aggregation capabilities of graph neural networks, reverse attribution analysis from end-efficacy to production processes is achieved; and by precisely outputting process feedback strategies, a closed-loop continuous optimization of Chinese medicine quality is realized, providing technical support for the precise manufacturing of Chinese herbal medicine pieces.

[0026] For further details, please refer to Figure 3 A data management system for the entire lifecycle of traditional Chinese medicine decoction pieces, used to implement a method for managing data throughout the entire lifecycle of traditional Chinese medicine decoction pieces; the system includes: The source tracing and blockchain module is responsible for connecting with agricultural IoT sensors to obtain basic and environmental data, performing resampling and unique thermal coding alignment, generating identity identifiers, and interacting with the consortium blockchain. The multimodal processing evaluation module, which incorporates a one-dimensional convolutional neural network and a multispectral image processing unit, is responsible for extracting color, texture, and temperature features from the processing container and running the processing quality benchmark evaluation network to output the initial quality baseline index. The dynamic environment attenuation calculation module integrates environmental parameter comparison logic and linear attenuation summation calculation unit, and is responsible for receiving warehousing and logistics environment data and updating the dynamic quality index in real time. The clinical efficacy quantification feedback module connects to the hospital's electronic medical record database and includes a named entity recognition unit and a biochemical indicator standardization conversion unit, which is responsible for outputting structured efficacy feedback vectors. The graph feedback optimization decision module is used to construct and store entity nodes and directed edges, run graph convolutional neural networks to mine high-quality batch features, and output specific process feedback optimization strategies. This invention addresses the issue of data tampering by establishing a closed-loop concept for data-driven quality improvement and constructing a trust chain throughout the entire lifecycle using blockchain technology. It also addresses the biodegradability of bioactive substances in traditional Chinese medicine by designing a dynamic quality index model based on environmental dynamics experiments, shifting quality management from static to dynamic. Furthermore, it tackles the challenge of passing on processing experience by utilizing multimodal AI technology to digitize and objectify this experience. Finally, it transforms clinical value into production momentum by mining deep nonlinear correlations between clinical efficacy and production parameters using graph neural networks. This invention not only improves regulatory efficiency but also enables the scientific iteration of traditional Chinese medicine processing techniques through a data feedback mechanism.

[0027] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0028] The preferred embodiments of the invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A method for managing the entire lifecycle data of traditional Chinese medicine decoction pieces, characterized in that, Includes the following steps: S1. Obtain source production environment monitoring data and basic attribute data of the target batch of Chinese medicinal materials, perform time series alignment and outlier removal preprocessing, generate source feature vector, and upload it to the consortium blockchain node after binding it with the identity identifier of the target batch of Chinese medicinal materials. S2. During the processing stage of the target batch of Chinese medicinal materials, the operating time parameters of the processing equipment, the time series data of ambient temperature, and the color gradient characteristics and texture roughness characteristics of the sliced ​​medicinal materials are collected simultaneously. After feature fusion, the data are input into the processing quality benchmark evaluation network to obtain the initial quality baseline index of the target batch of Chinese medicinal materials, and then added to the consortium blockchain. S3. During the storage and logistics of the Chinese herbal medicine pieces, real-time storage environment temperature and humidity data and light intensity data are obtained, and combined with the initial quality baseline index, the cumulative loss value of the effective ingredients is calculated, and then the dynamic quality index is updated and generated in real time. S4. In the clinical application stage of the Chinese herbal medicine, obtain the prescription dispensing record, perform association matching based on the identity identifier, extract the subjective symptom improvement entity and the change range of laboratory test index data from the electronic medical record text data of the patient taking the medicine, and calculate and generate the efficacy feedback vector. S5. Using the identity identifier as a node, and the source feature vector, initial quality baseline index, dynamic quality index, and efficacy feedback vector as associated attribute nodes, construct a knowledge graph of the entire chain of traditional Chinese medicine. By using graph neural networks to perform multi-hop path aggregation and high-dimensional feature mining on the entire knowledge graph of traditional Chinese medicine, the optimal parameter combination of the processing stage corresponding to the optimal interval of the target efficacy feedback vector is found, and the process feedback optimization strategy for the target traditional Chinese medicine is output.

2. The method for managing the entire lifecycle data of traditional Chinese medicine decoction pieces according to claim 1, characterized in that, S1 includes the following steps: S11. Obtain the average daily precipitation data, soil pH data, and sunshine duration data throughout the entire growth cycle of the target Chinese medicinal material, obtain the original species category and harvesting season information of the target Chinese medicinal material, and obtain the initial source dataset. S12. By detecting outliers, the daily average precipitation data in the initial source dataset is smoothly replaced with the noise data in the soil pH data. S13. The smoothed replacement data is resampled and aligned according to the time step, and then spliced ​​with the original species category and harvest season information that have undergone one-hot encoding to generate a source feature vector. The source feature vector is then calculated with the current timestamp using a secure hash algorithm to generate an identity identifier.

3. The method for managing the entire lifecycle data of traditional Chinese medicine decoction pieces according to claim 2, characterized in that, S2 includes the following steps: S21. Capture multispectral image data of the sliced ​​medicinal herbs at preset time intervals, and extract the pixel mean values ​​of the red-green and yellow-blue channels as color gradient features. The contrast and energy values ​​of the multispectral image data in different directions are calculated using the gray-level co-occurrence matrix, and the average value is used as the texture roughness feature. S22. Obtain the time series data of ambient temperature inside the processing container and extract the local temperature fluctuation features to obtain the temperature evolution feature vector; S23. Linearly concatenate the color gradient features, texture roughness features, running time parameters, and temperature evolution feature vectors to construct a multimodal processing feature matrix; S24. Input the multimodal processing feature matrix into the processing quality benchmark evaluation network to obtain the initial quality baseline index.

4. The method for managing the entire lifecycle data of traditional Chinese medicine decoction pieces according to claim 3, characterized in that, The training process of the processing quality benchmark evaluation network in S24 is as follows: S241. Collect multimodal processing feature matrices of historical processing batches as training input samples; perform destructive chemical component extraction and testing on the slices of medicinal materials from historical processing batches to obtain training output labels; S242. Construct an initial processing quality benchmark evaluation network and randomly initialize the connection weights and bias parameters of the fully connected hidden layers. S243. Input the training input samples into the initial processing quality benchmark evaluation network to obtain the predicted quality index; calculate the mean square error between the predicted quality index and the training output label; S244. Calculate the gradient direction of the mean squared error relative to each connection weight and bias parameter using the backpropagation algorithm, and update and adjust the connection weights and bias parameters accordingly. S245. Repeat S244 until the mean square error is less than the convergence threshold. Save the current network parameters to obtain the processing quality benchmark evaluation network.

5. The method for managing the entire lifecycle data of traditional Chinese medicine decoction pieces according to claim 1, characterized in that, S3 includes the following steps: S31. Set standard ideal storage conditions, including standard temperature value, standard relative humidity value, and standard light intensity value; obtain real-time actual temperature value, actual relative humidity value, and actual light intensity value; S32. Calculate the temperature difference between the actual temperature value and the standard temperature value; when the temperature difference is greater than zero, set the temperature attenuation weight to the temperature difference multiplied by the temperature sensitivity coefficient; otherwise, set the temperature attenuation weight to zero. S33. Calculate the humidity difference between the actual relative humidity value and the standard relative humidity value; when the humidity difference is greater than zero, set the humidity attenuation weight to the humidity difference multiplied by the humidity sensitivity coefficient; otherwise, set the humidity attenuation weight to zero. S34. Multiply the actual light intensity value by the light sensitivity coefficient to obtain the light attenuation intensity; add the temperature attenuation weight, humidity attenuation weight, light attenuation intensity and the preset baseline natural attenuation rate to obtain the instantaneous attenuation rate at the current sampling moment; multiply the duration between two adjacent sampling moments by the instantaneous attenuation rate to obtain the single environmental fluctuation loss value. S35. Subtract the sum of all single environmental fluctuation losses during the entire logistics flow stage from the initial quality baseline index to obtain the real-time updated dynamic quality index; when the dynamic quality index is lower than the scrap interception threshold, generate an early warning blocking instruction and broadcast it to the consortium blockchain.

6. The method for managing the entire lifecycle data of traditional Chinese medicine decoction pieces according to claim 1, characterized in that, S4 includes the following steps: S41. Based on the patient number containing the identity identifier in the prescription dispensing record, retrieve the electronic medical record text data and laboratory test indicator data; S42. Use the named entity recognition model to segment and tag the electronic medical record text data, and extract entity words belonging to the symptom description category; by comparing the entity words of the symptom description category before and after medication, count the frequency of symptom disappearance or relief, and map them to a symptom improvement score between zero and one. S43. Extract the target physiological and biochemical indicators corresponding to the target Chinese medicinal material from the laboratory test index data; calculate the percentage decrease or increase of the target physiological and biochemical indicator value after medication relative to the target physiological and biochemical indicator value before medication; convert the percentage into a biochemical score between zero and one according to the medical evaluation standard table. S44. Assign a first diagnostic weight to the symptom improvement score and a second diagnostic weight to the biochemical score; multiply the symptom improvement score by the first diagnostic weight and add the product of the biochemical score multiplied by the second diagnostic weight to obtain the comprehensive efficacy score; combine the symptom improvement score, biochemical score and comprehensive efficacy score to generate a structured efficacy feedback vector.

7. The method for managing the entire lifecycle data of traditional Chinese medicine decoction pieces according to claim 6, characterized in that, The quantification determination rules for the first diagnostic weight and the second diagnostic weight in S44 include the following steps: S441. Compile all medical disputes and complaints of poor efficacy related to the target Chinese medicinal materials in the historical database. S442. Calculate the proportion of complaint records due to unrelieved subjective symptoms to the total number of complaint records, as the initial value of the first diagnostic weight; calculate the proportion of complaint records due to unmet biochemical indicators to the total number of complaint records, as the initial value of the second diagnostic weight. S443. If the target Chinese medicinal material is clearly labeled as having the risk of toxic side effects, then the initial value of the second diagnostic weight will be forcibly increased by a fixed safety penalty value, and the initial value of the first diagnostic weight will be reduced accordingly; otherwise, the initial value will be kept. A weight boundary verification mechanism will be set simultaneously: if the initial value of the second diagnostic weight plus the fixed safety penalty value is greater than or equal to 1, then the initial value of the second diagnostic weight will be directly set to 1 and the initial value of the first diagnostic weight will be set to 0. S444. Normalize the initial values ​​of the first diagnostic weight and the second diagnostic weight in S443 so that their sum equals one, and finally obtain the first diagnostic weight and the second diagnostic weight.

8. The method for managing the entire lifecycle data of traditional Chinese medicine decoction pieces according to claim 1, characterized in that, S5 includes the following steps: S51. Create entity nodes in the graph database. Entity node types include medicinal material batch nodes, source feature nodes, processing parameter nodes, environmental attenuation nodes, and efficacy feedback nodes. S52. Based on the flow path of the identity identifier, establish directed edges between various entity nodes; use the directed edges to connect isolated entity nodes and construct a knowledge graph of the entire chain of traditional Chinese medicine. S53. Use graph convolutional neural networks to process the knowledge graph of the entire chain of traditional Chinese medicine; calculate the arithmetic mean of the feature vector of the entity node itself and the feature vector of the neighboring nodes as the aggregate feature, and perform nonlinear mapping through the activation function to update the hidden state vector of the entity node in the next layer. S54. After multiple iterations of graph convolutional layers, extract the final hidden state vectors of all medicinal herb batch nodes; sort the medicinal herb batch nodes in descending order according to the comprehensive efficacy score in the efficacy feedback nodes, and select a set of high-quality medicinal herb batch nodes. S55. Perform a statistical averaging operation on the running time parameters and temperature evolution feature vectors in the processing parameter nodes associated with the set of high-quality medicinal material batch nodes to generate a recommended processing time and recommended processing temperature range for the target Chinese medicinal material, which is then output as a process feedback optimization strategy.

9. A data management system for the entire lifecycle of traditional Chinese medicine decoction pieces, used to implement the data management method for the entire lifecycle of traditional Chinese medicine decoction pieces as described in any one of claims 1-8; the system comprises: The source tracing and on-chain module acquires basic and environmental data, performs resampling and one-hot encoding alignment, generates identity identifiers, and interacts with the consortium blockchain; The multimodal processing evaluation module is responsible for extracting color, texture and temperature features from the processing container and running the processing quality benchmark evaluation network to output the initial quality baseline index. The dynamic environment attenuation calculation module is responsible for receiving warehousing and logistics environment data and updating the dynamic quality index in real time. The clinical efficacy quantitative feedback module is responsible for outputting structured efficacy feedback vectors; The graph feedback optimization decision module is used to construct and store entity nodes and directed edges, run graph convolutional neural networks to mine high-quality batch features, and output specific process feedback optimization strategies.

10. A data storage medium for the entire life cycle of traditional Chinese medicine decoction pieces, wherein a program is stored thereon; when the program is executed, it is used to implement the data management method for the entire life cycle of traditional Chinese medicine decoction pieces as described in any one of 1-8.

Images