Risk assessment method and system for pesticide residues of traditional Chinese medicinal materials

Abstract

The invention discloses a traditional Chinese medicine pesticide residue-oriented risk assessment method and system, particularly relates to the technical field of traditional Chinese medicine quality control, and is used for solving the problem of inaccurate assessment caused by neglecting the influence of a processing technology on pesticide form transformation and synergistic toxicity of multiple pesticides in the prior art. The method comprises the following steps: acquiring pesticide residue detection data and processing technological parameters; determining a conversion form and a conversion rate by associating a pesticide chemical conversion database; quantifying a toxicity weight based on biological network topology data and a graph convolution model; correcting the actual residual quantity by combining the chaos theory and the recurrent neural network with environmental disturbance; dynamically optimizing the weight aiming at the synergistic effect of multiple pesticide residues; generating cross-batch risk early warning in combination with historical data; through multi-dimensional dynamic coupling analysis, the pesticide residue toxicity evaluation precision is remarkably improved, and the safety risk of pesticide residues of traditional Chinese medicinal materials is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of traditional Chinese medicine quality control. More specifically, the present invention relates to a risk assessment method and system for pesticide residues in Chinese medicinal materials. Background Art

[0002] In the production process of Chinese medicinal materials, processing and preparation are often required to enhance the medicinal effects or reduce toxicity, and pesticide residues are one of the key factors affecting their safety; during the processing and preparation of Chinese medicinal materials, the chemical forms of pesticide residues may change. Currently, the risk assessment of pesticide residues in Chinese medicinal materials mainly relies on detecting the total residue of pesticides and directly comparing the detection data of the total residue of pesticides with a preset threshold to determine whether the medicinal materials meet the safety standards. However, the potential impact of the processing and preparation technology on the toxicity of pesticide residues is not fully considered, and the impact of the processing and preparation technology on the toxicity of residual pesticides is not incorporated into the assessment logic, resulting in a possible deviation between the assessment result and the actual safety of the medicinal materials.

[0003] In the prior art, the determination of pesticide residue risk is only based on the detection results of the total residue, without distinguishing the toxicity differences between the original forms of residual pesticides and the transformed forms that may be generated after processing and preparation. This leads to some Chinese medicinal materials with significantly reduced actual toxicity after processing and preparation being misjudged as high-risk due to excessive total residue, which not only affects the reasonable utilization of the medicinal materials but also may cause inaccurate assessment results and unnecessary waste of resources. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a risk assessment method and system for pesticide residues in Chinese medicinal materials to solve the problems raised in the above background art.

[0005] To achieve the above object, the present invention provides the following technical solutions: A risk assessment method for pesticide residues in Chinese medicinal materials, comprising the following steps: S1. Obtain pesticide residue detection data including the pesticide name and total residue in the Chinese medicinal materials to be evaluated; extract processing and preparation process parameters including processing temperature and duration; S2. According to the pesticide name and processing and preparation process parameters, associate with a pre-stored pesticide chemical transformation database to determine the transformed form of the pesticide under the processing and preparation process and the corresponding conversion rate; S3. Calculate the node influence of the transformed form through a graph convolutional model based on the pre-stored biological network topology data to generate a toxicity weight value; S4. Based on the micro-perturbation data of the processing and preparation environment collected by the sensor, quantify the influence of the non-linear perturbation on the conversion rate through chaos theory, and dynamically correct the actual residue amount in combination with a recurrent neural network; S5. If there are two or more types of pesticide residues, dynamically adjust the toxicity weight value according to the synergistic effect relationship associated with the action mechanism type; S6. Generate a risk assessment result based on the actual residue amount and the adjusted toxicity weight value, and perform cross-batch risk prediction in combination with historical batch data.

[0006] In a preferred embodiment, S1 includes: S1a. Obtain pesticide residue detection data of the Chinese herbal medicine planting area through Internet of Things sensors, and the pesticide residue detection data includes the pesticide name and the corresponding total residue amount; S1b. Match the pre-stored pesticide detection data cleaning rule library according to the pesticide name, and perform outlier removal and standardization processing on the pesticide residue detection data; S1c. Extract the time series data of the processing temperature and duration from the control system of the processing equipment to generate processing technology parameters; S1d. Associate the Chinese herbal medicine production batch code according to the pesticide name and the processing technology parameters, and establish a batch data index.

[0007] In a preferred embodiment, S2 includes: S2a. Match the reaction path template in the pesticide chemical conversion database according to the pesticide name and the processing temperature, and screen the candidate conversion forms corresponding to the processing technology; S2b. Calculate the conversion rate of each conversion form through a conditional probability model according to the duration and the kinetic parameters of the candidate conversion form; S2c. Based on the processing stage label, associate the experimental verification data in the pesticide chemical conversion database, and perform confidence correction on the conversion rate; S2d. Store the corrected conversion form and conversion rate in the pesticide chemical conversion database according to the batch data index.

[0008] In a preferred embodiment, S3 includes: S3a. Construct a metabolism-toxicity association network, with nodes being conversion forms and human metabolites, and edges being toxicity transmission paths; S3b. Extract node feature vectors based on biological network topology data, including chemical descriptors of conversion forms and toxicity values of metabolites; S3c. Aggregate neighborhood node features through multiple convolutional layers of a graph convolutional model to generate node embedding representations; S3d. Calculate the topological centrality score based on the node embedding representation, and generate a toxicity weight value in combination with the toxicity value; S3e. Associatively store the toxicity weight value in the pesticide chemical conversion database according to the batch data index.

[0009] In a preferred embodiment, S4 includes: S4a. Collect the time - series data of temperature, humidity, and air - flow velocity in the collection and processing equipment as the micro - disturbance data of the processing environment, and store them in an associated manner according to the batch - data index. S4b. Analyze the chaotic characteristics of the micro - disturbance data in stages based on the Lyapunov exponent. Quantify the instantaneous non - linear disturbance intensity in the first stage, and extract the disturbance cumulative effect coefficient of historical batches in the second stage. S4c. Through a recurrent neural network, fuse the instantaneous non - linear disturbance intensity and the disturbance cumulative effect coefficient to generate a dynamic correction factor, and synchronously associate the conversion rate in step S2 with the toxicity weight value in step S3. S4d. Bidirectionally correct the total residue amount based on the dynamic correction factor to generate the actual residue amount. S4e. Store the actual residue amount in the pesticide chemical conversion database in an associated manner according to the batch - data index, and mark the type of correction path.

[0010] In a preferred embodiment, the bidirectional correction includes: forward correction amplifies the residue amount of high - toxicity forms based on the dynamic factor, and reverse correction suppresses the conversion rate error with low confidence.

[0011] In a preferred embodiment, S5 includes: S5a. If there are two or more kinds of pesticide residues, determine the type of action mechanism of the pesticides according to the pesticide names. S5b. Based on the type of action mechanism, match the pre - stored synergistic effect rule library to screen the synergistic action modes of pesticide combinations. S5c. Extract the metabolic - path cross - nodes according to the metabolic - toxicity association network in step S3, and quantify the competitive inhibition intensity of the pesticide combination on the cross - nodes. S5d. Based on the competitive inhibition intensity and the topological centrality score in step S3, construct a dynamic attenuation model of the toxicity propagation path to generate a comprehensive toxicity correction coefficient. S5e. Apply the comprehensive toxicity correction coefficient to the toxicity weight value in step S3 to generate an adjusted comprehensive toxicity weight, and store it in the pesticide chemical conversion database in an associated manner according to the batch - data index.

[0012] In a preferred embodiment, the synergistic action modes include competitive inhibition of metabolic paths and superposition of target - organ toxicity.

[0013] In a preferred embodiment, S6 includes: S6a. Based on the actual residue amount and the adjusted comprehensive toxicity weight in step S5e, construct a time - series data set of the risk index. S6b. Combine the time - series data set of historical batch data, and extract the cross - batch risk trend characteristics through grey relational analysis. S6c. Generate a dynamic risk warning model based on the node influence of cross-batch risk trend characteristics and transformation forms; S6d. Associatively store the output results of the dynamic risk warning model according to batch data indexes to generate a cross-batch risk prediction report; S6e. Update the parameters of the dynamic risk warning model according to real-time batch data and adjust the risk warning level threshold.

[0014] On the other hand, the present invention provides a risk assessment system for pesticide residues in Chinese medicinal materials, including the following modules: Data acquisition module: Obtain pesticide residue detection data including the pesticide name and total residue amount in the Chinese medicinal materials to be evaluated; Extract the processing technology parameters including processing temperature and duration; Transformation analysis module: Associate with the pre-stored pesticide chemical transformation database according to the pesticide name and processing technology parameters to determine the transformation form and corresponding conversion rate of the pesticide under the processing technology; Weight generation module: Calculate the node influence of the transformation form through a graph convolution model based on the pre-stored biological network topology data to generate a toxicity weight value; Dynamic correction module: Based on the micro-perturbation data of the processing environment collected by the sensor, quantify the influence of non-linear perturbation on the conversion rate through chaos theory, and dynamically correct the actual residue amount in combination with a recurrent neural network; Cooperative adjustment module: If there are two or more pesticide residues, dynamically adjust the toxicity weight value according to the cooperative effect relationship associated with the action mechanism type; Prediction and evaluation module: Generate a risk assessment result according to the actual residue amount and the adjusted toxicity weight value, and perform cross-batch risk prediction in combination with historical batch data.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. Through multi-dimensional dynamic coupling analysis, the accuracy and adaptability of the risk assessment of pesticide residues in Chinese medicinal materials are significantly improved. By associating the processing technology parameters with the pesticide chemical transformation database, the influence of pesticide form transformation on toxicity during the processing process is directly quantified, solving the problem of misjudgment of toxicity caused by traditional methods ignoring the processing technology; Through conversion rate calculation and dynamic adjustment of toxicity weight, the actual toxicity risk is accurately identified, avoiding misjudgment caused by simply relying on the total residue amount; At the same time, metabolic-toxicity network topology analysis is introduced, and the graph convolution model is combined to quantify the propagation influence of the transformation form in the biological network, making the toxicity assessment more in line with the actual biological effect.

[0016] 2. By integrating environmental micro-perturbation data through chaos theory and recurrent neural networks, dynamically correcting the pesticide residue levels and toxicity weights, it solves the problem that traditional static models cannot adapt to the dynamic changes of the processing environment; through real-time perturbation analysis and dynamic correction, it significantly improves the robustness of residue prediction; for the synergistic effects of multi-pesticide residues, based on the coupling analysis of the action mechanism type and the cross-interference of metabolic network nodes, it realizes the adaptive adjustment of toxicity weights, avoids the error accumulation of traditional linear superposition models, and combines the historical data trend and real-time parameter update for cross-batch risk prediction to form a risk assessment-warning-feedback closed loop, providing continuously optimized decision-making support for the safety control of medicinal materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flowchart of a risk assessment method for pesticide residues in Chinese medicinal materials according to the present invention; Figure 2 It is a schematic structural diagram of a risk assessment system for pesticide residues in Chinese medicinal materials according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0019] Embodiment 1: Figure 1 A risk assessment method for pesticide residues in Chinese medicinal materials according to the present invention is given, including the following steps: S1. Obtain pesticide residue detection data including the pesticide name and total residue amount in the Chinese medicinal materials to be evaluated; extract the processing technology parameters including the processing temperature and duration; S2. According to the pesticide name and the processing technology parameters, associate with the pre-stored pesticide chemical conversion database to determine the conversion form and corresponding conversion rate of the pesticide under the processing technology; S3. Calculate the node influence of the conversion form through a graph convolutional model based on the pre-stored biological network topology data to generate a toxicity weight value; S4. Based on the micro-perturbation data of the processing environment collected by the sensor, quantify the influence of non-linear perturbation on the conversion rate through chaos theory, and dynamically correct the actual residue amount in combination with a recurrent neural network; S5. If there are two or more pesticide residues, dynamically adjust the toxicity weight value according to the synergistic effect relationship associated with the action mechanism type; S6. Generate a risk assessment result according to the actual residue amount and the adjusted toxicity weight value, and perform cross-batch risk prediction in combination with historical batch data.

[0020] In step S1a, high-precision Internet of Things sensors deployed at key nodes in the traditional Chinese medicine planting area are used to collect pesticide residue detection data in real time. The Internet of Things sensors include gas chromatography sensors, mass spectrometry sensors, and hyperspectral imaging sensors. Among them, the gas chromatography sensors are used to detect the chemical characteristic peaks corresponding to the pesticide names, the mass spectrometry sensors are used to measure the total pesticide residue, and the hyperspectral imaging sensors are used to assist in locating the hot spots of pesticide residue distribution. The pesticide residue detection data is uploaded to the central data processing platform through an encrypted transmission protocol. The pesticide residue detection data includes the pesticide name, the total residue, and the corresponding sampling timestamp. The pesticide name is matched to a standardized name through the international pesticide coding library, and the total residue is stored in milligrams per kilogram.

[0021] In step S1b, the pre-stored pesticide detection data cleaning rule library sets different outlier judgment rules according to pesticide categories. For example, the allowable fluctuation range of the detection value of organophosphorus pesticides is ±5% when the temperature is higher than 30 °C, and the detection value of pyrethroid pesticides needs to be logarithmically transformed when the humidity exceeds 70%. The cleaning rule library is generated through training with historical data, including the following specific operations: perform time series analysis on the total residue in the pesticide residue detection data to identify outlier data points that exceed the preset fluctuation threshold. For example, if the total residue suddenly increases by 50% at adjacent time points during a certain detection, it is marked as an outlier. The outlier data points are corrected using the sliding window mean interpolation method, and the window size is set according to the pesticide degradation half-life. For example, pesticides with a half-life less than 7 days use a 3-hour window, and pesticides with a half-life greater than 30 days use a 24-hour window. The standardization process includes unifying the dimensions of the data collected by different sensors. For example, converting the ion intensity value output by the mass spectrometry sensor to the same concentration unit as the gas chromatography sensor.

[0022] In step S1c, the processing equipment control system is an intelligent temperature control device with a data interface, including a PLC control system or an Internet of Things-based distributed temperature controller. The time series data of the processing temperature and duration are extracted from the device control log, and the data format includes CSV, JSON, or time series database records. After extraction, the temperature data is smoothed, for example, using the exponentially weighted moving average method to eliminate the device temperature measurement noise, and the duration data is converted into an accumulated value in minutes. The processing temperature and duration are stored in segments according to the processing stage. For example, separating the temperature data in the frying stage from the temperature data in the steaming and boiling stage and associating the corresponding process stage labels.

[0023] In step S1d, the production batch code of traditional Chinese medicinal materials consists of the year, month, and day, the production area code, and the production line number. For example, 20231001 - GZB - 003 represents the batch produced by production line No. 3 in the Guizhou production area on October 1, 2023; the association operation includes establishing a mapping relationship between the pesticide name, processing temperature, and duration according to the batch code. For example, creating a batch data index table in a relational database, and the index fields include batch code, pesticide name, temperature range, and duration interval; the index table is associated with the pesticide chemical transformation database through a foreign key. For example, the batch code is used as the primary key to link to the transformation form table in the subsequent steps to ensure traceability of cross - step data; after the index is generated, aggregate multiple groups of pesticide residue detection data under the same batch code. For example, taking the arithmetic mean of the total residue data at different sampling points of the same batch to avoid duplicate calculations.

[0024] In step S2a, the reaction path template in the pesticide chemical transformation database is constructed based on the predefined correspondence between pesticide categories and temperature ranges. The pesticide categories are classified according to the chemical structure of the pesticide name. For example, pesticides containing an organophosphorus group are classified as organophosphorus pesticides, and pesticides containing a pyrethroid group are classified as pyrethroid pesticides; the temperature ranges are divided according to the common temperature ranges of the processing techniques. For example, the temperature range for the stir - frying process is 100 degrees Celsius to 150 degrees Celsius, and the temperature range for the steaming and boiling process is 80 degrees Celsius to 120 degrees Celsius; the matching operation includes querying the pesticide category according to the pesticide name and screening the corresponding reaction path template by combining the interval where the processing temperature is located; the reaction path template contains the main reaction path and the side reaction path. For example, in the hydrolysis reaction path template of organophosphorus pesticides in the 100 - degree Celsius to 150 - degree Celsius interval, the main reaction generates phosphate ester compounds, and the side reaction generates thiophosphate ester compounds; the candidate transformation forms are extracted from the product list of the template. For example, when the pesticide name is "chlorpyrifos" and the processing temperature is 120 degrees Celsius, the candidate transformation forms include phosphate ester A, phosphate ester B, and the unreacted chlorpyrifos original drug.

[0025] In step S2b, the kinetic parameters include the reaction rate constant and the activation energy. Among them, the reaction rate constant is determined through kinetic experiments simulating the processing technique in the laboratory. For example, setting a 120 - degree Celsius environment in an incubator, sampling once every 30 minutes, detecting the change in chlorpyrifos concentration by mass spectrometry, and calculating the hydrolysis rate constant to be 0.05h -1The activation energy parameter is obtained by differential scanning calorimetry. For example, the chlorpyrifos sample is placed in a differential scanning calorimeter, and the thermal decomposition curve is tested at different heating rates. The activation energy obtained by fitting is 50 kJ / mol. The conditional probability model is constructed based on the Arrhenius equation. The specific logic is as follows: The rate constant corrected by temperature is calculated according to the processing temperature and the activation energy. The formula is that the rate constant is equal to the pre-exponential factor multiplied by the negative ratio of the activation energy to the product of the gas constant and the absolute temperature in the natural exponential function. The reaction progress is calculated according to the product of the duration and the corrected rate constant. For example, the reaction progress with a duration of 2 hours is 0.05 h -1 ×2 h = 0.1. The conversion rates of each candidate transformation form are allocated according to the reaction progress and the molar ratio of the products. For example, the molar ratio of phosphate A, phosphate B, and the unreacted original drug is 6:3:1, so their conversion rates are 60%, 30%, and 10% respectively. Finally, normalization is performed to ensure that the sum of all conversion rates is 100%.

[0026] In step S2c, the experimental verification data is from the test results of laboratory-simulated actual processing technology. For example, under the conditions of the same processing temperature of 120 °C and a duration of 2 hours, the peak area of phosphate A is separated and detected by high-performance liquid chromatography, and the actual conversion rate is calculated to be 58%, with a 2% deviation from the calculated value of 60% in step S2b. The confidence correction rule is set according to the deviation range, and the deviation range is determined by the absolute difference between the calculated value and the experimental value. For example, when the deviation is less than 5%, it is a high confidence level; when the deviation is 5% to 10%, it is a medium confidence level; when the deviation is greater than 10%, the manual review process is triggered. The correction operation includes taking the weighted average of the calculated conversion rate and the experimental value as the final conversion rate, and the weights are set according to the confidence level. For example, when the confidence level is high, the weight of the experimental value is 0.8 and the weight of the calculated value is 0.2; when the confidence level is medium, the weight of the experimental value is 0.5 and the weight of the calculated value is 0.5. The corrected conversion rate is reserved to two decimal places. For example, the corrected conversion rate of phosphate A is 59.2%, and it is associated with the experimental verification data and stored in the pesticide chemical transformation database.

[0027] In step S2d, the batch data index is composed of the batch code generated in step S1d and the pesticide name. For example, the combination of batch code 20231001 - GZB - 003 and pesticide name "chlorpyrifos" is 20231001 - GZB - 003 - chlorpyrifos; The storage operation includes creating two data tables in the pesticide chemical transformation database: 1) The transformation form table, whose fields include batch data index, candidate transformation form name, chemical formula and CAS number. For example, the chemical formula of phosphate A corresponding to the batch data index 20231001 - GZB - 003 - chlorpyrifos is C10H12O4PS; 2) The conversion rate table, whose fields include batch data index, candidate transformation form name, calculated conversion rate, corrected conversion rate, confidence level and update timestamp. For example, the corrected conversion rate of phosphate A is 59.2% and the confidence level is high; The data tables establish a foreign key association with the batch data index table in step S1d through the batch data index. The foreign key association is implemented through SQL statements of a relational database. For example, execute "ALTER TABLE conversion rate table ADD FOREIGN KEY (batch data index) REFERENCES batch data index table(batch data index)"; When storing, the multiple groups of conversion rate data under the same batch data index are sorted in ascending order of timestamp to form a time - series data set for cross - batch trend analysis in subsequent steps.

[0028] It should be noted that the pre - stored pesticide chemical transformation database is constructed by integrating laboratory simulation processing data, literature - disclosed data and public toxicology databases (such as PubChem, KEGG), and specifically includes the following steps: 1) Determine the transformation forms of pesticides at different temperatures, humidities and durations through controlled - environment experiments. For example, organophosphorus pesticides generate phosphate - type transformation products under the frying process at 120°C, and the conversion rate is quantified by high - performance liquid chromatography (HPLC). The data is stored according to the pesticide name - process parameter - transformation form triple; 2) Integrate the verified pesticide transformation path data in the literature. For example, the oxidation metabolites and their half - lives of pyrethroid pesticides under the steaming process at 80°C; 3) Associate the chemical structure information and toxicity data in the public database. For example, map the CAS number of the transformation product to the ToxCast toxicity value; 4) Remove conflicting data (such as outliers with a conversion rate difference > 20% under the same conditions) through data cleaning and standardize and store them as relational database tables. The table structure includes pesticide name, process parameters (temperature, time), transformation form, conversion rate and data source label.

[0029] In step S3a, the metabolism-toxicity association network is constructed based on pre-stored biological network topology data, which is derived from the integrated data of the KEGG metabolic pathway database and the ToxCast toxicity database; the nodes include transformation forms and human metabolites. For example, the pesticide transformation form "phosphate A" is used as node A, and its metabolite "hydroxy phosphate" is used as node B; the edges represent the toxicity propagation paths. For example, the edge between node A and node B is defined as the transformation path of phosphate A metabolized into hydroxy phosphate, and the edge between node B and the toxicity target organ "liver cells" is defined as the toxicity action path of hydroxy phosphate on the liver; when constructing the network, the relationship between nodes and edges is stored through the graph database Neo4j. The node attributes include chemical formula and toxicity level, and the edge attributes include reaction type and toxicity action intensity. The reaction types are divided into hydrolysis, oxidation, and reduction. The toxicity action intensity is divided into high (<10 μM), medium (10 - 100 μM), and low (>100 μM) according to the IC50 value.

[0030] In step S3b, the extraction of node feature vectors is performed based on the biological network topology data. The chemical descriptors of the transformation forms are calculated by the open-source cheminformatics toolkit RDKit. The chemical descriptors include molecular weight, lipophilicity-hydrophilicity partition coefficient LogP, and the number of hydrogen bond donors. For example, the molecular weight of phosphate A is 246.2 g / mol, the lipophilicity-hydrophilicity partition coefficient LogP is 2.3, and the number of hydrogen bond donors is 2; the toxicity values of the metabolites are derived from the in vitro cytotoxicity experimental data of the ToxCast database. For example, the IC50 value of hydroxy phosphate is 10 μM; the feature vectors are generated by concatenating the standardized values of the chemical descriptors and the toxicity values. For example, the chemical descriptor vector of phosphate A is [246.2, 2.3, 2], the toxicity value vector is

[10] , and the concatenated feature vector is [246.2, 2.3, 2, 10].

[0031] In step S3c, the multi-layer convolutional layers of the graph convolutional model are used to aggregate the neighborhood node features. The first-layer convolutional operation multiplies the node feature vectors by the adjacency matrix to generate the primary embedding representation. The adjacency matrix is defined according to the connection relationship between nodes. For example, if there is an edge between phosphate A and hydroxy phosphate, the corresponding matrix position value is 1, otherwise it is 0; the second-layer convolutional operation performs a non-linear transformation on the primary embedding representation, and the activation function uses the ReLU function to generate the final node embedding representation with a dimension of 64. For example, the embedding representation of phosphate A is a floating-point vector with a length of 64.

[0032] In step S3d, the topological centrality score is calculated by the PageRank algorithm. The algorithm parameters are set conventionally according to the literature. The damping factor is set to 0.85, and the number of iterations is 100 times. For example, the PageRank score of phosphate A is 0.032, and the PageRank score of hydroxyphosphate is 0.045. The generation of the toxicity weight value is achieved by multiplying the topological centrality score by the toxicity value. For example, the toxicity weight value of hydroxyphosphate is 0.045 × 10 = 0.45. The final toxicity weight value is mapped to the interval from 0 to 1 through max-min normalization. For example, the original weight value 0.32 of phosphate A is normalized to 0.72.

[0033] In step S3e, the batch data index is composed of the batch code generated in step S1d and the pesticide name. For example, the combination of batch code 20231001 - GZB - 003 and pesticide name "chlorpyrifos" is 20231001 - GZB - 003 - chlorpyrifos. The storage operation includes creating a toxicity weight table in the pesticide chemical transformation database. The fields of the toxicity weight table include batch data index, transformation form name, toxicity weight value, and update timestamp. For example, the toxicity weight value of phosphate A corresponding to the batch data index 20231001 - GZB - 003 - chlorpyrifos is 0.72. The toxicity weight table is associated with the conversion rate table in step S2d through the batch data index. The association operation is achieved by the SQL statement "SELECT * FROM toxicity weight table JOIN conversion rate table ON toxicity weight table.batch data index = conversion rate table.batch data index". When storing, the toxicity weight values under the same batch data index are sorted in ascending order of timestamp to form a time series dataset.

[0034] In step S4a, the time-series data of the temperature, humidity, and air velocity inside the processing equipment are collected in real time by high-precision Internet of Things sensors deployed inside the equipment. The Internet of Things sensors include a thermocouple temperature sensor, a capacitive humidity sensor, and an ultrasonic air velocity sensor. The thermocouple temperature sensor is installed in the core area of the heat source of the processing equipment, for example, fixedly installed at the center point of the bottom of the frying pan, with a measurement range of 0 degrees Celsius to 300 degrees Celsius and an accuracy of ±0.5 degrees Celsius. The capacitive humidity sensors are distributed at multiple monitoring points in the medicinal material stacking area, for example, one sensor is set at the upper, middle, and lower layers of the medicinal material pile, with a measurement range of 0% to 100% relative humidity and an accuracy of ±2%. The ultrasonic air velocity sensor is fixed at the equipment ventilation opening, with a measurement range of 0 m / s to 10 m / s and an accuracy of ±0.1 m / s. The collected micro-disturbance data are stored by associating with the batch data index. The batch data index is composed of the batch code generated in step S1d and the pesticide name. For example, the combination of batch code 20231001-GZB-003 and pesticide name "chlorpyrifos" is 20231001-GZB-003-chlorpyrifos. The storage format is a time-series database record. Each record contains a timestamp, a temperature value (unit: degree Celsius), a humidity value (unit: percentage), and an air velocity value (unit: m / s). The data are transmitted to the central database through an encryption protocol and stored by partitioning according to the batch data index.

[0035] In step S4b, the phased analysis operation of the Lyapunov exponent is divided into the initial stage of quantifying the instantaneous disturbance intensity and the secondary stage of extracting the historical cumulative effect. The initial stage analysis is aimed at the current batch of micro-disturbance data. The Lyapunov exponent of the time series is calculated by the sliding window method. The window size is 10 minutes, and the step size is 1 minute. The window size is set according to the typical reaction time of the processing technology. For example, in the frying process, the reaction time of pesticide conversion is about 10 minutes, so the window size is set to 10 minutes. When calculating, the Wolf algorithm is used to estimate the maximum Lyapunov exponent by tracking the divergence rate of adjacent orbits. For example, in the micro-disturbance data with a temperature fluctuation variance of 2.5, the calculated Lyapunov exponent is 0.35, and it is determined that the instantaneous non-linear disturbance intensity is high. The secondary stage analysis extracts the cumulative disturbance effect of the same origin in the historical batch data. The disturbance intensity decay coefficient is calculated by the exponentially weighted moving average method. The number of historical batch data is set to the most recent 5 batches according to data availability. The decay coefficient is set according to the pesticide stability. For example, the decay coefficient of pesticides with a half-life less than 7 days is set to 0.8, and that of pesticides with a half-life greater than 30 days is set to 0.5. The cumulative effect coefficient calculation formula is the weighted average of the historical batch disturbance intensity multiplied by the decay coefficient. For example, if the weighted average of the temperature fluctuation variances of the past 5 batches is 2.0 and the decay coefficient is 0.8, then the cumulative effect coefficient is 2.0×0.8 = 1.6.

[0036] In step S4c, the recurrent neural network adopts a long short-term memory network (LSTM) structure. The input layer receives the concatenated vector of the instantaneous non-linear perturbation intensity and the cumulative effect coefficient. For example, the instantaneous perturbation intensity of 0.35 and the cumulative effect coefficient of 1.6 are concatenated as [0.35, 1.6]. The hidden layer is set with two LSTM units, and the time step is aligned with the time stamp of the micro-perturbation data. For example, there is one time step per minute, and a processing technology lasting for 2 hours corresponds to 120 time steps. The output layer generates a dynamic correction factor through the Sigmoid activation function. The dynamic correction factor is a scalar value between 0 and 1. For example, the output value is 0.72. The synchronous correlation operation includes multiplying the dynamic correction factor by the conversion rate in step S2 and the toxicity weight value in step S3. For example, when the conversion rate is 60% and the toxicity weight value is 0.72, the correction factor 0.72 acts on the product of the two (60%×0.72 = 43.2%) to generate a comprehensive correction coefficient.

[0037] In step S4d, the two-way correction includes forward correction and reverse correction. The forward correction is for the residual amount of the highly toxic form. It amplifies the residual amount of the form with a toxicity weight value higher than the preset threshold (for example, ≥0.5) through the dynamic correction factor. For example, the residual amount of phosphate A with a toxicity weight value of 0.72 is corrected from 100 mg / kg to 100×0.72 = 72 mg / kg. The reverse correction is for the conversion rate error with low confidence. It suppresses the conversion rate deviation with a confidence level of medium in step S2 through the dynamic correction factor. For example, when the confidence level is medium, the conversion rate deviation of 5% is reduced to 5%×0.72 = 3.6% according to the correction factor of 0.72. The formula for the actual residual amount is the total residual amount multiplied by the corrected conversion rate. For example, the total residual amount of 200 mg / kg is corrected to 200×43.2% = 86.4 mg / kg. During the correction process, a balance check is performed on the amplification and suppression operations. For example, if the forward correction causes the residual amount to exceed the safety threshold, an artificial review process is triggered.

[0038] In step S4e, the actual residue amount is stored in the pesticide chemical conversion database according to the batch data index. The storage table structure includes fields such as batch data index, pesticide name, actual residue amount, correction path type, and update timestamp. The correction path type is marked according to the two-way correction rule. For example, the forward correction is marked as "high toxicity amplification", and the reverse correction is marked as "low confidence suppression". The association operation connects the actual residue amount table with the conversion rate table in step S2d and the toxicity weight table in step S3e through SQL statements. For example, execute "SELECT * FROM actual residue amount table JOIN conversion rate table ON actual residue amount table.batch data index = conversion rate table.batch data index JOIN toxicity weight table ON actual residue amount table.batch data index = toxicity weight table.batch data index". When storing, a time-series data set is generated by sorting according to the timestamp. For example, the correction data at 14:30:00 on October 1, 2023 is arranged in order with the data at the previous and subsequent time points to form a continuous trend curve for cross-batch risk prediction in subsequent steps.

[0039] In step S5a, the system matches the pre-stored pesticide toxicology database according to the pesticide name to determine the type of action mechanism of the pesticide. The pesticide toxicology database is constructed based on the pesticide classification standard of the World Health Organization (WHO). The types of action mechanisms are divided into neurotoxicity, metabolic toxicity, and endocrine disruption. For example, the pesticide name "chlorpyrifos" corresponds to neurotoxicity, the pesticide name "cypermethrin" corresponds to metabolic toxicity, and the pesticide name "DDT" corresponds to endocrine disruption. The determination operation includes inputting the pesticide name through the database query interface and returning the corresponding action mechanism type label. For example, after inputting "chlorpyrifos", "neurotoxicity" label is returned, and after inputting "cypermethrin", "metabolic toxicity" label is returned.

[0040] In step S5b, the system combines and queries the pre-stored synergistic effect rule library based on the type of action mechanism to screen the synergistic action mode of the pesticide combination. The synergistic effect rule library is constructed by integrating the research literature on the interaction of known pesticides. The synergistic action modes include competitive inhibition of metabolic pathways and superposition of target organ toxicity. For example, when the types of action mechanisms of two pesticides are neurotoxicity and metabolic toxicity respectively, the "competitive inhibition of metabolic pathways" mode is matched, which is defined as the two pesticides competing for the active site of the same metabolic enzyme. When both pesticides are of metabolic toxicity, the "superposition of target organ toxicity" mode is matched, which is defined as the cumulative toxic effects of the two pesticides on the same target organ (such as the liver). The screening operation includes using the type of action mechanism of the pesticide combination as the joint key to query the rule library. For example, inputting "neurotoxicity + metabolic toxicity" returns the "competitive inhibition of metabolic pathways" mode.

[0041] In step S5c, the system extracts the metabolic pathway cross-nodes according to the metabolic-toxicity association network constructed in step S3a. The extraction method includes traversing the shared metabolic pathway nodes in the network. For example, the metabolic pathway of pesticide A is node A → node B → node C, and the metabolic pathway of pesticide B is node D → node B → node E, then the cross-node is node B. The competitive inhibition intensity is calculated through the enzyme kinetic parameters of the cross-node. The enzyme kinetic parameters include the inhibition constant Ki value. For example, the Ki value of pesticide A for node B is 10 μM, and the Ki value of pesticide B for node B is 20 μM. The formula for calculating the competitive inhibition intensity is 1 / Ki_A + 1 / Ki_B, that is, 1 / 10 + 1 / 20 = 0.15. The calculation result is normalized to the range of 0 to 1 according to a ratio. For example, the maximum competitive inhibition intensity of 0.15 corresponds to the normalized value of 1.0, and the minimum intensity of 0 corresponds to 0.

[0042] In step S5d, the system constructs a dynamic decay model of the toxicity propagation path based on the competitive inhibition intensity and the topological centrality score calculated in step S3d. The topological centrality score is derived from the PageRank value in step S3d. For example, the PageRank value of cross-node B is 0.045. The formula for calculating the decay coefficient of the dynamic decay model is α = 1 / (1 + PR(v)), where PR(v) is the PageRank value of node v. For example, the decay coefficient of node B is 1 / (1 + 0.045) ≈ 0.957. The comprehensive toxicity correction coefficient is calculated by the product of the competitive inhibition intensity and the decay coefficient. For example, if the competitive inhibition intensity is 0.8 and the decay coefficient is 0.957, then the correction coefficient is 0.8 × 0.957 ≈ 0.766. The correction coefficient is mapped to the range of 0 to 1 through the Sigmoid function. For example, when the input is 0.766, the output is 0.68, and the mapping parameters are determined by training with historical co-toxicity experiment data.

[0043] In step S5e, the system applies the comprehensive toxicity correction coefficient to the toxicity weight value in step S3 to generate an adjusted comprehensive toxicity weight. For example, if the original toxicity weight value is 0.72 and the correction coefficient is 0.68, the adjusted weight is 0.72 × 0.68 ≈ 0.49. The storage operation includes creating a comprehensive toxicity weight table in the pesticide chemical transformation database, with fields including batch data index, pesticide combination name, adjusted comprehensive toxicity weight, and update timestamp. For example, the adjusted weight corresponding to the batch data index 20231001 - GZB - 003 - chlorpyrifos - cypermethrin is 0.49. The association operation connects the comprehensive toxicity weight table with the toxicity weight table in step S3e and the actual residue table in step S4e through SQL statements. For example, execute "SELECT * FROM comprehensive toxicity weight table INNER JOIN toxicity weight table ON comprehensive toxicity weight table.batch data index = toxicity weight table.batch data index INNER JOIN actual residue table ON comprehensive toxicity weight table.batch data index = actual residue table.batch data index". When storing, a time - series dataset is generated in ascending order of timestamps for cross - batch trend analysis in subsequent steps.

[0044] The determination of "two or more pesticide residues" in step S5 is based on the pesticide residue detection data in step S1. The pesticide residue detection data includes the pesticide name and the corresponding total residue amount. The determination operation includes traversing the number of pesticide names in the detection data. If the number ≥ 2, the synergistic effect correction process is triggered. For example, when the pesticide names in the detection data are "chlorpyrifos" and "cypermethrin", the number of pesticides is 2, and it is determined that there are two pesticide residues. The pesticide name and the total residue amount are from the standardized processing results in step S1a, and the data storage format is a key - value pair of "pesticide name - total residue amount", such as "chlorpyrifos: 100mg / kg, cypermethrin: 80mg / kg".

[0045] In step S5, by combining the pesticide action mechanism type, metabolic pathway cross - nodes, and topological centrality scores, a dynamic decay model is constructed to correct the toxicity weight value, solving the evaluation deviation problem caused by traditional methods ignoring the multi - pesticide synergistic effect. Compared with the prior art that only relies on a single action mechanism or simple toxicity superposition, step S5 introduces two - dimensional analysis of metabolic pathway competitive inhibition and topological decay, significantly improving the quantification accuracy of synergistic toxicity. For example, the fusion of the enzyme competition intensity and network topology influence of metabolic pathway cross - nodes can accurately identify high - transmission - risk toxicity forms, avoiding underestimation of high - toxicity residues. The dynamic decay model is trained with historical data to adaptively correct the weight assignment, reducing the error of low - confidence conversion rates. Step S5 dynamically associates the topological characteristics of the metabolic network with the synergistic effect, realizing a qualitative change in risk assessment from "single - linear" to "multi - dimensional coupling", which fits the complex toxicity interaction characteristics of traditional Chinese medicine processing technology.

[0046] In step S6a, the system constructs a time series data set of risk indices based on the actual residue amount in step S4e and the adjusted comprehensive toxicity weight in step S5e; the actual residue amount is read from the actual residue amount table in the pesticide chemical transformation database, and the adjusted comprehensive toxicity weight is read from the comprehensive toxicity weight table; the risk index is generated by multiplying the actual residue amount by the adjusted comprehensive toxicity weight. For example, when the actual residue amount is 86.4 mg / kg and the comprehensive toxicity weight is 0.49, the risk index is 86.4×0.49≈42.3; the fields of the time series data set include batch data index, pesticide combination name, risk index, and time stamp. For example, the risk index corresponding to batch data index 20231001-GZB-003-chlorpyrifos-cypermethrin is 42.3; the data set is sorted by time stamp and stored in the time series database, and the time stamp granularity is accurate to minutes. For example, the risk index 42.3 is recorded at 14:30:00 on October 1, 2023.

[0047] In step S6b, the system combines the time series data set of historical batch data and extracts cross-batch risk trend characteristics through grey relational analysis; the time range of the historical batch data is the most recent 12 months, and the data is queried from the risk index table in the pesticide chemical transformation database; grey relational analysis is realized by calculating the correlation degree between the current batch and the historical batch risk index sequences. For example, the correlation degree between the current batch risk index sequence [42.3, 45.1, 40.8] and the historical batch sequence [38.5, 41.2, 39.7] is 0.85; the correlation degree calculation formula is the sum of the reciprocals of the absolute values of the differences at each time point of the two sequences, normalized to the interval from 0 to 1. For example, when the differences are 3.8, 3.9, 1.1, the sum of the reciprocals is 1 / 3.8 + 1 / 3.9 + 1 / 1.1≈1.43, and after normalization, it is 1.43 / 3≈0.48; the risk trend characteristics include correlation degree threshold, trend slope, and fluctuation variance. For example, when the correlation degree ≥0.7, it is determined as a high-risk trend, and a positive trend slope indicates an increase in risk.

[0048] In step S6c, the system generates a dynamic risk warning model based on the cross-batch risk trend characteristics and the node topology influence (i.e., PageRank value) in step S3d; the node topology influence is read from the topological centrality score table of the metabolism-toxicity association network. For example, the PageRank value of the cross-node B is 0.045; the input of the model is the concatenated vector of the risk trend characteristics and the node influence. For example, the correlation degree of 0.85, the trend slope of 0.12, the fluctuation variance of 1.5, and the PageRank value of 0.045 are concatenated as [0.85, 0.12, 1.5, 0.045]; the model fits the historical risk data and the current trend characteristics through linear regression, and the weight parameters are optimized according to the least squares method. For example, the weight vector is [0.6, 0.3, 0.1, 0.2]; the output is the risk warning level, and the level is divided into three levels: low, medium, and high. For example, the output value of 0.72 corresponds to a high risk.

[0049] In step S6d, the system stores the output result of the dynamic risk warning model in the pesticide chemical conversion database according to the batch data index association; the storage table structure includes the batch data index, the pesticide combination name, the warning level, and the update timestamp. For example, the warning level corresponding to the batch data index 20231001-GZB-003-chlorpyrifos-cypermethrin is high risk; the cross-batch risk prediction report is generated through SQL joint query. For example, execute "SELECT batch data index, warning level FROM risk warning table WHERE warning level = 'high risk'" and associate with the actual residue amount table in step S4e; the report format is a structured table, including the batch code, the pesticide name, the risk index, and the warning level. For example, the record in the table is "20231001-GZB-003-chlorpyrifos-cypermethrin, high risk, 42.3".

[0050] In step S6e, the system updates the parameters of the dynamic risk warning model according to the real-time batch data, and the real-time data is updated by the sensor in step S4a; the parameter adjustment adopts the sliding window method, and the window size is the data of the last 3 months. For example, when the proportion of high-risk batches in the window exceeds 20%, the risk level threshold is lowered from 0.7 to 0.6; the adjustment operation includes recalculating the grey correlation degree weight and the regression coefficient. For example, the original weight vector [0.6, 0.3, 0.1, 0.2] is adjusted to [0.65, 0.25, 0.1, 0.15]; the updated parameters are stored according to the timestamp version. For example, the version number 20231001-1 corresponds to the first adjustment on October 1, 2023.

[0051] Embodiment 2: Figure 2 The structural schematic diagram of a risk assessment system for pesticide residues in traditional Chinese medicine materials according to the present invention is given. A risk assessment system for pesticide residues in traditional Chinese medicine materials includes the following modules: Data acquisition module: Obtain pesticide residue detection data including the names of pesticides and the total residue amounts in traditional Chinese medicines to be evaluated; extract processing technology parameters including processing temperature and duration. Transformation analysis module: Correlate with a pre-stored pesticide chemical transformation database according to the names of pesticides and processing technology parameters to determine the transformation forms and corresponding conversion rates of pesticides under the processing technology. Weight generation module: Calculate the node influence of the transformation form through a graph convolutional model based on pre-stored biological network topology data to generate toxicity weight values. Dynamic correction module: Based on the micro-perturbation data of the processing environment collected by sensors, quantify the influence of non-linear perturbations on the conversion rate through chaos theory, and dynamically correct the actual residue amount in combination with a recurrent neural network. Cooperative adjustment module: If there are residues of two or more pesticides, dynamically adjust the toxicity weight values according to the cooperative effect relationship associated with the type of action mechanism. Prediction and evaluation module: Generate a risk assessment result based on the actual residue amount and the adjusted toxicity weight values, and perform cross-batch risk prediction in combination with historical batch data.

[0052] The calculations involved in the embodiments are all dimensionless numerical calculations. The preset parameters and threshold selections in the calculations are set by those skilled in the art according to the actual situation.

[0053] It should be noted that the present invention can be deployed on the device itself to achieve embedded applications, or can also run on a PC or other terminals with a user interface, so as to meet various hardware environments and usage requirements.

[0054] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more collections of available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, or magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0055] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and modules described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0056] In several embodiments provided in the present application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there can be other division methods in actual implementation. For example, multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of the devices or modules can be in electrical, mechanical, or other forms.

[0057] The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules. They can be located in one place or distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0058] In addition, in each embodiment of the present application, each functional module can be integrated into a processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0059] If the above-mentioned function is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0060] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present application, and all of them should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0061] Finally: The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A risk assessment method for pesticide residues in Chinese medicinal materials, characterized in that, It includes the following steps: S1. Obtain pesticide residue detection data including the pesticide names and total residue amounts in the Chinese medicinal materials to be evaluated; extract the processing technology parameters including the processing temperature and duration; S2. According to the pesticide names and the processing technology parameters, associate with the pre-stored pesticide chemical conversion database to determine the conversion forms and corresponding conversion rates of the pesticides under the processing technology; S3. Calculate the node influence of the conversion forms through a graph convolutional model based on the pre-stored biological network topology data to generate toxicity weight values; S4. Based on the micro-perturbation data of the processing environment collected by sensors, quantify the influence of non-linear perturbations on the conversion rate through chaos theory, and dynamically correct the actual residue amount in combination with a recurrent neural network; S5. If there are two or more pesticide residues, dynamically adjust the toxicity weight values according to the synergistic effect relationship associated with the action mechanism type; S6. Generate a risk assessment result according to the actual residue amount and the adjusted toxicity weight values, and perform cross-batch risk prediction in combination with historical batch data.

2. The risk assessment method for pesticide residues in Chinese medicinal materials according to claim 1, characterized in that, S1 includes: S1a. Obtain pesticide residue detection data in the planting area of Chinese medicinal materials through Internet of Things sensors. The pesticide residue detection data includes the pesticide names and the corresponding total residue amounts; S1b. Match the pre-stored pesticide detection data cleaning rule library according to the pesticide names, and perform outlier removal and standardization processing on the pesticide residue detection data; S1c. Extract the time series data of the processing temperature and duration from the processing equipment control system to generate processing technology parameters; S1d. According to the pesticide names and the processing technology parameters, associate with the production batch codes of Chinese medicinal materials to establish batch data indexes.

3. The risk assessment method for pesticide residues in Chinese medicinal materials according to claim 1, characterized in that S2 It includes: S2a. According to the pesticide names and the processing temperature, match the reaction path templates in the pesticide chemical conversion database, and screen the candidate conversion forms corresponding to the processing technology; S2b. According to the duration and the kinetic parameters of the candidate conversion forms, calculate the conversion rates of each conversion form through a conditional probability model; S2c. Based on the processing stage labels, associate with the experimental verification data in the pesticide chemical conversion database to correct the confidence level of the conversion rates; S2d. Store the corrected conversion forms and conversion rates in the pesticide chemical conversion database according to the batch data indexes.

4. The risk assessment method for pesticide residues in Chinese medicinal materials according to claim 1, characterized in that S3 It includes: S3a. Construct a metabolism-toxicity association network, with nodes being the conversion forms and human metabolites, and edges being the toxicity transmission paths; S3b. Extract node feature vectors based on the biological network topology data, including the chemical descriptors of the conversion forms and the toxicity values of the metabolites; S3c. Aggregate the neighborhood node features through the multi-layer convolutional layers of the graph convolutional model to generate node embedding representations; S3d. Calculate the topological centrality scores based on the node embedding representations, and generate toxicity weight values in combination with the toxicity values; S3e. Store the toxicity weight values in the pesticide chemical conversion database according to the batch data indexes.

5. The risk assessment method for pesticide residues in Chinese medicinal materials according to claim 1, characterized in that S4 It includes: S4a. Collect the time series data of the temperature, humidity and air flow velocity in the processing equipment as the micro-perturbation data of the processing environment, and store it in association according to the batch data indexes; S4b. Analyze the chaotic characteristics of the micro-perturbation data in stages based on the Lyapunov exponent, quantify the intensity of the instantaneous non-linear perturbation in the first stage, and extract the perturbation cumulative effect coefficient of the historical batches in the second stage; S4c. Generate a dynamic correction factor by fusing the instantaneous non - linear perturbation intensity and the perturbation cumulative effect coefficient through a recurrent neural network, and synchronously correlate the conversion rate in step S2 with the toxicity weight value in step S3; S4d. Perform two - way correction on the total residue amount based on the dynamic correction factor to generate the actual residue amount; S4e. Index and store the actual residue amount in the pesticide chemical conversion database according to the batch data, and mark the correction path type.

6. The risk assessment method for pesticide residues in Chinese medicinal materials according to claim 5, characterized in that, The two - way correction includes: forward correction amplifies the residue amount of high - toxicity forms based on the dynamic factor, and reverse correction suppresses the conversion rate error of low confidence.

7. A risk assessment method for pesticide residues in Chinese medicinal materials according to claim 1, characterized in that S5 Including: S5a. If there are two or more pesticide residues, determine the type of action mechanism of the pesticides according to the pesticide names; S5b. Match the pre - stored synergistic effect rule library based on the type of action mechanism to screen the synergistic action modes of pesticide combinations; S5c. Extract the metabolic path cross - nodes according to the metabolism - toxicity association network in step S3, and quantify the competitive inhibition intensity of the pesticide combination on the cross - nodes; S5d. Based on the competitive inhibition intensity and the topological centrality score in step S3, construct a dynamic attenuation model of the toxicity propagation path to generate a comprehensive toxicity correction coefficient; S5e. Apply the comprehensive toxicity correction coefficient to the toxicity weight value in step S3 to generate an adjusted comprehensive toxicity weight, and index and store it in the pesticide chemical conversion database according to the batch data.

8. A risk assessment method for pesticide residues in Chinese medicinal materials according to claim 7, characterized in that The synergistic action modes include competitive inhibition of metabolic paths and superposition of target organ toxicity.

9. A risk assessment method for pesticide residues in Chinese medicinal materials according to claim 1, characterized in that S6 Including: S6a. Based on the actual residue amount and the adjusted comprehensive toxicity weight in step S5e, construct a time - series data set of risk indices; S6b. Combine the time - series data set of historical batch data, and extract the cross - batch risk trend characteristics through grey relational analysis; S6c. Based on the cross - batch risk trend characteristics and the node influence of the conversion form, generate a dynamic risk warning model; S6d. Index and store the output results of the dynamic risk warning model according to the batch data to generate a cross - batch risk prediction report; S6e. Update the parameters of the dynamic risk warning model according to the real - time batch data and adjust the risk warning level threshold.

10. A risk assessment system for pesticide residues in Chinese medicinal materials, used to implement a risk assessment method for pesticide residues in Chinese medicinal materials according to any one of claims 1-9, characterized in that Including the following modules: Data acquisition module: Obtain the pesticide residue detection data including the pesticide names and the total residue amount in the Chinese medicinal materials to be evaluated; extract the processing technology parameters including the processing temperature and duration; Conversion analysis module: Correlate the pre - stored pesticide chemical conversion database according to the pesticide names and the processing technology parameters to determine the conversion forms of the pesticides under the processing technology and the corresponding conversion rates; Weight generation module: Calculate the node influence of the conversion form through a graph convolution model based on the pre - stored biological network topology data to generate the toxicity weight value; Dynamic correction module: Based on the micro - perturbation data of the processing environment collected by sensors, quantify the influence of non - linear perturbation on the conversion rate through chaos theory, and dynamically correct the actual residue amount in combination with a recurrent neural network; Synergistic adjustment module: If there are two or more pesticide residues, dynamically adjust the toxicity weight value according to the synergistic effect relationship associated with the type of action mechanism; Prediction and evaluation module: Generate a risk assessment result according to the actual residue amount and the adjusted toxicity weight value, and perform cross - batch risk prediction in combination with historical batch data.

Images