ai-based rapid screening and risk assessment method for contaminants in imported recycled pulp

Abstract

This invention relates to the field of pulp contaminant screening and risk assessment technology, and discloses an AI-based method for rapid screening and risk assessment of contaminants in imported recycled pulp. The method involves collecting multi-dimensional feature data from samples through a detection chamber integrating multiple types of sensor probes. After processing by an outlier identification and correction algorithm, the data is input into a recycled pulp contaminant gradient enhancement decision model for feature extraction and weight allocation. This is then combined with a pulp contaminant spatiotemporal distribution prediction model to generate contaminant distribution prediction results. Preliminary judgment results are generated through multi-dimensional parameter coupling analysis, and cross-module verification and optimization are completed through the collaborative interaction of intelligent detection, AI analysis, and decision support modules. This invention achieves efficient screening and accurate risk assessment of contaminants in imported recycled pulp through feature processing, spatiotemporal distribution prediction, and a multi-module collaboration mechanism, adapting to the needs of large-scale trade customs clearance and pollution control, and improving the intelligence and precision of the detection process.

Description

Technical Field

[0001] This invention relates to the field of pulp contaminant screening and risk assessment technology, and in particular to an AI-based method for rapid screening and risk assessment of contaminants in imported recycled pulp. Background Technology

[0002] As the global trade in recycled pulp continues to expand, pollutants such as heavy metals and toxic organic compounds in imported recycled pulp have become key hidden dangers affecting the safe production and ecological environment of the paper industry. Establishing an efficient and accurate pollutant screening and risk assessment mechanism has become an urgent need for the industry. Currently, the testing of imported recycled pulp largely relies on traditional manual sampling and laboratory analysis, which suffers from long testing cycles and limited coverage, making it difficult to meet the rapid screening needs of large-scale trade clearance. The integrated application of AI technology and multi-sensor detection technology provides technical support for achieving efficient pollutant identification and risk assessment. Screening solutions based on integrated detection hardware and intelligent algorithm models can integrate multi-dimensional feature data to achieve accurate pollutant identification and risk level determination, aligning with the practical application scenarios of pollution prevention and control in imported recycled pulp.

[0003] Existing technologies have two significant drawbacks: First, traditional detection methods rely on single-sensor data acquisition, lacking the collaborative application of multi-dimensional sensing technologies such as near-infrared, X-ray, and fluorescence. Furthermore, they lack dedicated sensor probe fixing and scanning structures designed for the detection of pollutants in recycled pulp, resulting in insufficient comprehensiveness and accuracy of feature data acquisition, and an inability to fully capture the complex attribute information of pollutants. Second, there is a lack of an intelligent analysis system that deeply integrates gradient enhancement decision-making, spatiotemporal distribution prediction, and outlier correction. Existing algorithms are mostly applied separately to data processing or prediction stages, without forming a parameter coupling analysis mechanism for multi-model collaboration. At the same time, the lack of standardized communication protocols and circuit connection designs for data interaction between modules leads to insufficient accuracy and consistency in screening results and risk assessments, making it difficult to meet the refined requirements for pollution control of imported recycled pulp. Summary of the Invention

[0004] To overcome the shortcomings and deficiencies of existing technologies, this invention provides an AI-based method for rapid screening and risk assessment of contaminants in imported recycled pulp.

[0005] The technical solution adopted in this invention is an AI-based method for rapid screening and risk assessment of contaminants in imported recycled pulp, comprising the following steps: S1, collecting multidimensional contaminant characteristic data of imported recycled pulp samples through the physical layout and connection structure of a detection chamber integrating near-infrared, X-ray, and fluorescence sensor probes, wherein the sensor probes perform full-area coverage data acquisition of the samples through a special fixing and scanning structure; S2, using an outlier identification and correction algorithm to identify and correct anomalies in the collected multidimensional characteristic data, and locating the abnormal data interval through correlation analysis between the characteristic data and preset thresholds; S3, inputting the corrected characteristic data into a gradient enhancement decision model for recycled pulp contaminants, and classifying contaminants through the gradient enhancement iteration process in the model. S4. Based on the extracted feature parameters, a spatiotemporal distribution prediction model input matrix for pulp contaminants is constructed, and the contaminant distribution prediction results are generated through the spatiotemporal dimension correlation mapping mechanism in the model. S5. Combining the feature weights output by the gradient enhancement decision model with the spatiotemporal distribution prediction results, a parameter system for rapid screening and risk assessment of contaminants in imported recycled pulp is constructed, and preliminary screening and risk level determination results are generated through multi-dimensional parameter coupling analysis. S6. Through the circuit connection and preset communication protocol between the intelligent detection module, AI analysis module, and decision support module, data interaction and collaboration are carried out to verify and optimize the preliminary determination results across modules, and the final contaminant screening results and risk assessment report are output.

[0006] Furthermore, the feature weight allocation expression for the recycled pulp contaminant gradient enhancement decision model is as follows: ,in, This represents the weight value of the i-th feature for the j-th type of pollutant. Represents the loss function. Indicates the preceding The output of the model in each iteration, Indicates the step size parameter. This represents the output of the base learner corresponding to the j-th type of pollutant. This represents the k-th feature value of the i-th sample. This represents the characteristic threshold of the j-th type of pollutant. Represents the feature correlation function. This represents the regularization parameter.

[0007] Furthermore, the distribution prediction expression of the spatiotemporal distribution prediction model for pulp contaminants is as follows: ,in, Represents the coordinates at time t The probability of pollutant distribution at a location, Represents the distribution coefficient. Let represent the contribution coefficient of the p-th feature. Let represent the mapping function of the p-th feature at time t. This represents the p-th input feature parameter. The time characteristic factor at time t. Indicates the spatial attenuation coefficient. This represents the spatial coordinates corresponding to the p-th feature. This indicates the error correction term.

[0008] Furthermore, the outlier correction expression of the detection data outlier identification and correction algorithm is as follows: ,in, This represents the corrected k-th feature data. This represents the original k-th feature data. Represents the mean of the feature data. This represents the anomaly correction function. Indicates the standard deviation of the feature data. This indicates the number of neighboring data points involved in the correction. This indicates the correction smoothing coefficient.

[0009] Furthermore, the risk level determination expression for the rapid screening and risk assessment of contaminants in imported recycled pulp is as follows: ,in, Indicates the risk level value. This represents the risk adjustment factor. This represents the weight of pollutant class j. This represents the screening result value for pollutant category j. Indicates the number of pollutant categories. This represents the risk amplification factor. This represents the spatiotemporal distribution density of pollutant type j.

[0010] Furthermore, the data interaction efficiency expression of the intelligent detection module, AI analysis module, and decision support module is as follows: ,in, This represents the data interaction efficiency value. Indicates the communication protocol adaptation coefficient. This represents the size of the data packet in the i-th data exchange. This represents the priority weight of the i-th data interaction. Indicates the number of data interactions. This represents the transmission time of the i-th data interaction. Indicates the link loss coefficient. This represents the length of the transmission link for the i-th data interaction.

[0011] Further, step S3 includes the following sub-steps: S31, based on the initialization parameters of the recycled pulp contaminant gradient enhancement decision model, the corrected multidimensional feature data is grouped according to contaminant categories to establish a preliminary mapping relationship between feature data and contaminant categories; S32, through the gradient enhancement iteration mechanism in the model, the contribution of different features to contaminant identification in each iteration is calculated, and a subset of calibrated features is selected based on the contribution ranking results; S33, each feature in the calibrated feature subset is assigned a dynamic weight value, and the weight value is dynamically adjusted based on the correlation strength between the feature and the contaminant category and the error change rate during the iteration process; S34, the adjusted calibrated features and corresponding weights are input into the decision layer of the model, and the contaminant category and content correlation features are accurately extracted through hierarchical decision rules to form feature extraction results.

[0012] Further, S4 includes the following sub-steps: S41, based on the feature parameters extracted in S3, combined with the collection time and location information of the imported recycled pulp samples, construct a model input matrix including spatiotemporal dimension information, and clarify the feature types and spatiotemporal attributes corresponding to different elements in the matrix; S42, import the input matrix into the pulp contaminant spatiotemporal distribution prediction model, and mine the potential mapping relationship between feature parameters and spatiotemporal dimensions through the spatiotemporal correlation algorithm in the model; S43, construct a contaminant spatiotemporal distribution prediction equation based on the mapping relationship, and obtain the preliminary predicted value of contaminant distribution at different spatiotemporal nodes by solving the equation; S44, perform spatiotemporal consistency verification on the preliminary predicted value, eliminate prediction results that do not conform to the spatiotemporal distribution law, and generate the final contaminant spatiotemporal distribution prediction result.

[0013] Further, S5 includes the following sub-steps: S51, collecting the feature weight data output by the gradient enhancement decision model and the calibration parameters in the spatiotemporal distribution prediction results to determine the calibration parameter indicators for rapid screening and risk assessment of pollutants in imported recycled pulp; S52, standardizing the calibration parameter indicators, clarifying the value range and judgment criteria of different parameters, and constructing a multi-dimensional parameter system framework; S53, classifying different parameter indicators according to screening and risk assessment functions, calculating the correlation strength between different categories of parameters through a parameter coupling algorithm, and forming a parameter coupling matrix; S54, based on the parameter coupling matrix and the preset risk level classification rules, screening for pollutants and determining the risk level of imported recycled pulp samples to generate preliminary judgment results.

[0014] A rapid screening and risk assessment method for contaminants in imported recycled pulp based on AI is implemented through different units, including: a multi-dimensional feature data acquisition and transmission unit, which is physically connected to the detection chamber integrating near-infrared, X-ray, and fluorescence sensor probes. This unit acquires multi-dimensional feature data of the samples through a special fixing and scanning structure of the sensor probes and transmits the data to subsequent units via a pre-set communication interface; a detection data anomaly correction and processing unit, which is circuitically connected to the multi-dimensional feature data acquisition and transmission unit. This unit calls an anomaly identification and correction algorithm to identify and correct anomalies in the received data and outputs the corrected data; and a contaminant feature extraction and weight allocation unit, which is communicatively connected to the detection data anomaly correction and processing unit. This unit runs a recycled pulp contaminant gradient enhancement decision model to apply the corrected data... The system performs feature extraction and weight allocation to generate feature weight data; the pollutant spatiotemporal distribution prediction calculation unit interacts with the pollutant feature extraction and weight allocation unit through a preset communication protocol, performs distribution prediction based on the pulp pollutant spatiotemporal distribution prediction model and feature weight data, and outputs the prediction results; the multi-module collaborative screening and evaluation unit is connected to the pollutant spatiotemporal distribution prediction calculation unit, intelligent detection module, AI analysis module, and decision support module respectively, integrates the prediction results with data from different modules, and completes the preliminary judgment through the rapid screening and risk assessment parameter system for imported recycled pulp pollutants; the final result output and storage unit is circuit-connected to the multi-module collaborative screening and evaluation unit, receives and stores the final pollutant screening results and risk assessment report, and provides an external output interface.

[0015] Beneficial Effects: This invention proposes an AI-based method for rapid screening and risk assessment of contaminants in imported recycled pulp. Utilizing a detection chamber with integrated near-infrared, X-ray, and fluorescence sensors, and a unique fixing and scanning structure, it achieves comprehensive, multi-dimensional feature data acquisition covering the entire sample area. This solves the problem of insufficient comprehensiveness and accuracy in traditional single-sensor data acquisition, fully capturing the complex attributes of contaminants. In the data processing and analysis stage, it integrates an outlier identification and correction algorithm, a gradient-enhanced decision model for recycled pulp contaminants, and a spatiotemporal distribution prediction model for pulp contaminants. A multi-model collaborative parameter coupling analysis mechanism is constructed, improving the accuracy of screening and assessment through hierarchical feature extraction, spatiotemporal dimension correlation mapping, and multi-dimensional parameter coupling analysis. Simultaneously, by leveraging standardized circuit connections and communication protocols between the intelligent detection module, AI analysis module, and decision support module, it achieves efficient data interaction and cross-module verification and optimization, overcoming the shortcomings of insufficient module collaboration and poor result consistency in existing technologies. This invention forms a complete technical chain from data acquisition and anomaly correction to feature extraction, distribution prediction, collaborative assessment, and result output, significantly improving screening efficiency and the precision of risk assessment, fully meeting the practical needs of pollution prevention and control in imported recycled pulp. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the overall process of the method of the present invention.

[0017] Figure 2 This is a flowchart of method step S3 of the present invention;

[0018] Figure 3 This is a flowchart of method step S4 of the present invention;

[0019] Figure 4 This is a flowchart of step S5 of the method of the present invention;

[0020] Figure 5 This is a diagram showing the unit composition for implementing the method of the present invention. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] like Figure 1 As shown, the AI-based rapid screening and risk assessment method for contaminants in imported recycled pulp includes the following steps:

[0023] S1, through the physical layout and connection structure of the detection chamber integrating near-infrared, X-ray and fluorescence sensing probes, collects multi-dimensional pollutant characteristic data of imported recycled pulp samples. The sensing probes perform full-area coverage data collection of the samples through a special fixing and scanning structure.

[0024] Specifically, step S1 utilizes a detection chamber integrating near-infrared, X-ray, and fluorescence sensors to collect multi-dimensional contaminant characteristic data from imported recycled pulp samples. The detection chamber employs a layered physical layout design, with the sensor probes fixed to the internal beam structure according to a triangular distribution principle. The spacing between adjacent probes is set to 15 cm to ensure comprehensive coverage. The sensor probes are equipped with an electrically operated sliding rail scanning structure, with a scanning speed set at 3 cm / s. The scanning path uses a serpentine trajectory, driven by a stepper motor to achieve full-area data acquisition along the sample surface, collecting at least 8 data points per square centimeter of sample area. The detection chamber and sensor probes are mechanically and electrically connected via waterproof and dustproof connectors. The connection interface uses a snap-fit ​​fixing structure, coupled with elastic sealing gaskets to ensure structural stability during the detection process. The near-infrared sensor probe operates in a wavelength range of 900-1700 nm, the X-ray probe tube voltage is adjustable from 30-80 kV, and the fluorescence probe excitation wavelength is set to 254 nm. All three probes are started simultaneously for data acquisition, with the data sampling frequency uniformly set to 100 Hz. Multi-dimensional feature data is received and stored synchronously through a multi-channel data acquisition card. During the acquisition process, the working status is monitored in real time by the temperature sensor built into the probe to ensure the stability and reliability of data acquisition, providing comprehensive and accurate raw data support for subsequent pollutant characteristic analysis.

[0025] S2, using an outlier detection and correction algorithm to identify and correct outliers in the collected multidimensional feature data, and locating outlier data ranges through correlation analysis between feature data and preset thresholds;

[0026] Specifically, step S2 uses an outlier detection and correction algorithm to process the collected multidimensional feature data. The algorithm first establishes independent data distribution models for near-infrared, X-ray, and fluorescence sensor data, setting normal value ranges for each dimension based on the 3σ principle. The preset threshold ranges are 0.12-0.86 for near-infrared data, 0.08-0.74 for X-ray data, and 0.15-0.92 for fluorescence data. By calculating the deviation of each data point from the mean of the corresponding dimension, and combining this with the preset thresholds, correlation analysis is performed. When the deviation exceeds ±30% of the threshold, it is identified as outlier data and the corresponding data interval is marked. For the marked outlier data intervals, a neighborhood data interpolation method is used for correction. Ten consecutive data points before and after the outlier data point are selected as reference neighborhoods. The weighted average of the neighborhood data is used to replace the outlier point. The weight coefficients decrease linearly with distance from the outlier point, with closer points having higher weights. The highest weight value is set to 0.8, and the lowest to 0.2. During the correction process, the location, original value, and corrected value of the abnormal data are recorded simultaneously to form an abnormal data processing log, ensuring the traceability of data correction and providing high-quality, interference-free feature data for subsequent model analysis.

[0027] S3, input the corrected feature data into the recycled pulp pollutant gradient enhancement decision model, and extract and assign weights to the pollutant category and content correlation features through the gradient enhancement iteration process in the model;

[0028] Specifically, in step S3, the corrected multidimensional feature data is input into the gradient enhancement decision model for pollutants in recycled pulp. During model initialization, the number of decision trees is set to 120, with a maximum depth of 8 layers per tree and a minimum number of samples per leaf node of 15. Feature extraction and weight allocation are performed through the gradient enhancement iterative mechanism within the model, with 80 iterations. In each iteration, the loss value between the current model prediction and the actual pollutant characteristics is calculated. A new base learner is constructed based on the negative gradient direction of the loss value, using the CART decision tree algorithm. The learning rate of the new base learner in each iteration is set to 0.05. For pollutant categories (heavy metals, toxic organics, suspended particulate matter, etc.) and content-related features, extraction is performed hierarchically. The first layer extracts basic features such as data intensity and spectral peaks; the second layer extracts derived features such as feature change rate and peak interval; and the third layer extracts coupled and correlated features of multidimensional features. By calculating the contribution value of each feature during the iteration process, the features are weighted. The contribution value is calculated based on the proportion of the feature's contribution to the reduction of the loss function. The higher the contribution, the larger the weight value. The weight value range is set from 0.01 to 0.98, with the highest weight value for heavy metal-related features not exceeding 0.92 and the highest weight value for toxic organic matter-related features not exceeding 0.88. The sum of all feature weight values ​​is normalized to 1.0. The final output includes a feature set including hierarchical features and corresponding weights, providing core input parameters for subsequent spatiotemporal distribution prediction.

[0029] S4. Construct the input matrix of the spatiotemporal distribution prediction model for pulp contaminants based on the extracted feature parameters, and generate the contaminant distribution prediction results through the spatiotemporal dimension correlation mapping mechanism in the model.

[0030] Specifically, step S4 constructs an input matrix for the spatiotemporal distribution prediction model of pulp contaminants based on the feature parameters extracted in S3. The input matrix has a dimension of 120×80, with rows corresponding to 120 core feature parameters and columns corresponding to 80 spatiotemporal nodes. Each element in the matrix corresponds to the value of a specific feature parameter at a specific spatiotemporal node. The spatiotemporal nodes are divided according to a time interval of 5 minutes and a spatial interval of 20 centimeters, covering the entire time range and spatial region of sample detection. The input matrix is ​​imported into the spatiotemporal distribution prediction model of pulp contaminants. The spatiotemporal dimension correlation mapping mechanism in the model is implemented through the spatiotemporal covariance function. The time dimension correlation coefficient is set to 0.75, the spatial dimension correlation coefficient is set to 0.82, and the spatiotemporal cross-correlation coefficient is set to 0.68. By calculating the correlation of feature parameters between different spatiotemporal nodes, the potential mapping relationship between feature parameters and spatiotemporal dimensions is explored. The sliding window method is used in the mapping relationship construction process, with the window size set to 10×10 (10 time window intervals and 10 spatial window intervals). A spatiotemporal distribution prediction equation for pollutants is constructed based on the mapping relationship. The preliminary predicted value of pollutant distribution for each spatiotemporal node is obtained through iterative solution. During the solution process, the convergence condition is set as the difference between two adjacent iterations being less than 0.001, and the maximum number of iterations is set to 50. The preliminary predicted values ​​are then checked for spatiotemporal consistency. The check standard is that the rate of change of predicted values ​​between adjacent spatiotemporal nodes does not exceed ±25%. Predicted values ​​exceeding this range are deemed inconsistent with the spatiotemporal distribution pattern and are replaced by the average of the predicted values ​​of three adjacent spatiotemporal nodes. Finally, a pollutant distribution prediction result including all spatiotemporal nodes is generated.

[0031] S5, combining the feature weights output by the gradient enhancement decision model with the spatiotemporal distribution prediction results, constructs a parameter system for rapid screening and risk assessment of pollutants in imported recycled pulp, and generates preliminary screening and risk level determination results through multi-dimensional parameter coupling analysis;

[0032] Specifically, step S5 combines the feature weights output by the gradient-enhanced decision model with the spatiotemporal distribution prediction results to construct a parameter system for rapid screening and risk assessment of contaminants in imported recycled pulp. This parameter system includes eight dimensions: feature importance parameters, spatiotemporal distribution density parameters, contaminant concentration parameters, and category identification confidence parameters. Each dimension includes 15-20 specific parameter indicators. First, the top 50 high-weight features from the feature weight data and key parameters such as peak distribution, mean distribution, and variance distribution from the spatiotemporal distribution prediction results are collected, identifying a total of 120 core parameter indicators. These core parameter indicators are then standardized, mapping all parameter values ​​to the 0-1.0 range. The standardization of feature importance parameters is based on direct normalization of weight values; spatiotemporal distribution parameter standardization is based on maximum / minimum value scaling; and contaminant concentration parameter standardization is based on industry standard threshold conversion. The value range and judgment criteria for each parameter are clearly defined. For example, a feature importance parameter value ≥0.6 indicates high importance, a spatiotemporal distribution density parameter value ≥0.7 indicates a high-density region, and a contaminant concentration parameter value ≥0.8 indicates a high concentration level. The parameters are divided into two categories based on their screening and risk assessment functions: 70 parameters for screening and 50 parameters for risk assessment. A parameter coupling algorithm is used to calculate the correlation strength between parameters of different categories. The correlation strength calculation is based on the Pearson correlation coefficient; an absolute value of ≥0.65 is considered a strong correlation, 0.35-0.65 a moderate correlation, and ≤0.35 a weak correlation, forming a 120×120 parameter coupling matrix. Based on the parameter coupling matrix and preset risk level classification rules (low risk, medium risk, high risk, and extremely high risk), a comprehensive risk value is calculated through multi-dimensional parameter coupling analysis. The comprehensive risk value range is set from 0 to 10.0, where 0-3.0 is low risk, 3.0-5.5 is medium risk, 5.5-8.0 is high risk, and 8.0-10.0 is extremely high risk. Finally, preliminary screening results and risk level determination results are generated.

[0033] The S6 uses circuit connections and preset communication protocols between the intelligent detection module, AI analysis module, and decision support module to interact and collaborate on data, perform cross-module verification and optimization of preliminary judgment results, and output the final pollutant screening results and risk assessment report.

[0034] Specifically, step S6 achieves data interaction and collaboration through standardized circuit connections and preset communication protocols between the intelligent detection module, AI analysis module, and decision support module. The three modules are connected via a PCIe 4.0 interface with a transmission rate set at 32GB / s. Redundant circuit connections ensure data transmission stability. A custom industrial Ethernet protocol is used for communication, with a data frame format of frame header (8 bytes) + data length (4 bytes) + data content (variable length) + checksum (4 bytes). The communication baud rate is set to 1000Mbps, and data transmission latency is controlled within 50 milliseconds. The intelligent detection module first encapsulates the feature data and prediction result data from the preliminary judgment results according to the protocol format and transmits them to the AI ​​analysis module via circuit connection. The AI ​​analysis module receives the data and performs secondary feature verification. The verification process uses feature matching degree calculation, with a matching degree threshold set at 0.85. Features are considered valid if the matching degree is higher than the threshold. The AI ​​analysis module transmits the verified valid data to the decision support module. The decision support module, combined with its built-in industry pollution prevention and control standard library (including 150 national and industry standards), performs compliance verification on the preliminary judgment results. Verification items include 12 aspects such as pollutant category compliance, concentration limit compliance, and distribution range compliance. Through cross-module data interaction and collaboration, the preliminary judgment results are verified from multiple dimensions. When the consistency of the verification results from the three modules reaches over 90%, the final result is directly output. When the consistency is below 90%, a second iteration optimization is initiated, re-calling the core algorithms of steps S3-S5 for parameter adjustment, with no more than three adjustments. After optimization, a final screening result and risk assessment report are generated, including pollutant category, concentration range, spatiotemporal distribution location, risk level, and compliance conclusion. The report data is stored in XML format and simultaneously transmitted to storage devices and display terminals through the module's built-in output interface, ensuring the results are queryable and displayable.

[0035] Preferably, the feature weight allocation expression of the recycled pulp contaminant gradient enhancement decision model is as follows: ,in, This represents the weight value of the i-th feature for the j-th type of pollutant. Represents the loss function. Indicates the preceding The output of the model in each iteration, Indicates the step size parameter. This represents the output of the base learner corresponding to the j-th type of pollutant. This represents the k-th feature value of the i-th sample. This represents the characteristic threshold of the j-th type of pollutant. Represents the feature correlation function. This represents the regularization parameter.

[0036] Specifically, the feature weight allocation design of the recycled pulp contaminant gradient enhancement decision model aims to accurately quantify the contribution of different features to the identification of various contaminants, ensuring the targetedness and effectiveness of the model's feature extraction. During implementation, basic model parameters are first set: the loss function adopts a squared loss type, the step size is controlled within the range of 0.01-0.1, and the regularization parameter is fixed at 0.005 to limit model complexity and avoid overfitting. For different types of contaminants in imported recycled pulp, such as heavy metals and toxic organic compounds, specific feature thresholds are set. The feature association function adopts a non-linear mapping form, outputting the association value by calculating the matching degree between feature data and the corresponding threshold. During model training, based on the output results of previous iterations, the partial derivative of the loss function with respect to this output result is calculated to reflect the current trend of model error. Simultaneously, the feature value of each sample is associated with the corresponding contaminant feature threshold. The sum of these values ​​is then compared with the square of the sum plus the square root of the regularization parameter to obtain the feature association strength coefficient. Multiplying the partial derivative by the feature correlation strength coefficient yields the weight value of each feature corresponding to the pollutant category. The weight value is strictly controlled between 0.01 and 0.98. Through this dynamic allocation process, core features that are highly correlated with pollutant identification receive higher weights, significantly improving the model's accuracy in identifying various pollutants and providing reliable feature weight support for subsequent risk assessment.

[0037] Preferably, the distribution prediction expression of the spatiotemporal distribution prediction model for pulp contaminants is: ,in, Represents the coordinates at time t The probability of pollutant distribution at a location, Represents the distribution coefficient. Let represent the contribution coefficient of the p-th feature. Let represent the mapping function of the p-th feature at time t. This represents the p-th input feature parameter. The time characteristic factor at time t. Indicates the spatial attenuation coefficient. This represents the spatial coordinates corresponding to the p-th feature. This indicates the error correction term.

[0038] Specifically, the spatiotemporal distribution prediction model for pulp contaminants is designed to establish a deep correlation between feature parameters and spatiotemporal dimensions, enabling accurate prediction of contaminant distribution probabilities at different spatiotemporal nodes and providing data support for defining the pollution range. During implementation, the distribution coefficient is fixed at 0.85, the spatial attenuation coefficient is set at 0.7, and the error correction term is controlled within ±0.05 to ensure prediction accuracy. The total number of features is determined based on the core feature parameters extracted in step S3. Contribution coefficients are assigned according to feature importance ranking: the top 20% of features have contribution coefficients of 0.8-1.0, the middle 60% have contribution coefficients of 0.4-0.7, and the bottom 20% have contribution coefficients of 0.1-0.3. The time feature factor is set according to the data acquisition interval; when the acquisition interval is 5 minutes, this factor increases sequentially in increments of 0.1, 0.2…1.0. The feature mapping function uses a polynomial mapping method to couple feature parameters with the time feature factor, strengthening the correlation between the spatiotemporal dimensions and feature parameters. Spatial coordinates are determined based on the physical layout of the detection chamber and assigned according to the grid division results of the sample detection area, with each grid node corresponding to unique coordinate information. By calculating the product of the feature parameters processed by the mapping function and the spatial attenuation term, and then weighting and summing according to the contribution coefficient, and finally superimposing the error correction term, the probability of pollutant distribution at a specific coordinate at a specific time is obtained. Through spatiotemporal collaborative modeling, the distribution pattern of pollutants in different spatiotemporal scenarios is comprehensively and accurately characterized.

[0039] Preferably, the outlier correction expression of the outlier identification and correction algorithm for the detected data is: ,in, This represents the corrected k-th feature data. This represents the original k-th feature data. Represents the mean of the feature data. This represents the anomaly correction function. Indicates the standard deviation of the feature data. This indicates the number of neighboring data points involved in the correction. This indicates the correction smoothing coefficient.

[0040] Specifically, the outlier detection and correction algorithm aims to eliminate abnormal interference information in multidimensional feature data, ensuring the data quality of the input model and providing a reliable foundation for subsequent analysis. During implementation, the number of neighboring data points involved in the correction is fixed at 10, the correction smoothing coefficient is set to 0.02, and the outlier correction function uses an S-shaped function to ensure a smooth transition during the correction process and avoid data distortion. First, the mean and standard deviation of all feature data are calculated. Based on the 3σ principle, the outlier judgment boundary is determined. When the absolute value of the deviation of a feature data point from the mean exceeds three times the standard deviation, the data is judged as outlier data. Five consecutive normal data points before and after the outlier data are selected as neighboring reference data. The difference between each neighboring data point and the mean is calculated. Simultaneously, the correction weight of each neighboring data point on the outlier data is calculated using the outlier correction function. The correction weight decreases as the distance between the neighboring data and the outlier data increases; the correction weight for the closest neighboring data is 0.8, and the correction weight for the farthest neighboring data is 0.2. The difference between neighboring data and the mean is multiplied by the corresponding correction weight and summed. Then, the sum is divided by the sum of the correction weights and the smoothing coefficient to obtain the correction increment of the outlier data. This increment is added to the original outlier data to obtain the corrected feature data. Through this precise correction process, it is ensured that the corrected data conforms to the overall data distribution pattern, effectively improving the reliability and usability of the data.

[0041] Preferably, the risk level determination expression for the rapid screening and risk assessment of contaminants in imported recycled pulp is as follows: ,in, Indicates the risk level value. This represents the risk adjustment factor. This represents the weight of pollutant class j. This represents the screening result value for pollutant category j. Indicates the number of pollutant categories. This represents the risk amplification factor. This represents the spatiotemporal distribution density of pollutant type j.

[0042] Specifically, the design for rapid screening and risk assessment of contaminants in imported recycled pulp is used to integrate multi-dimensional parameters to achieve quantitative assessment of risk levels, providing an intuitive basis for pollution prevention and control decisions. During implementation, the risk correction coefficient is fixed at 0.9, the risk amplification coefficient is set at 1.2, and the number of contaminant categories is controlled between 8 and 12 categories based on actual testing needs. The weight of each contaminant category is the mean of the characteristic weight allocation results. The screening result value is determined by comparing the characteristic data with the standard threshold for the contaminant. When the characteristic data is lower than the standard threshold, the screening result value is 0.1-0.3; when the characteristic data is equal to the standard threshold, the screening result value is 0.4-0.6; and when the characteristic data is higher than the standard threshold, the screening result value is 0.7-1.0. The spatiotemporal distribution density is calculated through distribution prediction results, statistically analyzing the mean distribution probability of each contaminant category at all spatiotemporal nodes. The mean ranges from 0 to 1.0; a higher mean indicates a greater distribution density of the contaminant. By calculating the weighted average screening results of various pollutants, multiplying them by the logarithm of (1 + risk amplification coefficient × spatiotemporal distribution density), and then multiplying by the risk correction coefficient, the final risk level value is obtained. The risk level value is set to range from 0 to 10.0. Through the coupled calculation of multiple parameters, the risk level is accurately quantified, making the risk assessment results more valuable and providing scientific data support for the prevention and control of pollution from imported recycled pulp.

[0043] Preferably, the data interaction efficiency expression of the intelligent detection module, AI analysis module, and decision support module is: ,in, This represents the data interaction efficiency value. Indicates the communication protocol adaptation coefficient. This represents the size of the data packet in the i-th data exchange. This represents the priority weight of the i-th data interaction. Indicates the number of data interactions. This represents the transmission time of the i-th data interaction. Indicates the link loss coefficient. This represents the length of the transmission link for the i-th data interaction.

[0044] Specifically, the data interaction efficiency design of the intelligent detection module, AI analysis module, and decision support module aims to optimize data transmission and collaboration processes between modules, thereby improving the overall system's operational efficiency. During implementation, key parameters are set: the communication protocol compatibility coefficient is fixed at 0.95, and the link loss coefficient is set at 0.03. The number of data interactions is determined based on the detection process, with a fixed 8 interactions between modules per detection cycle. The size of each data packet is allocated according to the data type: feature data packets are set to 2MB, prediction result packets to 1MB, and judgment result packets to 0.5MB. Priority weights are allocated according to data importance: core feature data interactions have a priority weight of 0.9, prediction result data interactions have a priority weight of 0.7, and routine state data interactions have a priority weight of 0.3. Transmission time is determined through inter-module communication link test results, requiring a single transmission time not exceeding 50 milliseconds. The transmission link length is measured based on the physical installation location of the modules and converted to a unified unit. By calculating the sum of the products of the data packet size and priority weight of each data interaction, multiplying it by the communication protocol adaptation coefficient, and then dividing it by the sum of the transmission time and link loss terms, the data interaction efficiency value is obtained. This design provides a clear basis for optimizing the communication protocol and circuit connection structure by quantifying the interaction efficiency between modules, ensuring the high efficiency and stability of data interaction between modules, and guaranteeing the smooth operation of the overall screening and evaluation process.

[0045] Preferred, such as Figure 2 As shown, step S3 includes the following sub-steps: S31, based on the initialization parameters of the recycled pulp pollutant gradient enhancement decision model, the corrected multidimensional feature data is grouped according to pollutant categories to establish a preliminary mapping relationship between feature data and pollutant categories; S32, through the gradient enhancement iteration mechanism in the model, the contribution of different features to pollutant identification in each iteration is calculated, and a subset of calibrated features is selected based on the contribution ranking results; S33, each feature in the calibrated feature subset is assigned a dynamic weight value, and the weight value is dynamically adjusted based on the correlation strength between the feature and the pollutant category and the error change rate during the iteration process; S34, the adjusted calibrated features and corresponding weights are input into the decision layer of the model, and the pollutant category and content correlation features are accurately extracted through hierarchical decision rules to form feature extraction results.

[0046] Specifically, step S3 utilizes a gradient-enhanced decision model for recycled pulp contaminants to achieve accurate feature extraction and dynamic weight allocation, providing high-quality feature support for subsequent analysis. During implementation, initial model parameters are set: 120 decision trees, a maximum depth of 8 layers per tree, and a minimum of 15 samples per leaf node. In step S31, based on these initial parameters, the multidimensional feature data corrected in step S2 is explicitly grouped according to contaminant categories, establishing a preliminary mapping relationship between feature data and various contaminants to ensure targeted data classification. In step S32, the gradient enhancement iteration mechanism in the model is activated, with 80 iterations. In each iteration, the contribution of each feature to contaminant identification is precisely calculated. Based on the ranking of contributions from highest to lowest, the top 80% of features are selected to form a core feature subset, eliminating redundant information. In S33, each feature in the core feature subset is assigned a dynamic weight value with an initial range of 0.01-0.98. The weight values ​​are dynamically adjusted in real time based on the correlation strength between the feature and its corresponding pollutant category, as well as the rate of change of model error during iteration. Features with higher correlation strength and more significant error reduction have higher weight values. Finally, in S34, the adjusted core features and corresponding weight values ​​are input into the model's decision layer. Using hierarchical decision rules, the model progressively delves into basic features, derived features, and coupled correlation features to accurately extract features related to pollutant categories and concentrations, ultimately forming a structured feature extraction result.

[0047] Preferred, such as Figure 3 As shown, step S4 includes the following sub-steps: S41, based on the feature parameters extracted in S3, combined with the collection time and location information of the imported recycled pulp samples, a model input matrix including spatiotemporal dimension information is constructed, clarifying the feature types and spatiotemporal attributes corresponding to different elements in the matrix; S42, the input matrix is ​​imported into the pulp contaminant spatiotemporal distribution prediction model, and the potential mapping relationship between feature parameters and spatiotemporal dimensions is mined through the spatiotemporal correlation algorithm in the model; S43, based on the mapping relationship, a contaminant spatiotemporal distribution prediction equation is constructed, and the preliminary predicted values ​​of contaminant distribution at different spatiotemporal nodes are obtained by solving the equation; S44, the preliminary predicted values ​​are spatiotemporally consistent, and prediction results that do not conform to the spatiotemporal distribution law are eliminated to generate the final contaminant spatiotemporal distribution prediction result.

[0048] Specifically, step S4 accurately depicts the distribution patterns of pollutants at different spatiotemporal nodes using a pulp contaminant spatiotemporal distribution prediction model. Implementation is carried out systematically in stages. Step S31 first integrates the collection time and location information of imported recycled pulp samples based on the core feature parameters extracted in S3, constructing a 120×80 model input matrix. The matrix's rows correspond to 120 core feature parameters, and the columns correspond to 80 spatiotemporal nodes. The feature type and spatiotemporal attributes of each element in the matrix are clearly labeled to ensure the standardization of the input data. Step S42 imports the constructed input matrix into the pulp contaminant spatiotemporal distribution prediction model, activating the spatiotemporal correlation algorithm within the model. The time dimension correlation coefficient is set to 0.75, the spatial dimension correlation coefficient to 0.82, and the spatiotemporal cross-correlation coefficient to 0.68. The algorithm deeply mines the potential mapping relationships between feature parameters and spatiotemporal dimensions. Based on the mapping relationship obtained through mining, S43 uses the sliding window method (window size set to 10×10) to construct a pollutant spatiotemporal distribution prediction equation. Preliminary predictions of pollutant distribution at different spatiotemporal nodes are obtained through iterative solution. The convergence condition for the iteration is set as the difference between two adjacent iterations being less than 0.001, with a maximum of 50 iterations. S44 performs spatiotemporal consistency verification on the preliminary predictions. A threshold of ±25% for the rate of change of predictions between adjacent spatiotemporal nodes is set. Prediction results exceeding this threshold or not conforming to the spatiotemporal distribution pattern are discarded, and the mean of the predictions from three adjacent spatiotemporal nodes is used for replacement. Finally, a complete and reliable pollutant spatiotemporal distribution prediction result is generated.

[0049] Preferred, such as Figure 4 As shown, S5 includes the following sub-steps: S51, collecting the feature weight data output by the gradient enhancement decision model and the calibration parameters in the spatiotemporal distribution prediction results to determine the calibration parameter indicators for rapid screening and risk assessment of pollutants in imported recycled pulp; S52, standardizing the calibration parameter indicators, clarifying the value range and judgment criteria of different parameters, and constructing a multi-dimensional parameter system framework; S53, classifying different parameter indicators according to screening and risk assessment functions, calculating the correlation strength between different categories of parameters through a parameter coupling algorithm, and forming a parameter coupling matrix; S54, based on the parameter coupling matrix and the preset risk level classification rules, screening for pollutants and determining the risk level of imported recycled pulp samples to generate preliminary judgment results.

[0050] Specifically, step S5 constructs a multi-dimensional parameter system and conducts coupled analysis to achieve accurate screening and risk level determination of contaminants from imported recycled pulp. The implementation process strictly follows a step-by-step procedure. S51 first comprehensively collects the feature weight data output from the gradient enhancement decision model in step S3, as well as the key parameters such as peak distribution, mean distribution, and variance distribution from the spatiotemporal distribution prediction results in step S4. From these, 120 core parameter indicators for rapid screening and risk assessment of contaminants from imported recycled pulp are selected. S52 standardizes these core parameter indicators, using methods such as maximum / minimum value scaling and weight normalization to uniformly map all parameter values ​​to the 0-1.0 range. The value range and judgment criteria for each parameter are clearly defined; for example, a feature importance parameter value ≥0.6 indicates high importance, and a spatiotemporal distribution density parameter value ≥0.7 indicates a high-density region, thus constructing a clearly structured multi-dimensional parameter system framework. S53 divides the 120 core parameter indicators into two categories based on their screening and risk assessment functions: 70 parameters for screening and 50 parameters for risk assessment. A parameter coupling algorithm is used to calculate the correlation strength between parameters of different categories, and the degree of correlation is determined based on the Pearson correlation coefficient, forming a 120×120 parameter coupling matrix. S54, based on the constructed parameter coupling matrix and the preset low, medium, high, and extremely high risk level classification rules, calculates the comprehensive risk value (range 0-10.0) through multi-dimensional parameter coupling analysis. Based on the comprehensive risk value, the pollutant screening results and risk levels of the imported recycled pulp samples are determined, generating preliminary judgment results and providing a foundation for subsequent cross-module verification.

[0051] like Figure 5As shown, an AI-based rapid screening and risk assessment method for contaminants in imported recycled pulp is implemented through different units, including: a multi-dimensional feature data acquisition and transmission unit, which is connected to the physical structure of the detection chamber integrating near-infrared, X-ray, and fluorescence sensor probes. This unit acquires multi-dimensional feature data of the samples through a special fixing and scanning structure of the sensor probes and transmits the data to subsequent units via a preset communication interface; a detection data anomaly correction and processing unit, which is connected to the multi-dimensional feature data acquisition and transmission unit via a circuit. This unit calls an anomaly identification and correction algorithm to identify and correct anomalies in the received data and outputs the corrected data; and a contaminant feature extraction and weight allocation unit, which is communicatively connected to the detection data anomaly correction and processing unit. This unit runs a recycled pulp contaminant gradient enhancement decision model to correct the contaminants. The system performs feature extraction and weight allocation on the data to generate feature weight data. The pollutant spatiotemporal distribution prediction calculation unit interacts with the pollutant feature extraction and weight allocation unit through a preset communication protocol. Based on the pulp pollutant spatiotemporal distribution prediction model and feature weight data, it performs distribution prediction and outputs the prediction results. The multi-module collaborative screening and evaluation unit is connected to the pollutant spatiotemporal distribution prediction calculation unit, the intelligent detection module, the AI ​​analysis module, and the decision support module, respectively. It integrates the prediction results with data from different modules and completes the preliminary judgment through the rapid screening and risk assessment parameter system for imported recycled pulp pollutants. The final result output and storage unit is circuitically connected to the multi-module collaborative screening and evaluation unit. It receives and stores the final pollutant screening results and risk assessment report, and provides an external output interface.

[0052] This AI-based method for rapid screening and risk assessment of contaminants in imported recycled pulp utilizes a detection chamber integrating multiple types of sensors and a dedicated fixing and scanning structure to achieve comprehensive collection of multi-dimensional contaminant characteristic data from imported recycled pulp samples, ensuring the comprehensiveness and representativeness of the data sources. A dedicated outlier identification and correction algorithm precisely processes the collected data, effectively eliminating interfering information and providing high-quality data support for subsequent analysis. Simultaneously, a gradient-enhanced decision model, through hierarchical feature extraction and dynamic weight allocation, accurately captures core features related to contaminant category and content. Combined with a spatiotemporal distribution prediction model for in-depth analysis of contaminant distribution patterns, this achieves intelligent analysis throughout the entire process from feature extraction to distribution prediction, significantly improving the accuracy of contaminant identification.

[0053] This method constructs an efficient data interaction mechanism through standardized circuit connections and communication protocols for three major modules: intelligent detection, AI analysis, and decision support. This enables cross-module verification and optimization of preliminary judgment results, ensuring the reliability and consistency of the final output. The multi-functional units have clear divisions of labor and close connections, forming a complete technical chain from data acquisition, anomaly correction, feature extraction, distribution prediction to collaborative evaluation and result output. The process design is scientific and efficient. Furthermore, the method deeply integrates multiple core algorithms with hardware architecture to construct a dedicated parameter system. This not only meets the efficiency requirements of rapid screening of large-scale imported recycled pulp but also enables refined risk level assessment, adapting to practical pollution control application scenarios and possessing strong practicality and operability.

[0054] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0055] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An AI-based rapid screening and risk assessment method for contaminants in imported recycled pulp, characterized in that, Includes the following steps: S1, through the physical layout and connection structure of the detection chamber integrating near-infrared, X-ray and fluorescence sensing probes, collects multi-dimensional pollutant characteristic data of imported recycled pulp samples. The sensing probes perform full-area coverage data collection of the samples through a special fixing and scanning structure. S2, using an outlier detection and correction algorithm to identify and correct outliers in the collected multidimensional feature data, and locating outlier data ranges through correlation analysis between feature data and preset thresholds; S3, input the corrected feature data into the recycled pulp pollutant gradient enhancement decision model, and extract and assign weights to the pollutant category and content correlation features through the gradient enhancement iteration process in the model; S4. Construct the input matrix of the spatiotemporal distribution prediction model for pulp contaminants based on the extracted feature parameters, and generate the contaminant distribution prediction results through the spatiotemporal dimension correlation mapping mechanism in the model. S5, combining the feature weights output by the gradient enhancement decision model with the spatiotemporal distribution prediction results, constructs a parameter system for rapid screening and risk assessment of pollutants in imported recycled pulp, and generates preliminary screening and risk level determination results through multi-dimensional parameter coupling analysis; The S6 uses circuit connections and preset communication protocols between the intelligent detection module, AI analysis module, and decision support module to interact and collaborate on data, perform cross-module verification and optimization of preliminary judgment results, and output the final pollutant screening results and risk assessment report.

2. The AI-based rapid screening and risk assessment method for contaminants in imported recycled pulp according to claim 1, characterized in that, The feature weight allocation expression for the recycled pulp contaminant gradient enhancement decision model is as follows: ， in, This represents the weight of the i-th feature with respect to the j-th type of pollutant. Represents the loss function. Indicates the preceding The output of the model in each iteration, Indicates the step size parameter. This represents the output of the base learner corresponding to the j-th type of pollutant. This represents the k-th feature value of the i-th sample. This represents the characteristic threshold of the j-th type of pollutant. Represents the feature correlation function. This represents the regularization parameter.

3. The AI-based rapid screening and risk assessment method for contaminants in imported recycled pulp according to claim 1, characterized in that, The distribution prediction expression of the spatiotemporal distribution prediction model for pulp contaminants is as follows: ， in, Represents the coordinates at time t The probability of pollutant distribution at a location, Represents the distribution coefficient. Let represent the contribution coefficient of the p-th feature. Let represent the mapping function of the p-th feature at time t. This represents the p-th input feature parameter. The time characteristic factor at time t. Indicates the spatial attenuation coefficient. This represents the spatial coordinates corresponding to the p-th feature. This indicates the error correction term.

4. The AI-based rapid screening and risk assessment method for contaminants in imported recycled pulp according to claim 1, characterized in that, The outlier correction expression for the outlier identification and correction algorithm in the detected data is: ， in, This represents the corrected k-th feature data. This represents the original k-th feature data. Represents the mean of the feature data. This represents the anomaly correction function. Indicates the standard deviation of the feature data. This indicates the number of neighboring data points involved in the correction. This indicates the correction smoothing coefficient.

5. The AI-based rapid screening and risk assessment method for contaminants in imported recycled pulp according to claim 1, characterized in that, The risk level determination expression for the rapid screening and risk assessment of contaminants in imported recycled pulp is as follows: ， in, Indicates the risk level value. This represents the risk adjustment factor. This represents the weight of pollutant class j. This represents the screening result value for pollutant category j. Indicates the number of pollutant categories. This represents the risk amplification factor. This represents the spatiotemporal distribution density of pollutant type j.

6. The AI-based rapid screening and risk assessment method for contaminants in imported recycled pulp according to claim 1, characterized in that, The data interaction efficiency expression of the intelligent detection module, AI analysis module, and decision support module is as follows: ， in, This represents the data interaction efficiency value. Indicates the communication protocol adaptation coefficient. This represents the size of the data packet in the i-th data exchange. This represents the priority weight of the i-th data interaction. Indicates the number of data interactions. This represents the transmission time of the i-th data interaction. Indicates the link loss coefficient. This represents the length of the transmission link for the i-th data interaction.

7. The AI-based rapid screening and risk assessment method for contaminants in imported recycled pulp according to claim 1, characterized in that, S3 includes the following steps: S31, Based on the initialization parameters of the recycled pulp pollutant gradient enhancement decision model, the corrected multidimensional feature data are grouped according to pollutant categories to establish a preliminary mapping relationship between feature data and pollutant categories; S32, through the gradient boosting iterative mechanism in the model, calculates the contribution of different features to pollutant identification in each iteration, and selects a subset of calibrated features based on the contribution ranking results; S33, assign a dynamic weight value to each feature in the calibration feature subset. The weight value is dynamically adjusted based on the correlation strength between the feature and the pollutant category and the error change rate during the iteration process. S34. The adjusted calibration features and corresponding weights are input into the decision layer of the model. The hierarchical decision rules are used to accurately extract the correlation features between pollutant categories and contents, forming the feature extraction results.

8. The AI-based rapid screening and risk assessment method for contaminants in imported recycled pulp according to claim 1, characterized in that, S4 includes the following steps: S41, based on the feature parameters extracted in S3, combined with the collection time and location information of the imported recycled pulp samples, constructs a model input matrix that includes spatiotemporal dimension information, and clarifies the feature types and spatiotemporal attributes corresponding to different elements in the matrix; S42, import the input matrix into the spatiotemporal distribution prediction model of pulp contaminants, and use the spatiotemporal correlation algorithm in the model to mine the potential mapping relationship between feature parameters and spatiotemporal dimensions; S43, Construct a spatiotemporal distribution prediction equation for pollutants based on the mapping relationship, and obtain preliminary prediction values ​​of pollutant distribution at different spatiotemporal nodes by solving the equation; S44. Perform spatiotemporal consistency verification on the preliminary prediction values, eliminate prediction results that do not conform to the spatiotemporal distribution pattern, and generate the final spatiotemporal distribution prediction results of pollutants.

9. The AI-based rapid screening and risk assessment method for contaminants in imported recycled pulp according to claim 1, characterized in that, S5 includes the following steps: S51, collect the feature weight data and spatiotemporal distribution prediction results from the gradient enhancement decision model output to determine the calibration parameter indicators for rapid screening and risk assessment of pollutants in imported recycled pulp; S52 standardizes the calibration parameters, clarifies the value range and judgment criteria of different parameters, and constructs a multi-dimensional parameter system framework. S53 classifies different parameter indicators according to screening and risk assessment functions, and calculates the correlation strength between different categories of parameters through parameter coupling algorithm to form parameter coupling matrix; S54, based on the parameter coupling matrix and the preset risk level classification rules, performs pollutant screening and risk level determination on imported recycled pulp samples, and generates preliminary judgment results.

10. The AI-based rapid screening and risk assessment method for contaminants in imported recycled pulp according to any one of claims 1-9, characterized in that, This method is implemented through different units, including: The multidimensional feature data acquisition and transmission unit is connected to the physical structure of the detection chamber that integrates near-infrared, X-ray, and fluorescence sensing probes. It acquires multidimensional feature data of the sample through the special fixing and scanning structure of the sensing probes and transmits the data to the subsequent units through a preset communication interface. The detection data anomaly correction processing unit is connected to the multi-dimensional feature data acquisition and transmission unit via a circuit. It calls the detection data anomaly identification and correction algorithm to identify and correct anomalies in the received data and outputs the corrected data. The pollutant feature extraction and weight allocation unit communicates with the detection data anomaly correction and processing unit, runs the recycled pulp pollutant gradient enhancement decision model to extract features and allocate weights to the correction data, and generates feature weight data. The pollutant spatiotemporal distribution prediction calculation unit interacts with the pollutant feature extraction and weight allocation unit through a preset communication protocol. Based on the pulp pollutant spatiotemporal distribution prediction model and feature weight data, it performs distribution prediction and outputs the prediction results. The multi-module collaborative screening and assessment unit is connected to the pollutant spatiotemporal distribution prediction and calculation unit, the intelligent detection module, the AI ​​analysis module, and the decision support module, respectively. It integrates the prediction results with data from different modules and completes the preliminary judgment through the rapid screening and risk assessment parameter system for pollutants from imported recycled pulp. The final result output and storage unit is connected to the multi-module collaborative screening and evaluation unit circuitry to receive and store the final pollutant screening results and risk assessment report, while also providing an external output interface.

Images