Online monitoring method for product quality in solid waste recycling process

Abstract

The invention provides an online monitoring method for product quality in a solid waste recycling process, and belongs to the technical field of solid waste recycling. A parameter acquisition system is constructed to synchronously acquire a multi-parameter matrix, a sampling rate determination function is applied to realize dynamic adjustment of a sampling frequency, and a solid waste component migration equation is utilized to analyze a substance conversion rule; a spectrum fusion optimization function is used for processing various spectrum data, an FWM model is used for achieving product quality multi-dimensional evaluation, the heavy metal content is monitored in combination with an electrochemical sensor network, a comprehensive quality evaluation result is finally formed, an updating sampling rate is output through the FWM model, a function weight coefficient is determined, online monitoring closed-loop control is formed, and the comprehensive quality evaluation result is obtained. Real-time, accurate and comprehensive monitoring of the quality of solid waste recycling products is realized, the problem of key process transition point data missing caused by fixed frequency sampling is effectively solved, and reliable technical support is provided for optimization of the solid waste recycling process.

Description

Technical Field

[0001] The present invention belongs to the technical field of solid waste resource utilization, and in particular relates to an online monitoring method for product quality in a solid waste resource utilization process. Background Art

[0002] Solid waste resource utilization is an important approach to achieving solid waste reduction and resource recycling, and is of great significance for promoting green development and building an ecological civilization. In traditional solid waste resource utilization processes, product quality monitoring primarily relies on periodic manual sampling and analysis. With the advancement of automated control technology, some systems have adopted online monitoring equipment based on fixed-frequency sampling, such as spectrometers and gas chromatographs, enabling real-time collection of product quality parameters. However, most existing solid waste resource utilization product quality monitoring systems utilize a constant sampling frequency design, collecting data at pre-set fixed time intervals. This monitoring approach may be sufficient for industrial solid wastes with stable components and smooth reaction processes, but it exhibits significant shortcomings when dealing with complex and heterogeneous solid wastes, such as municipal solid waste and hazardous waste. In particular, product quality often exhibits dynamic changes during the solid waste resource utilization process due to factors such as fluctuations in raw material composition, changes in reaction conditions, and adjustments to process parameters. Fixed-frequency sampling strategies cannot flexibly adjust sampling intervals based on the dynamic characteristics of the treatment process, resulting in redundant sampling and waste of resources during the stable process phase. At critical transition points, such as sudden temperature changes, catalyst deactivation, or drastic changes in input material composition, untimely sampling may result in the loss of important data, making it impossible to accurately capture quality fluctuations, affecting the accuracy and timeliness of product quality assessments. In other words, the existing technology for monitoring product quality in the solid waste resource recovery process has a fixed sampling frequency, which can lead to data loss at key process transition points, thereby affecting the accuracy of product quality monitoring. Summary of the Invention

[0003] In view of this, the present invention provides an online monitoring method for product quality in the solid waste resource utilization process, which can solve the technical problem in the existing technology that the sampling frequency of product quality monitoring in the solid waste resource utilization process is fixed, resulting in data missing at key process transition points.

[0004] The present invention is implemented as follows: The present invention provides an online monitoring method for product quality in a solid waste resource utilization process, including: constructing a parameter acquisition system to acquire a multi-parameter matrix; applying a sampling rate determination function to calculate a real-time sampling rate, wherein the sampling rate determination function is jointly determined based on the variable component of the multi-parameter matrix, the processing temperature, the material flow rate, the solid waste component heterogeneity index and the FWM model output result; using the solid waste component migration equation to analyze the material transformation law; using a spectral fusion optimization function to process spectral data; using the FWM model to realize product quality assessment; monitoring the heavy metal content through an electrochemical sensor network; updating the weight coefficient in the sampling rate determination function according to the FWM model output result to form an online monitoring closed-loop control.

[0005] Among them, the step of constructing a parameter acquisition system to collect a multi-parameter matrix is specifically to initialize the sampling rate to 5 times per minute, and synchronously collect spectral data and physical and chemical parameters in the solid waste treatment process to form a multi-parameter matrix.

[0006] Among them, the multi-parameter matrix refers to a data set consisting of real-time monitoring parameters such as temperature, pressure, humidity, redox potential, pH value and flow rate during the solid waste treatment process.

[0007] Among them, the sampling rate determination function specifically calculates the weight coefficient by calculating the change rate and standard deviation of each parameter in the multi-parameter matrix. When the weight coefficient exceeds the preset threshold, the sampling frequency is increased, otherwise the sampling frequency is reduced, and the real-time sampling rate is dynamically adjusted in combination with the output results of the FWM model.

[0008] When the FWM model has not yet output a result, the sampling rate determination function sets the weight coefficient of the FWM model output result to 0.

[0009] Among them, the solid waste component heterogeneity index is a quantitative indicator characterizing the heterogeneity of solid waste, which is calculated by analyzing the fluctuation range of solid waste components, the degree of uneven distribution and the intensity of interaction between components.

[0010] Among them, the solid waste component migration equation specifically expresses the physical and chemical behaviors of pollutants such as phase change, adsorption and desorption, oxidation-reduction and complexation precipitation during the solid waste resource utilization process. By solving the solid waste component migration equation group, the concentration distribution of each component in the final product is obtained.

[0011] Among them, the input parameters of the solid waste component migration equation include temperature field distribution, redox potential, reactant concentration, catalyst activity and pH value in the multi-parameter matrix, and the output is the predicted value of the target component concentration.

[0012] The spectral fusion optimization function is used to perform noise reduction, baseline correction and feature extraction on Raman spectrum, near-infrared spectrum and X-ray fluorescence spectrum data, and realizes the optimized fusion of different spectral information through weight distribution.

[0013] The input parameters of the spectrum fusion optimization function include wavelength range, signal-to-noise ratio threshold, characteristic peak position, peak intensity ratio and baseline offset, and the output is a spectrum feature vector.

[0014] The FWM model is structured as a multi-layer encoder-decoder architecture, comprising a self-attention mechanism layer, a residual connection layer, and a normalization layer. The encoder layer extracts deep features from spectral feature vectors, target component concentration predictions, and a multi-parameter matrix. The decoder layer reconstructs product quality evaluation indicators, while the intermediate connection layer utilizes a cross-attention mechanism to facilitate information exchange between data from different modalities. The input parameters of the FWM model include the spectral feature vectors output by the spectral fusion optimization function, the multi-parameter matrix, and the target component concentration predictions output by the solid waste component migration equation. The self-attention mechanism is a multi-head attention structure, and its weight distribution is determined by a dynamic scaling function that considers the current sampling rate, parameter change trends, data stability index, and spectral feature similarity. The electrochemical sensor network is a system that monitors heavy metal ion concentration in real time by measuring changes in redox currents using selective electrodes placed at key nodes in the processing flow. The comprehensive quality evaluation result is calculated by normalizing the heavy metal ion content data measured by the electrochemical sensor network with the spectral data processed by the spectral fusion optimization function, and then establishing a hierarchical weighting model to calculate a comprehensive product quality index, enabling dynamic assessment of resource-based product quality.

[0015] The present invention realizes real-time adjustment of the sampling frequency according to parameters such as the variable component of the multi-parameter matrix, processing temperature, material flow rate and solid waste component heterogeneity index by constructing a sampling rate determination function, appropriately reducing the sampling frequency in the process stability stage and automatically increasing the sampling density at parameter fluctuations or process transition points. The present invention combines the solid waste component migration equation and spectral fusion optimization technology to establish a FWM model as the core algorithm. The model, through a multi-layer encoder-decoder architecture and self-attention mechanism, can deeply explore the intrinsic correlation between multiple parameters and realize effective processing and fusion of data at different sampling frequencies. In particular, the dynamic scaling function and cross-attention mechanism enable the model to automatically adjust the attention weight according to data stability and parameter change trends, ensuring that key information can be obtained even in the low-frequency sampling stage. Through the intelligent dynamic adjustment of the sampling rate and the deep data fusion of the FWM model, the present invention effectively solves the problem of missing data at key process transition points caused by fixed-frequency sampling, realizes full-process and accurate monitoring of the quality of solid waste resource products, improves monitoring efficiency and reduces system resource consumption, and provides reliable technical support for quality control and process optimization of solid waste resource processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0018] like Figure 1 FIG. 1 is a flow chart of a method for online monitoring of product quality in a solid waste resource recovery process provided by the present invention. The method comprises the following steps:

[0019] S01. Build a parameter acquisition system with an initial sampling rate of 5 times per minute, synchronously collect spectral data and physical and chemical parameters in the solid waste treatment process, and form a multi-parameter matrix;

[0020] S02. Calculate a real-time sampling rate using a sampling rate determination function, wherein the sampling rate determination function is determined based on the variable component of the multi-parameter matrix, the processing temperature, the material flow rate, the solid waste component heterogeneity index, and the output of the FWM model, and includes an initial sampling rate and a real-time variable sampling rate. When the FWM model has not yet output a result, the sampling rate determination function sets the weight coefficient of the FWM model output result to 0;

[0021] S03. Analyzing the transformation patterns of substances in the resource recovery process through a solid waste component migration equation to predict the content of key components in the product, wherein the input parameters of the solid waste component migration equation include the temperature field distribution, redox potential, reactant concentration, catalyst activity, and pH value in the multi-parameter matrix;

[0022] S04. Processing Raman spectrum, near-infrared spectrum, and X-ray fluorescence spectrum data using a spectrum fusion optimization function, wherein input parameters of the spectrum fusion optimization function include wavelength range, signal-to-noise ratio threshold, characteristic peak position, peak intensity ratio, and baseline offset;

[0023] S05. Implement a multidimensional product quality assessment using the FWM model, where the input parameters of the FWM model include the spectral feature vector output by the spectral fusion optimization function, the multi-parameter matrix, and the target component concentration prediction value output by the solid waste component migration equation, wherein the dimension of the attention mechanism parameter matrix is determined by the dimension of the multi-parameter matrix, the sampling rate, and the solid waste component heterogeneity index;

[0024] S06. Real-time monitoring of the heavy metal ion content in the solid waste resource product based on the electrochemical sensor network, and fusing the spectral data processed by the spectral fusion optimization function to form a comprehensive quality evaluation result;

[0025] S07, updating the weight coefficient in the sampling rate determination function according to the output result of the FWM model, returning to step S02 to recalculate the real-time sampling rate, and forming an online monitoring closed-loop control;

[0026] The multi-parameter matrix refers to a data set consisting of real-time monitoring parameters such as temperature, pressure, humidity, redox potential, pH value and flow rate during the solid waste treatment process;

[0027] The sampling rate determination function specifically calculates a weight coefficient based on the change rate and standard deviation of each parameter in the multi-parameter matrix, increases the sampling frequency when the weight coefficient exceeds a preset threshold, and decreases the sampling frequency otherwise, and dynamically adjusts the real-time sampling rate based on the output result of the FWM model;

[0028] The solid waste component heterogeneity index is a quantitative indicator of solid waste heterogeneity calculated by analyzing the fluctuation range of solid waste components, the degree of uneven distribution and the intensity of interaction between components.

[0029] Among them, the solid waste component migration equation specifically expresses the physicochemical behaviors of pollutants in the process of solid waste resource utilization, such as phase change transformation, adsorption and desorption, redox and complex precipitation. By solving the solid waste component migration equation group, the concentration distribution of each component in the final product is obtained, providing a theoretical basis for quality assessment. The input of the solid waste component migration equation includes the reactant concentration, temperature field distribution, redox potential, catalyst activity and pH value in the multi-parameter matrix, and the output is the predicted concentration value of the target component. The predicted concentration value of the target component serves as one of the input parameters of the FWM model and a component of the comprehensive quality evaluation result;

[0030] The spectral fusion optimization function is used to perform noise reduction, baseline correction, and feature extraction on the raw data obtained by various spectral analysis techniques, and optimize the fusion of different spectral information through weight distribution to improve detection accuracy. The input of the spectral fusion optimization function includes the wavelength range, signal-to-noise ratio threshold, characteristic peak position, peak intensity ratio, and baseline offset of each spectrum, and the output is the spectral feature vector, which serves as one of the input parameters of the FWM model.

[0031] The specific structure of the FWM model is a multi-layer encoder-decoder architecture, including a self-attention mechanism layer, a residual connection layer, and a normalization layer. The encoder part is responsible for extracting the spectral feature vector, the concentration prediction value of the target component, and the deep features in the multi-parameter matrix. The decoder part is responsible for reconstructing the product quality evaluation index. The intermediate connection part adopts a cross-attention mechanism to realize information interaction of different modal data. The output result of the FWM model is used as one of the input parameters of the sampling rate determination function.

[0032] The steps for establishing the training data set for the FWM model specifically include collecting historical monitoring data from the resource recovery process of solid waste from different sources, labeling the corresponding product quality indicators for each set of data, using data augmentation technology to simulate parameter fluctuations under different working conditions, constructing standard data pairs containing input parameter matrices and corresponding quality evaluation results, and randomly dividing them into training sets, validation sets, and test sets;

[0033] The FWM model training steps specifically include first conducting unsupervised training on large-scale solid waste treatment historical data to learn the parameter distribution law and the intrinsic structure of the data, then conducting supervised fine-tuning training based on the solid waste type, optimizing the model parameters using an adaptive learning rate and weight reduction strategy, and evaluating the model generalization ability through cross-validation. Finally, the model parameters with the best performance on the validation set are selected;

[0034] The electrochemical sensor network refers to a system that measures changes in redox current by arranging selective electrodes at key nodes in the processing flow to achieve real-time monitoring of heavy metal ion concentrations. The measurement results of the electrochemical sensor network serve as an integral part of the comprehensive quality evaluation results.

[0035] Among them, the comprehensive quality evaluation result is obtained by standardizing the heavy metal ion content data measured by the electrochemical sensor network and the spectral data processed by the spectral fusion optimization function, establishing a hierarchical weight model to calculate the product quality comprehensive index, and realizing a dynamic evaluation of the quality of resource-based products. The product quality comprehensive index and the concentration predicted value of the target component together constitute the comprehensive quality evaluation result.

[0036] The self-attention mechanism is a multi-head attention structure, and its weight distribution is determined by a dynamic scaling function. The dynamic scaling function takes into account the current sampling rate, parameter change trend, data stability index, and spectral feature similarity. The final attention weight distribution is obtained by calculating the dot product of the query matrix, key matrix, and value matrix and performing softmax normalization.

[0037] The specific implementation of the above steps is described in detail below. The specific implementation of step S01 involves constructing a parameter acquisition system that synchronously collects various spectral data and physicochemical parameters from the solid waste treatment process at an initial frequency of five times per minute. In this implementation, spectral probes and physicochemical parameter sensors, including Raman, near-infrared, and X-ray fluorescence probes, as well as temperature, pressure, humidity, redox potential, pH, and flow rate sensors, are installed at key nodes in the solid waste treatment production line. An industrial-grade data acquisition network is then established, employing time-division multiplexing to ensure synchronous data acquisition from each sensor, while industrial-grade fieldbus technology is used to enable real-time data transmission. A real-time database is then used to store the collected information, forming a multi-parameter matrix containing parameters such as temperature, pressure, humidity, redox potential, pH, and flow rate. Finally, a data preprocessing unit filters and denoises the raw data, eliminating electromagnetic interference and outliers, and achieving data standardization. The purpose of this step is to ensure that the monitoring system can promptly and accurately capture changes in key parameters during the solid waste treatment process, providing basic data support for subsequent analysis.

[0038] The specific implementation of step S02 is to calculate the real-time sampling rate using a sampling rate determination function. This function is determined based on the variable component of the multi-parameter matrix, the processing temperature, the material flow rate, the solid waste component heterogeneity index, and the output of the FWM model. In specific implementation, the rate of change and standard deviation of each parameter in the multi-parameter matrix within a specified time window are first calculated. When the parameter change rate exceeds 12% or the standard deviation exceeds a set threshold (the temperature parameter threshold is ±3°C, the pH parameter threshold is ±0.4, and the redox potential parameter threshold is ±25mV), the weight coefficient of that parameter is increased by 0.2; otherwise, it is reduced by 0.1. The temperature impact factor is then calculated based on the deviation of the processing temperature from the theoretical optimal reaction temperature. When the temperature deviation exceeds ±15°C, the sampling rate is appropriately increased. The flow rate impact factor is then determined based on the ratio of the material flow rate to the standard flow rate. When the flow rate exceeds 120% or is less than 80% of the standard flow rate, the sampling rate is increased by at least 30%. The basic sampling rate is then adjusted according to the heterogeneity index of solid waste components. When the heterogeneity index is greater than 0.75, the sampling rate is increased to at least 8 times per minute. Finally, the quality fluctuation prediction value output by the FWM model is comprehensively considered. When the prediction value shows an increased risk of quality fluctuation, the sampling rate is further increased. The final real-time sampling rate is calculated by multiplying the initial sampling rate by the weighted sum of each influencing factor. When the FWM model has not yet output a result, its weight coefficient is set to 0. The purpose of this step is to achieve intelligent allocation of monitoring resources, increase the monitoring frequency when the process fluctuation is large or the product quality risk increases, ensure monitoring accuracy, and reduce the monitoring frequency when the process is stable to save system resources.

[0039] The specific implementation of step S03 is to analyze the transformation law of substances in the resource utilization process through the solid waste component migration equation and predict the content of key components in the product. In the specific implementation, first construct a migration equation group that describes the physicochemical behavior of pollutants in the solid waste resource utilization process, such as phase change, adsorption and desorption, redox and complex precipitation. This equation group covers the three aspects of mass conservation, momentum conservation and energy conservation. Then, using the temperature field distribution, redox potential, reactant concentration, catalyst activity and pH value in the multi-parameter matrix as input parameters, the solid waste component migration equation group is discretized and solved by finite element method. Then, the adaptive step-size Runge-Kutta integration algorithm is used for numerical solution to calculate the concentration distribution of pollutants under different working conditions. Then, combined with the thermodynamic equilibrium constant and reaction kinetic parameters, the speciation distribution of heavy metal elements under different pH and redox conditions is predicted. Finally, the concentration prediction value of the target component is output as one of the input parameters of the FWM model and as a component of the comprehensive quality evaluation results. The error in predicted concentrations of typical metal ions, such as lead, cadmium, mercury, and arsenic, is controlled within ±7%, and the error in predicted concentrations of organic pollutants is controlled within ±12%. This step aims to accurately predict the transformation behavior of each component in the solid waste recycling process and the concentration distribution in the final product through physical and chemical theoretical models, providing a theoretical basis for quality assessment.

[0040] The specific implementation of step S04 is to use the spectrum fusion optimization function to process Raman spectrum, near-infrared spectrum and X-ray fluorescence spectrum data. In the specific implementation, each spectrum data is first pre-processed, including denoising, baseline correction and normalization. Among them, the Raman spectrum uses wavelet transform to denoise, and the signal-to-noise ratio threshold is set to 4.5; the near-infrared spectrum uses standard normal variable transformation to correct baseline drift; and the X-ray fluorescence spectrum uses Compton scattering peak normalization method. Then perform feature extraction, and the Raman spectrum focuses on 400-1800cm -1 Characteristic peaks within the wavenumber range are selected for near-infrared spectroscopy. Absorption peaks within the wavelength range of 1100 to 2500 nm are selected, while X-ray fluorescence spectroscopy focuses on the positions and intensity ratios of characteristic peaks of typical elements. Principal component analysis is then used for dimensionality reduction, retaining principal components that explain 95% of the variance. A multispectral fusion model is then constructed, and a feature-level fusion algorithm based on a convolutional neural network is used to optimize the fusion of different spectral information. Finally, a spectral feature vector is output as one of the input parameters of the FWM model. The inputs to this spectral fusion optimization function include the wavelength range, signal-to-noise ratio threshold, characteristic peak position, peak intensity ratio, and baseline offset of each spectrum. The purpose of this step is to fully leverage the advantages of different spectral technologies, extract complementary information, improve detection accuracy, and provide multidimensional spectral features for product quality assessment.

[0041] The specific implementation of step S05 is to use the FWM model to achieve multidimensional product quality assessment. In this implementation, the spectral feature vectors and multi-parameter matrices output by the spectral fusion optimization function, as well as the target component concentration predictions output by the solid waste component migration equation, are first used as input parameters for the FWM model. The FWM model's encoder component then uses a multi-layer perceptron and self-attention mechanism to extract deep feature representations of the input data. The self-attention mechanism adopts a multi-head attention structure with 8 heads and 512 hidden layer dimensions. In the intermediate connection component, a cross-attention mechanism is used to enable information exchange between different modal data. The attention weights are determined by a dynamic scaling function that comprehensively considers the current sampling rate, parameter change trends, data stability index, and spectral feature similarity. The FWM model's decoder component then uses residual connections and layer normalization techniques to reconstruct product quality evaluation indicators. The decoder layer is set to 6, each layer comprising a self-attention sublayer, a cross-attention sublayer, and a feedforward neural network sublayer. Finally, the multidimensional quality assessment results of the product are output, including purity, safety, stability, and applicability indicators. The purpose of this step is to integrate multi-source heterogeneous data through deep learning models to achieve a comprehensive and accurate assessment of the quality of solid waste resource products and provide a decision-making basis for the production process.

[0042] The specific implementation of step S06 involves real-time monitoring of heavy metal ion content in solid waste recycling products using an electrochemical sensor network. This is then combined with spectral data processed by a spectral fusion optimization function to generate a comprehensive quality assessment result. In this implementation, a selective electrode array, comprising an ion-selective electrode, a modified electrode, and a reference electrode, is deployed at key nodes in the processing flow to form an electrochemical sensor network. Changes in the redox current of each electrode are then measured, and heavy metal ion concentrations are quantitatively analyzed using voltammetry and potentiometry, focusing on toxic heavy metals such as lead, cadmium, mercury, arsenic, and chromium, with a detection limit of 1 μg / L. A multivariate correction algorithm is then employed to eliminate cross-interference between ions and improve measurement accuracy, achieving measurement accuracy within ±5% for typical heavy metal ions. The heavy metal ion content data measured by the electrochemical sensor network and the spectral data processed by the spectral fusion optimization function are then normalized to eliminate dimensional differences. Finally, a hierarchical weighting model is established to calculate a comprehensive product quality index. This model assigns weights to each indicator based on its importance in different application scenarios, generating a comprehensive quality assessment result. The purpose of this step is to achieve accurate monitoring of heavy metal pollutants in products through the complementary combination of electrochemical methods and spectroscopic methods, thereby ensuring the environmental safety of resource-based products.

[0043] The specific implementation of step S07 involves updating the weight coefficients in the sampling rate determination function based on the output of the FWM model, returning to step S02 to recalculate the real-time sampling rate, and establishing an online monitoring closed-loop control. In specific implementation, the product quality assessment results output by the FWM model are first analyzed to calculate the deviation of each quality indicator from the target value. Next, the weight coefficients of each parameter in the sampling rate determination function are adjusted based on the magnitude and trend of the deviation. When the quality indicator deviation increases, the weight coefficient of the relevant parameter is increased; when the quality indicator stabilizes within the target range, the weight coefficient is appropriately reduced. The weight coefficient adjustment step size is then set, typically between 0.05 and 0.15, and dynamically adjusted based on the magnitude of quality fluctuations. The updated weight coefficients are then applied to recalculate the real-time sampling rate and apply the new sampling rate to the parameter acquisition system. Finally, a closed-loop feedback mechanism for quality monitoring and sampling control is established, achieving intelligent allocation and optimal utilization of monitoring resources. The purpose of this step is to enable the monitoring system to adaptively adjust its monitoring strategy based on product quality status through a closed-loop control strategy, improving monitoring efficiency and ensuring monitoring quality.

[0044] The FWM model employs a multi-layer encoder-decoder architecture, including self-attention layers, residual connections, and normalization layers. The detailed structural design is as follows: The encoder consists of six stacked encoder layers, each of which includes a multi-head self-attention sublayer and a feedforward neural network sublayer. Each sublayer utilizes residual connections and layer normalization. The multi-head self-attention mechanism has eight heads and a hidden layer dimension of 512. The feedforward neural network utilizes a two-layer fully connected network with an intermediate layer dimension of 2048 and a GELU activation function. The decoder also consists of six stacked decoder layers, each of which includes a self-attention sublayer, a cross-attention sublayer, and a feedforward neural network sublayer. The cross-attention sublayer is responsible for associating the encoder output with the current decoder state, enabling information exchange between data from different modalities. The intermediate connection utilizes a cross-attention mechanism. Attention weights are calculated by calculating the dot product of the query matrix, key matrix, and value matrix, followed by softmax normalization, to effectively integrate data from different modalities. The model's output layer is a fully connected layer that maps the decoder output to the space of product quality evaluation metrics. The self-attention mechanism adopts a multi-head attention structure, and its weight distribution is determined by a dynamic scaling function. This function takes into account the current sampling rate, parameter change trend, data stability index and spectral feature similarity. The final attention weight distribution is obtained by calculating the dot product of the query matrix, key matrix and value matrix and performing softmax normalization.

[0045] The detailed steps for constructing the training dataset for the FWM model are as follows: First, historical monitoring data from the resource recovery process of solid waste from various sources, including municipal solid waste, industrial solid waste, and agricultural waste, were collected. The data covered at least 100 complete production cycles. Each data set was then cleaned and annotated. The cleaning process included outlier detection and processing, missing value interpolation, and data consistency verification. The annotation process included labeling corresponding product quality indicators based on laboratory analysis results, such as purity, hazardous substance content, and stability. Data augmentation techniques were then used to simulate parameter fluctuations under different operating conditions. These techniques included adding Gaussian noise to simulate sensor errors, random scaling to simulate parameter fluctuations, and time shifting to simulate process delays, expanding the dataset size to 3–5 times the original data size. A standard data pair consisting of an input parameter matrix and corresponding quality evaluation results was then constructed. The input parameter matrix had a dimension of time step × number of parameters, while the quality evaluation results had a dimension of number of indicators. Finally, the dataset was partitioned into training, validation, and test sets using stratified random sampling, with a ratio of 70%, 15%, and 15%, respectively, ensuring a consistent distribution of data for different solid waste types within each subset. The FWM model training process first conducts unsupervised pre-training on large-scale solid waste treatment historical data, using a masked autoencoder structure to learn the parameter distribution law and the intrinsic structure of the data. Then, supervised fine-tuning training is performed for the current solid waste type, and the loss function comprehensively considers the mean square error and weighted cross entropy. An adaptive learning rate strategy is used during training, with the initial learning rate set to 1×10 -4 When the validation set loss does not improve for 5 consecutive epochs, the learning rate is reduced to 0.8 times the original value; at the same time, the weight reduction regularization technique is introduced, and the reduction coefficient is set to 5×10 -5 , to prevent model overfitting; evaluate the generalization ability of the model through 5-fold cross-validation, and finally select the model parameters with the best performance in the validation set as the final model.

[0046] It's important to note that the sampling rate determination function incorporates the FWM model output as a parameter, dynamically adjusting the sampling strategy based on product quality assessment results, thus implementing a two-way feedback mechanism between data collection and quality assessment. When the system detects fluctuations in product quality, the FWM model outputs a higher quality coefficient of variation, and the sampling rate determination function increases the sampling frequency accordingly, ensuring sufficient and intensive data acquisition during critical quality transitions. This adaptive sampling mechanism, based on quality assessment results, overcomes the limitations of traditional fixed-interval sampling and equips the monitoring system with intelligent, quality-oriented decision-making capabilities.

[0047] When the FWM model has not yet output results, its weight coefficient is set to 0, ensuring rationality and smooth transition during the system's startup phase. In the initial monitoring phase, the system relies on basic parameters such as the variable components of the multi-parameter matrix and the processing temperature to determine the sampling frequency, avoiding the failure of the sampling strategy due to a lack of model output. As the FWM model begins to produce evaluation results, the system gradually increases its weight influence, achieving a smooth transition from pure parameter-driven to quality assessment-driven. This progressive weight adjustment mechanism ensures the stability and continuity of system operation and adapts to the parameter learning and model optimization stages of the solid waste resource utilization process.

[0048] The multi-head attention mechanism uses a dynamic scaling function to determine weight distribution, enabling the FWM model to automatically adjust its focus on different features based on real-time operating conditions. This dynamic scaling function comprehensively considers the current sampling rate, parameter change trends, data stability index, and spectral feature similarity. At high sampling rates, it prioritizes short-term patterns, while at low sampling rates, it emphasizes long-term trends. This adaptive attention allocation mechanism addresses the issue of uneven information density at different sampling frequencies and improves the model's ability to handle sparsely sampled data.

[0049] The above three key technologies form a complete closed-loop adaptive system. The sampling rate determination function receives the evaluation results of the FWM model to adjust the sampling strategy, while the FWM model adapts to the data characteristics of different sampling frequencies through a dynamically scaled multi-head attention mechanism. The two promote and optimize each other. In the early stage of system operation, a smooth start is ensured by setting the weight coefficient to zero; as the operating time increases, the system gradually establishes a closed-loop control of sampling-analysis-evaluation-feedback, optimizing system resource utilization while ensuring monitoring quality. This collaborative mechanism enables the monitoring system to automatically evolve according to the dynamic characteristics of the solid waste treatment process, forming the optimal monitoring strategy that adapts to the current process, solving the problem that traditional fixed-frequency sampling cannot cope with process fluctuations.

[0050] Specifically, the principle of the present invention is: the core technical principle of the present invention lies in the organic combination of the adaptive sampling frequency adjustment mechanism and the deep integration of multimodal data. First of all, the sampling rate determination function is the key to realizing adaptive sampling. This function calculates the weight coefficient of the change rate and standard deviation of each parameter in the multi-parameter matrix in real time, and comprehensively determines the real-time sampling rate based on factors such as processing temperature, material flow rate, and solid waste component heterogeneity index. When it is monitored that the parameter change rate exceeds the preset threshold, the system will automatically increase the sampling frequency to ensure that sufficiently dense data is obtained at the process transition point; in the parameter stabilization stage, the sampling frequency will be appropriately reduced to save computing resources and storage space.

[0051] Secondly, the solid waste component migration equation provides a theoretical basis for adaptive sampling. This equation describes the physicochemical behavior of pollutants during the solid waste resource utilization process, including phase transformation, adsorption and desorption, redox, and complex precipitation. By analyzing the influence of parameters such as temperature field distribution, redox potential, and reactant concentration on component migration, it can predict the potential impact of process parameter changes on product quality, providing a scientific basis for adjusting sampling frequency. When the migration equation predicts a significant change in component concentration, the system will increase the sampling density in advance to prevent data loss.

[0052] The spectral fusion optimization function optimizes the integration of different spectral information by performing noise reduction, baseline correction, and feature extraction on Raman, near-infrared, and X-ray fluorescence spectral data. This multi-spectral fusion method can improve the information density of a single sample while reducing the sampling frequency through spectral complementarity, thereby reducing the overall number of required samples without compromising monitoring quality.

[0053] The FWM model, the core of the proposed algorithm, employs a multi-layer encoder-decoder architecture and processes data acquired at different sampling frequencies through a self-attention mechanism. This model is unique in that its self-attention mechanism uses a dynamic scaling function to determine weight allocation, taking into account factors such as the current sampling rate, parameter change trends, and data stability index. This allows the model to automatically adjust its attention allocation strategy based on changes in data acquisition frequency. During low-frequency sampling, the model focuses more on historical data and changing trends; during high-frequency sampling, it prioritizes capturing transient correlations between parameters.

[0054] Finally, the FWM model output is used to update the weight coefficients in the sampling rate determination function, forming a complete closed-loop control system. This feedback mechanism enables the sampling strategy to continuously optimize as the monitoring system operates, gradually forming a sampling pattern that best suits the current solid waste treatment process. This fundamentally solves the problem that fixed-frequency sampling cannot cope with dynamic process changes.

[0055] A specific embodiment 1 of the present invention is provided below. The specific implementation of each step in this embodiment 1 is described in detail as follows.

[0056] The specific implementation method of step S01 is to construct a parameter acquisition system, initialize the sampling rate to 5 times per minute, and synchronously collect spectral data and physical and chemical parameters in the solid waste treatment process to form a multi-parameter matrix. First, install spectral probes and physical and chemical parameter sensors at the key nodes of the solid waste treatment production line, including Raman spectral probes, near-infrared spectral probes, X-ray fluorescence spectral probes, as well as temperature sensors, pressure sensors, humidity sensors, redox potential sensors, pH sensors and flow rate sensors. Then establish an industrial-grade data acquisition network, use time-division multiplexing technology to ensure the synchronous acquisition of data from each sensor, and use industrial-grade field bus technology to realize real-time data transmission. The acquisition frequency is initialized to 5 times per minute. Then use a real-time database to store the collected information to form a multi-parameter matrix M, which can be expressed as:

[0057]

[0058] Where, T ij represents the temperature value of the i-th sampling point at the j-th time step, in °C; P ij Indicates the pressure value of the i-th sampling point at the j-th time step, in kPa; H ij Indicates the humidity value of the i-th sampling point at the j-th time step, in %; E ij Indicates the redox potential value of the i-th sampling point at the j-th time step, in mV; pH ij represents the pH value of the i-th sampling point at the j-th time step; V ij Indicates the velocity value of the i-th sampling point at the j-th time step, in m / s; S ij，k represents the kth spectral value at the jth time step of the i-th sampling point; m represents the number of sampling points; and n represents the spectral data dimension. Finally, the raw data is filtered and denoised by the data preprocessing unit to eliminate electromagnetic interference and outliers and achieve data standardization. The wavelet transform denoising algorithm is used to preprocess the spectral data. The expression of wavelet transform denoising is:

[0059]

[0060] Where, represents the spectral data after denoising; W represents the wavelet transform operator; w -1 represents the inverse wavelet transform operator; δ λ The threshold function is represented by λ, where the threshold is set to 4.5 times the noise standard deviation, which is estimated using the first-order difference method. The purpose of this step is to ensure that the monitoring system can capture changes in key parameters during solid waste treatment in a timely and accurate manner, providing basic data support for subsequent analysis.

[0061] The specific implementation of step S02 is to calculate the real-time sampling rate using a sampling rate determination function. This function is determined based on the multi-parameter matrix variation component, processing temperature, material flow rate, solid waste component heterogeneity index, and the FWM model output. The specific expression of the sampling rate determination function is:

[0062]

[0063] Where R real Indicates the real-time sampling rate, in times / minute; R init Indicates the initial sampling rate, which is 5 times / minute; w i It represents the weight coefficient of the i-th influencing factor. The initial values are set as w1=0.3, w2=0.25, w3=0.2, w4=0.15, w5=0.1 according to experience, and satisfy f i Represents the i-th influencing factor, including parameter variation factor f1, temperature deviation factor f2, flow rate deviation factor f3, component heterogeneity factor f4 and FWM model output factor f5. The calculation method of each influencing factor is as follows:

[0064] The calculation formula of parameter change factor f1 is:

[0065]

[0066] Where ΔP represents the rate of change vector of each parameter in the multi-parameter matrix within the specified time window; σ P Represents the standard deviation vector of each parameter in the multi-parameter matrix within the specified time window; σ thresh represents the standard deviation threshold vector, corresponding to the temperature parameter threshold of 3°C, the pH parameter threshold of 0.4, and the redox potential parameter threshold of 25mV. The calculation formula for the temperature deviation factor f2 is:

[0067]

[0068] Where, T represents the current processing temperature, in °C; T opt It represents the theoretical optimal reaction temperature in °C, which is determined according to different solid waste types and treatment processes, with a typical value of 400-800 °C. The calculation formula for the flow rate deviation factor f3 is:

[0069]

[0070] Where, V represents the current material flow rate in m / s; V std It represents the standard flow rate in m / s, which is determined according to the process requirements, and the typical value is 0.5 to 2 m / s. The calculation formula of the component heterogeneity factor f4 is:

[0071]

[0072] Where, I het It represents the heterogeneity index of solid waste components, with a value range of 0 to 1. It is calculated by analyzing the fluctuation range of solid waste components, the degree of uneven distribution, and the interaction strength between components:

[0073]

[0074] Where C max，i represents the maximum concentration of the i-th component; C min，i represents the minimum concentration of the i-th component; represents the average concentration; σ C represents the standard deviation of concentration; r ij represents the correlation coefficient between the i-th component and the j-th component; m represents the number of components. The calculation formula of the FWM model output factor f5 is:

[0075]

[0076] Where Q risk This represents the quality fluctuation risk value output by the FWM model, ranging from 0 to 1. This step aims to achieve intelligent allocation of monitoring resources. When process fluctuations are large or product quality risks increase, the monitoring frequency is increased to ensure monitoring accuracy, and when the process is stable, the monitoring frequency is reduced to conserve system resources.

[0077] The specific implementation of step S03 is to analyze the transformation law of the material in the resource recovery process through the solid waste component migration equation and predict the content of key components in the product. The solid waste component migration equation is constructed from three aspects: conservation of mass, conservation of momentum, and conservation of energy, and is specifically expressed as follows:

[0078] The mass conservation equation:

[0079]

[0080] Where C i represents the concentration of the i-th component in mg / L or mg / kg; t represents time in seconds; represents the velocity vector, in m / s; represents the gradient operator; D i represents the diffusion coefficient of the i-th component, in m 2 / s;r ij It represents the reaction rate constant of the conversion of the jth component into the ith component, in units of s -1 ;S i Represents the source and sink term of the i-th component, with the unit of mg / (L·s) or mg / (kg·s). Momentum conservation equation:

[0081]

[0082] Where ρ represents density, unit is kg / m 3 ; p represents pressure, unit is Pa; τ represents stress tensor; Represents the gravity acceleration vector, in m / s 2 . Energy conservation equation:

[0083]

[0084] Where c p represents specific heat capacity in J / (kg·K); T represents temperature in K; k represents thermal conductivity in W / (m·K); r j represents the reaction rate of the jth reaction, in mol / (L·s) or mol / (kg·s); ΔH j The reaction rate r is the heat of reaction of the jth reaction in J / mol. j It is described by the modified Arrhenius equation:

[0085]

[0086] Where A j It represents the pre-exponential factor of the jth reaction, and its unit is related to the reaction order; E j represents the activation energy of the jth reaction, in J / mol; R represents the gas constant, which is 8.314 J / (mol·K); α ij represents the reaction order of the i-th component in the j-th reaction; f(pH, E h ) represents the effect of pH value and redox potential on the reaction rate, and the expression is:

[0087]

[0088] Wherein, k1 and k2 represent the influence coefficients of pH value and redox potential, respectively, which are determined by experiments and the typical values are 0.5 and 0.01 respectively; pH opt Indicates the optimal pH value, which is determined according to different reaction types, with a typical value of 5 to 9; E h，optrepresents the optimal redox potential (mV), determined by the reaction type, with a typical value ranging from -400 to 400 mV. The finite element method (FEM) is used to discretize the solid waste component migration equations. The Galerkin weighted residual method is used for spatial discretization, and the adaptive step-size Runge-Kutta integration algorithm is used for temporal discretization to obtain predicted concentrations of the target components. By substituting the temperature field distribution, redox potential, reactant concentration, catalyst activity, and pH value from the multi-parameter matrix into the solution, the speciation distribution of heavy metal elements under different pH and redox conditions is predicted. The error in the predicted concentrations of typical metal ions such as lead, cadmium, mercury, and arsenic is controlled within ±7%, while the error in the prediction of organic pollutants is controlled within ±12%. This step aims to accurately predict the transformation behavior of each component in the solid waste resource utilization process and the concentration distribution in the final product through physical and chemical theoretical models, providing a theoretical basis for quality assessment.

[0089] The specific implementation of step S04 is to use the spectrum fusion optimization function to process Raman spectrum, near-infrared spectrum and X-ray fluorescence spectrum data. The spectrum fusion optimization function first pre-processes each spectrum data, including denoising, baseline correction and normalization. Raman spectrum uses wavelet transform to denoise, near-infrared spectrum uses standard normal variable transformation to correct baseline drift, and X-ray fluorescence spectrum uses Compton scattering peak normalization method. Then perform feature extraction, Raman spectrum focuses on 400~1800cm -1 Characteristic peaks within the wavenumber range. Near-infrared spectroscopy selects absorption peaks within the wavelength range of 1100 to 2500 nm, while X-ray fluorescence spectroscopy focuses on the characteristic peak positions and intensity ratios of typical elements. The mathematical expression of the spectrum fusion optimization function is:

[0090] F(S)=W R ·F R (S R )+W N ·F N (S N )+W X ·F X (S X );

[0091] Where F(S) represents the fused spectral feature vector; S R 、S N and S X Represent the raw data of Raman spectroscopy, near-infrared spectroscopy and X-ray fluorescence spectroscopy respectively; F R 、F N and F X Represent the feature extraction functions of Raman spectrum, near-infrared spectrum and X-ray fluorescence spectrum respectively; W R 、W N and W XRepresent the weight coefficients of Raman spectroscopy, near-infrared spectroscopy and X-ray fluorescence spectroscopy, respectively, satisfying W R +W N +W X =1, the initial value is set to W R =0.4,W N =0.3,W X =0.3, and then dynamically adjusted according to the feature correlation. The feature extraction function uses the principal component analysis method, and the expression is:

[0092]

[0093] Where i represents the spectrum type index, which can be R, N or X; P i represents the principal component loading matrix; μ i Represents the mean vector of spectral data. The dynamic adjustment method of weight coefficient is:

[0094]

[0095] Where r i represents the correlation coefficient between the i-th spectral feature and the concentration of the target component; β represents the adjustment parameter, which is set to 5. The purpose of this step is to fully utilize the advantages of different spectral techniques, extract complementary information, improve detection accuracy, and provide multidimensional spectral features for product quality assessment.

[0096] The specific implementation of step S05 is to use the FWM model to achieve multi-dimensional product quality assessment. The FWM model adopts a multi-layer encoder-decoder architecture, including a self-attention mechanism layer, a residual connection layer, and a normalization layer. The input parameters of the model include the spectral feature vector output by the spectral fusion optimization function, the multi-parameter matrix, and the target component concentration prediction value output by the solid waste component migration equation. The self-attention mechanism adopts a multi-head attention structure, and its weight distribution is determined by a dynamic scaling function. The expression of this function is:

[0097]

[0098] Where, α ij represents the attention weight of the i-th query vector to the j-th key vector; s ij Represents the similarity score, and the calculation formula is:

[0099]

[0100] Where Q i represents the i-th query vector; K j represents the jth key vector; d k represents the dimension of the key vector; g(R real ,ΔP,I stab , Ssim ) represents a dynamic scaling function, taking into account the current sampling rate R real , parameter change trend ΔP, data stability index I stab and spectral feature similarity S sim , the expression is:

[0101] g(R real ,ΔP,I stab , S sim )=a·log(1+R real )+b·‖ΔP‖2+c·(1-I stab )+d·S sim ;

[0102] Where a, b, c, and d are weight coefficients, satisfying a+b+c+d=1, and the initial values are set to a=0.2, b=0.3, c=0.25, and d=0.25; ||ΔP||2 represents the L2 norm of the parameter change trend vector; I stab Indicates the data stability index, with a value range of 0 to 1, and the calculation formula is:

[0103]

[0104] Where σ i represents the standard deviation of the i-th parameter; μ i represents the mean of the i-th parameter; m represents the number of parameters. sim Indicates the similarity of spectral features, with a value range of 0 to 1, and the calculation formula is:

[0105]

[0106] Where, F(S) represents the current spectral feature vector; F(S) ref represents the reference spectral feature vector. The output layer of the FWM model is a fully connected layer, which maps the decoder output to the product quality evaluation index space and outputs the product's multidimensional quality assessment results, including purity, safety, stability, and suitability indicators. The purpose of this step is to integrate multi-source heterogeneous data through a deep learning model to achieve a comprehensive and accurate assessment of the quality of solid waste resource recovery products, providing a basis for decision-making in the production process.

[0107] The specific implementation method of step S06 is to monitor the heavy metal ion content in solid waste resource products in real time based on the electrochemical sensor network, and fuse the spectral data processed by the spectral fusion optimization function to form a comprehensive quality evaluation result. First, a selective electrode array is arranged at the key nodes of the processing flow, including ion-selective electrodes, modified electrodes and reference electrodes, to form an electrochemical sensor network. Then, the redox current changes of each electrode are measured, and the heavy metal ion concentration is quantitatively analyzed by voltammetry and potentiometry. The calculation formula for heavy metal ion concentration is:

[0108] C M =k·(I peak -I baseline )·f corr ;

[0109] Where C M Indicates the concentration of heavy metal ions in mg / L; I peak Indicates the peak current in μA; I baseline represents the baseline current in μA; k represents the correction factor in mg / (L·μA), which is determined by the standard curve method; f corr It represents the interference correction factor, and its calculation formula is:

[0110]

[0111] Where k i represents the interference coefficient of the i-th interfering ion; C i represents the concentration of the i-th interfering ion, in mg / L; C i，ref represents the reference concentration of the i-th interfering ion, expressed in mg / L. A multivariate correction algorithm is then used to eliminate cross-interference between ions and improve measurement accuracy. The measurement accuracy of typical heavy metal ions is controlled within ±5%. The heavy metal ion content data measured by the electrochemical sensor network is then normalized with the spectral data processed by the spectral fusion optimization function to eliminate dimensional differences. The normalization method is:

[0112]

[0113] Where, X norm represents the standardized data; X represents the original data; μ X represents the mean; σ X Represents the standard deviation. Finally, a hierarchical weight model is established to calculate the comprehensive index of product quality. The expression of this model is:

[0114]

[0115] Where Q total Indicates the comprehensive index of product quality, with a value range of 0 to 1; wi Represents the weight coefficient of the i-th category indicator, satisfying w ij Represents the weight coefficient of the jth sub-indicator under the i-th category indicator, satisfying q ij It represents the normalized score of the jth sub-indicator under the i-th indicator, with a value range of 0 to 1, and m represents the number of indicator categories; n i The number of sub-indicators included in the i-th category indicator is represented by the number of sub-indicators. The purpose of this step is to achieve accurate monitoring of heavy metal pollutants in products through the complementary combination of electrochemical and spectroscopic methods, thereby ensuring the environmental safety of resource-based products.

[0116] The specific implementation of step S07 is to update the weight coefficient in the sampling rate determination function based on the output of the FWM model, return to step S02 to recalculate the real-time sampling rate, and form an online monitoring closed-loop control. First, analyze the product quality assessment results output by the FWM model and calculate the deviation of each quality indicator from the target value:

[0117] ΔQ i =|Q i -Q i，target |;

[0118] Where ΔQ i represents the deviation of the i-th quality index; Q i represents the actual value of the i-th quality index; Q i，target Represents the target value of the i-th quality indicator. Then, according to the deviation size and change trend, adjust the sampling rate to determine the weight coefficient of each parameter in the function:

[0119]

[0120] Where, Represents the updated weight coefficient; Represents the weight coefficient before update; Δw i Indicates the adjustment amount of the weight coefficient, and the calculation formula is:

[0121]

[0122] Where η represents the learning rate, which ranges from 0.01 to 0.1; sign represents the sign function; Represents the partial derivative of the quality index deviation with respect to the weight coefficient; Δw max Indicates the maximum step size of weight adjustment, ranging from 0.05 to 0.15. After weight adjustment, normalization is required to ensure

[0123]

[0124] Where, Represents the normalized weight coefficient. Then the updated weight coefficient is applied to recalculate the real-time sampling rate and the new sampling rate is applied to the parameter acquisition system:

[0125]

[0126] Where, represents the updated real-time sampling rate, measured in times per minute. This ultimately forms a closed-loop feedback mechanism for quality monitoring and sampling control, enabling intelligent allocation and optimized utilization of monitoring resources. This step aims to enable the monitoring system to adaptively adjust its monitoring strategy based on product quality status through a closed-loop control strategy, improving monitoring efficiency and ensuring quality.

[0127] The detailed structure of the FWM model adopts a multi-layer encoder-decoder architecture, including a self-attention mechanism layer, a residual connection layer, and a normalization layer. The encoder part is composed of 6 layers of encoders, each of which contains a multi-head self-attention sublayer and a feedforward neural network sublayer. Each sublayer uses residual connections and layer normalization. The mathematical expression of the multi-head self-attention mechanism is:

[0128] MultiHead(Q,K,V)=Concat(head1,head2,...,head h )W O ;

[0129] Where MultiHead(Q, K, V) represents the output of multi-head attention; head i represents the output of the i-th attention head, and the calculation formula is:

[0130]

[0131] Where, represents the learnable parameter matrix; d model Indicates the model dimension, the value is 512; d k and d v Indicates the dimension of key and value, the value is 64; h indicates the number of heads, the value is 8. The calculation formula of the attention function Attention(Q, K, V) is:

[0132]

[0133] Where Q represents the query matrix; K represents the key matrix; V represents the value matrix; d k Represents the dimension of the key; softmax represents the softmax function. The calculation formula for residual connection and layer normalization is:

[0134] LayerNorm(x+Sublayer(x));

[0135] Where x represents the input of the sublayer; Sublayer(x) represents the function of the sublayer; LayerNorm represents the layer normalization function, and the calculation formula is:

[0136]

[0137] Where γ and β represent the learnable scale and offset parameters; μ and σ represent the mean and standard deviation, respectively; ∈ represents a small constant to prevent division by zero errors, with a value of 1×10 -6 The calculation formula of the feedforward neural network sublayer is:

[0138] FFN(x)=max(0,xW1+b1)W2+b2;

[0139] Where, b1 and b2 represent learnable parameters; d ff The hidden layer dimension of the feedforward network is 2048. The decoder is also composed of six stacked decoder layers, each of which contains a self-attention sublayer, a cross-attention sublayer, and a feedforward neural network sublayer. The calculation formula for the cross-attention sublayer is the same as that for the multi-head attention, but the query matrix comes from the decoder input, and the key matrix and value matrix come from the encoder output. The model output layer is a fully connected layer that maps the decoder output to the product quality evaluation indicator space:

[0140] Y=XW+b;

[0141] Where Y represents the output vector; X represents the output of the decoder; W and b represent the learnable weight matrix and bias vector.

[0142] The steps for establishing the training dataset for the FWM model are as follows: First, historical monitoring data on the resource utilization of solid waste from different sources is collected, including treatment data for different types of solid waste, such as municipal solid waste, industrial solid waste, and agricultural waste, with the data volume covering at least 100 complete production cycles. Each set of data is then cleaned and labeled. The cleaning process includes outlier detection and processing, missing value interpolation, and data consistency verification. Outlier detection uses an improved local outlier factor algorithm, calculated as follows:

[0143]

[0144] Where LOF k (p) represents the local outlier factor of point p; N k (p) represents the k-nearest neighbor set of point p; lrd k (p) represents the local reachability density of point p, and the calculation formula is:

[0145]

[0146] Where reach-dist k (p, o) represents the reachable distance from point p to point o, and the calculation formula is:

[0147] reach-dist k (p, o) = max {k-distance (o), d (p, o)};

[0148] Where k-distance(o) represents the distance from point o to its kth nearest neighbor; d(p, o) represents the Euclidean distance from point p to point o. Missing value interpolation uses the multivariate linear regression method, and the calculation formula is:

[0149]

[0150] Where, represents the missing value to be interpolated; β0 represents the intercept term; β j represents the regression coefficient; x j Represents known variables; ε represents the random error term; and m represents the number of known variables. The labeling process involves labeling corresponding product quality indicators based on laboratory analysis results, such as purity, harmful substance content, stability, and other key indicators. Data enhancement technology is then used to simulate parameter fluctuations under different operating conditions, including adding Gaussian noise to simulate sensor errors, random scaling to simulate parameter fluctuations, and time shifting to simulate process delays, expanding the data set size to 3 to 5 times the original data. The noise addition formula is:

[0151]

[0152] Where x noisy represents the data after adding noise; x represents the original data; α represents the noise intensity, which ranges from 0.01 to 0.1; Indicates that the mean is 0 and the variance is σ 2 Gaussian noise, σ is determined according to the standard deviation of the original data, generally 5% to 10% of the standard deviation of the original data. The random scaling formula is:

[0153]

[0154] Where x scaled represents the scaled data; β represents the scaling strength, ranging from 0.05 to 0.2; represents a uniform distribution on the interval [-1, 1]. The time shift formula is:

[0155]

[0156] Where, represents the shifted time series data; Δt represents the time offset, which ranges from [-5, 5] time steps. A standard data pair consisting of an input parameter matrix and corresponding quality assessment results is then constructed. The input parameter matrix has the dimension of time step × number of parameters, and the quality assessment result has the dimension of the number of indicators. Finally, the dataset is partitioned into training, validation, and test sets using stratified random sampling, with a ratio of 70%, 15%, and 15%, respectively. The distribution of different types of solid waste data within each subset is ensured to be consistent.

[0157] The FWM model training process first performs unsupervised pre-training on large-scale solid waste treatment historical data, using a masked autoencoder structure to learn the parameter distribution law and the intrinsic structure of the data. The loss function of the masked autoencoder is:

[0158]

[0159] Where, L MAE represents the loss function of the masked autoencoder; N represents the number of samples; M represents the mask matrix, whose elements are 0 or 1; ⊙ represents the Hadamard product; X represents the original data matrix; represents the reconstructed data matrix; ‖·‖ F Denotes the Frobenius norm. Then supervised fine-tuning training is performed for the current solid waste type, and the loss function comprehensively considers the mean square error and weighted cross entropy:

[0160] L total =λ1L MSE +λ2L WCE ;

[0161] Where, L total Represents the total loss function; L MSE represents the mean square error loss; L WCE represents the weighted cross entropy loss; λ1 and λ2 represent weight coefficients, satisfying λ1+λ2=1, and the initial values are set to λ1=0.7 and λ2=0.3. The mean square error loss function is:

[0162]

[0163] Where y i represents the true label of the i-th sample; Represents the predicted value of the i-th sample. The weighted cross entropy loss function is:

[0164]

[0165] Where C represents the number of categories; w j represents the weight coefficient of the jth class; y ij Indicates the true probability that the i-th sample belongs to the j-th class; Indicates the predicted probability that the i-th sample belongs to the j-th class. An adaptive learning rate strategy is used during training, and the initial learning rate is set to 1×10 -4 , when the validation set loss does not improve for 5 consecutive epochs, the learning rate is reduced to 0.8 times the original:

[0166]

[0167] Where, lr new Indicates the updated learning rate; lr old Indicates the learning rate before updating; n no_improve Indicates the number of epochs in which the validation set loss has not improved. At the same time, the weight reduction regularization technique is introduced, and the reduction coefficient is set to 5×10 -5 , to prevent the model from overfitting:

[0168]

[0169] Where, L reg represents the loss function after adding regularization; γ represents the decreasing coefficient, which is 5×10 -5 ; P represents the total number of parameters; W i The model generalization ability is evaluated through 5-fold cross-validation, and the model parameters with the best performance on the validation set are finally selected as the final model.

[0170] To better understand and implement the present invention, Example 2, a specific application scenario, is provided below: Researchers developed an online quality monitoring system for products from the fly ash recycling process of municipal solid waste, applied to a production line for fly ash-based building materials. Based on a proposed online quality monitoring method for solid waste recycling, this system enables parameter collection, quality prediction, and assessment throughout the entire fly ash recycling process.

[0171] First, a parameter acquisition system was constructed. Spectroscopic probes and physicochemical parameter sensors were installed at key production line processes, including pretreatment, stabilization, molding, and curing. The spectral probes included three Raman spectrometers, two near-infrared spectrometers, and two X-ray fluorescence spectrometers. The physicochemical parameter sensors included 12 temperature sensors, eight pressure sensors, six humidity sensors, five redox potential sensors, four pH sensors, and three flow rate sensors. The initial sampling rate was set to five times per minute, and real-time data transmission was achieved via an industrial fieldbus, generating a multi-parameter matrix containing 38 parameters.

[0172] The system dynamically adjusts the monitoring frequency based on the sampling rate determination function. In actual operation, when the stabilization process temperature was detected to fluctuate from the set value of 790°C to 830°C, the temperature deviation factor f2 value increased from 0.5 to 1.0, and the sampling rate increased from 5 times per minute to 7.5 times. The system also monitored that the material flow rate dropped from the standard value of 1.2m / s to 0.85m / s, and the flow rate deviation factor f3 value was 1.0, further supporting the increase in the sampling rate. The fly ash component analysis showed that its heterogeneity index I het The value of the component heterogeneity factor f4 is 0.68, and the corresponding component heterogeneity factor f4 is 1.0. Taking into account the influence of various factors, the real-time sampling rate is adjusted to 8.5 times per minute to ensure sufficient monitoring data during process fluctuations.

[0173] Solid waste component migration equations are used to analyze the transformation patterns of heavy metals in fly ash during the stabilization process. This example focuses on the migration behavior of four heavy metal elements: lead, cadmium, chromium, and zinc. Parameters such as temperature field distribution, redox potential, and pH value are substituted into the corresponding equations. By solving the component migration equations, the system predicts the heavy metal concentrations in the stabilized product, as shown in Table 1:

[0174] Table 1 Heavy metal concentrations predicted by solid waste component migration equations

[0175]

[0176] The spectrum fusion optimization function is used to process three types of spectral data. Raman spectroscopy focuses on 400-1800cm -1 Characteristic peaks within the wavenumber range are selected. Near-infrared spectroscopy selects absorption peaks in the wavelength range of 1100 to 2500 nm, while X-ray fluorescence spectroscopy focuses on the positions and intensity ratios of characteristic peaks of heavy metal elements. By calculating the correlation coefficient between spectral features and the concentration of target components, the system dynamically adjusts the weighting coefficients of each spectrum: 0.38 for Raman spectroscopy, 0.27 for near-infrared spectroscopy, and 0.35 for X-ray fluorescence spectroscopy. The fused and optimized spectral data is converted into a 128-dimensional feature vector, which serves as the input for the FWM model.

[0177] The FWM model is based on a multi-layer encoder-decoder architecture, and the self-attention mechanism uses an 8-head attention structure. The model input includes a 128-dimensional spectral feature vector, a 38-dimensional parameter matrix, and a 4-dimensional heavy metal concentration prediction value. The model is trained on a training set containing 5000 sets of historical data, with an initial learning rate of 1×10 -4 When the validation set loss did not improve for 5 consecutive epochs, the learning rate was reduced to 0.8 times the original rate. After 100 epochs of training, the model achieved a mean square error of 0.035 on the test set and an accuracy of 94.7%.

[0178] Electrochemical sensor networks are used to monitor the heavy metal ion content of products in real time. Researchers installed six selective electrode arrays at the end of the production line, targeting heavy metal ions such as lead, cadmium, chromium, and zinc. Table 2 shows the electrochemical monitoring results for a batch of products compared with laboratory analysis:

[0179] Table 2 Comparison of electrochemical sensor network monitoring results and laboratory analysis

[0180] heavy metal elements Electrochemical sensor measurement value (mg / L) Laboratory analysis value (mg / L) Relative error (%) Lead (Pb) 0.43 0.45 4.4 Cadmium (Cd) 0.07 0.075 6.7 Chromium (Cr) 0.36 0.39 7.7 Zinc (Zn) 1.59 1.67 4.8

[0181] The system integrates the electrochemical sensor network measurement results with the spectral data to form a comprehensive quality evaluation result, including safety indicators, mechanical performance indicators, and durability indicators. Table 3 shows the comprehensive quality index of a batch of products calculated by the system:

[0182] Table 3 Evaluation results of product quality comprehensive index

[0183] Quality indicator categories Weight coefficient Review score Quality Status Safety indicators 0.4 0.86 good Mechanical properties 0.35 0.92 excellent Durability index 0.25 0.78 qualified Comprehensive Quality Index 1.0 0.86 good

[0184] According to the quality fluctuation risk value Q output by the FWM model risk The system sets the FWM model output factor f5 to 0.5 and updates the weight coefficients in the sampling rate determination function accordingly, with the learning rate η set to 0.05. After the update, the value of w1 is adjusted from 0.3 to 0.32, the value of w2 from 0.25 to 0.27, the value of w3 from 0.2 to 0.18, the value of w4 from 0.15 to 0.14, and the value of w5 from 0.1 to 0.09. Based on the updated weight coefficients, the real-time sampling rate is recalculated to 6.3 times per minute, forming a closed-loop monitoring control.

[0185] Traditional quality monitoring of solid waste resource products mainly relies on offline laboratory analysis, which has problems such as monitoring lag, insufficient sampling representativeness and high monitoring costs. The online monitoring method adopted in this embodiment solves these core problems: on the one hand, real-time monitoring of multi-dimensional parameters is achieved through multi-parameter matrix and spectral fusion technology, and the monitoring delay is shortened from the traditional hours to minutes; on the other hand, the dynamic sampling rate adjustment mechanism improves the representativeness of sampling, especially increasing the sampling frequency during process fluctuations, which can more accurately capture quality fluctuations. This method not only improves monitoring efficiency and accuracy, but also provides a decision-making basis for quality control of solid waste resource products, and promotes the development of solid waste resource technology.

[0186] It should be noted that the variables involved in the present invention are explained in detail as shown in Tables 4, 5, 6 and 7 below.

[0187] Table 4 Variable Explanation Table (Part 1)

[0188]

[0189] Table 5 Variable Explanation Table (Part 2)

[0190]

[0191]

[0192] Table 6 Variable Explanation Table (Part 3)

[0193]

[0194]

[0195] Table 7 Variable Explanation Table (Part 4)

[0196]

[0197] The above description is only a specific embodiment of the present invention, but the protection scope of the present invention includes any technical personnel familiar with this technical field within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for online monitoring of product quality in a solid waste resource recovery process, characterized in that: include: Construct parameter acquisition system to collect multi-parameter matrix; A sampling rate determination function is applied to calculate the real-time sampling rate, and the sampling rate determination function is jointly determined based on the variable component of the multi-parameter matrix, the processing temperature, the material flow rate, the solid waste component heterogeneity index, and the output results of the FWM model; the solid waste component migration equation is used to analyze the material transformation law; the spectral fusion optimization function is used to process the spectral data; the FWM model is used to realize product quality assessment; the heavy metal content is monitored through an electrochemical sensor network; and the weight coefficient in the sampling rate determination function is updated according to the output results of the FWM model to form an online monitoring closed-loop control.

2. The method according to claim 1, characterized in that The step of constructing a parameter acquisition system to acquire a multi-parameter matrix specifically involves initializing a sampling rate of 5 times per minute, synchronously acquiring spectral data and physical and chemical parameters in the solid waste treatment process, and forming a multi-parameter matrix.

3. The method according to claim 2, characterized in that The multi-parameter matrix refers to a data set consisting of real-time monitoring parameters during the solid waste treatment process, including temperature, pressure, humidity, redox potential, pH value and flow rate.

4. The method according to claim 3, characterized in that The sampling rate determination function specifically calculates the weight coefficient by calculating the change rate and standard deviation of each parameter in the multi-parameter matrix. When the weight coefficient exceeds a preset threshold, the sampling frequency is increased, otherwise the sampling frequency is reduced, and the real-time sampling rate is dynamically adjusted in combination with the output results of the FWM model.

5. The method according to claim 4, characterized in that When the FWM model has not yet output a result, the sampling rate determination function sets the weight coefficient of the FWM model output result to 0.

6. The method according to claim 5, characterized in that The solid waste component heterogeneity index is a quantitative indicator that characterizes the heterogeneity of solid waste and is calculated by analyzing the fluctuation range of solid waste components, the degree of uneven distribution, and the intensity of interaction between components.

7. The method according to claim 6, characterized in that The solid waste component migration equation specifically expresses the physical and chemical behavior of pollutants in the solid waste resource utilization process, including phase change transformation, adsorption and desorption, redox and complex precipitation. By solving the solid waste component migration equation group, the concentration distribution of each component in the final product is obtained.

8. The method according to claim 7, characterized in that The input parameters of the solid waste component migration equation include temperature field distribution, redox potential, reactant concentration, catalyst activity and pH value in a multi-parameter matrix, and the output is the predicted value of the target component concentration.

9. The method according to claim 8, characterized in that The spectrum fusion optimization function is used to perform noise reduction, baseline correction and feature extraction on Raman spectrum, near-infrared spectrum and X-ray fluorescence spectrum data, and realizes the optimized fusion of different spectral information through weight distribution.

10. The method according to claim 9, characterized in that The input parameters of the spectrum fusion optimization function include wavelength range, signal-to-noise ratio threshold, characteristic peak position, peak intensity ratio and baseline offset, and the output is a spectrum feature vector.

Images