Data management system for recycling resources based on internet of things

Abstract

The present application relates to the technical field of industrial platform data analysis, in particular to a renewable resource recycling data management system based on the Internet of Things, which comprises synchronously acquiring the weight, spectral and microwave characteristic data of the measured resource through a data acquisition module, performing moisture weight decoupling operation by using a resource matching module, outputting standard dry basis asset weight with cross-environment consistency, then, the inventory evolution module uses an evolution model integrated with long and short-term memory network to track the performance loss of the resource in real time, generates resource attenuation weight, and combines real-time inventory saturation to construct a two-dimensional decision matrix containing scheduling priority and pretreatment intensity level instructions, finally, the clean-up and dispatching module executes dynamic pruning and bidirectional optimization through a resource scheduling model, accurately allocates the loading share and access order of each node, and generates a dynamic instruction set containing delivery order. The present application realizes homogeneous feeding through multi-dimensional collaborative analysis of industrial data.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of industrial platform data analysis, in particular to a renewable resource recycling data management system based on Internet of Things. BACKGROUND

[0002] With the wide application of the Internet of Things technology in the field of renewable resources, the amount of material attribute data generated in the recycling link increases explosively. In the current renewable resource recycling management process, the management system usually collects the resource weight, basic category and storage time of the recycling node in real time, and executes inventory statistics and preliminary planning of the clearing logistics on the basis.

[0003] However, renewable resources have a wide range of sources and strong physical heterogeneity. The internal material composition and water content of the resources often change with the change of the storage environment and the space-time span, and at the same time, the automated production line of the terminal processing plant has strict stability requirements for the quality of the raw materials. Any lag or deviation of single-dimensional data will cause fluctuations in the composition of the raw materials.

[0004] Therefore, how to accurately align and coordinate the heterogeneous data of renewable resources at the supply and demand ends in a complex fluctuation environment to ensure the homogenization of the raw materials in the resource circulation process is a key technical problem in the field of renewable resource recycling data management.

[0005] Therefore, the renewable resource recycling data management system based on the Internet of Things is proposed. SUMMARY

[0006] The purpose of the application is to provide a renewable resource recycling data management system based on the Internet of Things, which realizes the homogenization of raw materials through multi-dimensional collaborative analysis of industrial data. The weight, spectral and microwave characteristic data of the measured resources are synchronously acquired through the data acquisition module, the resource matching module is used to perform water weight decoupling operation, and the standard dry basis asset weight with consistent cross-environment is output. Then, the inventory evolution module uses the evolution model integrated with long and short-term memory network to track the performance loss of the resources in real time, generates the resource decay weight, and combines the real-time inventory saturation to build a two-dimensional decision matrix containing scheduling priority and pretreatment intensity level instructions. Finally, the clearing and scheduling module performs dynamic pruning and bidirectional optimization through the resource scheduling model, accurately allocates the loading share and access order of each node, and generates a dynamic instruction set containing the delivery sequence.

[0007] To achieve the above purpose, the application provides the following technical scheme:

[0008] The renewable resource recycling data management system based on the Internet of Things comprises:

[0009] a data acquisition module, which acquires real-time total weight, spectral reflection sequence, microwave phase shift, microwave amplitude attenuation, environmental temperature value and humidity value of the measured paper;

[0010] a resource matching module, which extracts features of the microwave phase shift and the microwave amplitude attenuation and extracts characteristic peaks of the spectral reflection sequence, obtains deep water content of the paper package and fiber grade features, and performs water weight decoupling operation in combination with environmental humidity compensation to output standard dry basis asset weight data;

[0011] an inventory evolution module, which inputs the standard dry basis asset weight, deep water content of the paper package, fiber grade features and environmental temperature and humidity into an inventory evolution model to obtain resource attenuation weight, generates quality grade data based on the resource attenuation weight and the fiber grade features, and constructs a decision matrix in combination with real-time inventory saturation of a recycling node;

[0012] a collection and transportation scheduling module, which calls production line feeding quality threshold requirements of a terminal processing plant, inputs the decision matrix, standard dry basis asset weight and quality grade data of each recycling node into a resource scheduling model to output a delivery order instruction set including logistics triggering time, allocation path coordinates and requirements meeting terminal processing plant feeding homogenization.

[0013] Preferably, the acquisition of the real-time total weight, spectral reflection sequence, microwave phase shift, microwave amplitude attenuation, environmental temperature value and humidity value of the measured paper comprises: scanning and collecting the surface of the measured paper by using a near-infrared array sensor to generate the spectral reflection sequence; transmitting a penetrating signal to the measured paper by using a microwave transceiving assembly symmetrically arranged inside a recycling bin, and acquiring the microwave phase shift and the microwave amplitude attenuation by analyzing the phase deviation and energy loss of the penetrating signal after passing through the paper medium; acquiring the real-time total weight by using a weighing unit to sense the force value of the measured paper, and acquiring the real-time environmental temperature value and real-time environmental humidity value by using an environmental perception unit integrated inside the bin.

[0014] Preferably, the process of outputting the standard dry basis asset weight data comprises: extracting the slope of the phase change with frequency of the microwave signal propagating in the paper package medium to obtain a group delay parameter; calculating the unit decibel ratio of the microwave phase shift and the microwave amplitude attenuation, combining the unit decibel ratio and the group delay parameter into a normalized vector, and mapping the modulus value of the normalized vector to obtain the deep water content data of the paper package;

[0015] performing standard normal variable transformation on the spectral reflection sequence to eliminate background baseline noise caused by optical path difference; positioning local extreme points of the transformed sequence by using a sliding window, extracting wavelength coordinates and peak intensities corresponding to the local extreme points, and constructing the wavelength coordinates and peak intensities into a fiber grade feature vector.

[0016] An intermediate weight value is obtained by subtracting the product of the real-time total weight and the paper package deep layer moisture content data from the real-time total weight; a corresponding equilibrium moisture compensation operator is retrieved from a preset fiber equilibrium moisture mapping table according to the ambient humidity value, the intermediate weight value is multiplied by the equilibrium moisture compensation operator, and the standard dry basis asset weight data is output.

[0017] Preferably, the inventory evolution model comprises: an environmental stress layer that receives real-time ambient temperature and humidity as input, generates an environmental partial pressure difference feature by calculating the difference between the saturated vapor pressure and the actual vapor pressure under the current temperature and humidity;

[0018] A moisture migration layer inputs a historical moisture content time series constructed from the paper package deep layer moisture content and the environmental partial pressure difference feature into a long short-term memory network, converts discrete moisture content fluctuations into continuous internal energy state latent vectors, and maps and calculates the environmental partial pressure difference feature, to output a non-equilibrium potential difference feature;

[0019] A structure calibration layer inputs the fiber grade feature and the non-equilibrium potential difference feature into a multi-layer perceptron, learns the effective diffusion resistance distribution of different fiber grades at different aging stages, performs weight coupling operation on the potential difference feature and the resistance coefficient, and generates an effective evolution rate feature;

[0020] A boundary constraint layer receives the standard dry basis asset weight and the effective evolution rate feature as input, takes the average of the effective evolution rates at the current sampling time and the previous sampling time, and performs multiplication operation with the sampling step, to output an instantaneous performance loss value;

[0021] A space-time accumulation layer receives the instantaneous performance loss value as the current input, and obtains the historical loss accumulation value stored at the previous sampling time, performs first-order recursive superposition operation, generates an asset performance total loss value, maps the asset performance total loss value to [0, 1] through nonlinear mapping logic, and outputs a resource attenuation weight.

[0022] Preferably, the generation process of the quality grade data comprises: performing proportional deduction calculation on the fiber grade feature using the resource attenuation weight, to determine the residual performance quantization value of the paper under the current environmental influence; aligning the residual performance quantization value with a plurality of preset quality evaluation benchmarks, to determine the quality management interval in which the residual performance quantization value is located; extracting the grade label corresponding to the quality management interval, to confirm the quality grade data of the recycled resources.

[0023] Preferably, the process of constructing the decision matrix comprises: extracting the real-time inventory saturation of the recycling node as the first decision dimension, and extracting the resource attenuation weight as the second decision dimension, to construct a two-dimensional attribute mapping space; establishing an orthogonal judgment boundary line in the two-dimensional attribute mapping space according to a preset inventory capacity critical constraint and a resource quality threshold, and dividing the mapping space into a plurality of rectangular attribute management regions through the orthogonal judgment boundary line; wherein each rectangular attribute management region corresponds to a group of warehouse urgency and resource loss state combinations determined by the saturation interval and the preset resource attenuation weight interval defined by the orthogonal judgment boundary line; for each rectangular attribute management region, a corresponding physical intervention parameter sequence is configured respectively; the physical intervention parameter sequence includes a scheduling priority reflecting the transit timeliness and a pretreatment intensity level reflecting the quality maintenance demand; the real-time coordinate point of the current recycling resource in the two-dimensional attribute mapping space is topologically matched with the rectangular attribute management region to determine the target region to which the recycling resource belongs, and the physical intervention parameter sequence associated with the region is extracted to generate a decision matrix including resource survival control state and pretreatment intensity level instructions.

[0024] Preferably, the processing process of the collection and transportation scheduling module comprises: real-time calling the production line feeding quality threshold demand of the terminal processing plant as the target guide benchmark of global scheduling; synchronously monitoring the standard dry basis asset weight and quality grade data of each recycling node, combining the real-time state in the decision matrix, and completing multi-dimensional data alignment based on the time stamp to construct a dynamic scheduling task pool; monitoring the resource survival control state in the decision matrix, and when the state changes and / or the inventory capacity touches the threshold, automatically extracting the decision matrix and the feeding quality threshold as physical boundary constraints to activate the resource scheduling model; according to the model calculation result, outputting a real-time dynamic instruction set including the logistics trigger time, the deployment path coordinates and the delivery sequence.

[0025] Preferably, the resource scheduling model comprises: an attribute mapping layer: mapping the input standard dry basis asset weight and quality grade data of each recycling node to a supply state vector in a high-dimensional feature space;

[0026] a demand guide layer: converting the input terminal processing plant production line feeding quality threshold demand into a target center vector, and establishing a convergence guide direction of the supply state vector in the feature space;

[0027] a constraint conversion layer: analyzing the input decision matrix to convert the resource survival control state into a boundary restriction operator acting on the algorithm logic, and realizing dynamic pruning by real-time removing node branches that do not meet the control state in the dynamic scheduling task pool to reduce the search range of the candidate scheme;

[0028] The two-way optimization layer: under the constraint of the boundary restriction operator, the spatial distance between the supply state vector and the target center vector is used to measure the homogenization deviation, and the candidate scheme of double minimization of path cost and quality deviation is locked through iterative calculation;

[0029] The instruction generation layer: the quality deviation of each recycling node in the candidate scheme is extracted, and the deviation offset logic containing the stability adjustment benchmark is used to allocate the loading share; the linear alignment of the total quality and the feed demand is realized by matching the mixed proportion of positive and negative deviation resources, and after simulation verification and error fine-tuning, the delivery order instruction set containing the logistics trigger time, the deployment path coordinates and meeting the terminal processing plant feed homogenization requirements is generated.

[0030] Compared with the prior art, the beneficial effects of the present application are:

[0031] 1. Through the synergistic extraction of microwave penetration characteristics and infrared spectrum peak value, combined with environmental humidity mapping compensation logic, the depth stripping of moisture and dry basis weight in the physical properties of the measured paper is realized, the paper moisture absorption and weight error caused by environmental humidity fluctuation is effectively corrected, the bottom physical sensing signal is converted into standard dry basis data reflecting the actual asset value, the fluctuation amplitude of the original data disturbed by the environment is significantly reduced, a highly consistent data starting point is provided for the management system, the data source difference caused by material heterogeneity is effectively alleviated, and a reliable quantitative basis is laid for the precise alignment of the supply and demand ends.

[0032] 2. By constructing an inventory evolution model integrating long and short term memory networks and structure calibration layers, the performance loss of renewable resources caused by temperature and humidity stress in the warehousing link is realized in real time dynamic tracking, discrete sensor data is converted into continuous resource attenuation weight, the quality evolution law of resources in time and space span is quantified, the information deviation between static warehouse entry label and actual warehouse exit quality is significantly reduced, and the management system can realize real-time perception of the physical performance state of the supply side resources, and provide dynamic parameter support with timeliness for real-time optimization of scheduling decisions in complex fluctuation environment.

[0033] 3. By mapping the real-time inventory saturation and resource attenuation weight to a two-dimensional attribute space and establishing orthogonal orthogonal determination boundary lines, the fast conversion from multi-dimensional heterogeneous data to standardized decision instructions is realized, the complex logistics control requirements are converted into a decision matrix containing scheduling priority and pretreatment intensity, the fine regional topological matching of the recycling node state is realized, the management path of the system in processing large-scale heterogeneous resources is optimized, which helps to improve the response speed and accuracy of management decisions, and provides standardized logical support for realizing cross-node and multi-target collaborative scheduling.

[0034] 4. By using dynamic pruning logic in the resource scheduling model and an adaptive weight allocation algorithm based on quality deviation offsetting logic, precise coupling between fluctuating resources at the recycling end and the high-precision production needs of the processing plant is achieved. By utilizing the dynamic balance compensation mechanism of quality attributes, the mixing ratio of resources of different qualities is dynamically adjusted and transformed into specific logistics triggering time and delivery sequence instructions. This scheduling method effectively offsets the interference of the physical heterogeneity of recycled resources on the stability of terminal feeding, making the instantaneous component ratio of delivered resources closer to the preset production threshold, thus providing a systematic management means for the goal of homogenization of feeding in the process of large-scale resource circulation. Attached Figure Description

[0035] Fig. 1 This is a schematic diagram of the Internet of Things-based data management system for recycling of renewable resources according to the present invention;

[0036] Fig. 2 This is a schematic diagram of the inventory evolution model of the present invention;

[0037] Fig. 3 This is a schematic diagram of the resource scheduling model of the present invention. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] Please see Figs. 1 to 3 This invention provides an Internet of Things-based data management system for the recycling of renewable resources, and the technical solution is as follows.

[0040] Example 1:

[0041] The IoT-based data management system for recycled resource recovery has the following structure: Fig. 1 As shown, it includes:

[0042] The data acquisition module obtains the real-time total weight, spectral reflectance sequence, microwave phase shift, microwave amplitude attenuation, ambient temperature, and humidity of the tested paper.

[0043] The resource matching module extracts features from microwave phase offset and microwave amplitude attenuation, and extracts feature peaks from spectral reflectance sequences to obtain the deep moisture content and fiber grade characteristics of the paper package. It also performs moisture weight decoupling calculations in conjunction with environmental humidity compensation to output standard dry basis asset weight data.

[0044] The inventory evolution module inputs the standard dry basis asset weight, paper package deep moisture content, fiber grade characteristics and environmental temperature and humidity into an inventory evolution model to obtain resource decay weights, generates quality grade data based on the resource decay weights and fiber grade characteristics, and constructs a decision matrix in combination with real-time inventory saturation of the recycling node;

[0045] The clean-up scheduling module calls the production line feeding quality threshold demand of the terminal processing plant, inputs the decision matrix of each recycling node, the standard dry basis asset weight and the quality grade data into a resource scheduling model, and outputs a delivery sequence instruction set including a logistics triggering time, a deployment path coordinate and a requirement meeting terminal processing plant feeding homogenization.

[0046] Further, the acquisition of the real-time total weight, the spectral reflection sequence, the microwave phase shift amount, the microwave amplitude attenuation amount, the environmental temperature value and the humidity value of the measured paper includes: scanning and collecting the surface of the measured paper by using a near-infrared array sensor to generate the spectral reflection sequence; transmitting a penetrating signal to the measured paper by using a microwave transceiving assembly symmetrically arranged in the recycling bin, and acquiring the microwave phase shift amount and the microwave amplitude attenuation amount by analyzing the phase deviation and energy loss of the penetrating signal after passing through the paper medium; acquiring the real-time total weight by using a weighing unit to sense the force value of the measured paper, and acquiring the real-time environmental temperature value and the real-time environmental humidity value by using an environmental perception unit integrated in the bin.

[0047] In the spectral data acquisition stage, the surface of the measured paper is scanned by using a near-infrared array sensor installed above the feeding port of the recycling bin or on the top of the conveying device, and the sensing wavelength range covers 900nm to 1700nm. When the measured paper passes through the sensing area, the sensor performs linear scanning on the surface of the paper at a preset sampling frequency, and generates a continuous spectral reflection sequence by capturing the diffuse reflection intensity of the specific waveband light of the paper fiber, which contains the characteristic information of the paper surface fiber type, coating and surface humidity; in the deep physical characteristic acquisition stage, a penetrating detection is performed by using a microwave transceiving assembly symmetrically arranged on the inner walls of both sides of the recycling bin, which includes a microwave transmitting antenna and a microwave receiving antenna. The transmitting antenna transmits a penetrating microwave signal of a preset frequency to the measured paper. When the microwave signal penetrates through the paper medium with a certain thickness, the phase and amplitude will change due to the influence of the internal polar molecules (such as water) of the paper. By comparing and analyzing the signal waveforms of the transmitting end and the receiving end, the phase deviation and energy loss after penetrating the paper medium are analyzed, so as to respectively acquire the microwave phase shift amount and the microwave amplitude attenuation amount. This penetrating sensing method can effectively acquire the average moisture content inside the paper package, avoiding the data distortion caused by only detecting the surface moisture.

[0048] In the physical weight collection stage, the force value of the measured paper is sensed by using a weighing unit installed at the bottom of the recycling bin, which is usually composed of multiple high-precision pressure sensors. The pressure exerted by the measured paper and its carrier on the bin bottom is sensed, and the pressure analog signal is converted into a digital weight signal. The force value is extracted in real time and zero-point drift calibration is performed to obtain the real-time total weight of the measured paper. In the environmental benchmark collection stage, the real-time environmental state data is obtained by using an environmental perception unit integrated in the non-contact area inside the bin body. The environmental perception unit includes an integrated digital thermometer and hygrometer for monitoring the real-time environmental temperature and humidity values of the microenvironment inside the recycling bin.

[0049] The collected real-time total weight, spectral reflectance sequence, microwave phase shift, microwave amplitude attenuation, and environmental temperature and humidity values are uploaded in real time to the central processing unit through the industrial bus or wireless transmission network of the data acquisition module for subsequent feature extraction and weight decoupling operations by the resource matching module.

[0050] The data acquisition module, after obtaining the spectral reflectance sequence, microwave phase shift, and microwave amplitude attenuation, further includes a multi-scale wavelet packet decomposition logic: using a preset wavelet basis function to perform time-frequency domain deconstruction on the original signal, separating the high-frequency noise component reflecting environmental interference and the low-frequency feature component reflecting the internal structure of the material; performing energy suppression on the noise frequency interval and reconstructing the signal through a threshold shrinkage algorithm to eliminate baseline drift and shot noise interference generated by the dynamic electromagnetic environment in the recycling bin, and inputting the reconstructed spectral reflectance sequence and microwave characteristic waveform to the resource matching module.

[0051] Specifically, in actual application scenarios, due to the alternating electromagnetic field generated by the operation of high-power motors on site and the signal multipath interference caused by the metal recycling bin body to the electromagnetic wave, the original signal obtained by the data acquisition module is often superimposed with complex non-stationary noise. In order to ensure the accuracy of subsequent feature extraction, the embodiment introduces a multi-scale wavelet packet decomposition preprocessing process after obtaining the spectral reflection sequence, microwave phase offset and microwave amplitude attenuation. First, the system uses a preset wavelet basis function (in this embodiment, Daubechies wavelet basis db4) to decompose the spectral signal collected by the near-infrared array sensor and the microwave original signal generated by the microwave transceiver component into multiple orthogonal frequency band subspaces. During the decomposition process, the algorithm maps the signal to the approximation coefficient covering the low-frequency trend of the material fiber structure, and the burst high-frequency detail coefficient generated by the start-stop of the motor or environmental vibration. Second, for the high-frequency noise components obtained by decomposition, the system uses a threshold shrinkage algorithm to perform energy suppression. By setting a scientific hard threshold or soft threshold function, the system can accurately identify and reduce the shot noise hidden in the detail coefficient and the baseline drift generated by the internal environment of the recycling bin. This step effectively reduces the energy amplitude of the background electromagnetic interference while preserving the physical nature of the material, and then the system performs inverse wavelet transform on the processed frequency band coefficients to restore the reconstructed spectral reflection sequence and the reconstructed microwave feature waveform with high signal-to-noise ratio. The reconstructed waveform removes non-characteristic burrs and offsets, making the waveform features more prominent and directly improving the data quality input to the resource matching module.

[0052] Through signal reconstruction, closed-loop calibration, and logistics time efficiency coordination, the purity of the underlying data and the long-term accuracy of the system are enhanced, and the unloading smoothness is improved, which helps to improve the coordination efficiency and flow stability of resource allocation in the whole process on the basis of realizing material homogenization.

[0053] Through the multi-source sensing matrix, the physical properties of the measured paper and the environmental parameters are synchronously perceived, and the single-dimensional physical weighing is converted into multi-dimensional data description containing surface fiber characteristics, deep water content state, and environmental influence factors. This collection mode effectively alleviates the problem of incomplete data representation caused by the physical heterogeneity of recycled resources. Through the coordination of penetrating microwave signals and real-time environmental temperature and humidity, the interference of external environmental fluctuations on material property determination is significantly reduced. This not only provides high timeliness and cross-environment consistency for the subsequent asset value quantification and attribute decoupling, but also supports the precise alignment of heterogeneous data on both supply and demand ends from the perception level, thereby helping to achieve the goal of material homogenization management in a dynamic flow environment.

[0054] Further, the process of outputting the standard dry basis asset weight data comprises: extracting a slope of a phase change with frequency of the microwave signal propagating in the paper package medium to obtain a group delay parameter; calculating a unit decibel ratio of the microwave phase shift amount and the microwave amplitude attenuation amount, and performing normalized vector combination on the unit decibel ratio and the group delay parameter, and obtaining the paper package deep moisture content data through a modulus value mapping of the normalized vector;

[0055] Performing a standard normal variable transformation on the spectral reflection sequence to eliminate background baseline noise caused by optical path difference; positioning a local extreme point of the transformed sequence through a sliding window, extracting a wavelength coordinate and a peak intensity corresponding to the local extreme point, and constructing the wavelength coordinate and the peak intensity as the fiber grade feature vector;

[0056] Obtaining an intermediate weight value by subtracting a product of the real-time total weight and the paper package deep moisture content data from the real-time total weight; retrieving a corresponding equilibrium moisture compensation operator in a preset fiber equilibrium moisture mapping table according to the environmental humidity value, multiplying the intermediate weight value by the equilibrium moisture compensation operator, and outputting the standard dry basis asset weight data.

[0057] Specifically, the resource matching module receives the multi-dimensional original signal transmitted by the data acquisition module, and performs feature extraction and weight decoupling through the following logical steps to realize the quantitative conversion of the measured paper physical asset value:

[0058] In the process of outputting the standard dry basis asset weight data, the system first extracts a slope of a phase change with frequency of the microwave signal propagating in the paper package medium, thereby calculating the group delay parameter to represent the equivalent thickness and density characteristics of the medium. Simultaneously, the system calculates a unit decibel ratio of the microwave phase shift amount and the microwave amplitude attenuation amount by analyzing the energy loss of the microwave in the medium, which reflects the modulation intensity of the water polarity in the unit volume to the microwave signal. Then, the system performs normalized vector combination on the unit decibel ratio and the group delay parameter, and maps the modulus value of the normalized vector to a pre-stored modulus-humidity mapping curve, thereby outputting the paper package deep moisture content data. This mapping method of density compensation through group delay ensures that the moisture content data can reflect the real deep moisture state inside the paper package. The modulus-humidity mapping curve is obtained by the following method: a plurality of paper package samples with different densities and fiber grades are preselected, the true moisture content is measured by the oven drying method as the true value, and the normalized vector modulus value generated by the corresponding microwave signal is recorded. The least squares method is used to fit the modulus value and the true moisture content value to generate a continuous modulus-humidity mapping curve.

[0059] The normalization vector combination is specifically: firstly, since the unit decibel ratio and the group time delay parameter belong to different categories of physical quantities, the system utilizes maximum minimum normalization logic to respectively map the original values of the two parameters to the dimensionless interval between 0 and 1, secondly, the normalized unit decibel ratio is taken as a longitudinal characteristic component, and the normalized group time delay parameter is taken as a transverse characteristic component, so that the two discrete physical characteristics are combined and defined as a two-dimensional normalized vector pointing to a specific coordinate point in the characteristic coordinate system, then, the system obtains a comprehensive characteristic modulus reflecting the internal physical properties of the resource by calculating the geometric span of the two-dimensional normalized vector from the origin in the coordinate space (i.e. calculating the arithmetic square root of the sum of the squares of the two components), and finally, the system retrieves the preset modulus-moisture content mapping curve through the modulus to output the paper package deep moisture content data with density compensation characteristics.

[0060] In the specific process of constructing the fiber grade feature vector, the system performs the following standardized data dimension reduction and feature mapping operations: the system first performs a standard normal variable transformation on the collected spectral reflection sequence. This process eliminates the optical path changes and baseline noise caused by the differences in the surface physical structure of different tested samples by subtracting the sequence mean and dividing by the standard deviation, so that the spectral data of different batches are comparable.

[0061] The system uses a preset size sliding window to perform local extreme value retrieval in the transformed sequence, the size of the sliding window is set to 11 consecutive data points (corresponding to a wavelength width of about 8 nanometers), and the moving step is 1 data point. The system compares the values of the center point and the adjacent points of the window to identify the wavelength position reflecting the fiber absorption characteristics.

[0062] In order to ensure the fixed dimension of the feature vector and the calculation stability of the subsequent model, the system accurately selects the top 8 significant extreme points from all the retrieved local extreme values in descending order of peak intensity. If the number of extreme points detected by the sample is less than 8, zero padding is performed at the end of the vector.

[0063] The final generated fiber grade feature vector is set to a fixed 16-dimensional structure. The vector is composed of the first 8-dimensional wavelength coordinate data (ranging from 900 nanometers to 1700 nanometers) and the last 8-dimensional peak intensity data (normalized values) spliced in order.

[0064] The feature points captured by the vector directly correspond to the key components of the paper fibers: for example, the peak near 970 nanometers reflects the moisture content of the paper; the feature points near 1200 nanometers reflect the polysaccharide structure of cellulose and hemicellulose; and the features near 1450 nanometers are associated with the hydrogen bond stretching vibration inside the fiber. The combination of these wavelengths and intensities constitutes the core basis for determining the fiber quality grade.

[0065] In the process of outputting the standard dry basis asset weight data, the system performs a two-stage operation of moisture stripping and environmental compensation. First, the system subtracts the product of the real-time total weight and the deep moisture content of the paper package from the real-time total weight to exclude the free moisture content from the physical total weight, obtaining the intermediate weight value. To offset the dynamic interference of external environmental humidity on the weight of the fiber, the system retrieves the corresponding equilibrium moisture compensation operator from a preset fiber equilibrium moisture mapping table according to the environmental humidity value. The mapping table presets the moisture absorption weight increase ratio of the fiber under different humidity benchmarks. Finally, the system multiplies the intermediate weight value by the equilibrium moisture compensation operator to output the standard dry basis asset weight data, which serves as an asset indicator with cross-environment consistency, effectively solving the asset accounting deviation in the recycling process of renewable resources. The fiber equilibrium moisture mapping table adopts a multi-dimensional structure composed of a group of sub-tables corresponding to different environmental temperature values. Before performing the compensation calculation, the system first locates the target temperature sub-table according to the real-time environmental temperature value obtained by the environmental perception unit, and then retrieves the corresponding equilibrium moisture compensation operator in the sub-table according to the real-time environmental humidity value. In this embodiment, the system presets the fiber equilibrium moisture mapping table in the database associated with the resource matching module. Table 1 shows part of the typical data of the mapping table.

[0066] Table 1: Example of Fiber Equilibrium Moisture Mapping

[0067]

[0068] The resource matching module also includes a self-calibration closed-loop feedback link that real-time acquires the material charging quality inspection true value output by the terminal processing plant and calculates the multi-dimensional residual error between the standard dry basis asset weight output value and the inspection true value. According to the characteristic distribution of the residual error, the gradient descent algorithm is used to dynamically adjust the correction coefficients of the equilibrium moisture compensation operator and the microwave-moisture content mapping curve for adaptive compensation of physical sensing errors.

[0069] In actual operation, due to sensor aging, environmental mutations or long-period drift of material types, fixed mapping parameters may cause asset quantization errors to gradually increase. In order to ensure the long-term reliability of the system, the resource matching module establishes a self-calibration mechanism connected to the feedback link of the terminal processing plant.

[0070] Firstly, when the recycled resources arrive at the terminal processing plant and enter the production line, the system obtains the true value of the batch of resources through the industrial interface. The true value is usually obtained by the factory laboratory through oven method or high-precision sample analysis. Then, the system performs bias traceability operation. The calculation module extracts the standard dry basis asset weight output by the resource matching module when the batch of resources is discharged from the warehouse and the true value of the batch of resources obtained by the factory laboratory through the oven method or high-precision sample analysis. By calculating the algebraic difference between the two, a multi-dimensional residual sequence is generated. The residual reflects the cumulative prediction bias caused by environmental humidity compensation error or microwave mapping deviation. Then, the system uses the gradient descent algorithm to perform parameter optimization. If the residual distribution shows a significant systematic deviation, the algorithm will adjust the correction coefficient in the balanced moisture compensation operator by iteratively calculating the residual as the objective function. For example, when it is detected that the actual dry basis weight is consistently lower than the predicted value, the system will lower the weight of the compensation operator by a preset step size until the residual converges within the preset error tolerance.

[0071] In the process of performing self-calibration closed-loop feedback, the acquisition of the preset step is not a fixed value, but a dynamic analysis result based on the residual intensity and the operator sensitivity. First, the system calls the resource matching module to record multiple sets of historical data in a preset period, calculates the average absolute value of the residual between the standard dry basis asset weight and the terminal feedback true value, and determines the systematic deviation level of the current sensor system by analyzing the distribution characteristics of the residual under different environmental humidity values. Second, the system performs a sensitivity test of the compensation operator, and calculates the displacement of the output standard dry basis asset weight when the balance moisture compensation operator changes by a unit increment, thereby establishing the influence factor of the operator on the final quality determination. The system defines this influence factor as the step reference to ensure that the adjustment action does not cause a dramatic jump in asset value. Subsequently, the preset step is obtained by multiplying the learning rate and the normalized residual. The system presets a convergence increment in the interval [0.001, 0.01] as the basic learning rate, and multiplies it by the normalized residual of the current batch. When the residual is large, the calculated preset step increases accordingly to accelerate error convergence. When the residual approaches zero, the step automatically shrinks to ensure that the balance moisture compensation operator can smoothly approach the true value in the search precision of the fiber balance moisture mapping table. Finally, to prevent numerical oscillation during the adjustment process, the system compares the preset step with the preset discrete step reference in the mapping table (such as the compensation change rate corresponding to a temperature step of 0.1°C) after obtaining the preset step. If the calculated step exceeds the preset multiple (such as 1.5 times) of the reference, the system performs a truncation speed limit, forcing the minimum index interval of the mapping table sub-table to be used as the current execution step. Through this closed-loop self-calibration process, the systematic deviation caused by changes in the physical sensing environment is dynamically corrected, improving the long-term accuracy of asset accounting and ensuring that the balance moisture compensation operator and the microwave-moisture content mapping curve can continuously adapt to complex flow environments, providing more deterministic pre-sequence data support for homogenization of raw materials.

[0072] By mapping the microwave group time delay characteristics and the decibel ratio value, and combining the spectral extreme point extraction after standard normal variable transformation, the deep analysis of the deep moisture content and fiber grade characteristics of recycled resources is realized. The coupling relationship of physical parameters is used to offset the interference of material density and surface noise, and the balance moisture compensation operator is multiplied by the intermediate weight value to reduce the random influence of environmental humidity fluctuations on asset weight determination, thereby improving the physical accuracy and management consistency of standard dry basis asset weight data and providing reliable quantitative support for precise alignment of heterogeneous data at both supply and demand ends.

[0073] Further, the inventory evolution model includes an environmental stress layer that receives real-time environmental temperature and humidity as input, generates an environmental partial pressure difference feature by calculating the difference between the saturated vapor pressure and the actual vapor pressure at the current temperature and humidity.

[0074] a moisture migration layer, inputting a historical moisture content time series constructed from the deep moisture content of the paper package and the environmental partial pressure difference feature into a long short-term memory network, converting discrete moisture content fluctuations into continuous internal energy state latent vectors, and mapping and calculating with the environmental partial pressure difference feature, outputting a non-equilibrium potential difference feature;

[0075] a structure calibration layer, inputting fiber grade features and the non-equilibrium potential difference feature into a multi-layer perceptron, learning the effective diffusion resistance distribution of different fiber grades at different aging stages, and performing weight coupling operations on the potential difference feature and the resistance coefficient to generate an effective evolution rate feature.

[0076] a boundary constraint layer, receiving the standard dry basis asset weight and the effective evolution rate feature as inputs, taking the average of the effective evolution rates at the current sampling time and the previous sampling time, and performing a multiplication operation with the sampling step to output an instantaneous performance loss value;

[0077] a space-time accumulation layer, receiving the instantaneous performance loss value as the current input, and obtaining the historical loss accumulation value stored at the previous sampling time, performing a first-order recursive superposition operation to generate an asset performance total loss value, mapping the asset performance total loss value to [0, 1] through a nonlinear mapping logic, and outputting a resource decay weight, the specific process is as shown in Fig. 2 .

[0078] Specifically, the environmental stress layer serves as the input interface of the model, and the system obtains real-time environmental temperature values and real-time environmental humidity values collected by the environmental perception unit in real time. In this layer, the system pre-stores a saturated vapor pressure table (in this embodiment, the pressure reference values in the range of 0°C to 50°C with a step of 0.1°C are recorded). The system first retrieves the corresponding saturated vapor pressure according to the current temperature, then multiplies the real-time environmental humidity value with the saturated vapor pressure to calculate the actual vapor pressure in the current atmosphere, and generates an environmental partial pressure difference feature by calculating the difference between the saturated vapor pressure and the actual vapor pressure. This partial pressure difference is defined as the escape potential energy that induces the internal moisture of the fiber to migrate outward or absorb external moisture, and is encapsulated as the environmental partial pressure difference feature to provide driving parameters for subsequent levels.

[0079] The moisture migration layer uses a long short-term memory network to process moisture content data with time evolution characteristics. The system inputs the historical moisture content time series constructed by the deep moisture content of the paper package in the past 12 hours and the environmental partial pressure difference feature into a LSTM network with a hidden layer dimension of 128. The LSTM maps the discrete fluctuation data to a 128-dimensional high-dimensional feature space through internal memory gating mechanisms, generating an internal energy state latent vector containing historical trend information. The vector encodes the dynamic evolution law of the internal moisture distribution of the fiber, reflecting the blocking characteristics of the fiber pore structure on moisture diffusion. The 128-dimensional latent vector is then coupled with the current environmental partial pressure difference feature through a fully connected layer for spatial mapping, compressing the high-dimensional physical state into a scalar non-equilibrium potential difference feature.

[0080] The structure calibration layer uses a multilayer perceptron with a three-layer fully connected architecture to realize dynamic calibration of resource physical resistance. The system sets two hidden layers with 64 and 32 neurons respectively to process complex material structure constraints. The system inputs the fiber grade features containing spectral feature points and the non-equilibrium potential difference feature as joint input. The system uses the pre-trained diffusion model parameters for fiber density, lignin content, and porosity stored in the multilayer perceptron to map abstract grade features to continuous physical parameters, thereby automatically fitting the effective diffusion resistance distribution corresponding to the material at the current aging stage, i.e., the resistance coefficient. For example, for paper with dense fiber structure, the model outputs a higher resistance coefficient (such as 0.85), and vice versa, which reflects weaker barrier ability. The system couples the non-equilibrium potential difference feature as an evolution driving force and performs weighted multiplication with the inverse of the resistance coefficient. This weight coupling process physically follows the non-equilibrium state migration law that the evolution rate is proportional to the driving force and inversely proportional to the diffusion resistance, and finally outputs the effective evolution rate feature that accurately quantifies the speed of quality degradation.

[0081] The boundary constraint layer is responsible for performing numerical integration. The system obtains the effective evolution rate at the current sampling time and the previous sampling time, calculates their average value, and multiplies it with the standard dry basis asset weight and the sampling step (such as every 10 minutes), to obtain the instantaneous performance loss value in the current period. The space-time accumulation layer performs first-order recursive superposition operation, i.e., superimposes the current instantaneous loss on the historical loss accumulation value stored in the system register to generate the total loss value of asset performance. Finally, the system uses nonlinear normalization logic (Sigmoid function in this embodiment) to map the total loss value to the [0, 1] interval and outputs the resource decay weight. The closer the weight value is to 1, the closer the resource is to the original quality, and the closer the weight value is to 0, the more serious the loss.

[0082] In order to enable the inventory evolution model to accurately capture the physical evolution law of the tested resource under different environmental stresses, the embodiment constructs a training dataset through offline experiments. First, regenerated paper samples covering different fiber grades (corresponding to the fiber grade feature vector) are preselected and placed in a simulated warehouse with high-precision temperature and humidity control capability. During the 360-hour accelerated aging evolution period, the system records the environmental partial pressure difference feature and the paper package deep moisture content time series data obtained through microwave sensing at a sampling step of 10 minutes. These multi-dimensional time series data constitute the feature space of the model input layer.

[0083] In order to obtain the target label for supervised learning, at each sampling step, the residual tensile strength and breaking length of the paper fiber are measured synchronously using high-precision laboratory instruments. The system normalizes these measured physical performance indicators to measured loss values between 0 and 1 through a pre-set standardization mapping logic. The measured loss value is used as the Ground Truth in the training phase and directly corresponds to the resource decay weight at the output end of the model. By using the measured physical failure data as the label, the weight output by the model has a clear physical evolution significance.

[0084] In the loss function design of model training, the system uses a joint loss function to perform backpropagation optimization. The main loss function uses a mean square error function to minimize the deviation between the predicted resource decay weight of the model and the measured loss label in the laboratory, ensuring the prediction accuracy of the model in the macro trend. In order to make the internal energy state latent vector have a real physical evolution logic, a physical consistency constraint term based on Fick's law is introduced into the loss function. This constraint term penalizes non-equilibrium potential difference feature jumps that do not comply with the moisture migration rate law, forcing the latent vector generated by LSTM to comply with the energy conservation and migration law in the moisture diffusion process.

[0085] Finally, through the above training framework, the internal energy state latent vector can spontaneously capture the potential energy state inside the fiber through the history moisture content sequence under the memory gating mechanism of the LSTM layer. The multi-layer perceptron in the structure calibration layer automatically fits the effective diffusion resistance distribution by learning the strength decay rate difference of different fiber grade samples at different aging stages. This supervised learning method with physical constraints enables the model to output resource decay weights with high signal-to-noise ratio and definite physical meaning when processing unseen heterogeneous data, thereby providing scientific support for subsequent decision matrix construction.

[0086] The data input and synchronization mechanism of the inventory evolution model is as follows: the system sets a unified global sampling clock with a sampling period of 10 minutes. Each data acquisition module (microwave phase offset, microwave amplitude attenuation, spectral reflectance sequence, ambient temperature and humidity) performs synchronous acquisition according to this global clock to ensure that all physical sensing signals are aligned on the time axis.

[0087] In the environmental stress layer, the system acquires real-time ambient temperature and relative humidity once every sampling period (10 minutes) and performs environmental stress calculation. If the environmental sensing unit fails to provide temperature and humidity data in a certain sampling period due to hardware failure or other reasons, the system adopts a forward filling strategy, that is, it temporarily continues to use the temperature and humidity values ​​of the previous sampling period and marks the data point as an interpolation state. After the hardware is restored, the data is filled in and the model is recalculated.

[0088] In the moisture migration layer, the system constructs a new historical moisture content time series every 10 minutes. This series contains moisture content data points from the past 12 hours, totaling 72 points (72 × 10 minutes = 12 hours). When the system has not accumulated 12 hours of data in the early stages of operation, a warm-up strategy is adopted. In the first 720 minutes, the system only inputs the existing valid data (less than 72 points) into the LSTM for rolling processing, while marking the sequence length as partial. After accumulating 12 hours, the system switches to the standard 72-point input mode.

[0089] The output (resource decay weight) of the inventory evolution model is also set to update every 10 minutes. That is, the system calculates a new resource decay weight value every 10 minutes and updates the corresponding resource status parameters stored in memory. This synchronization mechanism ensures that the input data of each level of the model are strictly aligned on the time axis.

[0090] By dynamically decoupling environmental stress and fiber aging logic through an inventory evolution model, the system achieves a digital transformation from static snapshots to dynamic quality evolution. By using a long short-term memory network to extract loss trends and combining diffusion resistance to calibrate the evolution rate, it can more objectively reflect the decay characteristics of fiber performance. This not only improves the real-time accuracy of asset accounting, but also provides a timely basis for decay weights for subsequent proportioning and scheduling, which helps to ensure the stability of material quality throughout the entire resource flow process.

[0091] Furthermore, the process of generating quality grade data includes: using the resource attenuation weight to perform a proportional deduction calculation on the fiber grade characteristics to determine the residual performance quantification value of the paper under the current environmental influence; aligning the residual performance quantification value with multiple preset quality evaluation benchmarks to determine the quality management interval in which the residual performance quantification value is located; extracting the grade label corresponding to the quality management interval to confirm it as quality grade data of recycled resources.

[0092] Specifically, first, the system performs a quantitative evaluation of residual performance, the resource matching module obtains the fiber grade characteristics (in this embodiment, it is represented as the initial quality score mapped based on the intensity of spectral characteristics, with a score interval of 0-100), and the resource attenuation weight (interval 0-1) is obtained. The system uses the resource attenuation weight to calculate the proportional deduction of the fiber grade characteristics, that is, through the multiplication operation of the two, the residual performance quantitative value of the paper under the current environmental influence is determined; then, the system performs quality interval alignment and determination logic, the system aligns the calculated residual performance quantitative value with the preset multiple quality evaluation benchmarks, and the specific quality evaluation benchmark mapping table is shown in Table 2. The quality evaluation benchmark is a score boundary preset based on industry standards. The system determines the specific numerical range in which the residual performance quantitative value falls, that is, determines the quality management interval to which it belongs, and finally, the system completes the extraction and confirmation of the grade label. The system extracts the corresponding grade label from the grade database according to the determined quality management interval, and confirms it as the quality grade data output of the batch of recycled resources. This grade data is then input into the subsequent decision matrix as the core decision basis.

[0093] Table 2 Quality evaluation benchmark mapping table

[0094]

[0095] By dividing the residual performance quantitative value after dynamic deduction into four core management intervals, the system maximizes the reduction of classification complexity in warehouse logistics under the premise of ensuring quality accuracy. This four-level differentiation mapping logic not only captures the real-time damage of the environment to the value of resources through two-level and three-level labels, but also avoids the logistics stagnation and calculation redundancy caused by too fine classification through reasonable benchmark setting. This provides standardized data indexing for subsequent generation of homogenization matching instructions with high executability, effectively solving the quality-price alignment problem in the whole process of recycled resource circulation.

[0096] Further, the process of constructing the decision matrix comprises: extracting the real-time inventory saturation of the recycling node as a first decision dimension, and extracting the resource attenuation weight as a second decision dimension, to construct a two-dimensional attribute mapping space; establishing orthogonal judgment boundary lines in the two-dimensional attribute mapping space according to a preset storage capacity critical constraint and a resource quality threshold, and dividing the mapping space into a plurality of rectangular attribute management areas through the orthogonal judgment boundary lines; wherein each rectangular attribute management area corresponds to a combination of a saturation interval and a preset resource attenuation weight interval determined by the orthogonal judgment boundary lines; for each rectangular attribute management area, a corresponding physical intervention parameter sequence is configured; the physical intervention parameter sequence includes a scheduling priority reflecting the transit timeliness and a preprocessing intensity level reflecting the quality maintenance demand; the real-time coordinate point of the current recycling resource in the two-dimensional attribute mapping space is topologically matched with the rectangular attribute management area to determine the target area to which the recycling resource belongs, and the physical intervention parameter sequence associated with the area is extracted to generate a decision matrix including a resource survival control state and a preprocessing intensity level instruction.

[0097] Specifically, first, the decision module obtains the real-time inventory saturation of the recycling node in real time, and defines it as the first decision dimension reflecting the warehouse physical pressure; at the same time, the resource attenuation weight output by the previous inventory evolution model is called, and it is defined as the second decision dimension reflecting the resource residual value; the system projects the above two dimensions orthogonally to construct a two-dimensional attribute mapping space in the logic layer, and in this space, each point represents the inventory-quality dual state coordinates of a certain resource in a certain node.

[0098] Secondly, the system establishes orthogonal judgment boundary lines in the two-dimensional attribute mapping space according to a preset storage capacity critical constraint and a resource quality threshold, which act as logical scales in the space. By cutting the continuous coordinate space into discrete rectangular attribute management areas, the semantic classification of complex physical states is realized, and each rectangular attribute management area logically locks a combination of warehouse urgency (determined by the saturation interval) and resource loss state (determined by the resource attenuation interval).

[0099] The saturation span refers to a discrete numerical interval preset for the real-time inventory saturation of the recycling node, used to represent the physical pressure state of the warehouse space. Specifically, the system divides the saturation horizontal axis into multiple spans with different management implications according to the preset storage capacity critical constraint: safe span ([0, 0.50]): represents that the warehouse space is adequate, at this time the warehouse urgency is low, and the scheduling priority is set to normal. Warning span ([0.50, 0.80]): represents that the inventory has approached the critical point, the warehouse urgency is improved, and the system starts to activate the resource scheduling model to execute potential scheme optimization. Storage capacity saturation span ([0.80, 1.0]): represents that the warehouse space is about to be exhausted, the warehouse urgency reaches the highest, and the corresponding forced transfer state in the decision matrix.

[0100] In the process of constructing the decision matrix, the embodiment logically divides the resource attenuation weight (whose value range is [0, 1]) into multiple resource attenuation weight regions. The resource attenuation weight region refers to a discrete numerical interval established in the vertical axis direction of the two-dimensional attribute mapping space based on the physical performance evolution law of the fiber structure of renewable resources (such as waste paper). Each region represents the specific performance attenuation state and value survival level of the material in the current storage environment. The specific definition is as follows: high value preservation region ([0.85, 1.0]): represents that the fiber structure of the material is complete and is affected little by environmental stress, and is in the most stable stage of physical performance. Performance fluctuation region ([0.60, 0.85]): represents that the internal moisture content fluctuation of the material has induced preliminary fiber aging or strength attenuation, and the performance is in a dynamic rheological state. Serious loss region (for example, [0, 0.60]): represents that the total loss value of asset performance has reached the critical point, and the industrial utilization value of the material faces a significant risk of decline.

[0101] Subsequently, for each rectangular attribute management region divided, the system pre-configures a corresponding physical intervention parameter sequence in the underlying association. The physical intervention parameter sequence is the basis for the execution layer to execute specific actions, including scheduling priority reflecting transit timeliness and preprocessing intensity level reflecting resource quality maintenance needs. In the real-time operation process, the system extracts the real-time coordinate point of the current recycling resource in space and uses spatial geometric relationships to perform region topology matching with each rectangular attribute management region. Once the target region to which the resource belongs is determined, the system immediately extracts the pre-associated physical intervention parameter sequence in the region.

[0102] Finally, the decision module logically encapsulates the matched scheduling priority and preprocessing intensity level to generate a decision matrix including resource survival control state (used to define the transfer urgency of the resource) and preprocessing intensity level instruction (used to standardize the maintenance action intensity). The decision matrix is finally output as a decision result and directly issued to the logistics scheduling terminal and station processing equipment through this complete process from coordinate mapping to topology matching.

[0103] By mapping the inventory saturation and resource attenuation weight to a two-dimensional attribute space, the quantitative correlation between physical pressure and quality risk is realized. The orthogonal boundary division and topological matching mechanism are used to alleviate the hysteresis and subjective factor interference in the decision-making process. The process makes the scheduling parameters more consistent with the resource status, improves the balance between safety and quality maintenance of inventory capacity, and improves the objectivity and real-time response capability of the decision-making process.

[0104] Further, the processing process of the collection and dispatching module includes: real-time calling the production line material quality threshold demand of the terminal processing plant as the target guide reference of global scheduling; real-time synchronizing the standard dry basis asset weight and quality grade data of each recycling node, combining the real-time state in the decision matrix, completing multi-dimensional data alignment based on timestamp, and constructing a dynamic scheduling task pool; monitoring the resource survival control state in the decision matrix, and when the state changes and / or the inventory capacity touches the threshold, automatically extracting the decision matrix and the material quality threshold as the physical boundary constraint, and activating the resource scheduling model; according to the model calculation result, outputting the delivery order instruction set containing the logistics trigger time, the allocation path coordinates and meeting the terminal processing plant material homogenization requirements.

[0105] Specifically, first, the disposal scheduling module real-time calls the production line feeding quality threshold demand of the terminal processing plant, which covers the minimum tolerance of the production link to the raw material fiber strength, average moisture content and impurity content. The system sets it as the target guiding benchmark of global scheduling, which is used to reverse screen the material packages that meet the conditions in the whole network, to ensure that the resources flowing out of each recycling node can accurately align the production process requirements when they arrive at the terminal; second, the system synchronizes the standard dry basis asset weight and corresponding quality grade data of each distributed recycling node in real time through the network layer, and associates it with the real-time state of the decision matrix. In order to eliminate the signal asynchronous problem caused by geographical distribution, the system uses a unified global clock pulse to perform alignment operation on the above multi-dimensional data based on time stamp. Specifically, the system encapsulates the material weight, grade, location and control urgency at the same time node into a structured data unit, and aggregates and builds into a dynamic scheduling task pool, providing real-time and high-fidelity data foundation for subsequent global optimization; then, the system continuously monitors the resource storage control state recorded in the decision matrix. Once it is monitored that the state of a node changes (for example, from regular turnover to forced disposal), or the real-time inventory of the node touches the storage threshold, the system will automatically extract the decision matrix parameters of the node and the feeding quality threshold of the terminal, and uniformly convert them into physical boundary constraints of the resource scheduling model. This process realizes the conversion from logical research and judgment to physical limitation, and changes the necessity of management into a hard boundary condition in model calculation; finally, under the limitation of the above physical boundary constraints, the system activates the resource scheduling model to perform path and ratio optimization calculation. According to the model calculation result, the disposal scheduling module finally outputs a real-time dynamic instruction set containing multiple dimensions, which specifically includes: logistics trigger time, which is used to specify the accurate time of vehicle departure or loading equipment start; deployment path coordinates, which are the optimal logistics driving trajectory composed of a series of geographical longitude and latitude; and delivery sequence, which clearly defines the logic of resources from different sources arriving at the processing plant in sequence. The instruction set is directly issued to the logistics execution terminal to drive the physical layer loading and unloading and transfer operations.

[0106] By real-time coupling the terminal feeding demand with the front-end resource storage state, using the multi-dimensional data alignment mechanism based on time stamp, the problem of information fragmentation and poor timeliness in the regenerated resource supply chain is effectively solved, ensuring the authenticity of the dynamic scheduling task pool. By monitoring the state of the decision matrix and automatically extracting the physical boundary constraints, the system can ensure the safety of the storage capacity while performing quality preservation disposal of high-value resources, significantly improving the configuration efficiency and stability of the delivery quality of regenerated resources in the flow network.

[0107] Further, the resource scheduling model comprises an attribute mapping layer: mapping the input standard dry basis asset weight and quality grade data of each recycling node to feature mapping, converting to a supply state vector in a high-dimensional feature space;

[0108] Demand guiding layer: convert the input terminal processing plant production line feed quality threshold demand into a target center vector, and establish the convergence guiding direction of the supply state vector in the feature space;

[0109] Constraint conversion layer: analyze the input decision matrix, convert the resource survival management state into a boundary restriction operator acting on the algorithm logic, realize dynamic pruning by removing node branches that do not meet the management state in the dynamic scheduling task pool, and reduce the search range of the candidate scheme;

[0110] Two-way optimization layer: under the constraint of the boundary restriction operator, the spatial distance between the supply state vector and the target center vector is used to measure the homogenization deviation, and the candidate scheme that minimizes the path cost and quality deviation is locked through iterative calculation;

[0111] Instruction generation layer: extract the quality deviation of each recycling node in the candidate scheme, and use the deviation offset logic containing the stability adjustment benchmark to allocate the loading share; by matching the mixed proportion of positive and negative deviation resources, the linear alignment of total quantity and feed demand is realized, and after simulation verification and error fine-tuning, the delivery order instruction set containing the logistics trigger time, deployment path coordinates and meeting the terminal processing plant feed homogenization requirements is generated, as shown in the specific process of Fig. 3 .

[0112] Specifically, first, the system performs normalization processing of the original data, and the attribute mapping layer receives standard dry basis asset weight and quality grade data from each recycling node. The algorithm first calls the system's preset attribute dimension benchmark. In this embodiment, the quality grade (first to fourth grade) is mapped to the preset grade weight distribution (first grade corresponds to 0.95, second grade corresponds to 0.80, third grade corresponds to 0.65, and fourth grade corresponds to 0.50), and the standard dry basis asset weight is divided by the historical maximum carrying capacity of the node, thereby converting different dimension original physical data into dimensionless normalized feature components in the [0, 1] interval.

[0113] Second, the demand guiding layer establishes the gravitational center of the algorithm optimization. The system obtains the feed quality threshold demand (such as the required fiber strength benchmark value) output by the terminal processing plant production line in real time, and converts it into a target center vector in the feature space. The role of this layer is to establish the convergence guiding direction of the supply state vector in the space, and all subsequent path selection and resource combination actions are driven by the core driving force of reducing the geometric deviation of each supply point vector from the target center vector. By calculating the cosine similarity between each node supply state vector and the target center vector, the system can preliminarily lock the candidate node set that meets the terminal production requirements.

[0114] Subsequently, the constraint conversion layer performs dynamic space pruning based on management logic. The system deeply analyzes the input decision matrix and converts the resource survival management state (such as forced flow, delayed processing or loss over-limit state) recorded therein into a boundary restriction operator acting on the algorithm optimization. Before performing optimization calculation, the system performs physical elimination on node branches that do not meet the management requirements in the dynamic scheduling task pool using the operator, that is, dynamic pruning. Through this mechanism, the system directly blocks paths that do not have clearance conditions, greatly reduces the search range of the global optimization candidate scheme, and ensures the real-time performance and effectiveness of decision generation in large-scale recycling networks.

[0115] When analyzing the decision matrix, the constraint conversion layer of the resource scheduling model further includes performing multi-node time alignment discrimination logic: based on the geographic coordinates of each recycling node, real-time traffic and vehicle driving characteristics, predicting the estimated arrival time window of each node; setting the queuing length threshold of the unloading area of the terminal processing plant, constructing a time overlap constraint operator reflecting the unloading frequency; by calculating the intersection of the estimated arrival time window of each node, performing secondary dynamic pruning on low-priority nodes that cause logistics congestion conflicts.

[0116] In the global logistics scheduling process, if multiple recycling nodes with clearance conditions trigger logistics instructions at the same time, it is easy to cause vehicle accumulation in the unloading area of the terminal processing plant, thereby causing logistics congestion and reducing the feeding time efficiency. In order to solve this problem, the constraint conversion layer of the resource scheduling model introduces multi-node time alignment discrimination logic to achieve time axis smoothing processing of logistics flow; the system calls the geographic coordinates of each recycling node in real time, and mounts an external real-time traffic interface to obtain traffic flow data. By combining the average driving speed of the clearance vehicle and the estimated loading operation time based on the standard dry basis asset weight of the material, the system performs comprehensive simulation calculation. This calculation not only calculates the theoretical driving time, but also adds the operation disturbance amount, thereby generating an estimated arrival time window including the earliest arrival time and the latest arrival time for each recycling node. The system establishes the discrimination standard of unloading conflicts, and pre-sets the queuing length threshold of the unloading area of the terminal processing plant to define the maximum carrying capacity of the area per unit time. Based on the threshold, a time overlap constraint operator reflecting the unloading frequency is constructed. This operator is used to monitor the intersection span of the estimated arrival time windows of different nodes. When the arrival times of multiple nodes overlap on the time axis and the overlapping time exceeds the average unloading period of the terminal, the system determines that there is a risk of logistics congestion conflict. When determining that there is a congestion risk, the system will call the decision matrix of each node, compare the scheduling priority recorded therein, and perform secondary dynamic pruning on the node branches that cause conflicts and have lower priority.

[0117] Through signal reconstruction, closed-loop calibration, and logistics timeliness coordination, the scheme enhances the purity of underlying data and the long-term accuracy of the system, improves the unloading smoothness, and helps to improve the collaborative efficiency of resource allocation and the stability of the flow process on the basis of achieving homogeneous feeding.

[0118] The flow of the bidirectional optimization layer first extracts the supply state vector of all recycling nodes involved in the scheme, and synthesizes the comprehensive feature center vector of the batch of materials according to the loading share of each node. The initial value of the loading share is determined by the proportion of the inventory of each node to the total demand, and is dynamically adjusted by the algorithm in the subsequent iteration process. Then, the system calculates the spatial distance between the vector and the target center vector pre-existing in the system. The target center vector is obtained by the system real-time docking the production execution system, according to the process standard of the current paper to be produced (such as the target tensile index, the allowable moisture content range, etc.), the ideal physical characteristics are mapped to the ideal point in the feature space. The spatial distance is specifically represented as the sum of the absolute values of the difference values of each physical feature component. The smaller the distance, the higher the degree of material homogenization. Secondly, in order to ensure the economy of the scheme, the system defines the logistics cost as the logical set of the physical displacement cost in the spatial dimension and the timeliness devaluation cost in the time dimension. The physical displacement cost is obtained by multiplying the real-time navigation mileage of the carrier and the unit energy consumption coefficient. The unit energy consumption coefficient is pre-calculated according to the nominal energy consumption level of the carrier and the real-time load ratio and stored in the database. The timeliness devaluation cost is the product of the effective evolution rate output by the aforementioned inventory evolution model and the predicted transportation time, which is used to quantify the performance value loss of resources during the journey due to fiber aging. The system performs weighted summation on the two through a preset cost balance coefficient to generate a path cost component reflecting the overall logistics expenditure. The cost balance coefficient is pre-calibrated through marginal revenue analysis of historical logistics expenditure and material degradation loss, and is used to dynamically adjust the preference degree of the system for transportation cost and quality preservation.

[0119] Then, the system performs iterative evolution under the constraint of the boundary restriction operator by using a non-dominated sorting genetic algorithm with strategy, which specifically includes an elitist strategy and a crowding degree allocation strategy: the elitist strategy ensures that the optimal non-dominated individuals in each generation are directly reserved to the next generation, preventing high-quality scheduling schemes from being lost in evolution; the crowding degree allocation strategy preferentially selects individuals with uniform distribution in the solution space by calculating the distribution density of the individuals in the solution space, to ensure that the search process covers more diverse path combinations and avoids the algorithm falling into a local optimal solution. During the evolution process, the algorithm continuously optimizes the performance indicators of the candidate schemes through multiple rounds of crossover and mutation operations. When performing optimization calculation, the system introduces the boundary restriction operator converted from the decision matrix in real time. This operator acts as a logical barrier in the evolution process, forcing the algorithm to search within the feasible region of the dynamic scheduling task pool. Any illegal solution containing node branches that have been removed through dynamic pruning will be identified and excluded by the algorithm. Finally, the algorithm locks a set of candidate scheme combinations in the feature space in a Pareto optimal state, which ensures that under the premise of minimizing the physical cost of the logistics path, the comprehensive physical characteristics of the mixed materials can be most accurately aligned with the ideal target center vector of the terminal processing plant, thereby generating the final delivery order instruction set.

[0120] The system retrieves the quality level scores of each recycling node in the candidate scheme (quantitative values from 0.50 to 1.0) one by one and compares them with the real-time feedback of the material quality threshold demand of the terminal processing plant production line. The system automatically divides all nodes into two categories by calculating the algebraic difference between the node quality score and the demand threshold: the high-quality surplus group with a positive difference and the low-quality deficit group with a negative difference. To prevent numerical overflow due to a small deviation during the calculation process, the system introduces a regularization constant (0.01 in this embodiment) as a stability adjustment benchmark. The system calculates the negative correlation influence factor of the sum of the absolute value of each node deviation and the benchmark value as the preliminary quality influence coefficient. The closer the quality of a node is to the demand, the higher the coefficient will be. The system then divides the individual node coefficient by the total sum of the coefficients of all nodes in the scheme to obtain the preliminary loading share percentage. The system performs a physical hedging logic to offset the surplus and the deficit and dynamically adjusts the preliminary share. The adjustment process follows the linear mixing law: the system calculates the average deviation span of the surplus group and the deficit group, taking the material quality threshold demand as the balance point. If the average quality of the surplus group is much higher than the threshold, the system will automatically lower the loading share percentage of the surplus group while proportionally increasing the load of the deficit group. The system retrieves the standard dry basis asset weight of each recycling node in the task pool in real time. If the calculated loading exceeds the real-time inventory of the node, the system will automatically lock the loading share of the node to its inventory upper limit and proportionally shift the remaining difference percentage to other nodes in the same group, ensuring that the adjustment result does not deviate from the real inventory physical boundary.

[0121] Before the instruction is formally generated, the system enters a closed-loop simulation verification mode. The system simulates the average value of the instantaneous components of each node after mixing according to the current share, and calculates the predicted residual error thereof and the target threshold value: if the absolute value of the predicted residual error is greater than a preset accuracy threshold value (in this embodiment, 0.02), the system will dynamically reduce the total percentage of the group on the side with greater deviation by a fixed step (such as 1%), and simultaneously increase the percentage of the other side, and the simulation operation loop is executed until the predicted residual error is reduced to the allowable range, at which time the locked percentage is the final loading share. Finally, the system encapsulates and generates a dynamic instruction set according to the finally determined loading amount share, in combination with the geographical position and real-time road conditions. The instruction set specifies the access sequence of each node, and uses the sequence to drive the resource to form a specific physical stacking layer inside the carrier. When unloading at the terminal, this spatial layering is converted into a controlled time flow sequence, ensuring that the material component ratio entering the production line is accurately aligned with the homogenization feeding requirements of the processing plant.

[0122] The dynamic pruning mechanism effectively reduces the calculation redundancy under a large-scale network, ensuring the real-time nature of the scheduling decision generation. By synergistically regulating the loading share and the access sequence, the system converts the static storage characteristics into a dynamically stable feeding sequence at the physical level, significantly reducing the impact of natural fluctuations in raw materials on the stability of the production line, and improving the level of fine control during the recycling of resources.

[0123] By integrating dynamic judgment of resource quality, inventory evolution early warning, and multi-dimensional scheduling optimization, the present application constructs a quantitative management system for the recycling process of renewable resources. The logical coupling of resource decay weight and inventory pressure helps to reduce the decision-making deviation caused by information lag, improves the quality certainty of discrete resources in storage and transportation, and the bidirectional optimization algorithm takes into account the homogenization feeding requirements of the terminal processing plant while coordinating the path cost, thereby alleviating the impact of natural fluctuations in raw material quality on the stability of the backend production, and improving the synergistic efficiency of resource allocation and the scientificity of decision-making in the whole process.

[0124] Embodiment Two:

[0125] In this embodiment, a waste paper recycling scenario is taken as an example, which includes two supply points (recycling stations A and B), one limited supply point (recycling station C), and one terminal processing plant (paper mill):

[0126] Firstly, the system performs the data perception, physical feature extraction and asset value quantification process of the recycling node. On-site at the recycling station A, the system uses a near-infrared array sensor to perform surface scanning on the stored 16 tons of old corrugated boxes, generating characteristic peak data reflecting the fiber structure. In combination with the microwave transceiver component, the system obtains the microwave phase shift and amplitude attenuation through a penetrating signal. The resource matching module extracts the slope of the microwave phase with respect to frequency to obtain the group delay parameter. The signal attenuation ratio is calculated to obtain the deep moisture content of the paper package. The measured environmental humidity is used to retrieve the fiber equilibrium moisture mapping table to obtain the compensation operator. Through the above operations, the system deducts the moisture interference from the total weight of 16 tons, outputs the standard dry basis asset weight of recycling station A, and preliminarily maps the quality grade to a score of 0.95 (corresponding to a first-class resource weight).

[0127] Secondly, the system performs the inventory evolution simulation and real-time quality grade determination process. The environmental perception unit detects that the local humidity at recycling station A is high. The environmental stress layer in the inventory evolution model calculates the saturated vapor pressure difference accordingly and generates the environmental partial pressure difference feature. The moisture migration layer uses a long short-term memory network to process the moisture content history sequence, combined with the fiber diffusion resistance output by the structure calibration layer, and finally calculates the resource decay weight as 0.90 by the spatio-temporal accumulation layer. Subsequently, the system uses the decay weight to perform proportional deduction on the initial quality score, calculates the residual performance quantization value as 0.855, and aligns it with the preset quality management interval. The quality grade data of recycling station A is dynamically corrected from the initial first-class to the second-class label, realizing the digital transformation from static snapshot to dynamic quality evolution.

[0128] Subsequently, the system performs the decision matrix construction and logical pruning process of the dynamic scheduling task pool. The decision module extracts the real-time inventory saturation of recycling station A as the first decision dimension, and the resource decay weight as the second decision dimension. In the two-dimensional attribute mapping space, the system performs topological matching by determining the boundary line orthogonally. Since the point falls into the high urgency flow area, the system automatically associates the corresponding physical intervention parameters to generate a decision matrix containing high scheduling priority and forced cleaning state. Before activating the resource scheduling model, the cleaning and scheduling module detects the state change and automatically excludes recycling station C that is in the containment state and does not meet the control conditions in the dynamic scheduling task pool through the boundary restriction operator, realizing the dynamic pruning of candidate schemes.

[0129] Finally, the system executes the bidirectional optimization matching and dynamic instruction generation process based on terminal demand. The paper mill processing line issues a feed quality threshold demand of 0.75. The demand guide layer establishes the target center vector accordingly. The bidirectional optimization layer uses the NSGA-II algorithm to find the Pareto optimal balance between path cost and quality deviation (Euclidean distance), and locks the combination scheme of recycling station A and recycling station B. The instruction generation layer compares the quality scores of each station with the demand threshold. Station A (score 0.855) belongs to the high-quality surplus group, with an absolute deviation of 0.105. Station B (score 0.60) belongs to the low-quality deficit group, with an absolute deviation of 0.15. The system introduces a stability adjustment benchmark (regularization constant 0.01) to calculate the quality influence coefficient: the coefficient of station A is 8.70, and the coefficient of station B is 6.25. The system follows the linear mixing law to perform physical hedging to offset the surplus and deficit. Preliminary calculation shows that the loading share of station A is about 58%, and the loading share of station B is about 42%. The system verifies the physical boundary in real time and confirms that the 16-ton inventory of station A is sufficient to support this share demand, without the need to perform share translation. Before issuing the formal instruction, the system enters the simulation verification mode. After calculation, the average value of the mixed components is 0.748, which meets the requirement of the target accuracy threshold (0.02). The system plans the sequence of visiting station A first and then station B, and uses the space stacking structure to convert it into a controlled time flow sequence at the unloading end. Finally, the system generates a dynamic instruction set and issues it to the execution end to drive the physical layer to complete the precise homogenization scheduling.

[0130] While embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and variations can be made to these embodiments without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An Internet of Things-based renewable resource recycling data management system, characterized in that, Comprise: A data acquisition module acquires real-time total weight, spectral reflectance sequence, microwave phase shift, microwave amplitude attenuation, ambient temperature value and humidity value of the measured paper; A resource matching module extracts features from the microwave phase shift and microwave amplitude attenuation, and extracts characteristic peaks from the spectral reflectance sequence, obtains paper package deep moisture content and fiber grade characteristics, and performs moisture weight decoupling operation combined with environmental humidity compensation, and outputs standard dry basis asset weight data; An inventory evolution module inputs the standard dry basis asset weight, paper package deep moisture content, fiber grade characteristics and environmental temperature and humidity into an inventory evolution model to obtain resource attenuation weight, the inventory evolution model comprising: an environmental stress layer receiving real-time environmental temperature and humidity as input, generating an environmental partial pressure difference feature by calculating the difference between the saturated vapor pressure and the actual vapor pressure under the current temperature and humidity; A moisture migration layer inputs the historical moisture content time series constructed by the paper package deep moisture content and the environmental partial pressure difference feature into a long short-term memory network, converts discrete moisture content fluctuations into continuous internal energy state latent vectors, and maps and calculates with the environmental partial pressure difference feature to output an unbalanced potential difference feature; A structure calibration layer inputs the fiber grade characteristics and the unbalanced potential difference feature into a multilayer perceptron to learn the effective diffusion resistance distribution of different fiber grades at different aging stages, and performs weight coupling operation on the potential difference feature and the resistance coefficient to generate an effective evolution rate feature; A boundary constraint layer receives the standard dry basis asset weight and the effective evolution rate feature as input, obtains the average value of the effective evolution rate at the current sampling time and the previous sampling time, and performs multiplication operation with the sampling step to output the instantaneous performance loss value; The space-time cumulative layer receives the instantaneous performance loss value as the current input, obtains the historical loss cumulative value stored at the last sampling time, performs a first-order recursive superposition operation to generate an asset performance total loss value, maps the asset performance total loss value to [0, 1] through nonlinear mapping logic, outputs a resource attenuation weight, and generates quality grade data based on the resource attenuation weight and the fiber grade feature. The process of generating quality grade data includes: using the resource attenuation weight to perform proportional deduction calculation on the fiber grade feature to determine the residual performance quantization value of the paper under the current environmental influence; aligning the residual performance quantization value with a plurality of preset quality evaluation benchmarks to determine the quality management interval in which the residual performance quantization value is located; extracting the grade label corresponding to the quality management interval to confirm the quality grade data of the recycled resources, combining the real-time inventory saturation of the recycling node to construct a decision matrix. The process of constructing the decision matrix includes: extracting the real-time inventory saturation of the recycling node as the first decision dimension and the resource attenuation weight as the second decision dimension to construct a two-dimensional attribute mapping space; establishing an orthogonal judgment boundary line in the two-dimensional attribute mapping space according to the preset storage capacity critical constraint and the resource quality threshold; dividing the mapping space into a plurality of rectangular attribute management areas through the orthogonal judgment boundary line; wherein each rectangular attribute management area corresponds to a group of storage urgency and resource loss state combinations determined by the saturation interval and the preset resource attenuation weight interval defined by the orthogonal judgment boundary line; for each rectangular attribute management area, a corresponding physical intervention parameter sequence is configured respectively; the physical intervention parameter sequence includes scheduling priority reflecting transit timeliness and pretreatment intensity level reflecting quality maintenance demand; the real-time coordinate point of the current recycled resource in the two-dimensional attribute mapping space is topologically matched with the rectangular attribute management area to determine the target area to which the recycled resource belongs, and the region-associated physical intervention parameter sequence is extracted to generate a decision matrix including resource survival control state and pretreatment intensity level instructions; The clean-up scheduling module calls the production line feeding quality threshold demand of the terminal processing plant, inputs the decision matrix of each recycling node, the standard dry basis asset weight, and the quality grade data into the resource scheduling model, and outputs a delivery sequence instruction set containing logistics trigger time, allocation path coordinates, and meeting the terminal processing plant feeding homogenization requirements. 2.The renewable resource recycling data management system based on the Internet of Things according to claim 1, wherein, The acquiring of the real-time total weight, the spectral reflection sequence, the microwave phase offset, the microwave amplitude attenuation, the ambient temperature value and the humidity value of the measured paper includes: scanning and collecting the surface of the measured paper by using a near-infrared array sensor to generate the spectral reflection sequence; transmitting a penetrating signal to the measured paper by using a microwave transceiving assembly symmetrically arranged in the recycling bin, and acquiring the microwave phase offset and the microwave amplitude attenuation by analyzing the phase deviation and energy loss of the penetrating signal after passing through the paper medium; acquiring the real-time total weight by using a weighing unit to sense the force value of the measured paper, and acquiring the real-time ambient temperature value and the real-time ambient humidity value by using an environment sensing unit integrated in the bin.

3. The renewable resource recycling data management system based on the Internet of Things according to claim 1, characterized in that, The process of outputting the standard dry basis asset weight data includes: extracting the slope of the phase change with frequency of the microwave signal propagating in the paper package medium to obtain a group delay parameter; calculating the unit decibel ratio of the microwave phase offset and the microwave amplitude attenuation, normalizing the unit decibel ratio and the group delay parameter to obtain a normalized vector combination, and mapping the modulus value of the normalized vector to obtain the paper package deep moisture content data; The spectral reflection sequence is subjected to a standard normal variable transformation to eliminate background baseline noise caused by optical path difference; the local extreme points of the transformed sequence are positioned by a sliding window, the wavelength coordinates and peak intensities corresponding to the local extreme points are extracted, and the wavelength coordinates and peak intensities are constructed into the fiber grade feature vector; An intermediate weight value is obtained by subtracting the product of the real-time total weight and the paper package deep moisture content data from the real-time total weight; a corresponding equilibrium moisture compensation operator is retrieved in a preset fiber equilibrium moisture mapping table according to the ambient humidity value, the intermediate weight value is multiplied by the equilibrium moisture compensation operator, and the standard dry basis asset weight data is output. 4.The renewable resource recycling data management system based on the Internet of Things according to claim 1, wherein, The processing process of the clearing and transportation scheduling module includes: real-time calling of the production line feeding quality threshold demand of the terminal processing plant as the target guide reference of global scheduling; real-time synchronization of the standard dry basis asset weight and the quality grade data of each recycling node, combination of the real-time state in the decision matrix, multi-dimensional data alignment based on timestamp, and construction of a dynamic scheduling task pool; monitoring the resource survival control state in the decision matrix, automatically extracting the decision matrix and the feeding quality threshold as physical boundary constraints when the state changes and / or the storage capacity threshold is touched, and activating the resource scheduling model; according to the model calculation result, outputting a delivery order instruction set containing the logistics trigger time, the deployment path coordinates and the requirements of the terminal processing plant feeding homogenization.

5. The renewable resource recycling data management system based on the Internet of Things according to claim 4, characterized in that, The resource scheduling model includes: an attribute mapping layer that maps the input standard dry basis asset weight and quality grade data of each recycling node to a supply state vector in a high-dimensional feature space; a demand guide layer that converts the input terminal processing plant production line feeding quality threshold demand into a target center vector, and establishes the convergence guide direction of the supply state vector in the feature space; Constraint conversion layer: analyze the input decision matrix, convert the resource storage and management state into boundary restriction operators that act on the algorithm logic, and achieve dynamic pruning by removing node branches that do not meet the management state in real time in the dynamic scheduling task pool, thereby reducing the search range of candidate solutions; Bidirectional optimization layer: under the constraint of boundary restriction operators, measure the homogenization deviation using the spatial distance between the supply state vector and the target center vector, and lock the candidate solution that minimizes both path cost and quality deviation through iterative calculation; Instruction generation layer: extract the quality deviation of each recovery node in the candidate solution, allocate the loading share using deviation cancellation logic containing stability adjustment benchmarks; achieve linear alignment of total quantity and demand by matching the mixed proportion of positive and negative deviation resources, and after simulation verification and error fine-tuning, generate a delivery order instruction set that includes logistics trigger time, deployment path coordinates, and meets the terminal processing plant's demand for homogenization.

Images