A landfill resource recycling potential prediction method and prediction system

Abstract

This invention discloses a method and system for predicting the resource recovery potential of landfills, suitable for the mining, sorting, pollution control, and land reclamation of aging landfills. The method collects data on site conditions, pollution risks, mining and sorting, waste composition, and resource recovery. It uses the analytic hierarchy process (AHP) to obtain a pollution risk index, define risk levels, and clarify pre-mining stabilization requirements. The process is divided into nodes according to the entire process of mining, multi-level sorting, and resource recovery. A dataset is constructed through feature filtering and dimensionality reduction, and a neural network incorporating multiple constraints is trained to predict the yield of various sorting products. The supporting system combines a resource recovery rule base and an economic calculation model to simultaneously output utilization paths, recovery potential, economic benefits, mining sequence, and reclamation plans. This method can significantly reduce the workload of on-site trial excavation and testing, providing reliable support for landfill remediation layout, process selection, pollution control, and site redevelopment.

Description

Technical Field

[0001] This invention relates to the field of landfill resource recovery technology, specifically to a method and system for predicting the resource recovery potential of landfills. Background Technology

[0002] Early municipal solid waste was mostly collected and landfilled in a mixed manner. Old landfills, after years of decomposition and mineralization, contain humus, plastics, metals, and bricks, posing both pollution risks and recycling value. During remediation, existing waste can be mined and sorted at multiple stages to produce RDF fuel, metals, humus, building materials, and residues. The utilization of these products is constrained by multiple factors, including the landfill's age, degree of decomposition, and pollution risk. Current mining schemes rely heavily on on-site trial excavations, sorting and screening tests, and experience-based assessments, which have drawbacks such as numerous sampling points, long cycles, high costs, and poor sample representativeness. When dealing with old landfills near residential areas, water bodies, or with high leachate risks, assessing mining value solely based on waste composition easily overlooks prior stabilization, pollution control, and safety management requirements.

[0003] Existing research has shown that the analytic hierarchy process (AHP) can be used to determine the weights of pollution risk factors in aging landfills and classify pollution risk levels, providing a basis for the sequence of remediation and the selection of remediation technologies. Studies on exploitation and remediation schemes also indicate that before landfill exploitation, the process route should be determined based on the degree of waste stabilization, screening characteristics, and resource utilization destinations, and pollution control and safety management requirements should be formulated. In engineering practice, situations involving large-scale existing waste, high pollution risk, numerous surrounding sensitive targets, and high unit treatment costs necessitate a comprehensive assessment of resource recovery yield, economic value, and pollution risk before exploitation.

[0004] Currently, conventional machine learning models can be used for data fitting, but without considering material flow, material balance, and pollution risk correction during mining and screening, they are prone to problems such as predicted yields not matching the engineering process, overestimating resource recovery benefits, or providing unreliable reclamation and utilization recommendations. Therefore, a prediction method is needed that is compatible with the actual mining and screening process and can simultaneously incorporate pollution risk, resource yield, and economic value.

[0005] Publication No. CN117610785A discloses a comprehensive assessment method for the excavation of municipal solid waste landfills. This method comprehensively assesses the feasibility of landfill excavation for resource utilization from the perspectives of overall sustainability, including environmental, economic, and energy aspects. A comprehensive assessment of landfill excavation from environmental, economic, and energy perspectives can also better evaluate the necessity of landfill excavation. The steps of this invention are: A) data analysis; B) site investigation; C) site sampling; D) site analysis; E) sample testing; F) model building; G) calculation and assessment. Landfill excavation assessment personnel input basic data into the model and solve for the comprehensive assessment indicators of landfill excavation, including environmental, economic, and energy aspects. This method does not consider pollution risk factors during the mining process, and the impact of process nodes on the prediction results is not considered during model building.

[0006] Publication No. CN117951509B discloses a method and system for predicting mineral composition content, belonging to the field of oil and gas field exploration and development. The method includes collecting measured rock mineral composition content data and well logging data within a preset depth range, and performing preprocessing; selecting and reducing features using the correlation coefficient method and principal component analysis; then constructing a multi-source fusion dataset and training a hybrid neural network model; establishing a loss function for mineral content prediction with custom constraints; and finally using the hybrid neural network model to predict mineral composition content. However, this method only involves the prediction and evaluation of rock mineral composition content, and its model evaluation task is relatively simple. It cannot be used for comprehensive prediction of various types of landfill data, and the model does not consider the impact of mining processes on the prediction results. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a method and system for predicting the resource recovery potential of landfills.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: A method for predicting the resource recovery potential of landfills includes the following steps: S1: Collect basic site data, pollution risk data, mining operation data, screening and sorting data, waste component testing data, and resource utilization data of the landfill to be evaluated; S2: Based on the pollution risk data, the weights of pollution risk factors are determined using the analytic hierarchy process (AHP), and the pollution risk score index R of the landfill to be evaluated is calculated. S3: Perform data consistency processing on the site basic data, pollution risk data, mining operation data, screening and sorting data, waste component detection data and resource utilization data to obtain sample data; S4: Based on the mining nodes, process nodes are mapped to the standardized data to construct process topology data and establish the material transfer relationship between each process node; S5: Extract structured feature variables based on the process topology data, and use the correlation coefficient method to screen key features related to the yield of resource products; S6: Perform principal component analysis on the selected feature variables to construct a multi-source fusion dataset, where the input is the principal component features of the process nodes and the output is the yield and economic value indicators of resource products. S7: Input the multi-source fusion dataset into the neural network model for training, and introduce an engineering constraint loss function; S8: Input the data of the landfill to be evaluated into the trained model to obtain the predicted yield of resource recovery products. Combined with the resource recovery path rule base and economic value model, output the resource utilization path, resource recovery potential value, economic value prediction value, mining priority and land reclamation suggestions.

[0009] Further, in step S2, the pollution risk score index R is calculated using the following formula: R = ; Wherein, Wi is the weight of the i-th pollution risk factor, Si is the score of the i-th pollution risk factor, and n is the number of pollution risk factors; the pollution risk factors include one or more of the following: degree of waste stabilization, leachate pollution risk, landfill gas release risk, groundwater pollution risk, surface water pollution risk, odor impact risk, sensitivity of surrounding residential areas, sensitivity of surrounding water bodies, soil cover integrity, and anti-seepage system integrity.

[0010] Furthermore, based on the pollution risk scoring index R, the pollution risk level, pre-mining stabilization requirements, pollution prevention and control measures, and safety management measures are determined. When the pollution risk scoring index reaches a preset high-risk threshold, the landfill to be evaluated is identified as a high-pollution-risk landfill, and stabilization treatment measures are configured before mining and utilization. The stabilization treatment measures include one or more of the following: aerobic aeration, ventilation and dehumidification, leachate collection and treatment, landfill gas drainage, odor collection and treatment, dust prevention and suppression, safety isolation, and zoned excavation.

[0011] Furthermore, in step S3, the data consistency processing includes unit unification, missing data repair, anomaly correction, structured coding, and benchmark conversion.

[0012] Furthermore, in step S4, the mining node includes waste mining, conveying, manual sorting, drum screening, magnetic separation, secondary screening, and air separation processes.

[0013] Furthermore, in step S7, the neural network model is a structured prediction model built based on the process topology. Its input structure is grouped and coded according to the waste mining and screening process nodes, including mining node features, manual sorting node features, drum screening node features, magnetic separation node features, air separation node features, and secondary screening node features.

[0014] Furthermore, in step S7, the neural network model training employs an iterative optimization method based on process conservation constraints, specifically including the following steps: S701: Initialize process consistency constraints, perform material conservation checks on the input-output relationships of each process node in the training samples, and correct deviations for samples that do not meet the conservation relationships; S702: The forward computation process of each node is carried out in sequence according to the topology, so that the output of the upstream node is used as part of the input of the downstream node for computation. S703: Multi-objective error construction process, which constructs a comprehensive error function based on the resource product yield error, economic value deviation and process conservation deviation, where the process conservation deviation is used to constrain the model output to meet the material flow balance relationship at each process node; S704: The parameter update process involves iteratively adjusting the parameter subspaces of each process node to make the model output gradually approximate the actual mining and screening results. S705: Consistency convergence determination process. When the material deviation output by each process node meets the preset threshold range, the iterative training stops and the converged process topology prediction model is obtained.

[0015] Further, in step S7, the constraint loss function includes a yield prediction error term, a yield non-negativity constraint term, a material balance constraint term, a process flow constraint term, a pollution risk constraint term, and an economic value constraint term; the constraint loss function L is expressed by the following formula: L = Lp + αLn + βLm + γLf + δLr + μLv; Where Lp is the yield prediction error term, Ln is the yield non-negative constraint term, Lm is the material balance constraint term, Lf is the process flow constraint term, Lr is the pollution risk constraint term, Lv is the economic value constraint term, and α, β, γ, δ, and μ are the weight coefficients of the corresponding constraint terms.

[0016] Furthermore, the material balance constraint is used to ensure that the sum of the yields of each resource-based product and the yield of the residue is close to the total amount of material entering the mining and screening system; the process flow constraint is used to ensure that the yield of light materials from air separation is not greater than the yield of oversize material entering the air separation equipment, the yield of ferrous metals from magnetic separation is not greater than the yield of undersize material entering the magnetic separation equipment, and the yield of undersize material from secondary screening is not greater than the yield of material entering the secondary screening equipment; the pollution risk constraint is used to increase pollution control costs or reduce the resource recovery potential level when the pollution risk score index increases; the economic value constraint is used to constrain the correspondence between the yield of resource-based products and the resource recovery benefits.

[0017] This invention also provides a landfill resource recovery potential prediction system, which employs the above-mentioned prediction method and includes: The data acquisition module is used to collect basic site data, pollution risk data, mining operation data, screening and sorting data, waste component detection data, and resource utilization data of the landfill to be evaluated. The pollution risk assessment module is used to determine the weights of pollution risk factors, calculate the pollution risk score index, and classify pollution risk levels using the analytic hierarchy process. The preprocessing module is used to clean, correct, convert, and standardize the collected data. The process node division module is used to divide sample data into nodes according to the destinations of mining, conveying, manual sorting, vibrating feed, drum screening, magnetic separation, secondary screening, air separation and resource utilization. The feature processing module is used to screen features using the correlation coefficient method and perform dimensionality reduction using principal component analysis. The model training module is used to train neural network models based on multi-source fusion datasets; The constraint optimization module is used to introduce non-negativity constraints on yield, material balance, process flow, pollution risk, and economic value during model training. The product prediction module is used to predict RDF fuel yield, ferrous metal recovery rate, humus yield, road construction material yield, backfill material yield, and residue yield. The path matching module is used to determine the paths for RDF fuel utilization, metal recycling, humus utilization, road construction material utilization, backfill utilization, land reclamation utilization, and residue disposal based on the resource utilization path rule base. The value calculation module is used to calculate the resource recovery potential and the predicted economic value. The results output module is used to output the predicted yield of resource-based products, the recommended resource-based pathways, the pollution risk level, the predicted economic value, the mining priority, and the suggestions for land reclamation and utilization.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention integrates pollution risk assessment into a resource recovery potential prediction system, enabling the clarification of site stabilization standards and pollution control requirements before mining. By constructing a process node model based on the entire process of mining, manual sorting, drum screening, magnetic separation, secondary screening, and air separation, it ensures that the prediction results are compatible with the actual engineering process. Furthermore, by incorporating multiple constraints such as non-negative yield, material balance, process flow, pollution risk, and economic benefits into the loss function, it eliminates prediction data that deviates from actual site conditions. This method can simultaneously output resource recovery product yield, material utilization pathways, project benefits, and site reclamation plans, supporting the planning of remediation sequences for old landfills, the selection of mining processes, and decisions related to site reuse. Attached Figure Description

[0019] Figure 1 Flowchart of the method for predicting the resource recovery potential of landfills in this invention; Figure 2 This invention provides a process flow diagram for landfill waste mining, screening, and resource utilization. Figure 3 : Flowchart of model training, path matching and value calculation in this invention. Detailed Implementation

[0020] To clearly illustrate the technical features of the present invention, the present invention will be described in detail below through specific embodiments and in conjunction with the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0022] Example 1 Please refer to Figure 1 This embodiment provides a method for predicting the resource recovery potential of landfills, including the following steps: S1: Collect basic site data, pollution risk data, mining operation data, screening and sorting data, waste component testing data, and resource utilization data of the landfill to be evaluated.

[0023] In step S1, the basic site data includes landfill age, landfill area, landfill depth, landfill volume, waste source, landfill method, closure time, historical operation records, distribution of surrounding residential areas, distribution of surrounding water bodies, groundwater depth, and planned land use; the pollution risk data includes leachate pollution, landfill gas release, odor impact, groundwater monitoring data, surface water monitoring data, soil cover integrity, anti-seepage system integrity, waste stabilization level, and distribution of environmentally sensitive targets; the mining operation data includes excavator excavation volume, mining area, mining depth, mining strata, conveyor transport volume, manual sorting volume, vibrating feeder feed volume, and unit mining cost; the screening and sorting data includes the yield of material oversize from the drum screen. The waste composition analysis data includes: yield of undersize material from drum screen, yield of ferrous metal from magnetic separation, yield of oversize material from secondary screening, yield of undersize material from secondary screening, yield of light material from air separation, yield of heavy material from air separation, yield of large pieces of material, and yield of residues; the waste composition analysis data includes humus content, organic matter content, moisture content, plastic component content, rubber and textile content, bamboo and wood combustible content, metal content, glass and ceramic content, brick and stone content, and heavy metal content; the resource utilization data includes the utilization destination of various resource products, resource utilization path matching relationship, and corresponding economic value data, wherein the resource products include RDF fuel, ferrous metal recyclables, humus-based soil amendment materials, road construction materials, backfill materials, and unusable residues.

[0024] S2: Based on the pollution risk data, the weights of pollution risk factors are determined using the analytic hierarchy process (AHP), and the pollution risk score index R of the landfill to be evaluated is calculated.

[0025] After data collection, the analytic hierarchy process (AHP) was used for pollution risk assessment. First, an evaluation structure consisting of a target layer, a criterion layer, and a factor layer was established. Then, a judgment matrix was built based on expert judgment or historical engineering data. After consistency testing, the weights of the pollution risk factors were obtained. The pollution risk score index R was calculated based on the on-site scores of each factor. The pollution risk score index R is calculated using the following formula: R = ; Wherein, Wi is the weight of the i-th pollution risk factor, Si is the score of the i-th pollution risk factor, and n is the number of pollution risk factors; the pollution risk factors include one or more of the following: degree of waste stabilization, leachate pollution risk, landfill gas release risk, groundwater pollution risk, surface water pollution risk, odor impact risk, sensitivity of surrounding residential areas, sensitivity of surrounding water bodies, soil cover integrity, and anti-seepage system integrity.

[0026] The pollution risk level, pre-mining stabilization requirements, pollution prevention and control measures, and safety management measures are determined based on the pollution risk score index R. When the pollution risk score index reaches the preset high-risk threshold, the landfill to be evaluated is identified as a high-pollution-risk landfill, and stabilization treatment measures are configured before mining and utilization. The stabilization treatment measures include aerobic aeration, ventilation and dehumidification, leachate collection and treatment, landfill gas drainage, odor collection and treatment, dust prevention and suppression, safety isolation, and zoned excavation.

[0027] S3: Perform data consistency processing on the site basic data, pollution risk data, mining operation data, screening and sorting data, waste component detection data, and resource utilization data to obtain sample data.

[0028] In step S3, to address the issues of complex and inconsistent data, consistency processing is performed to create standardized sample data. This involves processing the site foundation data, pollution risk data, mining operation data, screening and sorting data, waste component detection data, and resource utilization data. The consistency processing includes unit unification, missing data repair, anomaly correction, structured coding, and benchmark conversion. The unit mass of waste entering the conveying system is used as the unified accounting benchmark. All data is converted into standardized data based on material mass flow to form sample data. For example, the total amount of material entering the mining and screening system is used as the benchmark to calculate the yield at each node. For samples with significant differences in moisture content, a unified wet or dry basis conversion can be performed.

[0029] S4: Based on the mining nodes, map the standardized data to process nodes, construct process topology data, and establish the material transfer relationship between each process node.

[0030] This step maps the sample data in S3 to process nodes. This setting ensures that the model input matches the actual processing flow, avoiding the reliance on a single waste component as the basis for prediction. Please refer to... Figure 2The waste mining and screening process includes: Existing waste is mined by excavation equipment and then transported to a conveying system, where it is manually sorted. The manually sorted material is fed through a vibrating feeder into a drum screen. The material oversizes the drum screen and is then separated into light combustibles and heavy materials by air separation. The material undersizes the drum screen and is then separated into ferrous metals by magnetic separation. The magnetically separated material is then sent to a secondary screening system to separate the material oversizes and undersizes. Light combustibles are used in RDF (Refuse-Derived Fuel) utilization, ferrous metals are used in metal recycling, and humus from the secondary screening undersizes is used in landscaping nutrient soil, soil conditioners, or landfill cover. Stones, bricks, tiles, and glass are used in road construction materials, roadbed fillers, or backfill materials. The mining process can be divided into waste mining, conveying, manual sorting, drum screening, magnetic separation, secondary screening, and air separation.

[0031] S5: Extract structured feature variables based on the process topology data, and use the correlation coefficient method to screen key features related to the yield of resource-based products.

[0032] In step S5, the correlation coefficient method can be the Pearson correlation coefficient method, the Spearman correlation coefficient method, or the Kendall correlation coefficient method.

[0033] S6: Perform principal component analysis on the selected feature variables to construct a multi-source fusion dataset, where the input is the principal component features of the process nodes and the output is the yield and economic value indicators of resource-based products; the principal component analysis method is used to reduce redundant features and the effects of collinearity.

[0034] S7: Input the multi-source fusion dataset into the neural network model for training, and introduce an engineering constraint loss function.

[0035] In step S7, the neural network model is a structured prediction model built based on the process topology. Its input structure is grouped and coded according to the waste mining and screening process nodes, including mining node features, manual sorting node features, drum screening node features, magnetic separation node features, air separation node features, and secondary screening node features.

[0036] Specifically, the neural network model training employs an iterative optimization method based on process conservation constraints, which includes the following steps: S701: Initialize process consistency constraints, perform material conservation checks on the input-output relationships of each process node in the training samples, and correct deviations for samples that do not meet the conservation relationships; S702: The forward computation process of each node is carried out in sequence according to the topology, so that the output of the upstream node is used as part of the input of the downstream node for computation. S703: Multi-objective error construction process, which constructs a comprehensive error function based on the resource product yield error, economic value deviation and process conservation deviation, where the process conservation deviation is used to constrain the model output to meet the material flow balance relationship at each process node; S704: The parameter update process involves iteratively adjusting the parameter subspaces of each process node to make the model output gradually approximate the actual mining and screening results. S705: Consistency convergence determination process. When the material deviation output by each process node meets the preset threshold range, the iterative training stops and the converged process topology prediction model is obtained.

[0037] Furthermore, the constraint loss function includes a yield prediction error term, a yield non-negativity constraint term, a material balance constraint term, a process flow constraint term, a pollution risk constraint term, and an economic value constraint term; the constraint loss function L is expressed by the following formula: L = Lp + αLn + βLm + γLf + δLr + μLv; Where Lp is the yield prediction error term, Ln is the yield non-negative constraint term, Lm is the material balance constraint term, Lf is the process flow constraint term, Lr is the pollution risk constraint term, Lv is the economic value constraint term, and α, β, γ, δ, and μ are the weight coefficients of the corresponding constraint terms.

[0038] The yield prediction error term measures the deviation between the resource product yield predicted by the model and the yield obtained from actual mining and screening. Its calculation formula is: Lp =

[0039] in, j represents the predicted yield of the j-th type of resource-based product, yj represents the actual yield of the j-th type of resource-based product, and m represents the number of resource-based product categories. The resource-based products include one or more of RDF fuel, ferrous metals, humus, road construction materials, backfill materials, and residues. When the model simultaneously predicts six types of resource-based products—RDF fuel, ferrous metals, humus, road construction materials, backfill materials, and residues (the residues being the unusable products remaining after resource recovery)—m is set to 6. When the predicted categories are increased or decreased according to actual engineering needs, m is determined accordingly based on the number of resource-based product categories involved in the prediction.

[0040] The non-negativity yield constraint term Ln is used to ensure that the yield of each resource product output by the model is not less than 0, avoiding results such as negative yields that do not conform to engineering reality. Its calculation formula is: Ln = Σ[max(0, - j)]²; in, j represents the predicted yield of the j-th type of resource product. When the predicted yield is greater than or equal to 0, this term is set to 0; when the predicted yield is less than 0, the negative value is penalized.

[0041] The material balance constraint term Lm is used to ensure that the total amount of waste entering the system and the output of each resource recovery product and residue maintain the law of conservation of mass. Its calculation formula is: Lm = (Min - ΣMj)²; Where Min is the total amount of material entering the mining and screening system, and Mj is the predicted quality of the j-th type of resource product. When the model uses yield as the output, it can also be written as: Lm = (1 -Σ j)²; in, j represents the predicted yield of the j-th type of resource product relative to the total amount of material entering the system. If the yield is expressed as a percentage, then 1 in the formula can be replaced with 100.

[0042] The process flow constraint term Lf is used to ensure that the predicted results conform to the process relationship before and after waste mining and screening, avoiding situations where the downstream product quantity exceeds the upstream feed quantity. For example, light materials from air separation originate from the oversize material of the drum screen, ferrous metals from magnetic separation originate from the undersize material of the drum screen, and the undersize material from secondary screening originates from the fine undersize material after magnetic separation. Its calculation formula can be expressed as: Lf = Σ[max(0, Oq - Ip)]²; Where Ip is the upstream input of a certain process node, and Oq is the predicted output of that node or downstream node. This term is 0 when the downstream output does not exceed the upstream input; when the downstream output exceeds the upstream input, the excess is penalized. Lf may include the following constraints: predicted amount of air-separated light materials ≤ predicted amount of material oversize from drum screen; predicted amount of magnetically separated ferrous metals ≤ predicted amount of material undersize from drum screen; predicted amount of material undersize from secondary screening ≤ amount of material entering the secondary screening machine; predicted amount of RDF fuel ≤ predicted amount of air-separated light combustibles.

[0043] The pollution risk constraint term Lr reflects the corrective effect of pollution risk on resource recovery potential and economic value. When the pollution risk score is high, the model should increase pollution control costs or decrease the resource recovery potential evaluation value to avoid high-risk sites being incorrectly identified as high-value mining targets. The pollution risk constraint term can be calculated using the following formula: Lr = [max(0, R - R0)]² × Kr; Where R is the pollution risk score index, R0 is the preset risk threshold, and Kr is the pollution risk correction coefficient. When the pollution risk score index does not exceed the preset threshold, the pollution risk constraint term is 0; when the pollution risk score index exceeds the preset threshold, the excess portion is included in the loss function as a penalty term. In economic value calculation, the pollution risk constraint can also be reflected as a pollution control cost correction: Cp = C0 × [1 + η(R - R0) / R0]; where Cp is the corrected pollution control and safety management cost, C0 is the basic pollution control cost, and η is the risk amplification coefficient.

[0044] The economic value constraint term Lv is used to ensure a reasonable correspondence between the predicted resource product yield and economic value in the model. That is, when the yield of usable products such as RDF fuel, ferrous metals, and humus increases, the predicted economic value should increase accordingly, assuming the unit price and cost remain constant; when the proportion of residues or pollution control costs increase, the economic value should decrease accordingly. The economic value constraint term can be calculated using the following formula: Lv = ( - V)²; in, V represents the predicted economic value calculated by the model based on the predicted yield, while V represents the actual economic value or accounting economic value calculated based on the actual quality of the resource-based products, the unit price of the products, and the processing costs. The predicted economic value is... It can be calculated using the following formula: = Σ( j × Rj × Uj - j × Cj) - Ck - Cs - Cf - Ct - Cd - Cp; where j represents the predicted quality of the j-th type of resource product, Rj represents the resource utilization rate of the j-th type of resource product, Uj represents the unit price or alternative disposal value of the j-th type of resource product, Cj represents the unit processing cost of the j-th type of resource product, Ck represents the mining cost, Cs represents the screening cost, Cf represents the sorting and air separation cost, Ct represents the transportation cost, Cd represents the residue disposal cost, and Cp represents the pollution control and safety management cost.

[0045] like Figure 3As shown, a neural network model is trained using a multi-source fusion dataset, and a constraint loss function is introduced. In the constraint loss function, the material balance constraint term is used to ensure that the sum of the yields of each resource recovery product and the yield of the residue is close to the total amount of material entering the mining and screening system; the process flow constraint term is used to ensure that the yield of light materials from air separation is not greater than the yield of materials entering the air separation equipment, the yield of ferrous metals from magnetic separation is not greater than the yield of materials entering the magnetic separation equipment, and the yield of materials entering the secondary screening equipment is not greater than the yield of materials entering the secondary screening equipment; the pollution risk constraint term is used to increase pollution control costs or reduce the resource recovery potential level when the pollution risk score index increases; and the economic value constraint term is used to constrain the correspondence between the yield of resource recovery products and the resource recovery revenue.

[0046] By incorporating the above six loss functions, this invention simultaneously integrates prediction accuracy, yield legitimacy, material conservation relationship, process flow relationship, pollution risk correction, and economic value relationship into the model training process, making the resulting prediction model more consistent with the actual working conditions of landfill mining, screening, and resource utilization.

[0047] S8: Input the data of the landfill to be evaluated into the trained model to obtain the predicted yield of resource recovery products. Combined with the resource recovery path rule base and economic value model, output the resource utilization path, resource recovery potential value, economic value prediction value, mining priority and land reclamation suggestions.

[0048] In step S8, the resource recovery potential value P is used to comprehensively evaluate the resource output capacity, economic feasibility, and pollution risk impact of the landfill to be evaluated. Its calculation formula is: P = λ1Pr + λ2Pe - λ3Rr; Wherein, Pr is the resource yield evaluation value, Pe is the economic value evaluation value, Rr is the pollution risk correction value, and λ1, λ2, and λ3 are the weight coefficients corresponding to resource yield, economic value, and pollution risk, respectively. The resource yield evaluation value Pr is calculated using the following formula: Pr = Σ Aj × Qj; Wherein, Qj is the predicted yield of the j-th type of resource product, and Aj is the weighting coefficient of the j-th type of resource product. The weighting coefficient is determined based on the unit value, recycling difficulty, market utilization conditions, and process stability of the resource product.

[0049] The economic value evaluation value Pe is obtained by dimensionless processing of the economic value prediction value V, and is used to represent the economic feasibility of resource recovery from the landfill to be evaluated. The dimensionless processing can be performed in the following manner: Pe = (V - Vmin) / (Vmax - Vmin); Wherein, V is the predicted economic value of the landfill to be evaluated, Vmax is the upper limit or preset upper limit of the economic value of similar landfill resource extraction projects, and Vmin is the lower limit or preset lower limit of the economic value of similar landfill resource extraction projects.

[0050] The pollution risk correction value Rr is calculated based on the pollution risk scoring index R, and is used to represent the degree to which pollution risk reduces resource recovery potential. Its calculation formula is as follows: Rr = R / 100; Alternatively, when setting the risk threshold R0, it can be calculated as follows: Rr = max(0, R - R0) / (100 - R0); Through the above calculations, the resource recovery potential value P can simultaneously reflect the resource yield, economic benefits, and pollution risks of the landfill to be evaluated, so that the priority judgment of resource recovery does not solely depend on the content of waste components or economic benefits.

[0051] After the model training is completed, the data of the landfill to be evaluated is input into the model to obtain the predicted yield of resource recovery products. The system then determines the utilization destination based on the resource recovery path rule base: light combustibles that meet the requirements for calorific value, moisture content and impurity content enter the RDF fuel utilization path; ferrous metals enter the metal recycling path; humus that meets the heavy metal and other limiting indicators enters the garden nutrient soil, soil conditioner or landfill cover soil utilization path; stones, bricks, tiles and glass materials enter the road construction materials, roadbed filler or backfill materials path; materials that do not meet the resource recovery conditions enter the residue disposal path.

[0052] Example 2 Based on the landfill resource recovery potential prediction method provided in Example 1, this example provides a landfill resource recovery potential prediction system, the system comprising: The data acquisition module is used to collect basic site data, pollution risk data, mining operation data, screening and sorting data, waste component detection data, and resource utilization data of the landfill to be evaluated. The pollution risk assessment module is used to determine the weights of pollution risk factors, calculate the pollution risk score index, and classify pollution risk levels using the analytic hierarchy process. The preprocessing module is used to clean, correct, convert, and standardize the collected data. The process node division module is used to divide sample data into nodes according to the destinations of mining, conveying, manual sorting, vibrating feed, drum screening, magnetic separation, secondary screening, air separation and resource utilization. The feature processing module is used to screen features using the correlation coefficient method and perform dimensionality reduction using principal component analysis. The model training module is used to train neural network models based on multi-source fusion datasets; The constraint optimization module is used to introduce non-negativity constraints on yield, material balance, process flow, pollution risk, and economic value during model training. The product prediction module is used to predict RDF fuel yield, ferrous metal recovery rate, humus yield, road construction material yield, backfill material yield, and residue yield. The path matching module is used to determine the paths for RDF fuel utilization, metal recycling, humus utilization, road construction material utilization, backfill utilization, land reclamation utilization, and residue disposal based on the resource utilization path rule base. The value calculation module is used to calculate the resource recovery potential and the predicted economic value. The results output module is used to output the predicted yield of resource-based products, the recommended resource-based pathways, the pollution risk level, the predicted economic value, the mining priority, and the suggestions for land reclamation and utilization.

[0053] Example 3 This embodiment uses the above-mentioned system to conduct pollution risk assessment for old landfills with high pollution risk. Pollution risk factors may include leachate pollution risk, landfill gas release risk, groundwater pollution risk, surface water pollution risk, odor impact risk, degree of waste stabilization, sensitivity of surrounding residential areas, and sensitivity of surrounding water bodies.

[0054] An old municipal solid waste landfill has a total waste volume of approximately 551,000 m³ and a total waste mass of approximately 501,000 tons, including 378,300 tons of humus, 65,100 tons of combustible waste, and 43,300 tons of bricks and stones. There are residential areas within a 500-meter radius and numerous water bodies nearby. A pollution risk assessment has determined the site to be at a high level of pollution risk. For this type of site, the system increases the pollution control and safety management cost (Cp) in the economic value calculation and recommends that stabilization measures be implemented before extraction.

[0055] In this embodiment, pre-mining stabilization measures include aerobic aeration, ventilation and dehumidification, landfill gas drainage, leachate collection and treatment, odor collection and treatment, and zoned closed management. After stabilization, construction is organized according to the process of mining, transportation, manual sorting, screening, and resource utilization. If the total investment of the project is 12.8304 million yuan, and the unit treatment cost is at least 53.2 yuan / t, then this investment and unit cost can be used as cost parameters in the economic value calculation model to calculate the predicted economic value of different treatment schemes.

[0056] Meanwhile, the total amount of resource conversion and the total amount of input waste basically satisfy the material conservation relationship, verifying the applicability and feasibility of the process topology and resource recovery potential prediction method described in this invention.

[0057] Example 4 This embodiment provides a method for correcting the resource utilization path when landfill humus soil after mining and screening is limited by heavy metal pollution.

[0058] After mining, transportation, manual sorting, drum screening, magnetic separation, secondary screening, and air separation, the undersize material from an old municipal solid waste landfill has a high proportion of humus. According to the prediction model described in this invention, the predicted output of humus from this landfill is 426,000 tons, which has the potential for resource utilization as cover soil, landscaping soil, or land reclamation soil.

[0059] However, on-site sampling and testing revealed that some heavy metal levels in the waste humus exceeded preset utilization limits. Direct use for landscaping, soil improvement, or reclamation and backfilling could pose a risk of heavy metal migration and diffusion. To avoid judging the resource utilization of humus solely based on yield and economic value, this invention uses the heavy metal pollution assessment results as one of the limiting conditions in the resource utilization path rule base, thus performing a secondary determination of the utilization path of waste humus.

[0060] In this embodiment, heavy metal pollution was detected in the undersize humus soil, including Cd, Pb, and Hg. The test results are shown in the table below:

[0061] As shown in the table above, Cd, Pb, and As in the waste humus soil exceed the limits to varying degrees, with Cd and Pb exceeding the limits more significantly. Based on the resource utilization path rule base, the system determines that this batch of humus soil is not suitable for direct use as greening nutrient soil or reclamation soil, and marks its resource utilization status as "requires stabilization treatment before utilization".

[0062] In response to the above situation, the system invokes the heavy metal pollution stabilization and remediation rules, recommending the use of a remediation agent prepared from sepiolite solid waste for in-situ stabilization treatment. This remediation agent is prepared from sepiolite solid waste through crushing, screening, drying, modification, and compounding. It can reduce the content of available heavy metals and the risk of leaching through adsorption, ion exchange, complexation precipitation, and alkalinity adjustment.

[0063] In this embodiment, the remediation agent based on sepiolite solid waste is added at 5% of the mass of the waste humus. Since the predicted humus yield is 426,000 tons, the theoretical dosage of the remediation agent is: Repair agent dosage = 42.6 × 5% = 21,300 tons.

[0064] The remediation agent was mechanically mixed with humus soil to a depth of 0-40 cm. After mixing, the soil was kept at a suitable moisture content and cured for 14 days. After curing, the treated humus soil was retested. The retest results are shown in the table below:

[0065] The reduction rate is calculated using the following formula: Reduction rate = (content before repair - content of effective state after repair) / content before repair × 100%.

[0066] The retest results show that after stabilization treatment with a remediation agent based on sepiolite solid waste, the bioavailable contents of Cd, Pb, and Hg in the humic soil all decreased significantly. Specifically, Cd decreased from 3.22 mg / kg to 0.76 mg / kg, a reduction rate of 76.40%; Pb decreased from 225.4 mg / kg to 57.50 mg / kg, a reduction rate of 74.49%; and Hg decreased from 2.17 mg / kg to 0.9 mg / kg, a reduction rate of 58.53%. After treatment, the risk of heavy metal migration in this batch of humic soil was reduced, meeting the requirements for restricted resource utilization.

[0067] The system re-enters the remediated heavy metal testing results into the resource utilization path rule base and updates the resource utilization path for this batch of humus. Before remediation, this batch of humus was deemed "unsuitable for direct use" due to excessive Cd and Pb levels; after remediation, the system adjusts its utilization path to "can be used as landfill cover soil, greening planting substrate, site reclamation soil, or low-risk backfill soil." Simultaneously, the system corrects the predicted economic value based on the cost of remediation agent application, mixing construction costs, curing period, and retesting fees.

[0068] In calculating its economic value, this batch of humus is no longer directly included in the usable resource quantity based on the total predicted output, but rather calculated according to the proportion of usable resources after restoration. Its effective utilization is calculated using the following formula: Mh = Mh0 × Rh; Where Mh represents the mass of humus that can be included in resource utilization after remediation, Mh0 represents the predicted output mass of humus, and Rh represents the resource utilization rate after remediation.

[0069] In this embodiment, the predicted yield of humus soil, Mh0, is 426,000 tons. After remediation and retesting, the system determines that 80% of it can be used as cover soil, landscaping soil, or reclamation soil. Therefore: Mh = 42.6 × 80% = 340,800 tons.

[0070] Therefore, in the economic value prediction and resource recovery potential calculation, the system only considers 340,800 tons as the effective resource utilization amount; the remaining 85,200 tons will be subject to further stabilization treatment, restricted backfilling, or standardized disposal based on the test results. This approach avoids simply including humus soil with excessive heavy metal content in usable resources, reducing the risk of overestimating the resource recovery potential of landfills.

[0071] If we calculate the value of the restored humus soil as a substitute for cover soil, landscaping soil, or reclamation soil at 60 yuan / t, then the resource utilization value that can be included after the restoration of this batch of humus soil is: 34.08 × 60 = 20.448 million yuan.

[0072] If heavy metal pollution assessment and stabilization remediation pathways are not introduced, and all 426,000 tons are directly included in resource utilization, then the theoretical value is: 42.6 × 60 = 25.56 million yuan.

[0073] The difference between the two is 5.112 million yuan. This difference reflects the impact of heavy metal pollution restrictions on the effective resource utilization value of humus, and also illustrates that the present invention, through pollution risk identification and remediation followed by reassessment, can avoid a simple overestimation of the recovery value of humus resources.

[0074] This embodiment demonstrates that the present invention does not merely predict the yield of landfill mining and screening products, but further introduces mechanisms for heavy metal pollution assessment, in-situ stabilization remediation, and re-determination of resource utilization pathways after remediation, based on the predicted yield. For materials such as landfill humus that have high yields but may be limited by heavy metal pollution, the present invention can adjust the resource utilization pathway based on pollution detection results and re-evaluate their utilization conditions after applying a remediation agent based on sepiolite solid waste, thereby realizing a continuous decision-making process of "resource yield prediction—pollution risk identification—remediation agent stabilization treatment—utilization pathway correction—economic value recalculation".

[0075] The above embodiments of the invention are merely examples for clearly illustrating the invention and are not intended to limit the implementation of the invention. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively describe all possible implementations. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the invention should be included within the scope of protection of the claims of this invention.

Claims

1. A method for predicting the resource recovery potential of a landfill, characterized in that, Includes the following steps: S1: Collect basic site data, pollution risk data, mining operation data, screening and sorting data, waste component testing data, and resource utilization data of the landfill to be evaluated; S2: Based on the pollution risk data, the weights of pollution risk factors are determined using the analytic hierarchy process (AHP), and the pollution risk score index R of the landfill to be evaluated is calculated. S3: Perform data consistency processing on the site basic data, pollution risk data, mining operation data, screening and sorting data, waste component detection data and resource utilization data to obtain sample data; S4: Based on the mining nodes, process nodes are mapped to the standardized data to construct process topology data and establish the material transfer relationship between each process node; S5: Extract structured feature variables based on the process topology data, and use the correlation coefficient method to screen key features related to the yield of resource products; S6: Perform principal component analysis on the selected feature variables to construct a multi-source fusion dataset; S7: Input the multi-source fusion dataset into the neural network model for training, and introduce an engineering constraint loss function; S8: Input the data of the landfill to be evaluated into the trained model to obtain the predicted yield of resource recovery products. Combined with the resource recovery path rule base and economic value model, output the resource utilization path, resource recovery potential value, economic value prediction value, mining priority and land reclamation suggestions.

2. The method for predicting the resource recovery potential of a landfill according to claim 1, characterized in that, In step S2, the pollution risk score index R is calculated using the following formula: R = ； Wherein, Wi is the weight of the i-th pollution risk factor, Si is the score of the i-th pollution risk factor (the evaluation refers to the "Recommended Method for Risk Assessment of Sudden Environmental Incidents in Administrative Regions" (Ministry of Ecology and Environment, January 2018), and the environmental risk index calculation method includes water environment risk index calculation, air environment risk index calculation and comprehensive environmental risk index calculation), and n is the number of pollution risk factors; the pollution risk factors include one or more of the following: degree of waste stabilization, leachate pollution risk, landfill gas release risk, groundwater pollution risk, surface water pollution risk, odor impact risk, sensitivity of surrounding residential areas, sensitivity of surrounding water bodies, soil cover integrity and anti-seepage system integrity.

3. The method for predicting the resource recovery potential of a landfill according to claim 2, characterized in that, The pollution risk level, pre-mining stabilization requirements, pollution prevention and control measures, and safety management measures are determined based on the pollution risk scoring index R. When the pollution risk scoring index reaches the preset high-risk threshold, the landfill to be evaluated is identified as a high-pollution-risk landfill, and stabilization treatment measures are configured before mining and utilization. The stabilization treatment measures include one or more of the following: aerobic aeration, ventilation and dehumidification, leachate collection and treatment, landfill gas drainage, odor collection and treatment, dust prevention and suppression, safety isolation, and zoned excavation.

4. The method for predicting the resource recovery potential of a landfill according to claim 1, characterized in that, In step S3, the data consistency processing includes unit unification, missing data repair, anomaly correction, structured coding, and benchmark conversion.

5. The method for predicting the resource recovery potential of a landfill according to claim 1, characterized in that, In step S4, the mining node includes waste mining, conveying, manual sorting, drum screening, magnetic separation, secondary screening and air separation processes.

6. The method for predicting the resource recovery potential of a landfill according to claim 1, characterized in that, In step S7, the neural network model is a structured prediction model built based on the process topology. Its input structure is grouped and coded according to the waste mining and screening process nodes, including mining node features, manual sorting node features, drum screening node features, magnetic separation node features, air separation node features, and secondary screening node features.

7. The method for predicting the resource recovery potential of a landfill according to claim 6, characterized in that, In step S7, the neural network model training adopts an iterative optimization method based on process conservation constraints, specifically including the following steps: S701: Initialize process consistency constraints, perform material conservation checks on the input-output relationships of each process node in the training samples, and correct deviations for samples that do not meet the conservation relationships; S702: The forward computation process of each node is carried out in sequence according to the topology, so that the output of the upstream node is used as part of the input of the downstream node for computation. S703: Multi-objective error construction process, which constructs a comprehensive error function based on the resource product yield error, economic value deviation and process conservation deviation, where the process conservation deviation is used to constrain the model output to meet the material flow balance relationship at each process node; S704: The parameter update process involves iteratively adjusting the parameter subspaces of each process node to make the model output gradually approximate the actual mining and screening results. S705: Consistency convergence determination process. When the material deviation output by each process node meets the preset threshold range, the iterative training stops and the converged process topology prediction model is obtained.

8. The method for predicting the resource recovery potential of a landfill according to claim 1, characterized in that, In step S7, the constraint loss function includes a yield prediction error term, a yield non-negativity constraint term, a material balance constraint term, a process flow constraint term, a pollution risk constraint term, and an economic value constraint term; the constraint loss function L is expressed by the following formula: L = Lp + αLn + βLm + γLf + δLr + μLv; Where Lp is the yield prediction error term, Ln is the yield non-negative constraint term, Lm is the material balance constraint term, Lf is the process flow constraint term, Lr is the pollution risk constraint term, Lv is the economic value constraint term, and α, β, γ, δ, and μ are the weight coefficients of the corresponding constraint terms.

9. The method for predicting the resource recovery potential of a landfill according to claim 8, characterized in that, The material balance constraint term is used to constrain the sum of the yield of each resource product and the yield of the residue to be close to the total amount of material entering the mining and screening system. The process flow constraint term is used to ensure that the yield of light materials from air separation is not greater than the yield of oversize material entering the air separation equipment, the yield of ferrous metals from magnetic separation is not greater than the yield of undersize material entering the magnetic separation equipment, and the yield of undersize material from secondary screening is not greater than the yield of material entering the secondary screening equipment; the pollution risk constraint term is used to increase pollution control costs or reduce the resource recovery potential level when the pollution risk score index increases; the economic value constraint term is used to constrain the correspondence between the yield of resource-based products and the revenue from resource recovery.

10. A landfill resource recovery potential prediction system, wherein the system employs the landfill resource recovery potential prediction method as described in any one of claims 1 to 9, characterized in that, include: The data acquisition module is used to collect basic site data, pollution risk data, mining operation data, screening and sorting data, waste component detection data, and resource utilization data of the landfill to be evaluated. The pollution risk assessment module is used to determine the weights of pollution risk factors, calculate the pollution risk score index, and classify pollution risk levels using the analytic hierarchy process. The preprocessing module is used to clean, correct, convert, and standardize the collected data. The process node division module is used to divide sample data into nodes according to the destinations of mining, conveying, manual sorting, vibrating feed, drum screening, magnetic separation, secondary screening, air separation and resource utilization. The feature processing module is used to screen features using the correlation coefficient method and perform dimensionality reduction using principal component analysis. The model training module is used to train neural network models based on multi-source fusion datasets; The constraint optimization module is used to introduce non-negativity constraints on yield, material balance, process flow, pollution risk, and economic value during model training. The product prediction module is used to predict RDF fuel yield, ferrous metal recovery rate, humus yield, road construction material yield, backfill material yield, and residue yield. The path matching module is used to determine the paths for RDF fuel utilization, metal recycling, humus utilization, road construction material utilization, backfill utilization, land reclamation utilization, and residue disposal based on the resource utilization path rule base. The value calculation module is used to calculate the resource recovery potential and the predicted economic value. The results output module is used to output the predicted yield of resource-based products, the recommended resource-based pathways, the pollution risk level, the predicted economic value, the mining priority, and the suggestions for land reclamation and utilization.

Images