Background garbage classification resource optimization decision-making system based on big data

Abstract

The invention relates to the technical field of resource optimization, in particular to a background garbage classification resource optimization decision-making system based on big data, and the system comprises a garbage classification recognition module; a collection demand prediction module; a dynamic collection scheduling module; and a resource optimization decision module. In the scheme, through AI visual identification and multi-modal data fusion, the fused data comprises weight, spectrum and smell sensor data. And in combination with a deep learning model, accurate recognition of garbage categories and materials is realized, and particularly, directional recognition and sorting of high-value recoverable objects are enhanced. The system solves the problem of misjudgment of similar material garbage through a dynamic threshold value judgment and secondary verification mechanism. And meanwhile, a recycling closed loop is constructed for high-value garbage, the resource recycling efficiency is remarkably improved, and environmental pollution is reduced.

Description

Technical Field

[0001] This invention relates to the field of resource optimization technology, and in particular to a background waste sorting resource optimization decision-making system based on big data. Background Technology

[0002] With the acceleration of urbanization and the improvement of residents' living standards, the amount of urban waste generated is increasing year by year. Waste sorting, as a key means to improve resource utilization and reduce environmental pollution, has become an important issue for urban governance in terms of management efficiency and accuracy. Traditional waste sorting relies on manual sorting and fixed collection methods, which not only consume a lot of manpower and resources but also suffers from low sorting accuracy and incomplete resource recycling. The lack of systematic methods for identifying and recycling high-value recyclables such as lithium batteries and PET bottles leads to resource waste and increased environmental pressure.

[0003] In existing technologies, waste sorting and identification largely rely on single visual sensors or manual spot checks. This approach lacks accuracy in identifying similar materials in complex scenarios, such as different types of plastics, and struggles to achieve targeted sorting of high-value waste. Waste collection and transportation often follow fixed routes and frequencies, frequently resulting in "empty runs" or "overloads," leading to significant waste of transportation resources. Furthermore, the lack of end-to-end data linkage and intelligent decision-making mechanisms means resource allocation relies on experience-based judgment, making it difficult to cope with fluctuations in waste production and changes in the external environment, such as holidays and weather, resulting in low management efficiency and limited recycling benefits. Therefore, we propose a big data-based background waste sorting resource optimization decision-making system. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a big data-based background waste sorting resource optimization decision-making system, thereby solving the technical problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0006] A big data-based backend waste sorting resource optimization decision-making system includes:

[0007] Waste sorting and identification module: It is used to identify waste through the integration of multiple technologies, with a focus on intelligent sorting of high-value recyclable waste. It includes an AI vision recognition unit, a multimodal fusion recognition unit, a high-value waste sorting unit, and a real-time feedback and quality assessment unit.

[0008] Collection demand forecasting module: used to predict waste generation through sensor data and multi-factor models, providing data support for dynamic collection, including intelligent overflow monitoring unit and multi-factor demand forecasting unit;

[0009] Dynamic collection and scheduling module: used to dynamically plan collection routes based on prediction results, prioritizing the recycling efficiency of high-value waste, including dynamic route optimization unit and multi-objective resource balancing unit;

[0010] Resource optimization decision-making module: Based on big data analysis of the entire process of waste classification, identification, collection and transportation, it builds a resource optimization model, outputs the optimal resource allocation plan, and tracks the implementation effect of the plan. It includes a resource optimization modeling unit and a decision plan execution and iteration unit.

[0011] In one possible implementation, the AI ​​visual recognition unit includes:

[0012] If the confidence level of the garbage category identification output is Conf v If the value is ≥85%, it is determined to be high-value waste and enters the multimodal fusion identification unit;

[0013] If confidence level Conf v 70%≤Conf v If the percentage is between 85%, a second lightweight model verification will be triggered for this unit.

[0014] If confidence level Conf v If the confidence level is less than 70%, it will be marked as a "low confidence sample" and pushed to the manual review interface.

[0015] In one possible implementation, the multimodal fusion recognition unit includes:

[0016] If the high-value feature probability Pr is obtained by fusing visual data, weight data, spectral data, and odor data... hv-m If the value is ≥0.9, it is judged as "high confidence high value waste" and directly enters the high value waste sorting unit;

[0017] If the probability of high-value features after fusion is Pr hv-m If the value is less than 0.6, it is determined to be misscreened waste, the category is corrected to the multimodal fusion result, and it enters the regular waste treatment process;

[0018] If the probability of high-value features after fusion is Pr hv-m In 0.6≤Pr hv-m If the value is less than 0.7, the historical data comparison unit will be activated to retrieve the identification records of similar waste from the collection point in the past 30 days and calculate the prior probability. If the prior probability P ≥ 0.7, the historical high-frequency result will be output. Otherwise, it will be marked as "awaiting manual confirmation" and pushed to the management terminal. If it is not processed within 2 hours, it will be automatically processed as regular waste.

[0019] In one possible implementation, the high-value waste sorting unit includes:

[0020] For high-value waste with priority level, the robotic arm is automatically triggered to sort it. If the weight data of the second weighing and the identification stage deviates by more than 3%, the weight of the recycling work order is automatically corrected and the weight prediction error of the identification model is recorded.

[0021] For secondary priority waste, electronic tags are generated and linked to the recycling plant's appointment system.

[0022] In one possible implementation, the real-time feedback and quality assessment unit includes:

[0023] If the accuracy rate of high-value waste disposal is Acc hv ≥80% and the accuracy rate of routine community waste disposal (Rec) hv If the rate is ≥70%, the community is identified as a "high-value waste classification standard-compliant community," and the computing power allocation for high-value waste identification in that area will be reduced.

[0024] If the accuracy rate of high-value waste disposal is Acc hv <60% and the accuracy rate of community routine waste disposal Rec hv If the percentage is less than 50%, it will be identified as a "key optimization community" and a high-frequency identification process will be initiated for three consecutive days.

[0025] In one possible implementation, the intelligent overflow monitoring unit includes:

[0026] If the loading rate L of the ordinary trash can is ≥85%, the "ordinary trash pre-overflow" signal is triggered and the priority is marked as P1. If the loading rate L of the ordinary trash can is ≥95%, it is marked as P0 emergency priority and directly inserted into the dynamic collection task.

[0027] If the loading rate L of the high-value waste bin is ≥70% or the opening frequency C hv If the number of collections is ≥50 times per day, it is considered a "surge in demand for high-value waste collection", and an expedited collection work order is generated, which has a higher priority than ordinary bins in the same area.

[0028] If the frequency of opening the high-value waste bins is C during non-disposal periods hv If the frequency is greater than 3 times per hour, it will be marked as "suspicious activation".

[0029] In one possible implementation, the multi-factor demand forecasting unit includes:

[0030] If the predicted output H of high-value waste pre If the waste volume is ≥50kg / day and the lower limit of the fluctuation range is >30kg / day, it is identified as a "high-value waste high-production point" and a dedicated collection vehicle will be dispatched for collection.

[0031] If predicting T com-full- T hv-full ≥2 hours, of which T com-full For the time it takes for a regular trash can to overflow, Thv-full When the special bins for high-value waste are overflowing, priority will be given to collecting high-value waste to avoid overloading the bins and causing resource loss.

[0032] In one possible implementation, in the total resource optimization modeling unit, if a candidate solution satisfies all constraints and matches existing resources, it is determined to be a "feasible solution"; otherwise, it is directly eliminated. By weighting and scoring the feasible solutions, the one with the highest score is determined as the "execution solution".

[0033] Beneficial effects compared to existing technologies:

[0034] 1. This solution utilizes AI visual recognition and multimodal data fusion, incorporating data from weight, spectrum, and odor sensors. Combined with a deep learning model, it achieves accurate identification of waste type and material, with a particular focus on enhancing the targeted identification and sorting of high-value recyclables. The system employs dynamic threshold judgment and a secondary verification mechanism to resolve the issue of misclassification of waste of similar materials. Simultaneously, it establishes a closed-loop recycling system for high-value waste, significantly improving resource recycling efficiency and reducing environmental pollution.

[0035] 2. In this solution, a real-time overflow monitoring and multi-factor prediction model is used, with inputs including historical data, weather, and holidays. The model predicts the waste output and overflow time at each collection point. Combined with a dynamic route optimization algorithm (VRPTW model) and traffic data, on-demand collection is achieved. The system prioritizes the transportation of high-value waste, allocating capacity through priority weights to avoid empty runs and overflows. Simultaneously, through vehicle networking and real-time scheduling, the system balances the load on processing plants, reduces queuing time, and achieves refined and efficient resource allocation.

[0036] 3. This solution integrates data from the entire process of waste sorting, identification, collection, and transportation to construct a resource optimization decision-making model. A genetic algorithm is used to find the optimal configuration of equipment, manpower, and transportation capacity, and the model parameters are continuously iterated through execution feedback. The system thus achieves data-driven decision-making, dynamically adjusting the allocation of identification computing power, collection frequency, and resource input. It also implements tiered management based on differences in sorting quality, thereby improving the level of intelligence in waste sorting management. Attached Figure Description

[0037] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0038] Figure 1 This is a schematic diagram of the system framework of the present invention; Detailed Implementation

[0039] Preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention can also be implemented in various different forms, and therefore the present invention is not limited to the embodiments described below. In addition, for the purpose of more clearly describing the present invention, parts not connected to the invention will be omitted from the drawings.

[0040] The technical solutions in this application are designed to address the problems described in the background, and are generally as follows:

[0041] Example:

[0042] This embodiment introduces a big data-based background waste sorting resource optimization decision-making system, which realizes intelligent decision-making throughout the entire process of waste sorting identification, collection, and transportation through multiple modules. The system includes a waste sorting identification module, a collection demand prediction module, a dynamic collection scheduling module, and other modules.

[0043] I. Waste Sorting and Identification Module

[0044] This module achieves accurate waste identification through the integration of multiple technologies, with a focus on intelligent sorting of high-value recyclable waste to promote resource recycling. It includes an AI visual recognition unit, a multimodal fusion recognition unit, a high-value waste sorting unit, and a real-time feedback and quality assessment unit.

[0045] 1. AI Visual Recognition Unit

[0046] This unit uses a camera to capture real-time images / video streams of waste, utilizes a deep learning model to identify basic categories, and marks potentially high-value waste.

[0047] A 5-megapixel RGB camera is installed 30cm above the waste disposal opening, supporting high-definition video capture; panoramic cameras are deployed at transfer stations and processing plants, supporting 180° wide-angle shooting and autofocus, and the video stream is transmitted to the edge computing node via HDMI interface.

[0048] Edge computing nodes decode the video stream in real time, use an adaptive threshold segmentation algorithm to remove irrelevant background (such as the thrower's arm, ground debris), extract ROI (region of interest) images, uniformly scale them to 416×416 pixels, and transmit them to the backend server at a rate of 500KB / s via the RTSP protocol.

[0049] The backend server runs an improved YOLOv8n model (the weight file is pre-trained on the COCO dataset and then fine-tuned with 100,000 of our own high-value junk images). The model structure includes:

[0050] Input: Mosaic data augmentation (randomly stitching together 4 images), adaptive anchor box calculation;

[0051] Backbone network: CSPDarknet53 feature extraction layer, embedded with CBAM attention mechanism to enhance feature extraction of high-value waste (such as reflective surfaces of lithium batteries and transparent materials of PET bottles);

[0052] Detection head: Three-scale output (13×13, 26×26, 52×52), supports detection of 20 types of conventional waste + 10 types of high-value waste, outputs category labels and confidence scores. v (range 0-1) and high-value feature probability (Pr) hv The value ranges from 0 to 1 and is calculated through an independent fully connected layer. A higher value indicates a greater likelihood that the item belongs to high-value garbage.

[0053] If Conf v ≥85%, enters the multimodal fusion recognition unit; if 70% ≤Conf v If <85%, trigger the second lightweight model verification of this unit; if Conf v Samples with a confidence level of less than 70% are marked as "low confidence samples" and sent to the manual review interface.

[0054] If Pr hv If the value is ≥0.7 and belongs to the Hv category, a "high-value waste awaiting sorting" label is generated, and the waste is given priority to enter the high-value waste sorting unit; otherwise, it enters the multimodal fusion recognition unit according to the normal process.

[0055] 2. Multimodal fusion recognition unit

[0056] This unit integrates visual data, weight data, spectral data, and odor data to construct a multi-dimensional feature vector, solving the problem of identifying occluded and similar-material waste, while also verifying the initial screening results of high-value waste.

[0057] Weight sensor: Four pressure sensors are symmetrically installed at the bottom of the trash can. The voltage signal is collected by the Wheatstone bridge circuit and converted into a digital weight value (unit: kg) by the 24-bit ADC module (ADS1256). The sampling frequency is 10Hz. The data is transmitted to the edge computing node through the I2C bus. The impact noise of disposal is removed by Kalman filtering to obtain a stable weight W.

[0058] Near-infrared spectrometer: A miniature NIR sensor is installed on the side of the trash can. Each recognition triggers a spectral scan (takes 50ms) and outputs a 1024-dimensional reflectance spectral vector S, which is matched with a preset material spectral library (containing 200 common trash materials) through a cosine similarity algorithm.

[0059] Odor sensor: For kitchen waste and hazardous waste, an electronic nose is installed 10cm above the disposal port to collect volatile gas signals, which are then reduced to a 16-dimensional odor feature vector O by PCA.

[0060] Constructing a three-layer Transformer fusion network:

[0061] Input layer: Concatenate visual features V (128-dimensional feature vector output by YOLOv8), weight features W normalized to [0,1], spectral features S (the top 50 principal components after standardization), and odor features O to form an initial 400-dimensional feature vector;

[0062] Encoding layer: Contains two multi-head attention sub-layers (8 heads each), which learn the interaction relationships between different modalities (such as the correspondence between the visual contour of a PET bottle and the characteristic peaks of the NIR spectrum);

[0063] Output layer: Fully connected layer outputs fused confidence level. m (Regular waste categories) and high-value probability Pr hv-m (For garbage that is initially screened as high-value), the activation functions are Softmax and Sigmoid, respectively.

[0064] Output the merged category label (e.g., "polyethylene plastic"), material properties (e.g., "recyclable"), and processing recommendations (e.g., "send to plastic recycling plant");

[0065] Output sorting priority (first priority: Pr) hv-m ≥0.9), corresponding to lithium batteries, lithium cobalt oxide batteries, etc.; Secondary priority: (0.7≤Pr hv-m <0.9), corresponding to PET bottles, aluminum cans, etc.), and include the contribution of each modality data (e.g., "visual proportion 55%, spectral proportion 35%").

[0066] If Pr hv-m ≥Pr hv +0.1 (probability increased by 10% or more after fusion) is judged as "high confidence high value waste" and directly enters the high value waste sorting unit;

[0067] If Pr hv-m If the value is less than 0.6, it is judged as "misidentified waste", the initial screening conclusion is overturned, the category is corrected to the multimodal fusion result (such as "misidentifying a transparent soap box as a PET bottle"), and it enters the regular waste disposal process;

[0068] If 0.6≤Pr hv-m If the value is less than 0.7, activate the historical data comparison unit: retrieve the identification records of similar waste from this collection point over the past 30 days and calculate the prior probability. If P ≥ 0.7, output according to historical high-frequency results; otherwise, mark it as "awaiting manual confirmation" and push it to the management terminal (if not processed within 2 hours, it will be automatically processed as regular waste).

[0069] If Conf m≥90% is considered a "valid identification result" and proceeds to the real-time feedback and quality assessment unit; if 80% ≤Conf m If the result is less than 90%, a second verification of the sensor data is triggered (e.g., repeated scanning of the spectrometer to confirm material consistency). If the two results are consistent, the result is considered valid; otherwise, it is marked as a "low confidence result" and an exception code is recorded (for subsequent model iterations).

[0070] 3. High-value waste sorting unit

[0071] Based on the results of multimodal fusion, this unit accurately sorts high-value waste, generates independent recycling work orders, and connects with downstream resource recovery channels to achieve a closed loop of "identification-sorting-recycling".

[0072] Based on the results of multimodal fusion, if the waste is identified as first-priority high-value waste (such as lithium batteries and lithium cobalt oxide batteries), the intelligent sorting robotic arm (deployed at the transfer station or processing plant) is automatically triggered to perform physical sorting; for second-priority waste (such as PET bottles and aluminum cans), an electronic tag (containing material, weight, and estimated value) is generated and linked to the recycling and processing plant's reservation system.

[0073] Generate and output a "High-Value Waste Recycling Work Order", which includes the item name and weight W. hv The estimated recovery value, Value, is calculated using the formula: Value = W hv ×Price market ×0.9, where Price market The real-time market recycling price is 0.9, and the system's commission rate is 0.9, which is simultaneously pushed to the order-receiving platform of recycling companies.

[0074] The high-value waste after sorting is weighed a second time and compared with the weight data in the identification stage. If the deviation is >3%, the weight of the recycling work order is automatically corrected (based on the second weighing) and the weight prediction error of the identification model is recorded.

[0075] 4. Real-time feedback and quality assessment unit

[0076] This unit provides feedback on the sorting results to those who dispose of waste, assesses the quality of waste sorting in the community, and focuses on monitoring the correct disposal rate of high-value waste.

[0077] The system displays customized feedback such as "Your lithium battery has entered the dedicated recycling channel. Thank you for supporting resource recycling!" on smart terminals; the accuracy rate of high-value waste disposal is calculated using the following formula: It should be included in the community's performance evaluation indicators.

[0078] The "Daily Classification Quality Report" is generated, which includes the accuracy rate of regular waste and the recycling rate of high-value waste. The formula is as follows: If Rec hvIf the waste disposal rate is less than 60%, a special campaign to guide the disposal of high-value waste will be launched (e.g., a sorting animation will be played at the disposal point).

[0079] If Acc hv ≥80% and Rec hv If the waste classification rate is ≥70%, the community will be classified as a "high-value waste classification compliant community," and the computing power allocation for high-value waste identification in that area will be reduced (from real-time identification to interval identification).

[0080] If Acc hv <60% and Rec hv If the waste rate is less than 50%, it will be identified as a "key community for optimization". High-frequency identification (captured once every 5 minutes) will be initiated for 3 consecutive days, and a pop-up window for the "Guide to Disposing of High-Value Waste" will be pushed to residents.

[0081] II. Demand Forecasting Module

[0082] This module uses sensor data and multi-factor models to predict waste generation, providing data support for dynamic collection. It includes an intelligent overflow monitoring unit and a multi-factor demand prediction unit.

[0083] 1. Intelligent overflow monitoring unit

[0084] This unit collects real-time data on the loading rate of trash cans, distinguishing between the load status of special bins for high-value waste and ordinary bins.

[0085] Install a pressure sensor (measured value W) in a regular trash can. com High-value waste bins are additionally equipped with dual sensors to measure weight and the number of times the bin is opened (C). hv The data is uniformly converted into load factor.

[0086] Output real-time loading rate L and special bucket abnormal opening alarm (if C) hv If more than 3 times / hour are used outside of the designated delivery period (22:00-6:00), it will be marked as "suspicious activation".

[0087] Regular bin: If L≥85%, trigger the “regular waste pre-overflow” signal and mark it as P1 priority; if L≥95%, mark it as P0 emergency priority and directly insert it into the dynamic collection task.

[0088] Special container: If L ≥ 70% or C hv If the number of collections is ≥50 times per day (calibrated based on the regional population density), it is considered a "surge in demand for high-value waste collection" and an expedited collection work order is generated (with higher priority than regular bins in the same area).

[0089] 2. Multi-factor demand forecasting unit

[0090] This unit combines historical data, real-time load, and external factors to predict the overflow time and high-value waste output at each collection point.

[0091] Construct a hybrid prediction model with the following input variables:

[0092] Historical production of high-value waste H his Historical production of ordinary waste C his ;

[0093] Real-time load rate L now Loading rate change rate in the past 2 hours

[0094] External factors: Promotional event days (high-value waste production may increase by 30%), heavy rain (general waste production may decrease by 20%). An LSTM network is used to model the generation trends of high-value and general waste separately, outputting the overflow time T. full and predicted production of high-value waste H pre =H his ×(1-δ), where δ is the external factor adjustment coefficient.

[0095] Generate a "Collection Point Demand Forecast Table", which includes the estimated overflow time T for general waste. com-full Expected output of high-value waste H pre and fluctuation range

[0096] If H pre If the waste volume is ≥50kg / day and the lower limit of the fluctuation range is >30kg / day, it is identified as a "high-value waste high-production point" and a special collection vehicle (equipped with anti-crushing sorting device) is dispatched for collection.

[0097] If the predicted overflow time T com-full- T hv-full If the waste overflows within 2 hours (the regular bins overflow later than the special bins), priority will be given to collecting high-value waste to avoid overloading the special bins and causing resource loss.

[0098] III. Dynamic Collection and Scheduling Module

[0099] This module dynamically plans collection routes based on prediction results, prioritizing the recycling efficiency of high-value waste. It includes a dynamic route optimization unit and a multi-objective resource balancing unit.

[0100] 1. Dynamic route optimization unit

[0101] This unit combines overflow prediction, traffic data, and vehicle status to generate collection routes that balance efficiency with prioritizing high-value waste.

[0102] Construct a VRPTW model with priority, with the objective function being min(α·D+β·∑Delay).hv ), where D is the total driving distance, and Delay hv The high-value waste collection delay time is α = 0.6, β = 0.4 (high-value waste priority weight). Input parameters include collection point priority (high-value dedicated bin priority + 1 level) and real-time traffic congestion index CI, where CI > 80 indicates severe congestion.

[0103] Output the route for each vehicle, with high-value waste collection points marked in red, along with the estimated arrival time T. arr Loading limits (the amount collected in a single batch of a dedicated container shall be ≤ 80% of the rated capacity to avoid crushing and breakage).

[0104] If the T of high-value waste collection points arr ≤T hv-full -30 minutes (allowing for sorting time), and the T standard waste collection point arr ≤T com-full It was determined to be a valid route;

[0105] If a certain high-value special barrel's T arr >T hv-full This triggers a "local replanning": the route to that point is adjusted first, allowing a delay of ≤1 hour for ordinary garbage collection points. If the conditions are not met after 3 adjustments, a backup vehicle is called in (with priority given to high-value garbage collection tasks).

[0106] 2. Multi-objective resource balancing unit

[0107] This unit dynamically allocates high-value waste to the optimal recycling channels based on waste type and treatment plant characteristics.

[0108] Establish a matching matrix M for the processing plants, where M i,j Let represent the matching degree between the i-th type of waste (i=1 for high-value waste, i=2 for ordinary waste) and the j-th processing plant (values ​​range from 0 to 100, with a matching degree of 90 for high-value waste and 60 for ordinary waste). The Hungarian algorithm is used to solve for the optimal allocation, ensuring that high-value waste is preferentially allocated to specialized recycling plants with a matching degree ≥ 80.

[0109] Generate a "Processing Plant Allocation Order", mark high-value waste with a "Priority Processing" label, and require the processing plant to complete sorting within 4 hours of receipt (24 hours for ordinary waste).

[0110] If the target recycling plant has a matching degree of ≥80 and the remaining processing capacity is ≥ the amount transported, the allocation will be confirmed directly.

[0111] If the matching degree is <80% or the capacity is insufficient, switch to the second-best factory (poorest matching degree) in descending order of matching degree.

[0112] If none of the plants meet the requirement of ≤15), the “inter-regional transfer” will be initiated (the transportation distance is allowed to increase by ≤30%, but the total recycling value must be ≥70% of the local processing value).

[0113] IV. Resource Optimization Decision Module

[0114] This module, serving as the core decision-making layer of the system, leverages big data analysis across the entire process of waste sorting, identification, collection, and transportation to construct a resource optimization model. It outputs optimal allocation schemes for resources such as equipment and transportation capacity, and tracks the effectiveness of these schemes. It includes a resource optimization modeling unit and a decision scheme execution and iteration unit.

[0115] 1. Full Resource Optimization Modeling Unit

[0116] This unit integrates data from the entire process and constructs an optimization model aimed at minimizing resource consumption while maximizing recycling benefits, to solve for the optimal allocation of resources such as equipment and transportation capacity in waste sorting management.

[0117] The system receives data from the identification module, collection module, and transportation module, and after standardization, forms a model input matrix (with dimensions of "region × resource type × time").

[0118] Model building:

[0119] Objective function: minC + maxP ， Where C represents total resource consumption (equipment operation and maintenance + transportation costs + labor costs), and P represents total recycling benefits (economic benefits of high-value waste + environmental value of carbon emission reduction).

[0120] Constraints: Maximum equipment load (e.g., daily sorting capacity of robotic arm ≤ 500 kg), vehicle range limit (electric collection vehicle range ≤ 200 km per charge), and processing plant receiving time (high-value waste from collection to processing ≤ 4 hours).

[0121] Solution algorithm: An improved genetic algorithm is used to find the Pareto optimal solution through 100 population iterations, and output 2-3 candidate solutions (examples are "Solution A: Prioritize the allocation of high-value special vehicles; Solution B: Optimize the layout of sorting equipment").

[0122] Output the "Resource Optimization Candidate Solutions", which includes the resource allocation details of each solution (e.g., "add 1 robotic arm and allocate 2 special vehicles to each of the 3 high-value waste and high-yield communities"), expected benefits (e.g., "cost reduction of 8% and recycling rate increase of 15%)", and constraint satisfaction status.

[0123] If a candidate solution meets all constraints (e.g., "vehicle allocation does not exceed the range limit") and matches existing resources (e.g., "the equipment being called is in an idle state"), it is determined to be a "feasible solution"; otherwise, it is directly eliminated (e.g., "Solution B exceeds the daily capacity of the processing plant, so it is excluded").

[0124] Feasible options are scored using a weighted scoring system (40% cost and 60% benefit), and the option with the highest score is selected as the "implementation option". For example, "Option A scores 92 points, which is better than Option B's 85 points".

[0125] 2. Decision-making scheme execution and iteration unit

[0126] This unit breaks down the optimal resource optimization plan into executable tasks, assigns them to the corresponding departments for execution, and tracks the results, forming a closed loop of "decision-execution-feedback-iteration" to ensure that the goal of resource optimization is achieved.

[0127] The optimal solution is broken down into specific tasks, with clearly defined responsible departments and timelines, as follows:

[0128] Equipment task: "Install an NIR spectrometer for high-value special barrels in Chengdong Community within 5 days";

[0129] Transportation task: "Adjust the collection vehicle routes in the western part of the city to prioritize coverage of high-value waste and high-volume collection points."

[0130] Execution tracking: The system monitors task progress in real time (e.g., "equipment installation completion rate 70%)", sends alerts (SMS + system pop-up) for overdue tasks (delay > 24 hours), and links them to the specific person in charge.

[0131] Compare the core indicators before and after implementation. For example, "After implementation, the high-value waste recycling rate in Chengdong Community increased from 68% to 82%", and calculate the improvement rate. The analysis of reasons for not meeting expectations is exemplified by "the transportation time decreased due to traffic congestion after the route adjustment".

[0132] Output the "Decision Implementation Evaluation Report", which includes task completion rate, indicator improvement rate, and optimization suggestions, and feed the evaluation data back to the resource optimization modeling unit.

[0133] Execution effect determination:

[0134] Achievement criteria: If the improvement rate is ≥10% and the task completion rate is ≥90%, the implementation is deemed "effective," and the solution is included in the best practice library for direct adoption in the future.

[0135] Needs optimization: 5% ≤ improvement rate < 10% or task completion rate 80%-90%, adjust execution details as suggested;

[0136] Iteration required: If the improvement rate is less than 5% or the task completion rate is less than 80%, feedback should be sent to the resource optimization modeling unit to retrain the model. Specifically, the objective function weights should be adjusted and the model should be retrained.

[0137] Model iteration: Update model parameters based on feedback data, and complete a model iteration every quarter to improve decision-making accuracy.

[0138] Finally, it should be noted that the above embodiments are merely examples for clearly illustrating the present invention and are not intended to limit the implementation. 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 list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A background waste sorting resource optimization decision-making system based on big data, characterized in that: include: Waste sorting and identification module: It is used to identify waste through the integration of multiple technologies, with a focus on intelligent sorting of high-value recyclable waste. It includes an AI vision recognition unit, a multimodal fusion recognition unit, a high-value waste sorting unit, and a real-time feedback and quality assessment unit. Collection demand forecasting module: used to predict waste generation through sensor data and multi-factor models, providing data support for dynamic collection, including intelligent overflow monitoring unit and multi-factor demand forecasting unit; Dynamic collection and scheduling module: used to dynamically plan collection routes based on prediction results, prioritizing the recycling efficiency of high-value waste, including dynamic route optimization unit and multi-objective resource balancing unit; Resource optimization decision-making module: Based on big data analysis of the entire process of waste classification, identification, collection and transportation, it builds a resource optimization model, outputs the optimal resource allocation plan, and tracks the implementation effect of the plan. It includes a resource optimization modeling unit and a decision plan execution and iteration unit.

2. The background waste sorting resource optimization decision-making system based on big data as described in claim 1, characterized in that, In the AI ​​visual recognition unit: If the confidence level of the garbage category identification output is Conf v If the value is ≥85%, it is determined to be high-value waste and enters the multimodal fusion identification unit; If confidence level Conf v 70%≤Conf v If the percentage is between 85%, a second lightweight model verification will be triggered for this unit. If confidence level Conf v If the confidence level is less than 70%, it will be marked as a "low confidence sample" and pushed to the manual review interface.

3. The background waste sorting resource optimization decision-making system based on big data as described in claim 1, characterized in that, In the multimodal fusion recognition unit: If the high-value feature probability Pr is obtained by fusing visual data, weight data, spectral data, and odor data... hv-m If the value is ≥0.9, it is judged as "high confidence high value waste" and directly enters the high value waste sorting unit; If the probability of high-value features after fusion is Pr hv-m If the value is less than 0.6, it is determined to be misscreened waste, the category is corrected to the multimodal fusion result, and it enters the regular waste treatment process; If the probability of high-value features after fusion is Pr hv-m In 0.6≤Pr hv-m If the value is less than 0.7, the historical data comparison unit will be activated to retrieve the identification records of similar waste from the collection point in the past 30 days and calculate the prior probability. If the prior probability P ≥ 0.7, the historical high-frequency result will be output. Otherwise, it will be marked as "awaiting manual confirmation" and pushed to the management terminal. If it is not processed within 2 hours, it will be automatically processed as regular waste.

4. The background waste sorting resource optimization decision-making system based on big data as described in claim 1, characterized in that, In the high-value waste sorting unit: For high-value waste with priority level, the robotic arm is automatically triggered to sort it. If the weight data of the second weighing and the identification stage deviates by more than 3%, the weight of the recycling work order is automatically corrected and the weight prediction error of the identification model is recorded. For secondary priority waste, electronic tags are generated and linked to the recycling plant's appointment system.

5. The background waste sorting resource optimization decision-making system based on big data as described in claim 1, characterized in that, In the real-time feedback and quality assessment unit: If the accuracy rate of high-value waste disposal is Acc hv ≥80% and the accuracy rate of routine community waste disposal (Rec) hv If the rate is ≥70%, the area is classified as a "high-value waste classification compliant community," and the computing power allocation for high-value waste identification in that area will be reduced. If the accuracy rate of high-value waste disposal is Acc hv <60% and the accuracy rate of community routine waste disposal Rec hv If the percentage is less than 50%, it will be identified as a "key optimization community" and a high-frequency identification process will be initiated for three consecutive days.

6. The background waste sorting resource optimization decision-making system based on big data as described in claim 1, characterized in that, In the intelligent overflow monitoring unit: If the loading rate L of the ordinary trash can is greater than or equal to 85%, the "ordinary trash pre-overflow" signal is triggered and the priority is marked as P1. If the loading rate L of the ordinary trash can is greater than or equal to 95%, it is marked as P0 emergency priority and directly inserted into the dynamic collection task. If the loading rate L of the high-value waste bin is ≥70% or the opening frequency C hv If the number of collections is ≥50 times per day, it is considered a "surge in demand for high-value waste collection", and an expedited collection work order is generated, with a higher priority than ordinary bins in the same area. If the frequency of opening the high-value waste bins is C during non-disposal periods hv If the frequency is greater than 3 times per hour, it will be marked as "suspicious activation".

7. The background waste sorting resource optimization decision-making system based on big data as described in claim 1, characterized in that, In the multi-factor demand forecasting unit: If the predicted output H of high-value waste pre If the waste volume is ≥50kg / day and the lower limit of the fluctuation range is >30kg / day, it is identified as a "high-value waste high-production point" and a dedicated collection vehicle will be dispatched for collection. If predicting T com-full- T hv-full ≥2 hours, of which T com-full For the time it takes for a regular trash can to overflow, T hv-full When the special bins for high-value waste are overflowing, priority will be given to collecting high-value waste to avoid overloading the bins and causing resource loss.

8. The background waste sorting resource optimization decision-making system based on big data as described in claim 1, characterized in that, In the full resource optimization modeling unit, if a candidate solution meets all constraints and matches existing resources, it is determined to be a "feasible solution"; otherwise, it is directly eliminated. By weighting the feasible solutions, the one with the highest score is determined as the "execution solution".

Images