Intelligent garbage classification management method and system based on Internet of Things

Abstract

The invention discloses an intelligent garbage classification management method and system based on the Internet of Things, and the method comprises the steps: obtaining garbage information data of thrown garbage through a built-in Internet of Things device of an intelligent garbage can, and generating a classified garbage data set with a user throwing behavior tag; inputting the classified garbage data set with the user throwing behavior tag into a dynamic prediction model, and generating a prediction result including a garbage throwing peak period, an easy-to-mix throwing category and overflow early warning time of each can; starting a classification guidance-clearance linkage mechanism according to the prediction result, and generating a garbage clearance scheduling scheme with classification guidance content; and collecting classification accuracy data actually thrown by the user and clearance task data completed by the garbage truck according to the scheduling scheme, generating a garbage classification management report, and realizing visual management of a garbage classification whole process. By utilizing the embodiment of the invention, the whole-process intelligent linkage of garbage classification can be realized, and the classification accuracy, the clearing efficiency and the refined level of garbage management are improved.

Description

Technical Field

[0001] This invention belongs to the field of waste treatment technology, specifically an intelligent waste sorting and management method and system based on the Internet of Things. Background Technology

[0002] In recent years, with the acceleration of urbanization and the improvement of environmental awareness, waste sorting has become an important part of urban management and community services. Currently, common waste sorting management methods mainly rely on residents' voluntary sorting and scheduled collection, lacking the ability to accurately perceive and predict disposal behavior, making it difficult to effectively address issues such as mixed disposal and overflow. Existing intelligent solutions often focus on using image recognition and other technologies to make post-disposal judgments on waste, or statically plan collection routes based on historical data. These solutions cannot adapt to changes in pedestrian flow and disposal habits in real time, resulting in untimely sorting guidance, low collection efficiency, and difficulty in achieving dynamic optimization and closed-loop management throughout the entire process. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent waste sorting management method and system based on the Internet of Things to address the shortcomings of existing technologies. This system enables intelligent linkage of the entire process of waste sorting, guidance, and collection scheduling, thereby improving sorting accuracy, collection efficiency, and the level of precision in waste management.

[0004] One embodiment of this application provides an intelligent waste sorting and management method based on the Internet of Things, the method comprising: By acquiring waste information data from the IoT devices built into the smart trash cans and recording the timestamps of user disposal, and combining it with real-time pedestrian traffic data acquired from IoT devices around the trash cans, a behavior-waste association algorithm is used to label the types of waste that users easily mix and dispose of at different times, generating a classified waste dataset with user disposal behavior tags. Input the categorized waste dataset with user disposal behavior tags into the dynamic prediction model. By analyzing the correlation between the disposal ratio of various types of waste and time, predict the growth curve of categorized waste in each waste bin and the peak period of easy mixed disposal within a preset time period. Generate prediction results including peak waste disposal time, easily mixed disposal categories and overflow warning time of each bin. Based on the prediction results, the classification guidance-collection linkage mechanism is activated. During the peak periods of mixed disposal and before the peak disposal period, classification guidance animations for the corresponding mixed disposal categories are pushed through the display screen on the top of the garbage bins. At the same time, the classification and collection routes of garbage trucks are planned by combining the garbage growth curve of each bin with real-time traffic data, and a garbage collection and dispatch plan with classification guidance content is generated. The system collects data on the accuracy of actual waste sorting by users and data on the collection tasks completed by garbage trucks according to the scheduling plan. It integrates the collected data with the prediction results and the collection scheduling plan to generate a waste sorting management report that includes the waste sorting compliance rate, collection efficiency and overflow warning accuracy of each area. The report is then uploaded to the community waste sorting management platform to achieve visualized management of the entire waste sorting process.

[0005] Optionally, the process involves acquiring waste information data from the IoT devices built into the smart trash can, synchronously recording the user's disposal timestamp, and combining this with real-time pedestrian traffic data acquired from IoT devices around the trash can. A behavior-waste association algorithm is then used to label the types of waste that users easily mix and dispose of at different times, generating a categorized waste dataset with user disposal behavior tags, including: The smart trash can uses its built-in IoT device to collect raw trash information data in real time, including at least the weight of the trash and image features. Add a timestamp to each piece of raw garbage information data and synchronize it with the real-time pedestrian flow data collected by IoT devices around the garbage bins to generate a time-synchronized multi-source dataset. Density-based clustering algorithm was used to analyze the correlation between waste disposal characteristics and pedestrian traffic in different time periods, to explore user behavior patterns of mixing waste, and to generate a set of behavior-waste association rules. Based on the behavior-garbage association rule set, the time-synchronized multi-source dataset is labeled, and user disposal behavior tags are added to each garbage disposal record, ultimately generating a categorized garbage dataset with user disposal behavior tags.

[0006] Optionally, the process involves inputting the categorized waste dataset with user disposal behavior tags into a dynamic prediction model. By analyzing the correlation between the disposal ratio of various types of waste and time, the model predicts the growth curve of categorized waste in each bin and the peak periods of mixed disposal within a preset time frame. The model generates prediction results including peak disposal times, easily mixed categories, and overflow warning times for each bin, including: Feature engineering is performed on the categorized waste dataset with user disposal behavior labels to extract the time-series features of disposal volume and the distribution features of behavior labels for each type of waste, and to generate feature vectors for model input. The feature vector is input into an LSTM dynamic prediction model based on an attention mechanism. This model learns the temporal pattern of waste disposal by analyzing the nonlinear correlation between the proportion of various types of waste disposal and time in historical data. By using a trained dynamic prediction model to perform multi-step forward prediction, the growth curve of classified waste in each trash can and the probability distribution of easy-to-mix waste disposal behavior are generated within a preset time period in the future. Based on the waste growth curve, an overflow warning threshold is set, and the high-incidence period is determined by combining the probability distribution of easy-to-mix waste disposal behavior. Finally, a prediction result is generated that includes the peak waste disposal period, easy-to-mix waste categories, and the overflow warning time of each bin.

[0007] Optionally, the step of activating the classification guidance-collection linkage mechanism based on the prediction results involves pushing classification guidance animations for the corresponding easily mixed categories via the display screen on top of the garbage bins before the peak periods of mixed disposal and during periods of high incidence of mixed disposal. Simultaneously, by combining the garbage growth curves of each bin with real-time traffic data, the classification and collection routes of garbage trucks are planned, and a garbage collection and dispatch plan with classification guidance content is generated, including: Analyze the prediction results to identify the peak periods of mixed-use behavior and peak delivery times, and trigger the classification guidance mechanism in advance to generate classification guidance activation instructions; Based on the classification guidance start command and the information on easily mixed categories in the prediction results, the corresponding classification guidance animation content is retrieved from the multimedia library and sent to the display screen on top of each trash can through the Internet of Things network to generate a classification guidance playback task. Real-time data on waste growth curves for each trash can and traffic conditions in the area related to the trash can are acquired. A path optimization algorithm is used to calculate the optimal collection route and generate a preliminary collection route plan. By integrating the classification guidance playback task and the preliminary collection route plan, and taking into account the capacity of collection vehicles and the constraints of operation time, multi-objective optimization scheduling is carried out to finally generate a waste collection scheduling plan with classification guidance content.

[0008] Optionally, the data collected includes the actual sorting accuracy rate of user waste disposal and the data on the collection tasks completed by garbage trucks according to the scheduling plan. This data is then integrated with the prediction results and the collection scheduling plan to generate a garbage sorting management report that includes the garbage sorting compliance rate, collection efficiency, and overflow warning accuracy rate for each area. Simultaneously, the report is uploaded to the community garbage sorting management platform to achieve visualized management of the entire garbage sorting process, including: The IoT devices in smart trash cans collect data on the actual sorting accuracy of users' trash disposal, and at the same time, they obtain data on the completion of collection tasks from the garbage truck's onboard terminal to generate an actual execution dataset. By comparing and analyzing the actual execution dataset with the prediction results and the waste collection and dispatch plan, key performance indicators such as the waste classification compliance rate, waste collection efficiency and overflow warning accuracy rate of each region are calculated, and performance evaluation results are generated. Based on the performance evaluation results, an automatic report generation technology is used to construct a waste sorting management report that includes data visualization charts and text analysis, generating a complete management report document; By uploading management reports to the community waste sorting management platform through the Internet of Things (IoT) data interface, the platform's visual display content is updated, enabling visualized management of the entire waste sorting process.

[0009] Another embodiment of this application provides an intelligent waste sorting management system based on the Internet of Things, the system comprising: The acquisition module is used to acquire garbage information data of garbage disposal through the IoT device built into the smart trash can, synchronously record the user's disposal timestamp, combine it with the real-time traffic data acquired by the IoT device around the trash can, and use the behavior-garbage association algorithm to label the garbage categories that users are likely to mix in different time periods, and generate a classified garbage dataset with user disposal behavior tags. The prediction module is used to input the classified waste dataset with user disposal behavior tags into the dynamic prediction model. By analyzing the correlation between the disposal ratio of various types of waste and time, it predicts the growth curve of classified waste in each waste bin and the high incidence of mixed disposal behavior within a preset time period in the future, and generates prediction results including peak waste disposal time, easily mixed disposal categories and overflow warning time of each bin. The planning module is used to activate the classification guidance-collection linkage mechanism based on the prediction results. Before the peak period of easy mixed disposal and the peak disposal period, the module pushes classification guidance animations for the corresponding easy mixed disposal categories through the display screen on the top of the garbage bin. At the same time, it plans the classification and collection routes of garbage trucks by combining the garbage growth curve of each bin and real-time traffic data, and generates a garbage collection and dispatch plan with classification guidance content. The integration module is used to collect data on the actual classification accuracy of user waste disposal and the data on the collection tasks completed by garbage trucks according to the scheduling plan. It integrates the collected data with the prediction results and the collection scheduling plan to generate a garbage classification management report that includes the garbage classification compliance rate, collection efficiency and overflow warning accuracy of each area. The report is uploaded to the community garbage classification management platform at the same time to realize the visualization management of the entire garbage classification process.

[0010] Another embodiment of this application provides a storage medium storing a computer program, wherein the computer program is configured to execute the method described in any of the preceding claims when running.

[0011] Another embodiment of this application provides an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the method described in any of the preceding claims.

[0012] Compared with existing technologies, this invention provides an intelligent waste sorting management method based on the Internet of Things (IoT). It acquires waste information data from waste disposal via IoT devices built into smart trash cans, generating a waste sorting dataset tagged with user disposal behavior. This dataset is then input into a dynamic prediction model to generate prediction results including peak disposal times, easily mixed categories, and overflow warning times for each bin. Based on the prediction results, a sorting guidance-collection linkage mechanism is activated, generating a waste collection and dispatching plan with sorting guidance. Data on the actual sorting accuracy of user disposal and the collection tasks completed by garbage trucks according to the dispatching plan are collected to generate a waste sorting management report. This achieves visualized management of the entire waste sorting process, enabling intelligent linkage of waste sorting, guidance, and collection scheduling, thus improving sorting accuracy, collection efficiency, and the level of precision in waste management. Attached Figure Description

[0013] Figure 1 A hardware structure block diagram of a computer terminal for an intelligent waste sorting and management method based on the Internet of Things, provided in an embodiment of the present invention; Figure 2 A flowchart illustrating an intelligent waste sorting and management method based on the Internet of Things, provided as an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an intelligent waste sorting management system based on the Internet of Things, provided as an embodiment of the present invention. Detailed Implementation

[0014] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0015] This invention first provides an intelligent waste sorting and management method based on the Internet of Things. This method can be applied to electronic devices, such as computer terminals, specifically ordinary computers.

[0016] The following detailed explanation uses a computer terminal as an example. Figure 1 This is a hardware structure block diagram of a computer terminal for an intelligent waste sorting and management method based on the Internet of Things, provided as an embodiment of the present invention. Figure 1 As shown, the computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0017] The non-volatile storage medium can store an operating system and a computer program. This computer program includes program instructions that, when executed, cause the processor to perform any IoT-based intelligent waste sorting management method.

[0018] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0019] The internal memory provides an environment for the execution of computer programs in non-volatile storage media. When the computer program is executed by the processor, it enables the processor to execute any intelligent waste sorting management method based on the Internet of Things.

[0020] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 1 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0021] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0022] See Figure 2 The present invention provides an intelligent waste sorting and management method based on the Internet of Things, which may include the following steps: S201 obtains garbage information data through the IoT device built into the smart trash can, records the user's garbage disposal timestamp, and combines it with real-time traffic data obtained from IoT devices around the trash can. It uses a behavior-garbage association algorithm to label the garbage categories that users are likely to mix and dispose of at different times, and generates a classified garbage dataset with user disposal behavior tags. Specifically, the smart trash can can use its built-in IoT devices to collect raw trash information data in real time, including at least the weight of the trash and image features. The IoT devices built into smart trash cans need to meet the requirements of "high precision and low latency" in data collection. Core devices include weighing sensors, high-definition cameras, and edge computing modules. The selection of each device and the data collection logic need to be designed in accordance with the actual needs of the waste sorting scenario. Weight Data Acquisition: A ZEMIC H3C series load cell is used, with a range of 0-50kg (covering the average household waste disposal volume, typically <5kg), an accuracy of 0.1kg (distinguishing between lightweight waste such as paper scraps and heavier waste such as aluminum cans), and a sampling frequency of 1 time / second. When the user opens the trash can opening, the sensor triggers data acquisition. After the waste has completely fallen into the can and the weight has stabilized for 3 seconds (to avoid errors caused by shaking during disposal), the weight value is recorded. For example, when a user disposes of a plastic bag with vegetable leaves, the sensor detects the weight increase from 0kg to 0.23kg, records the weight data as "0.23kg" after stabilizing for 3 seconds; when disposing of an empty plastic bottle, the weight is recorded as "0.08kg".

[0023] Image Feature Acquisition: Equipped with a 1080P high-definition USB camera (30fps, 120° field of view), the lens is aimed at the "temporary recognition area" below the trash can's disposal opening (to avoid obstruction by piled-up trash). When the weight sensor detects the disposal action, the camera automatically captures three consecutive images (0.5 seconds apart) to ensure the capture of multi-angle features of the trash. Image feature extraction is performed by the edge computing module (using the RK3399 chip, supporting real-time image processing). The OpenCV library is used to extract color histograms (e.g., kitchen waste is mostly green and brown, recyclable plastics are mostly white and transparent), shape contours (e.g., aluminum cans are cylindrical, paper towels are irregularly shaped), and texture features (e.g., leather surfaces are rough, plastic bag surfaces are smooth). For example, when disposing of a plastic bag with vegetable leaves, the image features are "color: green + white, shape: irregular block, texture: locally rough (vegetable leaves) + locally smooth (plastic bag)"; when disposing of an empty plastic bottle, the image features are "color: transparent, shape: cylindrical, texture: smooth".

[0024] Raw data integration: The edge computing module binds the weight data of the same disposal action with the feature information of 3 images to generate "raw garbage information data". Each data entry contains a unique collection ID (such as "TRASH-20240520-001"), weight value, image feature description (color, shape, texture) and collection device number (to distinguish different garbage bins in the same area, such as "BIN-01"), ensuring data traceability.

[0025] Add a timestamp to each piece of raw garbage information data and synchronize it with the real-time pedestrian flow data collected by IoT devices around the garbage bins to generate a time-synchronized multi-source dataset. Timestamp marking must be "high-precision and synchronizeable" to avoid errors in correlation analysis due to time deviations. The specific implementation is as follows: Timestamp Generation: The smart trash can's built-in real-time clock (RTC) module uses a DS3231 chip with an accuracy of ±2ppm (annual error <1 minute). It synchronizes with the community's IoT gateway via NTP (Network Time Protocol) to ensure that the device's time deviates from the standard time by <100ms. Each piece of raw trash information data is automatically stamped with a UTC format timestamp "YYYY-MM-DD HH:MM:SS.fff" (accurate to milliseconds). For example, the timestamp for the data collected from the plastic bag with vegetable leaves is "2024-05-20 07:45:32.678", and the timestamp for the empty plastic bottle is "2024-05-20 18:20:15.345".

[0026] Real-time pedestrian flow data collection: HC-SR501 infrared human body sensors (detection distance 3-7 meters, angle 110°) are deployed within a 10-meter radius of the trash cans. The sampling frequency is set to 1 time / second, and "effective pedestrian flow" is counted every 5 minutes (excluding single signals triggered by sensor malfunctions; only two consecutive human movements are counted), generating pedestrian flow data for each time period, such as "Time period: 2024-05-20 07:40-07:45, pedestrian flow: 62 people; Time period: 2024-05-20 18:15-18:20, pedestrian flow: 45 people". The sensors also synchronize their time with the gateway via NTP to ensure consistency with the trash can equipment time.

[0027] Time synchronization alignment: A "time period mapping" strategy is adopted to map the precise timestamp of garbage data to the corresponding 5-minute traffic flow time period. If the timestamp of garbage data falls within a certain time period (e.g., 07:45:32.678 falls within 07:40-07:45), then the garbage data is bound to the traffic flow of that time period; if it is at a time period boundary (e.g., 07:45:01.234), then it belongs to the next time period (07:45-07:50). An example of a multi-source dataset is generated after alignment: "Collection ID: TRASH-20240520-001, Timestamp: 2024-05-20 07:45:32.678, Weight: 0.23kg, Image Features: Green + White, Irregular Blocks, Local Roughness + Smoothness, Corresponding Time Period: 07:40-07:45, Traffic Flow: 62 people," ensuring that each piece of garbage data is associated with traffic flow information for a given time period.

[0028] Density-based clustering algorithm was used to analyze the correlation between waste disposal characteristics and pedestrian traffic in different time periods, to explore user behavior patterns of mixing waste, and to generate a set of behavior-waste association rules. The density-based clustering algorithm chosen is DBSCAN (Density-Based Spatial Clustering of Applications with Noise). This algorithm does not require a pre-defined number of clusters and can automatically identify noise points and dense clusters, making it suitable for uncovering implicit correlations between "garbage characteristics - pedestrian traffic - time period". The specific implementation process is as follows: Feature dimension construction: Multi-source data is converted into 5-dimensional feature vectors, including: ① Garbage weight (normalized to 0-1, e.g., 0.23kg corresponds to 0.0046); ② Image feature PCA dimensionality reduction values ​​(reducing the 128-dimensional features of color, shape, and texture to 2 dimensions, denoted as PC1 and PC2, e.g., green + white features correspond to PC1=0.6 and PC2=0.3); ③ Time period encoding (morning peak 7:00-9:00=1, noon 12:00-14:00=2, evening peak 18:00-20:00=3, other time periods=0); ④ Pedestrian density (period pedestrian flow / period duration, e.g., 62 people / 5 minutes = 12.4 people / minute, normalized to 0-1).

[0029] DBSCAN parameter determination: Parameters are set using the "elbow rule" and domain knowledge: Eps (neighborhood radius) = 0.3 (samples with a distance less than 0.3 in the feature space are considered to be in the same neighborhood, ensuring that similar features cluster together); MinPts (minimum number of samples in the neighborhood) = 5 (to avoid a small number of outliers forming pseudo-clusters). For example, a sample with a morning rush hour (code 1), a pedestrian density of 0.8 (12 people / minute), a weight of 0.0046, PC1=0.6, and PC2=0.3 has 8 similar samples in its neighborhood, satisfying MinPts=5, forming a dense cluster.

[0030] Association Pattern Analysis: Interpretation of Clustering Results Grouped by "Time Period - Pedestrian Flow": Morning rush hour (7:00-9:00) + high pedestrian traffic (density > 0.6, i.e. > 10 people / minute): Clustering results show that 32% of the clusters contain both "kitchen waste characteristics" (PC1=0.6-0.8, PC2=0.2-0.4, weight 0.002-0.006) and "other waste characteristics" (PC1=0.3-0.5, PC2=0.5-0.7, weight 0.001-0.005), and the samples within the clusters are mostly "packaged kitchen waste" (such as vegetable leaves in plastic bags, leftover food from takeout boxes), indicating that users are prone to confusing kitchen waste with other waste during this period due to time constraints.

[0031] Evening peak (18:00-20:00) + medium pedestrian flow (density 0.3-0.6): 18% of the clusters contained "recyclable waste characteristics" (PC1=0.1-0.3, PC2=0.1-0.3, such as the transparency of plastic bottles) and "other waste characteristics". The samples were mostly "contaminated recyclables" (such as plastic lunch boxes with oil stains), indicating that users tend to ignore the cleanliness requirements of recyclables when disposing of them during the evening peak.

[0032] Behavioral pattern mining and rule generation: The above patterns are transformed into "behavior-waste association rules". Each rule includes "applicable time period, traffic flow range, easily mixed waste categories, and waste characteristic description", for example: Rule 1: Applicable time period = morning peak (7:00-9:00), traffic flow range > 10 people / minute, easily mixed disposal category = kitchen waste - other waste, characteristic description = weight 0.002-0.006, PC1=0.3-0.8, PC2=0.2-0.7 (kitchen waste with packaging); Rule 2: Applicable time period = evening peak (18:00-20:00), pedestrian flow range 5-10 people / minute, easily mixed disposal category = recyclable waste - other waste, characteristic description = weight 0.001-0.004, PC1=0.1-0.5, PC2=0.1-0.7 (contaminated recyclables); The final result is a set of behavioral-garbage association rules containing 8 core rules, covering different time periods and pedestrian traffic scenarios.

[0033] Based on the behavior-garbage association rule set, the time-synchronized multi-source dataset is labeled, and user disposal behavior tags are added to each garbage disposal record, ultimately generating a categorized garbage dataset with user disposal behavior tags.

[0034] The labeling process must follow the workflow of "rule matching - label assignment - anomaly detection" to ensure that the labels of each waste disposal record accurately reflect user behavior. The specific implementation is as follows: Rule matching logic: For each record in the time-synchronized multi-source dataset, match the rules in the behavior-garbage association rule set sequentially. The matching conditions are "consistent time period + matching traffic flow range + garbage features falling within the feature interval described by the rule". For example, if a record has the following information: time period = 7:35 (morning peak), traffic flow = 12 people / minute (>10 people / minute), weight = 0.004 (0.2kg), PC1 = 0.5, PC2 = 0.4 (falling within the feature interval of rule 1), then it matches rule 1; if a record has the following information: time period = 19:20 (evening peak), traffic flow = 8 people / minute (5-10 people / minute), weight = 0.002 (0.1kg), PC1 = 0.2, PC2 = 0.3 (falling within the feature interval of rule 2), then it matches rule 2.

[0035] Behavioral tag definition and assignment: Tags are divided into two categories: "Easily Misused Tags" and "Normal Targeting Tags". Easy-to-mix waste label: The format is "Easy-to-mix waste - Category A - Category B", which corresponds to the matching rules. For example, records matching rule 1 are labeled "Easy-to-mix waste - Kitchen waste - Other waste", and records matching rule 2 are labeled "Easy-to-mix waste - Recyclable waste - Other waste". Normal disposal label: If the record does not match any easily mixed disposal rules (such as time period = 14:30 (off-peak), traffic flow = 3 people / minute, weight = 0.006 (0.3kg), PC1 = 0.7, PC2 = 0.3 (pure vegetable leaves, pure kitchen waste characteristics)), then the waste category is determined according to the image characteristics and weight, and labeled "Normal disposal - kitchen waste"; if it is a pure plastic bottle (PC1 = 0.2, PC2 = 0.2, weight = 0.0016 (0.08kg)), it is labeled "Normal disposal - recyclable waste".

[0036] Anomaly Detection and Correction: A 10% sampling rate is used to verify the accuracy of labeled records by manually checking image features and weight. For example, if a record labeled "Easily Mixed - Kitchen Waste - Other" is found to be pure vegetable leaves (without packaging) upon manual image verification, it is corrected to "Normal Disposal - Kitchen Waste"; if a record labeled "Normal Disposal - Recyclable" is actually an oil-stained plastic container, it is corrected to "Easily Mixed - Recyclable - Other Waste," ensuring label accuracy > 95%.

[0037] Classified waste dataset generation: All labeled records are integrated according to the format of "collection ID, timestamp, waste weight (kg), image feature description, corresponding time period pedestrian flow (people / minute), disposal behavior label, waste category (predicted value)" to generate a classified waste dataset with user disposal behavior labels. Example data is as follows: "Collection ID: TRASH-20240520-001, Timestamp: 2024-05-20 07:45:32.678, Waste weight: 0.23, Image feature description: Green + White, Irregular blocks, Local rough + smooth (Plastic bag vegetable leaves), Corresponding time period pedestrian flow: 12, Disposal behavior tag: Easily mixed disposal - Kitchen waste - Other waste, Waste category: Kitchen waste (with risk of mixed disposal)"; "Collection ID: TRASH-20240520-002, Timestamp: 2024-05-20 19:20:15.345, Waste weight: 0.08, Image feature description: Transparent, cylindrical, smooth (oil-stained plastic bottle), Corresponding time period pedestrian flow: 8, Disposal behavior tag: Easily mixed disposal - Recyclable waste - Other waste, Waste category: Recyclable waste (with risk of mixed disposal)."

[0038] S202: Input the classified waste dataset with user disposal behavior tags into the dynamic prediction model. By analyzing the correlation between the disposal ratio of various types of waste and time, predict the growth curve of classified waste in each waste bin and the high incidence of mixed disposal behavior within a preset time period in the future, and generate prediction results including peak waste disposal time, easily mixed disposal categories and overflow warning time of each bin. Specifically, feature engineering can be performed on categorized garbage datasets with user disposal behavior tags to extract the time-series features of disposal volume and the distribution features of behavior tags for each type of garbage, and generate feature vectors for model input. The categorized waste dataset with user disposal behavior tags contains core information: waste bin number (e.g., "Residential Area Bin 1", "Commercial Area Bin 3"), disposal timestamp (accurate to the minute), waste category (kitchen waste, recyclable waste, other waste, hazardous waste), disposal weight (kg), disposal behavior tag (e.g., "Easily Mixed Disposal - Kitchen Waste - Other", "Normal Disposal - Recyclable"), and corresponding pedestrian traffic (people / minute). Feature engineering needs to extract key information from the "time series dimension" and "behavioral distribution dimension," eliminate data redundancy, and unify feature scales to provide structured data for model input.

[0039] 1. Extraction of time-series features of delivery volume: For each trash can and each type of trash, data was divided into "1-hour time windows" (balancing timeliness and data stability, ensuring continuity of trash disposal behavior within one hour), and four core time-series features were extracted: Statistical characteristics of waste disposal within the designated window include: the average disposal volume within one hour (reflecting the overall disposal level during that period), the standard deviation (reflecting the degree of fluctuation in disposal volume), and the maximum value (reflecting the intensity of peak disposal). For example, the historical data for kitchen waste disposal in bin #1 of the residential area during the morning peak hours of 7:00-8:00 AM shows volumes of 7.2 kg, 8.5 kg, 7.8 kg… over a 30-day period. The average is calculated to be 8.0 kg; the standard deviation is expressed by the formula “√[Σ(xᵢ-μ]”. 2 The calculation yielded 1.2 kg (xᵢ is the daily disposal volume, μ is the average, n=30); the maximum value was 10.5 kg (Monday morning rush hour, users accumulate more garbage over the weekend).

[0040] Moving average trend characteristics: The average amount of waste disposed of is calculated using a 5-minute moving window to capture short-term disposal trends. Then, the trend slope (reflecting the rate of increase or decrease in disposal volume) is calculated using linear regression. For example, 1.5 kg of food waste was disposed of from 7:00 to 7:05, 1.8 kg from 7:05 to 7:10, and 2.0 kg from 7:10 to 7:15… The 5-minute moving averages are 1.5, 1.8, and 2.0 respectively… The linear regression slope is 0.06 kg / minute (i.e., an increase of 0.3 kg every 5 minutes), indicating a rapid increase in disposal volume at the beginning of the morning peak.

[0041] Periodic characteristics: The daily (e.g., morning peak 7-9 am, evening peak 18-20 pm) and weekly (e.g., Monday's disposal volume is 15% higher than Sunday's) cycles are extracted using Fourier transform. The intensity of the cycles is quantified as a feature value of 0-1 (1 indicates a highly significant cycle, 0 indicates no cycle). For example, the daily cycle feature value of kitchen waste in bin #1 of the community is 0.9 (clear bimodal characteristic), and the weekly cycle feature value is 0.7 (10%-20% higher on weekdays than on weekends).

[0042] Pedestrian Traffic Correlation Characteristics: Calculate the Pearson correlation coefficient between the amount of product distributed and pedestrian traffic within one hour (reflecting the degree of linear correlation between the two, ranging from -1 to 1). For example, if the average pedestrian traffic is 12 people / minute from 7:00 to 8:00, the correlation coefficient between the amount of product distributed and pedestrian traffic is 0.8, indicating that the higher the pedestrian traffic, the higher the amount of product distributed (concentrated user distribution); if the pedestrian traffic is 0.2 people / minute from 2:00 to 3:00 AM, the correlation coefficient is 0.1 (almost no correlation).

[0043] 2. Extraction of behavioral label distribution features: For each 1-hour time window, the distribution pattern of the campaign behavior tags was analyzed, and 3 types of features were extracted: Percentage of easily mixed disposal labels: The proportion of records with easily mixed disposal labels in a certain type of waste out of the total number of records for that type of waste. For example, there are 50 records of kitchen waste in bin No. 1 of the community from 7:00 to 8:00, of which 18 are labeled "easily mixed disposal - kitchen waste - other", accounting for 18 / 50=36% (0.36 after standardization).

[0044] Distribution of Easily Mixed Items by Category: For records with the "easily mixed" label, the percentage of different easily mixed items by category is calculated. For example, among the 18 easily mixed items records in this period, 15 records are labeled "easily mixed - kitchen waste - other" (accounting for 83%) and 3 records are labeled "easily mixed - kitchen waste - hazardous waste" (accounting for 17%). The percentage of the main easily mixed item category (kitchen waste - other) is used as the feature value (0.83).

[0045] Label-weight correlation feature: Calculate the difference in average disposal weight between records with "easy to mix" labels and records with normal labels. For example, the average weight of kitchen waste with the "easy to mix" label is 0.25 kg, while the average weight of waste with the normal label is 0.32 kg, with a difference of -0.07 kg (-0.3 kg after standardization). This reflects the pattern that users tend to dispose of smaller amounts of waste due to hesitation when "easy to mix".

[0046] 3. Feature standardization and vector construction: Z-score standardization is used to eliminate dimensional differences. The formula is "Standardized value = (Original value - Mean of feature) / Standard deviation of feature". For example, the average amount of kitchen waste disposed of is 8.0 kg, the standard deviation is 2.0 kg, and the amount disposed of at a certain window is 9.0 kg. After standardization, it is (9.0 - 8.0) / 2.0 = 0.5. The average proportion of easily mixed waste disposed of is 30%, the standard deviation is 15%, and the proportion at a certain window is 36%. After standardization, it is (36 - 30) / 15 = 0.4.

[0047] The standardized temporal features and behavioral label distribution features are integrated according to "trash can - trash category - time window" to generate a feature vector. For example, the feature vector of kitchen waste in bin No. 1 of the community from 7:00 to 8:00 is: [mean amount disposed of 0.5, standard deviation -0.3, maximum value 0.8, trend slope 0.6, daily periodic feature 0.9, weekly periodic feature 0.7, correlation coefficient of pedestrian flow 0.8, proportion of easily mixed disposal 0.4, proportion of easily mixed disposal categories 0.83, label-weight difference -0.3], with 10 dimensions (adjusted according to the number of features).

[0048] The feature vector is input into an LSTM dynamic prediction model based on an attention mechanism. This model learns the temporal pattern of waste disposal by analyzing the nonlinear correlation between the proportion of various types of waste disposal and time in historical data. The Attention-LSTM model combines the temporal memory capability of LSTM with the key information focusing capability of the attention mechanism, which can accurately capture the non-linear relationship of "time-category-behavior" in garbage disposal. The model architecture consists of four parts: input layer, LSTM encoding layer, attention layer, and fully connected layer.

[0049] 1. Model Architecture Design: Input layer: Receives the feature vector generated in step one, with a sequence length of 24 (corresponding to 24 hours of continuous historical data, covering the complete daily cycle), and a feature dimension of 10 (consistent with the feature vector dimension). For example, inputting the feature vector sequence of kitchen waste from bin #1 in the community over the past 24 hours (May 20th, 00:00-23:59), a total of 24 time steps, each time step being a 10-dimensional vector.

[0050] LSTM encoding layer: Two hidden layers are used. The first layer has 64 neurons (capturing basic temporal features), and the second layer has 32 neurons (extracting higher-order correlation features). The activation function is tanh (output range -1 to 1, suitable for fluctuations in temporal data), and the dropout rate is set to 0.2 (randomly discarding 20% ​​of neurons to prevent overfitting). LSTM uses a gating mechanism to remember key temporal information: for example, for historical data from 7-9 am during the morning rush hour, the forget gate retains 80% of the information (important time period), while for data from 2-4 am, it retains 20% of the information (irrelevant time period). The input gate assigns high update weights (0.7-0.8) to key features such as "correlation coefficient of pedestrian flow" and "proportion of easily confused labels," and low weights (0.3) to "label-weight difference."

[0051] Attention Layer: A multi-head self-attention mechanism is adopted (4 attention heads, focusing on "distribution trend", "easily mixed distribution", "periodicity", and "pedestrian traffic correlation" respectively) to calculate the attention weight for each historical time step. The weight calculation formula is "αᵢ=exp (eᵢ) / Σexp (eⱼ)" (eᵢ is the relevance score of time step i, calculated through a fully connected layer). For example, in the historical sequence, the relevance score eᵢ of the time step 7-8 o'clock is 1.2, and the average eⱼ of other time steps is 0.3. The weight αᵢ=exp(1.2) / [exp (1.2)+23×exp (0.3)]≈3.32 / [3.32+23×1.35]≈3.32 / 34.37≈0.096 (although the weight of a single step is not high, the total weight of the three time steps in the morning peak reaches 0.28, which is significantly higher than other time periods). The output of the attention layer is a weighted sum of "weights × LSTM hidden states", which enhances information during key periods.

[0052] Fully connected layer: Maps the 32-dimensional features output by the attention layer to “prediction time step × feature dimension” (e.g., 10-dimensional features per hour for the next 24 hours). The output layer uses a linear activation function to directly output the predicted delivery volume and the probability of easy mixing at each future time step.

[0053] 2. Model training and learning of correlation patterns: Dataset partitioning: Labeled historical data is divided into a 7:2:1 ratio: training set (70%, used for model parameter learning), validation set (20%, used for hyperparameter tuning), and test set (10%, used for evaluating generalization ability). The training data covers different scenarios: residential areas (significant morning / evening rush hours), commercial areas (prominent midday rush hour), and office areas (significant differences between weekdays and weekends).

[0054] Training parameters: The optimizer uses Adam (adaptive learning rate, initial learning rate 0.001, decaying with each training round), batch size 32 (32 “24-hour historical sequences” are input each time), the loss function is mean squared error (MSE, which measures the deviation between the predicted value and the true value), and the iteration is 50 rounds. Training stops when the MSE of the validation set decreases by <0.001 for 5 consecutive rounds (to avoid overfitting).

[0055] Example of nonlinear correlation learning: The model learned the patterns in "kitchen waste from bin #1 in the community" during training: 35% of the daily waste was disposed of during the morning peak (7-9 AM), and the proportion of waste easily mixed with other waste showed a nonlinear positive correlation with pedestrian traffic (as pedestrian traffic increased from 5 people / minute to 15 people / minute, the proportion of easily mixed waste increased from 20% to 45%, with the growth rate gradually slowing down); 30% of the daily waste was disposed of during the evening peak (6-8 PM), and the proportion of easily mixed waste was strongly correlated with "whether it's a weekend" (12% higher on weekends than weekdays, because users are more casual about waste disposal during leisure time). On the test set, the mean absolute error (MAE) of waste volume prediction was 0.5 kg (relative error < 8%), and the accuracy of predicting the probability of easily mixed waste was 82%, meeting practical needs.

[0056] By using a trained dynamic prediction model to perform multi-step forward prediction, the growth curve of classified waste in each trash can and the probability distribution of easy-to-mix waste disposal behavior are generated within a preset time period in the future. The core of "multi-step forward prediction" is to output the prediction results of the next consecutive time steps in one go, avoiding the accumulation of errors in recursive prediction (calculating the next step based on the prediction value of the previous step). The preset duration is set to 24 hours according to the needs of waste classification management (covering the complete daily cycle and facilitating collection and dispatching), and the time interval is 1 hour (balancing accuracy and dispatching granularity).

[0057] 1. Generation of growth curve for sorted waste: Predictive input: Input the feature vector sequence of the previous 24 hours (e.g., 00:00-23:59 on May 20) into the trained model, and the model outputs the predicted value of the amount of various types of waste disposed of each hour for the next 24 hours (00:00-23:59 on May 21).

[0058] Curve plotting logic: Using "time (hours)" as the horizontal axis (0-23) and "amount disposed of (kg)" as the vertical axis, the 24 predicted values ​​for each type of waste are connected sequentially over time to form a growth curve. The curve shapes for different waste categories reflect the disposal patterns. The growth curve of kitchen waste in bin No. 1 of the community is as follows: 0-6 o'clock, 0.5-1.0 kg per hour (low peak), 7-8 o'clock, a sharp increase to 8.0 kg (morning peak), 9-11 o'clock, a decrease to 3.0-4.0 kg, 12-13 o'clock, a slight increase to 4.5 kg due to lunch, 18-19 o'clock, an increase to 7.0 kg (evening peak), and 20-23 o'clock, a gradual decrease to 1.0-2.0 kg, showing a "double peak" pattern.

[0059] The growth curve of recyclable waste in bin No. 3 in the commercial area is as follows: 0.3-0.8kg is disposed of from 0-9 am, rising to 5.0kg from 12-2 pm (peak of takeout box disposal after lunch), dropping to 2.0-3.0kg from 3-5 pm, rising to 4.5kg from 6-8 pm (peak after dinner), and dropping to 1.0kg from 9-11 pm, showing a pattern of "primarily lunchtime peak and secondarily evening peak".

[0060] Curve confidence labeling: Label the confidence interval of the hourly prediction value next to the curve (based on the prediction error during model training, such as ±0.5kg). For example, the predicted amount of food waste to be disposed of at 7-8 o'clock is 8.0kg, with a confidence interval of 7.5-8.5kg, reflecting the uncertainty of the prediction.

[0061] 2. Generation of probability distribution for easily confused betting behaviors: Probability calculation logic: The model outputs the probability of occurrence of a certain type of mixed application in the next hour (prediction based on the distribution characteristics of behavioral tags), with a probability range of 0-100%. The higher the value, the more likely the user is to engage in the corresponding mixed application behavior during that period.

[0062] Distribution curve plotting: Plot curves for each category of easily confused bets, with "Time (hours)" on the horizontal axis and "Probability of easy confusion (%)" on the vertical axis. The probability distribution of "Easily Mixed Disposal - Kitchen Waste - Other" in the No. 1 bin of the community is as follows: 5%-10% from 0-6 am, rising to 62% from 7-8 am (during the morning rush hour, users are in a hurry and are likely to mistakenly put plastic bags of vegetable leaves into the bin), dropping to 25%-30% from 9-11 am, rising to 58% from 18-19 am (during the evening rush hour, users are fatigued and are likely to ignore the sorting), and dropping to 15%-20% from 20-23 pm.

[0063] The probability distribution of "Easy to mix - Recyclable - Other" in bin No. 3 in the commercial area is as follows: 55% from 12:00 to 14:00 (after lunch, users are more likely to throw oily plastic lunch boxes into other waste), 52% from 18:00 to 20:00, and less than 30% at other times, which highly overlaps with the peak time for recyclable waste disposal.

[0064] Preliminary determination of high-incidence periods: Continuous time steps with a probability >50% are tentatively defined as high-incidence periods for easily mixed waste disposal. For example, the tentative high-incidence periods for "easily mixed waste disposal - kitchen waste - other" in bin No. 1 of the community are 7-8 am and 18-19 pm, which provides a basis for subsequent accurate determination.

[0065] Based on the waste growth curve, an overflow warning threshold is set, and the high-incidence period is determined by combining the probability distribution of easy-to-mix waste disposal behavior. Finally, a prediction result is generated that includes the peak waste disposal period, easy-to-mix waste categories, and the overflow warning time of each bin.

[0066] 1. Overflow warning threshold setting: The overflow warning threshold needs to be combined with the actual capacity of the smart trash can and the density of the trash (different trash densities vary greatly, so setting it directly by weight is more accurate). The core principle is to "reserve 20% capacity space to avoid trash overflow leading to difficulties in disposal or environmental pollution".

[0067] Basic parameters are determined: the common capacity of smart trash cans is 50L (mainstream models in residential / commercial areas). The density of different types of waste is determined by historical data: kitchen waste density is 1.2kg / L (high water content, high density), recyclable waste is 0.8kg / L (plastic and paper, low density), other waste is 1.0kg / L (mixed waste, medium density), and hazardous waste is 0.9kg / L (small amount, no overflow warning set for now).

[0068] Weight threshold calculation: Overflow warning weight = bin capacity × 80% × waste density. For example, the overflow threshold for a 50L bin is 50 × 0.8 × 1.2 = 48kg for food waste, 50 × 0.8 × 0.8 = 32kg for recyclables, and 50 × 0.8 × 1.0 = 40kg for other waste. If the bin capacity is 30L (small community), the food waste threshold is 30 × 0.8 × 1.2 = 28.8kg, and so on.

[0069] 2. Overflow warning time calculation: The overflow warning time is determined based on the "cumulative amount" of waste disposal based on the waste growth curve. The steps are as follows: Starting from the current moment, the predicted delivery volume for each future time period is accumulated in 1-hour time steps to obtain the cumulative delivery volume sequence; Find the time step when the cumulative delivery volume first reaches the overflow warning threshold; the start time of this time step is the overflow warning time. If the cumulative amount of waste disposal does not reach the threshold within 24 hours, the warning time will be marked as "XX o'clock the next day".

[0070] Example: The current time for bin #1 in the residential area is 00:00 on May 21st. The predicted amount of food waste disposed of according to the growth curve is: 0-1:00 0.5kg, 1-2:00 0.4kg…7-8:00 8.0kg, 8-9:00 4.0kg…18-19:00 7.0kg, 19-20:00 4.0kg. Cumulative disposal calculation: 0-7:00 cumulative 10.2kg, 7-8:00 cumulative 18.2kg, 8-9:00 cumulative 22.2kg…18-19:00 cumulative 45.1kg, 19-20:00 cumulative 49.1kg (first exceeding the 48kg threshold). Therefore, the overflow warning time is 19:00 (the starting time of 19-20:00).

[0071] 3. Determining the peak periods for mixed-use and mixed-use applications: High-incidence periods for mixed disposal: Periods with a probability greater than 50% and a continuous duration of ≥1 hour in the probability distribution of mixed disposal are defined as high-incidence periods, and the corresponding mixed disposal category is labeled accordingly. For example, in the No. 1 bin of the community, the probability of "Easily mixed disposal - Kitchen waste - Other" is 62% from 7-8 am and 55% from 8-9 am (>50% for 2 consecutive hours), so the high-incidence period is 7:00-9:00 am, and the easily mixed disposal category is "Kitchen waste - Other waste"; the probability of "Easily mixed disposal - Recyclable waste - Other" is 58% from 18:00-19:00 and 51% from 19:00-20:00, so the high-incidence period is 18:00-20:00, and the category is "Recyclable waste - Other waste".

[0072] Peak garbage disposal times: The periods in the growth curve where "hourly disposal volume > 1.5 times the average daily disposal volume" are defined as peak periods. Average daily disposal volume = 24-hour cumulative predicted disposal volume / 24. For example, the cumulative disposal volume of kitchen waste in bin No. 1 of the community is 52kg, with an average of 2.17kg. 1.5 times is 3.25kg. The periods with disposal volume > 3.25kg are 7:00-8:00 (8.0kg) and 18:00-19:00 (7.0kg), which are the peak periods.

[0073] 4. Integration of prediction results: The above information is structured and integrated according to "trash can number - trash category - prediction indicator" to generate the final prediction result. Example: "Trash can number: No. 1 in the community (in front of Building 1); Estimated duration: 2024-05-21 00:00-23:59."

[0074] Peak garbage disposal times: Kitchen waste: 7:00-8:00, 18:00-19:00 (hourly disposal capacity: 8.0kg, 7.0kg respectively); Recyclable waste: 12:00-13:00, 18:00-19:00 (hourly disposal limits: 5.0kg, 4.5kg). Other waste: 8:00-9:00, 19:00-20:00 (hourly disposal limits: 3.5kg, 3.2kg).

[0075] Easily confused categories and time periods: Kitchen waste - Other waste: 7:00-9:00 (probability 55%-62%) Recyclable waste - Other waste: 12:00-14:00 (probability 52%-55%), 18:00-20:00 (probability 51%-58%).

[0076] Overflow warning time: Kitchen waste: 19:00 (cumulative amount disposed of 49.1kg, threshold 48kg); Recyclable waste: 22:00 (cumulative amount disposed of 32.5kg, threshold 32kg); Other waste: 05:00 the next day (cumulative amount disposed of 40.2kg, threshold 40kg)".

[0077] S203. Based on the prediction results, the classification guidance-collection linkage mechanism is activated. Before the peak period of easy mixed disposal and the peak disposal period, the classification guidance animation of the corresponding easy mixed disposal category is pushed through the display screen on the top of the garbage can. At the same time, combined with the garbage growth curve of each can and real-time traffic data, the classification and collection routes of garbage trucks are planned, and a garbage collection and dispatch plan with classification guidance content is generated. Specifically, it can analyze the high-incidence periods of easily mixed-use behaviors and peak delivery periods in the prediction results, trigger the classification guidance mechanism in advance, and generate classification guidance start instructions; First, the structured format of the "prediction results" needs to be clearly defined, which includes core fields such as the unique identifier of the trash can (e.g., "Kitchen waste bin east of Building 1, Xingfu Community" and "Recyclable bin in Area B of Commercial Plaza"), information on easily mixed disposal behavior (category combination + high-incidence time period), peak disposal time (start time + end time), and overflow warning time. The parsing process needs to be completed in three steps: "field extraction - time period verification - advance calculation" to ensure that the guidance mechanism is accurately activated before the critical time period.

[0078] 1. Field extraction and time period confirmation: Extract target information from the prediction results: Taking the "Kitchen Waste Bin East of Building 1, Xingfu Community" as an example, the information on the tendency to mix waste is "Category Combination: Kitchen Waste - Other Waste, Peak Period: 7:00-9:00", with the peak disposal period being "7:30-8:30" (highly overlapping with the tendency to mix waste, as users concentrate on disposal during peak hours, leading to mixed waste). The reasonableness of the time period needs to be verified, for example, confirming whether 7:00-9:00 is the community's morning peak (based on historical pedestrian traffic data, the pedestrian traffic during this period reaches 15 people / minute, meeting peak characteristics), to avoid misjudgment of the time period due to prediction errors.

[0079] 2. Pre-triggered time setting: The categorized guidance should be initiated "before users begin their concentrated waste disposal" to ensure that users can see the instructions when disposing of their waste. The advance time is determined based on the average walking time from the community entrance to the trash can (approximately 5-10 minutes) and the animation loading time (approximately 2 minutes), and is uniformly set to 30 minutes. For example, during the peak period of mixed waste disposal, 7:00-9:00, the guidance should be triggered 30 minutes in advance, i.e., the start time is 6:30; during the peak disposal period of 7:30-8:30, since it is already included in the mixed waste disposal guidance period, there is no need to trigger it again, but the guidance frequency should be increased during peak periods.

[0080] 3. Generation of categorized boot instructions: The instructions must include key information such as "executing entity (trash can number), start time, easily mixed waste categories combination, guidance duration, and playback parameters," and must be in JSON format to ensure that IoT devices can parse them. For example, the instruction for the aforementioned food waste bin would be: { "Trash Can Number": "XF-SQ-1-D-CY", Startup Time: 2024-05-22 06:30:00 "Easily Mixed Category": "Kitchen Waste - Other Waste" "Duration of the guidance": "150 minutes", Playback parameters: Initial frequency 5 minutes / playback, peak hours (7:30-8:30) 3 minutes / playback. }

[0081] After the instruction is generated, it is sent to the control module of the corresponding trash can through the community IoT gateway. The gateway uses the MQTT protocol (a lightweight IoT protocol suitable for low-bandwidth devices with a transmission delay of <1 second) to ensure that all target trash cans receive the instruction and are ready to start before 6:30.

[0082] Based on the classification guidance start command and the information on easily mixed categories in the prediction results, the corresponding classification guidance animation content is retrieved from the multimedia library and sent to the display screen on top of each trash can through the Internet of Things network to generate a classification guidance playback task. 1. Structured storage design for multimedia libraries: The multimedia library stores guide animations categorized by "easily mixed-projection category combinations." Each animation file is associated with metadata such as "category tag, duration, resolution, and playback scene" to ensure quick matching and retrieval. For example: The category label "Kitchen Waste - Other Waste" corresponds to an animation: a 30-second short video demonstrating the difference between "vegetable leaves and leftovers (kitchen waste)" and "plastic bags and takeout boxes (other)", emphasizing that "packaged kitchen waste must be removed from the bag before disposal", with a resolution of 1920×1080 (adapted to the 10-inch display screen on top of the trash can), and the playback scenario is labeled "before morning / evening rush hour". The category label "Recyclable Waste - Other Waste" corresponds to an animation: a 25-second short video demonstrating the difference between "clean plastic bottles (recyclable)" and "oil-stained plastic boxes (other)," emphasizing that "recyclable materials must be clean and dry."

[0083] The animation files are encoded using H.264 (high compression rate, low bandwidth consumption) and stored on the cloud server of the community management platform. Fuzzy search by tag is supported (e.g., entering "kitchen waste - other" will match the corresponding animation).

[0084] 2. Animation content retrieval and distribution: Based on the "easily mixed categories" in the classification guidance startup instruction, the management platform uses an SQL query to match the corresponding animation file from the multimedia library. For example, if the category in the instruction is "kitchen waste - other waste", the query retrieves the animation file ID "DY-2024-CY-QT-001". The distribution process uses a "fragmented transmission + breakpoint resume" mechanism: since the animation file is approximately 5MB, considering the bandwidth of the trash can device (usually 2Mbps), the file is divided into 5 1MB fragments and transmitted via HTTP protocol; if the transmission is interrupted (e.g., due to network fluctuations), the device requests the unreceived fragments upon the next connection to avoid duplicate transmission.

[0085] The target of the command is the display screen corresponding to the "trash can number". After receiving the file, the display screen control module (equipped with an ARM Cortex-A7 processor) stores it in 8GB of local flash memory to ensure playback when offline. For example, the "XF-SQ-1-D-CY" trash can complete the animation reception before 6:25 and waits for the start time to trigger.

[0086] 3. Category-guided playback task generation: Playback tasks need to specify the "playback time range, frequency, duration, and priority," and are generated in conjunction with the "startup time, guidance duration, and playback parameters" in the instructions. Taking the "XF-SQ-1-D-CY" trash can as an example: Playback time range: 6:30-9:00 (start time 6:30 + duration 150 minutes); Playback frequency: 6:30-7:30 (off-peak hours) once every 5 minutes, 7:30-8:30 (peak hours) once every 3 minutes, 8:30-9:00 (after peak hours) resumes once every 5 minutes; Playback duration: 30 seconds per session (to avoid excessively long sessions affecting user engagement, and excessively short sessions resulting in undelivered information); Priority: Higher than other notifications (such as overflow alerts), ensuring that the onboarding animation is not interrupted.

[0087] After a task is generated, it is stored in the task queue of the trash can control module. The module executes the task automatically according to the time nodes and sends the playback status (such as "6:30 First playback successful") back to the management platform for easy monitoring.

[0088] Real-time data on waste growth curves for each trash can and traffic conditions in the area related to the trash can are acquired. A path optimization algorithm is used to calculate the optimal collection route and generate a preliminary collection route plan. 1. Real-time data acquisition and preprocessing: Waste growth curve data: Data is retrieved in real-time from the community waste sorting management platform. The platform updates the "current weight-time" curve for each waste bin every 5 minutes (based on data from the smart waste bin weighing sensors). The data format is "waste bin number, current time, current weight, cumulative weight, overflow threshold, and remaining capacity percentage". For example, the real-time data for bin "XF-SQ-1-D-CY" at 7:00 AM is: current weight 28kg, overflow threshold 48kg, remaining capacity 41.7%, and estimated overflow time 19:00 (consistent with the prediction result, verifying data reliability).

[0089] Traffic data is obtained through third-party map APIs (such as the "Route Planning - Real-time Traffic" interface of Gaode Map Open Platform), covering major roads in the area where the trash cans are located (such as "Xingfu Road" and "Hexie Street" around the community). The data includes "road name, current congestion level (0-5, 0 for smooth traffic and 5 for severe congestion), average speed (km / h), and estimated travel time (minutes)". For example, at 7:00, "Xingfu Road (Building 1 section)" has a congestion level of 1, an average speed of 30km / h, and a travel time of 2 minutes; "Hexie Street (Commercial Plaza section)" has a congestion level of 3, an average speed of 15km / h, and a travel time of 5 minutes.

[0090] During preprocessing, it is necessary to link the data of both: bind the "remaining capacity percentage" in the growth curve with the "congestion level of surrounding roads" in the road condition data according to the "trash can number". For example, the "XF-SQ-1-D-CY" can be bound to "Xingfu Road congestion level 1, remaining capacity 41.7%" to provide constraints for route optimization.

[0091] 2. Path optimization algorithm selection and parameter setting: A genetic algorithm was chosen (suitable for solving path optimization problems with multiple constraints and objectives, and capable of finding near-optimal solutions in a short time). The core objectives of the algorithm are "minimizing the total collection time and minimizing the total travel distance," with constraints including: Time constraints: The emptying time for each trash can must be completed before its "estimated overflow time - 1 hour" (allowing a 1-hour buffer to avoid overflow); Capacity constraints: Garbage trucks are loaded into separate containers according to garbage category (one container each for kitchen waste, recyclables, and other waste), with each container having a capacity of 5 tons, and the total weight of a single collection not exceeding 15 tons; Road condition constraints: Prioritize roads with congestion level ≤ 2. If a congested road section (level > 2) must be passed through, the congestion coefficient must be added to the travel time calculation (1.5 times for level 3, 2 times for level 4, and 3 times for level 5).

[0092] Algorithm parameter settings: population size 50 (generating 50 candidate routes in each iteration), number of iterations 30 (ensuring convergence, computation time < 5 minutes), crossover probability 0.8 (controlling the diversity of route combinations), mutation probability 0.1 (avoiding local optima).

[0093] 3. Preliminary cleanup route plan generated: Taking the early morning garbage collection task in Xingfu Community as an example (1 garbage truck, responsible for 5 garbage bins: A=XF-SQ-1-D-CY, B=XF-SQ-2-D-KHS, C=XF-SQ-3-D-QT, D = Commercial Plaza B Zone - CY, E = Commercial Plaza C Zone - KHS): Initialize candidate routes: Randomly generate 50 routes, such as [A→B→C→D→E], [B→A→D→C→E], etc. Calculate the fitness (objective function value) for each route: For example, the total mileage of route [A→B→C→D→E] is 12km, the total travel time (including loading and unloading time) is 45 minutes, and all constraints are met (A barrel clearing time 7:30 < 18:00, container loading capacity 3.2 tons < 5 tons). Iterative optimization: New routes are generated through crossover and mutation operations, routes with poor fitness are eliminated, and the route with the highest fitness [B→A→C→E→D] is finally selected. Its total distance is 10.5km, the total time is 40 minutes, and all garbage cans are emptied more than 1 hour before overflow.

[0094] The preliminary plan should include "route sequence, estimated arrival time of each trash can, estimated collection time, and routes and road conditions", for example: Garbage truck number: QY-01; Preliminary route: B (XF-SQ-2-D-KHS, estimated arrival time 7:10, 5 minutes of cleanup, passing through Xingfu Road (level 1 congestion)) → A (XF-SQ-1-D-CY, estimated arrival time 7:20, 5 minutes of cleanup, passing through Hexie Street (level 2 congestion)) → C (XF-SQ-3-D-QT, estimated arrival time 7:35, 5 minutes of cleanup) → E (Commercial Plaza C area - KHS, estimated arrival time 7:50, 5 minutes of cleanup) → D (Commercial Plaza B area - CY, estimated arrival time 8:05, 5 minutes of cleanup)".

[0095] By integrating the classification guidance playback task and the preliminary collection route plan, and taking into account the capacity of collection vehicles and the constraints of operation time, multi-objective optimization scheduling is carried out to finally generate a waste collection scheduling plan with classification guidance content.

[0096] 1. Integrate logic and constraint verification: The core of the integration is to "ensure that the sorting guidance animation is playing when the garbage truck arrives at the bin, allowing the driver to observe the user's disposal (helping to judge the accuracy of sorting), while avoiding interruption of the playback during the collection operation." First, the "estimated arrival time" of each garbage bin in the initial route needs to be matched with the "sorting guidance playback time range" for that bin: If the expected arrival time is within the playback time range (e.g., if bucket A arrives at 7:20, it falls within the playback range of 6:30-9:00), then the match is successful; If the expected arrival time exceeds the playback range (e.g., a certain bucket's playback range is 8:00-10:00, and the initial route arrival time is 10:10), then the route order or playback time needs to be adjusted (prioritize adjusting the route to avoid affecting user guidance).

[0097] Simultaneously, a second verification of the constraints on "vehicle capacity" and "operating time" is required. Capacity verification: The weight of the garbage in each bin is added up in the order of the initial route. For example, bin B (recyclable, current weight 15kg) + bin A (kitchen waste, 28kg) + bin C (other, 22kg) + bin E (recyclable, 18kg) + bin D (kitchen waste, 30kg). The total weight of recyclable garbage is 33kg, the total weight of kitchen waste is 58kg, and the total weight of other garbage is 22kg. All of these are much less than the 5-ton bin capacity, which meets the constraints. Time verification: Total operation time (40 minutes) + round trip time to the garage (20 minutes) = 60 minutes. The early shift's cleaning deadline is 9:00. If you depart at 7:00, you can complete the task before 8:00, which meets the time constraint.

[0098] 2. Multi-objective optimization scheduling: The optimization objective is to "maximize the overlap between the classification guidance playback period and the collection arrival time, and minimize the total operation time deviation (compared to the initial plan)." A weighted summation method is used to transform the multiple objectives into a single objective (overlap weight 0.6, time deviation weight 0.4). For example: In the preliminary route, Bucket E (Commercial Plaza C Zone - KHS) is expected to arrive at 7:50, with a playback time range of 7:00-9:00 and a 100% overlap; Bucket D (Commercial Plaza B Zone - CY) has a playback time range of 8:00-10:00, and is expected to arrive at 8:05, with a 100% overlap and no adjustment required. If a bucket initially arrives at 8:00, and the playback range is 7:30-8:00, with an overlap of only 5 minutes (8:00-8:05 is outside the range), then the route order should be adjusted so that the bucket arrives earlier at 7:55, with an overlap of 10 minutes. This improves the overlap while increasing the total time deviation by only 2 minutes, which meets the optimization goal.

[0099] 3. Final scheduling scheme generation: The plan must include "vehicle information, route details, category guidance information, timeframes, and execution requirements," as shown in the following structured format example: Xingfu Community Waste Sorting and Collection Scheduling Plan (May 22, 2024, Morning Shift); Vehicle Information: Vehicle Number: QY-01; Driver: Zhang XX; Container Capacity: 5 tons for kitchen waste, 5 tons for recyclables, 5 tons for other waste; Departure Time: 7:00 AM (Garage Location: West Gate of Xingfu Community); Route details are shown in Table 1 (including classification guidance): Table 1

[0100] Implementation requirements: Upon arriving at each trash can, the driver must confirm whether the sorting guidance animation is playing correctly, and report any abnormalities to the platform via the vehicle terminal. The cleanup operation must be completed within the estimated timeframe to avoid impacting subsequent routes; The loading capacity of each trash can must be recorded to ensure that the capacity is not exceeded.

[0101] After the plan is generated, it is distributed to the garbage truck's onboard terminal (which supports touch screen viewing) through the community management platform, and simultaneously uploaded to the community's garbage classification management platform to facilitate management personnel's monitoring of the implementation progress.

[0102] S204 collects data on the actual classification accuracy of user waste disposal and data on the collection tasks completed by garbage trucks according to the scheduling plan. It integrates the collected data with the prediction results and the collection scheduling plan to generate a garbage classification management report that includes the garbage classification compliance rate, collection efficiency and overflow warning accuracy of each area. The report is then uploaded to the community garbage classification management platform to achieve visualized management of the entire garbage classification process.

[0103] Specifically, the IoT devices of smart trash cans can be used to collect data on the actual sorting accuracy of users' trash cans, and at the same time, data on the completion of collection tasks can be obtained from the on-board terminals of garbage trucks to generate actual execution datasets. 1. Data collection on classification accuracy (smart trash can end): The smart trash can needs to collect classification accuracy data through a two-dimensional approach of "image recognition + weight verification". The core IoT devices include a 1080P high-definition camera (25fps frame rate, 110° field of view, aimed at the recognition area below the disposal port), a ZEMIC H3C weighing sensor (0.1kg accuracy, sampling frequency 1 time / second) and an RK3399 edge computing module (supporting real-time image inference).

[0104] Image Recognition and Classification: The edge computing module is equipped with the YOLOv5-Lite lightweight object detection algorithm (suitable for low-computing-power devices, inference speed <0.5 seconds / frame). The pre-training dataset contains 50,000 labeled images (e.g., vegetable leaves, plastic bottles, plastic bags, used batteries) of four major categories of waste (kitchen waste, recyclables, other, and hazardous waste). When a user disposes of waste, the camera captures three consecutive frames. The algorithm identifies the waste category and outputs a confidence score (range 0-1, confidence score >0.8 is considered a valid identification). For example, if a user throws a plastic bottle into the "kitchen waste bin," the algorithm identifies it as "recyclable waste, confidence score 0.92," and classifies it as "misclassified"; when throwing vegetable leaves, it identifies it as "kitchen waste, confidence score 0.95," and classifies it as "correctly classified."

[0105] Weight verification assistance: The weighing sensor records the weight of each disposal. If the identification result matches the type of trash can, the weight is recorded as "correct disposal weight"; otherwise, it is recorded as "incorrect disposal weight". For example, if 0.2kg of vegetable leaves are disposed of in the food waste can (correct), and 0.08kg of plastic bottles are disposed of subsequently (incorrect), then the current "correct weight 0.2kg, incorrect weight 0.08kg" for the can.

[0106] Accuracy calculation logic: Statistics are based on a "1-hour time window". Classification accuracy = (Number of correct disposals / Total number of disposals) × 100%, while also recording the "distribution of incorrect categories" (e.g., the percentage of recyclable waste and other waste incorrectly disposed of in the food waste bin). For example, if the food waste bin is disposed of 20 times in a certain hour, with 16 correct disposals, the accuracy rate = 16 / 20 × 100% = 80%. Of the incorrect disposals, 3 were for recyclable waste and 1 was for other waste.

[0107] 2. Data collection upon completion of waste collection task (garbage truck onboard terminal): The garbage truck's onboard terminal integrates a GPS positioning module (10-meter accuracy, update frequency 1 time / 30 seconds), a Mettler Toledo weighing sensor (range 0-10 tons, accuracy 0.1 tons), a 4G communication module, and task management software. The collected data includes: Task execution status: The terminal stores the "garbage bin number, estimated arrival time, and estimated collection weight" in the pre-stored collection and dispatch plan. GPS records the actual arrival time (error < 1 minute) and departure time, and the weighing sensor records the actual collection weight. For example, if the dispatch plan requires the garbage truck to arrive at the "XF-SQ-1-D-CY" food waste bin at 7:10, and the actual arrival time is 7:12 and the departure time is 7:17, with an actual collection weight of 28.5 kg (0.5 kg deviation from the dispatch plan's estimated 28 kg), the task is marked as "completed, slight deviation".

[0108] Abnormal Records: If the vehicle does not arrive as scheduled (delay > 10 minutes) or is not cleared (weight change < 0.5 kg), the terminal will automatically record the reason for the abnormality (e.g., "C bin not arrived at 7:30, reason: traffic congestion on Xingfu Road"), which will be uploaded after the driver confirms the information.

[0109] 3. Actual execution of dataset generation: The two types of data are integrated using a dual primary key of "timestamp - trash can number". Each record includes: collection time (accurate to the minute), trash can number, classification accuracy rate (%), correctly disposed weight (kg), incorrectly disposed weight (kg), error category distribution, garbage truck number, actual arrival time, actual departure time, actual collected weight (kg), and task status (completed / abnormal). Example record: "Collection time: 2024-05-22 08:00, trash can number: XF-SQ-1-D-CY, classification accuracy: 80%, correct disposal weight: 2.5kg, incorrect disposal weight: 0.6kg (0.4kg recyclable, 0.2kg other), garbage truck number: QY-01, actual arrival time: 07:12, actual departure time: 07:17, actual collection weight: 28.5kg, task status: completed."

[0110] The dataset is stored in CSV format to ensure it can be directly used for comparative analysis later.

[0111] By comparing and analyzing the actual execution dataset with the prediction results and the waste collection and dispatch plan, key performance indicators such as the waste classification compliance rate, waste collection efficiency and overflow warning accuracy rate of each region are calculated, and performance evaluation results are generated. 1. Determination of benchmark for comparative analysis: Prediction result baseline: Extract the "predicted classification accuracy (e.g., 85%) and overflow warning time (e.g., 19:00)" from the prediction results; Basic criteria for waste collection and dispatching plan: Extract the "estimated arrival time (e.g., 07:10), estimated waste collection weight (e.g., 28kg), and total planned waste collection time (e.g., 40 minutes)" from the plan; Area division rules: Divide the area according to the community's administrative units (such as "Happy Community Buildings 1-3 Area" and "Commercial Plaza BC Area"). Each area contains 3-5 trash cans to ensure that the characteristics of trash disposal are similar within the area.

[0112] 2. Key Performance Indicator Calculation (with examples): Regional waste sorting compliance rate: The weighted average of "sorting accuracy rate × total weight disposed of" for all waste bins in the region, with the weight being the total weight disposed of in each bin (bins with higher disposal volumes have a greater impact on the region). Formula: Compliance rate = (Σ(bin i accuracy rate × bin i total disposal weight) / Σ bin i total disposal weight) × 100%.

[0113] Example: In the Xingfu Community, buildings 1-3 contain three bins: A, B, and C. Bin A has an accuracy rate of 80% and a total disposal volume of 3.1 kg; Bin B has an accuracy rate of 85% and a total disposal volume of 4.2 kg; and Bin C has an accuracy rate of 75% and a total disposal volume of 2.7 kg. The compliance rate is calculated as follows: (80% × 3.1 + 85% × 4.2 + 75% × 2.7) / (3.1 + 4.2 + 2.7) × 100% = (2.48 + 3.57 + 2.025) / 10 × 100% = 80.75%.

[0114] At the same time, compare the predicted compliance rate of the region (e.g., 85%), calculate the deviation = actual compliance rate - predicted compliance rate = -4.25%, and analyze the reasons for the deviation (e.g., a problem with the playback of a certain bucket's guide animation).

[0115] Collection efficiency: Calculated from two dimensions: "time efficiency" and "weight efficiency". Time efficiency = (planned total collection time / actual total collection time) × 100% (>100% indicates completion ahead of schedule, <100% indicates delay); Weight efficiency = (actual total collection weight / planned total collection weight) × 100% (reflects whether the trash cans were emptied as planned).

[0116] Example: The planned total collection time for the route handled by vehicle QY-01 was 40 minutes, and the actual time was 38 minutes; the planned total collection weight was 120 kg, and the actual weight was 118.5 kg. Time efficiency = 40 / 38 × 100% ≈ 105.26% (completed ahead of schedule), weight efficiency = 118.5 / 120 × 100% = 98.75% (close to the plan, deviation due to a small amount of residual garbage), overall collection efficiency = (105.26% + 98.75%) / 2 ≈ 102.01%.

[0117] Overflow warning accuracy: Calculated as "number of correct warnings / total number of warnings". Correct warnings include two categories: "predicted overflow and actually overflowed" and "predicted not overflowed and actually not overflowed". The total number of warnings is the number of warning judgments for all trash cans in the area (one overflow warning judgment per can per day).

[0118] Example: In the Xingfu Community, buildings 1-3 have 5 buckets. The prediction for the day was that buckets A and B would overflow, while buckets C, D, and E would not overflow. In reality, buckets A and B overflowed (correct 2 times), bucket C did not overflow (correct 1 time), bucket D unexpectedly overflowed (incorrect 1 time), and bucket E did not overflow (correct 1 time). The total number of warnings was 5, with 4 correct predictions. The accuracy rate is 4 / 5 × 100% = 80%.

[0119] 3. Performance evaluation results generation: The results should include "region name, actual value of each indicator, predicted / planned value, deviation rate, and preliminary analysis of the reasons for the deviation," for example: Area Name: Xingfu Community Buildings 1-3; Waste sorting compliance rate: Actual 80.75%, predicted 85%, deviation -4.25%. Preliminary analysis of the cause: The sorting guidance animation for bin B was not played from 6:30 to 7:00 (equipment network failure). Cleanup efficiency: Actual 102.01%, planned 100%, deviation +2.01%, reason: the road congestion level along the route was lower than predicted (actual level 1, predicted level 2). Overflow warning accuracy: Actual 80%, target 90%, deviation -10%, reason: The actual amount disposed of in bin D was higher than predicted (due to weekend garbage accumulation not being included in the prediction model).

[0120] Based on the performance evaluation results, an automatic report generation technology is used to construct a waste sorting management report that includes data visualization charts and text analysis, generating a complete management report document; 1. Report Auto-Generation Technology Selection and Process: The architecture employs a "template engine + data visualization library" approach. The template engine used is Freemarker (supports dynamic text and chart placeholders, compatible with Word and PDF formats), and the data visualization library is ECharts (lightweight, supports line charts, bar charts, pie charts, etc., and can export reports with embedded images). The generation process is as follows: Read the performance evaluation results dataset (CSV format); Use the ECharts API to generate visual charts (PNG image format, 300dpi resolution); Pass the "text data + chart path" to the Freemarker predefined template and fill in the placeholders; Convert the template to Word / PDF format to generate a complete report document.

[0121] 2. Data visualization chart design (including examples): Charts should intuitively reflect indicator trends and comparisons. Core charts include: Regional Classification Compliance Rate Comparison Bar Chart: The horizontal axis represents each region of the community (e.g., Buildings 1-3, Buildings 4-6, Commercial Plaza), and the vertical axis represents the compliance rate (%). Each region category corresponds to two bars: "Actual Value" and "Predicted Value," clearly showing the deviation. For example, the actual bar height for Buildings 1-3 is 80.75, the predicted bar height is 85, and the deviation bar is marked as -4.25%.

[0122] The line chart shows the time trend of waste collection efficiency: the horizontal axis represents the date (last 7 days), and the vertical axis represents the waste collection efficiency (%). The line shows the daily efficiency changes, and the "planned efficiency line" (100%) is also marked. For example, the efficiency on May 22 was 102.01%, which is higher than the planned line, while the efficiency on May 20 was 98%, which is lower than the planned line. This allows for a visual observation of efficiency fluctuations.

[0123] Overflow warning accuracy pie chart: Divided into "Correct Warnings" and "Incorrect Warnings," with the proportions corresponding to accuracy and error rates. For example, 80% correct warnings (green) and 20% incorrect warnings (red). Among the incorrect warnings, "predicted overflow but not actually overflowed" accounts for 60%, and "predicted not overflow but actually overflowed" accounts for 40%.

[0124] 3. Textual Analysis and Report Structure: The report is structured into four parts: "Report Summary, Performance Indicator Details, Problem Analysis and Improvement Suggestions, and Appendix (Raw Data)". Textual analysis should be combined with data and real-world scenarios to avoid vague generalities. Report Summary: Briefly summarize the period (e.g., May 22, 2024), coverage area (5 areas of Xingfu Community + 2 areas of Commercial Plaza), and core conclusions (overall compliance rate 82.3%, lower than the predicted 85%; waste removal efficiency 101.5%, higher than planned; overflow warning accuracy rate 78%, prediction model needs optimization).

[0125] Performance Indicator Details: Each indicator is explained in separate chapters, with corresponding charts. For example, in the chapter "3.1 Waste Sorting Compliance Rate", a regional comparison bar chart is first shown, followed by an analysis of the specific reasons for "the largest deviation in Buildings 1-3" (equipment failure of Bin B and a surge in waste disposal during the morning peak), and actual data is cited (the proportion of incorrect waste disposal during the period when Bin B was malfunctioning was 35%).

[0126] Problem Analysis and Improvement Suggestions: Implementationable suggestions are proposed for the deviations, such as "For the overflow warning deviation of D bucket, it is recommended to add a 'weekend delivery volume correction coefficient' (weekend coefficient 1.2, weekday coefficient 1.0) to the prediction model; for the equipment failure of B bucket, it is recommended to conduct IoT device network connectivity tests every Friday."

[0127] 4. Generation of complete report document: The final report will be generated in both Word (for easy editing) and PDF (for easy archiving and sharing). Each document must include the community logo, the report's creation date, and the department that prepared it (e.g., the community property and environmental management department). Example report excerpt: "Waste Sorting Management Report of Xingfu Community, May 22, 2024" Compiled by: Community Property and Environmental Management Department; Reporting period: 2024-05-22 00:00-23:59.

[0128] I. Report Summary: This period covered 7 areas of the community, with a total of 32 smart trash cans and 2 collection vehicles. The overall waste sorting compliance rate was 82.3% (predicted 85%), the collection efficiency was 101.5% (planned 100%), and the overflow warning accuracy rate was 78% (target 90%). The core issues were equipment malfunctions in buildings 1-3 and deviations in the prediction model for bin D. Equipment maintenance and model optimization need to be completed next week.

[0129] II. Performance Indicator Details: 2.1 Regional classification compliance rate: For example, the compliance rate for Buildings 1-3 was 80.75% (lowest), while the compliance rate for the Commercial Plaza area was 88% (highest). The deviation in Buildings 1-3 was mainly due to the failure to play the introductory animation for Bucket B (XF-SQ-2-D-KHS) between 6:30 and 7:00. During this period, 0.8 kg of incorrectly delivered waste was disposed of, accounting for 42% of the total number of errors that day…

[0130] By uploading management reports to the community waste sorting management platform through the Internet of Things (IoT) data interface, the platform's visual display content is updated, enabling visualized management of the entire waste sorting process.

[0131] 1. IoT Data Interface Design and Upload Process The system uses a RESTful API interface (compliant with IoT device communication specifications, supporting HTTP / HTTPS protocols). Interface parameters include "Report ID (unique identifier, such as REP-20240522-XF), Report Name, File Format (Word / PDF), File Size (bytes), Upload Time, and Uploader Account." The upload process is as follows: Community environmental management staff log into the report generation system and select the report document to be uploaded; The system calls the API interface to send an "upload request" (including parameters and file stream), and the platform performs identity authentication (based on a JWT token, valid for 2 hours). The platform receives the file, stores it on the cloud server (path such as " / reports / 20240522 / XF-SQ-20240522.pdf"), and returns a "successful upload" confirmation. The system updates the local report status to "Uploaded" and records the platform storage path.

[0132] 2. Updated visualizations on the community waste sorting management platform: The platform adopts a "dashboard + details page" architecture, developed based on Vue.js + Element UI, and the updates include: Core Indicator Dashboard: Real-time display of "Overall Community Compliance Rate, Cleanup Efficiency, and Early Warning Accuracy" using digital cards and progress bars. For example, the card with "Overall Compliance Rate 82.3%" is filled with 82.3% (target line 90%), and the color changes with the indicator level (≥90% green, 80%-90% yellow, <80% red).

[0133] Area Details Page: Clicking on the area name in the dashboard will take you to the details page, which displays the area's "indicator trend chart, equipment status list, and report download button." For example, the details page for Buildings 1-3 displays a line chart of the compliance rate over the past 7 days, lists the "current online status, today's sorting accuracy rate, and whether there is a malfunction" of all trash cans in the area, and provides a download link for the daily report PDF.

[0134] The end-to-end tracking module displays the status of nodes according to the process of "data collection → prediction → guidance / collection → assessment → report". Each node is labeled with "completion time, responsible person, and key data". For example, the "data collection" node displays "completion time 05-22 08:00, responsible person: automatic data collection by the system, key data: data uploaded from 32 buckets"; the "report generation" node displays "completion time 05-22 09:30, responsible person: Zhang XX, key data: 7 regional reports generated".

[0135] 3. Full-process visual management implementation: The platform supports "real-time monitoring + historical traceability," allowing administrators to: Real-time monitoring of the status of each trash can (online / offline), the playback of sorting guidance animations, and the real-time location of the trash truck (GPS / BeiDou positioning); Tracing the forecast results, collection routes, and performance reports for a specific period (such as May 22), and clicking on any node to view the raw data (such as the disposal record of a certain bin or the driving route of a garbage truck). Export historical reports and data for monthly / quarterly summary analysis. For example, when generating the May monthly report, daily performance data can be exported from the platform and automatically summarized into a monthly trend chart, achieving full-process visual management "from daily reports to periodic summaries".

[0136] Another embodiment of the present invention provides an intelligent waste sorting management system based on the Internet of Things, see [link to relevant documentation]. Figure 3 The system may include: The acquisition module 301 is used to acquire garbage information data of garbage disposal through the IoT device built into the smart trash can, synchronously record the user's disposal timestamp, combine it with the real-time traffic data acquired by the IoT device around the trash can, and use the behavior-garbage association algorithm to label the garbage categories that users are likely to mix in different time periods, and generate a classified garbage dataset with user disposal behavior tags. The prediction module 302 is used to input the classified waste dataset with user disposal behavior tags into the dynamic prediction model. By analyzing the correlation between the disposal ratio of various types of waste and time, it predicts the growth curve of classified waste in each waste bin and the high incidence of mixed disposal behavior within a preset time period in the future, and generates prediction results including peak waste disposal time, easily mixed disposal categories and overflow warning time of each bin. The planning module 303 is used to activate the classification guidance-collection linkage mechanism based on the prediction results. Before the peak period of easy mixed disposal and the peak disposal period, the classification guidance animation of the corresponding easy mixed disposal category is pushed through the display screen on the top of the garbage bin. At the same time, combined with the garbage growth curve of each bin and real-time traffic data, the classification and collection routes of garbage trucks are planned, and a garbage collection and dispatch plan with classification guidance content is generated. The integration module 304 is used to collect data on the actual classification accuracy of user waste disposal and data on the collection tasks completed by garbage trucks according to the scheduling plan. It integrates the collected data with the prediction results and the collection scheduling plan to generate a garbage classification management report that includes the garbage classification compliance rate, collection efficiency and overflow warning accuracy of each area. The report is uploaded to the community garbage classification management platform at the same time to realize the visualized management of the entire garbage classification process.

[0137] This invention also provides a storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above method embodiments when running.

[0138] This invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0139] Specifically, the aforementioned electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the aforementioned processor, and the input / output device is connected to the aforementioned processor.

[0140] The above description, based on the embodiments shown in the figures, details the structure, features, and effects of the present invention. The above description is only a preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the figures. Any changes made in accordance with the concept of the present invention, or equivalent embodiments modified to have equivalent changes, that do not exceed the spirit covered by the specification and figures, should be within the protection scope of the present invention.

Claims

1. An intelligent waste sorting and management method based on the Internet of Things, characterized in that, The method includes: By acquiring waste information data from the IoT devices built into the smart trash cans and recording the timestamps of user disposal, and combining it with real-time pedestrian traffic data acquired from IoT devices around the trash cans, a behavior-waste association algorithm is used to label the types of waste that users easily mix and dispose of at different times, generating a classified waste dataset with user disposal behavior tags. Input the categorized waste dataset with user disposal behavior tags into the dynamic prediction model. By analyzing the correlation between the disposal ratio of various types of waste and time, predict the growth curve of categorized waste in each waste bin and the peak period of easy mixed disposal within a preset time period. Generate prediction results including peak waste disposal time, easily mixed disposal categories and overflow warning time of each bin. Based on the prediction results, the classification guidance-collection linkage mechanism is activated. During the peak periods of mixed disposal and before the peak disposal period, classification guidance animations for the corresponding mixed disposal categories are pushed through the display screen on the top of the garbage bins. At the same time, the classification and collection routes of garbage trucks are planned by combining the garbage growth curve of each bin with real-time traffic data, and a garbage collection and dispatch plan with classification guidance content is generated. The system collects data on the accuracy of actual waste sorting by users and data on the collection tasks completed by garbage trucks according to the scheduling plan. It integrates the collected data with the prediction results and the collection scheduling plan to generate a waste sorting management report that includes the waste sorting compliance rate, collection efficiency and overflow warning accuracy of each area. The report is then uploaded to the community waste sorting management platform to achieve visualized management of the entire waste sorting process.

2. The method according to claim 1, characterized in that, The process involves acquiring waste information data through the IoT devices built into the smart trash cans, synchronously recording the user's disposal timestamp, and combining this with real-time pedestrian traffic data acquired from IoT devices around the trash cans. A behavior-waste association algorithm is then used to label the types of waste that users easily mix and match at different times, generating a categorized waste dataset with user disposal behavior tags, including: The smart trash can uses its built-in IoT device to collect raw trash information data in real time, including at least the weight of the trash and image features. Add a timestamp to each piece of raw garbage information data and synchronize it with the real-time pedestrian flow data collected by IoT devices around the garbage bins to generate a time-synchronized multi-source dataset. Density-based clustering algorithm was used to analyze the correlation between waste disposal characteristics and pedestrian traffic in different time periods, to explore user behavior patterns of mixing waste, and to generate a set of behavior-waste association rules. Based on the behavior-garbage association rule set, the time-synchronized multi-source dataset is labeled, and user disposal behavior tags are added to each garbage disposal record, ultimately generating a categorized garbage dataset with user disposal behavior tags.

3. The method according to claim 2, characterized in that, The process involves inputting a dataset of categorized waste labeled with user disposal behavior tags into a dynamic prediction model. By analyzing the correlation between the disposal ratio of various types of waste and time, the model predicts the growth curve of categorized waste in each bin and the peak periods of mixed disposal within a preset timeframe. The model generates prediction results including peak disposal times, easily mixed categories, and overflow warning times for each bin, including: Feature engineering is performed on the categorized waste dataset with user disposal behavior labels to extract the time-series features of disposal volume and the distribution features of behavior labels for each type of waste, and to generate feature vectors for model input. The feature vector is input into an LSTM dynamic prediction model based on an attention mechanism. This model learns the temporal pattern of waste disposal by analyzing the nonlinear correlation between the proportion of various types of waste disposal and time in historical data. By using a trained dynamic prediction model to perform multi-step forward prediction, the growth curve of classified waste in each trash can and the probability distribution of easy-to-mix waste disposal behavior are generated within a preset time period in the future. Based on the waste growth curve, an overflow warning threshold is set, and the high-incidence period is determined by combining the probability distribution of easy-to-mix waste disposal behavior. Finally, a prediction result is generated that includes the peak waste disposal period, easy-to-mix waste categories, and the overflow warning time of each bin.

4. The method according to claim 3, characterized in that, The mechanism for activating the classification guidance-collection linkage based on prediction results involves pushing classification guidance animations for the corresponding easily mixed categories via the display screen on top of the garbage bins before peak periods of high incidence of mixed disposal. Simultaneously, it combines the garbage growth curves of each bin with real-time traffic data to plan the classified collection routes for garbage trucks, generating a garbage collection and dispatch plan with classification guidance content, including: Analyze the prediction results to identify the peak periods of mixed-use behavior and peak delivery times, and trigger the classification guidance mechanism in advance to generate classification guidance activation instructions; Based on the classification guidance start command and the information on easily mixed categories in the prediction results, the corresponding classification guidance animation content is retrieved from the multimedia library and sent to the display screen on top of each trash can through the Internet of Things network to generate a classification guidance playback task. Real-time data on waste growth curves for each trash can and traffic conditions in the area related to the trash can are acquired. A path optimization algorithm is used to calculate the optimal collection route and generate a preliminary collection route plan. By integrating the classification guidance playback task and the preliminary collection route plan, and taking into account the capacity of collection vehicles and the constraints of operation time, multi-objective optimization scheduling is carried out to finally generate a waste collection scheduling plan with classification guidance content.

5. The method according to claim 4, characterized in that, The system collects data on the actual sorting accuracy of user waste disposal and data on the collection tasks completed by garbage trucks according to the scheduling plan. This data is then integrated with prediction results and the collection scheduling plan to generate a garbage sorting management report. This report includes data on the garbage sorting compliance rate, collection efficiency, and overflow warning accuracy for each area. The report is simultaneously uploaded to the community garbage sorting management platform, enabling visualized management of the entire garbage sorting process, including: The IoT devices in smart trash cans collect data on the actual sorting accuracy of users' trash disposal, and at the same time, they obtain data on the completion of collection tasks from the garbage truck's onboard terminal to generate an actual execution dataset. By comparing and analyzing the actual execution dataset with the prediction results and the waste collection and dispatch plan, key performance indicators such as the waste classification compliance rate, waste collection efficiency and overflow warning accuracy rate of each region are calculated, and performance evaluation results are generated. Based on the performance evaluation results, an automatic report generation technology is used to construct a waste sorting management report that includes data visualization charts and text analysis, generating a complete management report document; By uploading management reports to the community waste sorting management platform through the Internet of Things (IoT) data interface, the platform's visual display content is updated, enabling visualized management of the entire waste sorting process.

6. An intelligent waste sorting and management system based on the Internet of Things, characterized in that, The system includes: The acquisition module is used to acquire garbage information data of garbage disposal through the IoT device built into the smart trash can, synchronously record the user's disposal timestamp, combine it with the real-time traffic data acquired by the IoT device around the trash can, and use the behavior-garbage association algorithm to label the garbage categories that users are likely to mix in different time periods, and generate a classified garbage dataset with user disposal behavior tags. The prediction module is used to input the classified waste dataset with user disposal behavior tags into the dynamic prediction model. By analyzing the correlation between the disposal ratio of various types of waste and time, it predicts the growth curve of classified waste in each waste bin and the high incidence of mixed disposal behavior within a preset time period in the future, and generates prediction results including peak waste disposal time, easily mixed disposal categories and overflow warning time of each bin. The planning module is used to activate the classification guidance-collection linkage mechanism based on the prediction results. Before the peak period of easy mixed disposal and the peak disposal period, the module pushes classification guidance animations for the corresponding easy mixed disposal categories through the display screen on the top of the garbage bin. At the same time, it plans the classification and collection routes of garbage trucks by combining the garbage growth curve of each bin and real-time traffic data, and generates a garbage collection and dispatch plan with classification guidance content. The integration module is used to collect data on the actual classification accuracy of user waste disposal and the data on the collection tasks completed by garbage trucks according to the scheduling plan. It integrates the collected data with the prediction results and the collection scheduling plan to generate a garbage classification management report that includes the garbage classification compliance rate, collection efficiency and overflow warning accuracy of each area. The report is uploaded to the community garbage classification management platform at the same time to realize the visualization management of the entire garbage classification process.

7. The system according to claim 6, characterized in that, The acquisition module is specifically used for: The smart trash can uses its built-in IoT device to collect raw trash information data in real time, including at least the weight of the trash and image features. Add a timestamp to each piece of raw garbage information data and synchronize it with the real-time pedestrian flow data collected by IoT devices around the garbage bins to generate a time-synchronized multi-source dataset. Density-based clustering algorithm was used to analyze the correlation between waste disposal characteristics and pedestrian traffic in different time periods, to explore user behavior patterns of mixing waste, and to generate a set of behavior-waste association rules. Based on the behavior-garbage association rule set, the time-synchronized multi-source dataset is labeled, and user disposal behavior tags are added to each garbage disposal record, ultimately generating a categorized garbage dataset with user disposal behavior tags.

8. The system according to claim 7, characterized in that, The prediction module is specifically used for: Feature engineering is performed on the categorized waste dataset with user disposal behavior labels to extract the time-series features of disposal volume and the distribution features of behavior labels for each type of waste, and to generate feature vectors for model input. The feature vector is input into an LSTM dynamic prediction model based on an attention mechanism. This model learns the temporal pattern of waste disposal by analyzing the nonlinear correlation between the proportion of various types of waste disposal and time in historical data. By using a trained dynamic prediction model to perform multi-step forward prediction, the growth curve of classified waste in each trash can and the probability distribution of easy-to-mix waste disposal behavior are generated within a preset time period in the future. Based on the waste growth curve, an overflow warning threshold is set, and the high-incidence period is determined by combining the probability distribution of easy-to-mix waste disposal behavior. Finally, a prediction result is generated that includes the peak waste disposal period, easy-to-mix waste categories, and the overflow warning time of each bin.

9. A storage medium, characterized in that, The storage medium stores a computer program, wherein the computer program is configured to execute the method of any one of claims 1-5 when it is run.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to run the computer program to perform the method of any one of claims 1-5.

Images