Smart sanitation waste recycling method and system

Abstract

The present invention provides a smart sanitation waste recycling method and system, which relates to the field of environmental protection technology. The method includes: step 1, parsing the container's intrinsic identification code through the penetrating radio frequency unit of the intelligent classification terminal, and obtaining the garbage quality data stream to generate an original spatiotemporal observation set; step 2, inputting the original spatiotemporal observation set into the edge computing node, converting it into a triple structure according to the preset sanitation ontology rules, and outputting a normalized sanitation data body; step 3, performing spatiotemporal alignment between the normalized sanitation data body and two types of key facility status parameters: the real-time capacity saturation rate of the regional transfer station and the historical full load rate attenuation factor of the mobile recycling vehicle, to form a sanitation state tensor with physical constraints. Through intelligent data processing and analysis throughout the entire process, the present invention realizes the collection, standardized integration, accurate prediction and optimized scheduling of garbage recycling data, effectively improving the operational efficiency of the smart sanitation system and reducing transportation costs and the risk of garbage retention.

Description

Technical Field

[0001] The present invention relates to the field of environmental protection technology, and in particular to a smart sanitation waste recycling method and system. Background Art

[0002] With the acceleration of urbanization and the continued growth of urban waste generation, the need for smart sanitation systems to optimize waste recycling processes and improve resource utilization efficiency is becoming increasingly prominent. Currently, some cities have attempted to deploy smart waste sorting equipment (such as classified waste bins with weighing and identification functions) and data management platforms to collect and analyze waste data to support recycling route optimization and resource allocation. However, existing technologies still have certain limitations in data processing:

[0003] Traditional data collection relies on manual assistance or simple sensing equipment, which is easily affected by environmental interference, resulting in insufficient data accuracy. In addition, the collection frequency is low, making it difficult to reflect garbage dynamics in real time. At the same time, data from each link are mostly stored in isolation, with inconsistent formats, making integration and sharing difficult, affecting the efficiency of comprehensive data analysis. In addition, the data analysis model has limited ability to integrate spatiotemporal characteristics and multi-source influencing factors, making it difficult to accurately predict changes in garbage load, resulting in insufficient scientificity and foresight in resource scheduling, and making it difficult to fully meet the needs of efficient operation of the smart sanitation system. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an intelligent sanitation waste recycling method and system to achieve real-time and stable acquisition of waste container identification and quality data.

[0005] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0006] In a first aspect, a smart sanitation waste recycling method is provided, the method comprising:

[0007] Step 1: The penetrating radio frequency unit of the intelligent classification terminal parses the container's intrinsic identification code and obtains the garbage quality data stream to generate the original spatiotemporal observation set;

[0008] Step 2: Input the original spatiotemporal observation set into the edge computing node, convert it into a triple structure according to the preset sanitation ontology rules, and output the normalized sanitation data body;

[0009] Step 3: Perform spatiotemporal registration between the normalized sanitation data volume and two key facility status parameters: the real-time capacity saturation rate of regional transfer stations and the attenuation factor of the historical full load rate of mobile recycling vehicles, to form a sanitation status tensor with physical constraints.

[0010] Step 4: Input the sanitation status tensor into the gated loop architecture, combine it with the regional population flow characteristics and the commercial activity intensity time series variables, and generate the garbage load probability distribution matrix;

[0011] Step 5: Solve the saddle point equilibrium of the transportation path cost functional and the detention risk functional based on the garbage load probability distribution matrix, and output resource scheduling instructions to the transportation vehicle terminal.

[0012] Secondly, the smart sanitation waste recycling system includes:

[0013] The parsing module is used to parse the container's intrinsic identification code through the penetrating radio frequency unit of the intelligent classification terminal, obtain the garbage quality data stream and generate the original spatiotemporal observation set;

[0014] The conversion module is used to input the original spatiotemporal observation set into the edge computing node, convert it into a triple structure according to the preset sanitation ontology rules, and output the normalized sanitation data body;

[0015] The tensor construction module is used to perform spatiotemporal registration of the normalized sanitation data volume with two key facility status parameters: the real-time capacity saturation rate of regional transfer stations and the attenuation factor of the historical full load rate of mobile recycling vehicles, to form a sanitation status tensor with physical constraints;

[0016] The load probability module is used to input the sanitation status tensor into the gated loop architecture, combine it with the regional population flow characteristics and commercial activity intensity time series variables, and generate the garbage load probability distribution matrix;

[0017] The scheduling instruction module is used to solve the saddle point equilibrium of the transportation path cost functional and the detention risk functional based on the garbage load probability distribution matrix, and output resource scheduling instructions to the transportation vehicle terminal.

[0018] According to a third aspect, a computer-readable storage medium stores a program, which implements the method described above when executed by a processor.

[0019] The above solution of the present invention includes at least the following beneficial effects:

[0020] Through penetrating radio frequency units and a dynamic gravity sensor array, real-time and stable acquisition of garbage container identification and quality data is achieved, reducing errors caused by environmental interference and manual operation. Edge computing nodes and triple structure conversion are used to unify data formats and effectively correlate multi-source information, breaking down data silos. A physically constrained state tensor is formed through spatiotemporal alignment, and a gated loop architecture is used to integrate variables such as population mobility and commercial activities to accurately generate a probability distribution for garbage loads. By balancing cost and risk based on the probability distribution matrix, the resulting dispatch instructions can reduce redundancy in removal routes, mitigate the risk of garbage stagnation, and improve the overall operational performance of the sanitation system. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1It is a flow chart of the smart sanitation waste recycling method provided by an embodiment of the present invention.

[0022] Figure 2 This is a schematic diagram of the smart sanitation waste recycling system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0023] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0024] like Figure 1 As shown, an embodiment of the present invention provides a smart sanitation waste recycling method, which includes the following steps:

[0025] Step 1: The penetrating radio frequency unit of the intelligent classification terminal parses the container's intrinsic identification code and obtains the garbage quality data stream to generate the original spatiotemporal observation set;

[0026] Step 2: Input the original spatiotemporal observation set into the edge computing node, convert it into a triple structure according to the preset sanitation ontology rules, and output the normalized sanitation data body;

[0027] Step 3: Perform spatiotemporal registration between the normalized sanitation data volume and two key facility status parameters: the real-time capacity saturation rate of regional transfer stations and the attenuation factor of the historical full load rate of mobile recycling vehicles, to form a sanitation status tensor with physical constraints.

[0028] Step 4: Input the sanitation status tensor into the gated loop architecture, combine it with the regional population flow characteristics and the commercial activity intensity time series variables, and generate the garbage load probability distribution matrix;

[0029] Step 5: Solve the saddle point equilibrium of the transportation path cost functional and the detention risk functional based on the garbage load probability distribution matrix, and output resource scheduling instructions to the transportation vehicle terminal.

[0030] In the embodiment of the present invention, by analyzing the identification code through penetrating radio frequency and obtaining the quality data stream in real time, the efficiency and accuracy of garbage data collection can be improved, and the errors caused by manual operation and environmental interference can be reduced.

[0031] Converting raw data into a normalized triple structure enables unified and standardized data formatting, enhancing data integration and sharing, and laying the foundation for data correlation analysis across all links. Forming a physically constrained sanitation status tensor through spatiotemporal registration can integrate multiple types of facility status parameters, enhancing the comprehensive application value of the data. Generating a garbage load probability distribution matrix based on variables such as population mobility and commercial activity can improve the ability to predict garbage generation trends. Solving the cost-risk trade-off and outputting scheduling instructions can optimize removal routes and resource allocation, reducing operating costs while minimizing the risk of garbage retention and improving the overall operational efficiency of the sanitation system.

[0032] In a preferred embodiment of the present invention, the above step 1, parsing the container intrinsic identification code by the penetrating radio frequency unit of the intelligent classification terminal and obtaining the garbage quality data stream to generate the original spatiotemporal observation set, may include:

[0033] Step 100: non-contact reading of the passive RFID tag embedded in the garbage container by a penetrating radio frequency unit to resolve the container's unique intrinsic identification code;

[0034] Step 101: Activate a dynamic gravity sensor array based on the container's intrinsic identification code to collect real-time data streams of continuously changing garbage mass associated with the container, and trigger a positioning module to obtain the container's real-time geographic coordinates and timestamp.

[0035] Step 102: encapsulate the container intrinsic identification code, the bound garbage quality data stream, the geographic coordinates and the timestamp into a structured original spatiotemporal observation set, wherein the original spatiotemporal observation set includes a container ID field, a quality value field, a longitude field, a latitude field and a timestamp field.

[0036] In an embodiment of the present invention, the penetrating radio frequency unit of the intelligent sorting terminal first activates the radio frequency transmission module to transmit ultra-high frequency (UHF) radio waves with a frequency of 860-960 MHz. This frequency band can penetrate stains, dust or plastic / metal obstructions with a thickness of no more than 5 mm on the surface of the garbage container, ensuring that the signal coverage range is stable within 0.5-1 meter around the container.

[0037] When a passive RFID tag enters the signal coverage area, the tag's built-in antenna receives radio wave energy and converts the RF energy into DC power (approximately 3-5V) through a rectifier circuit, activating the tag's internal microchip. The chip then retrieves the stored coded information (encoded using the EPC Global Class 1 Gen 2 standard). This code contains 128 bits of data: the first 8 bits are the Header (identifying the encoding version), the next 28 bits are the EPC Manager (identifying the tag issuing organization), the next 24 bits are the Object Class (identifying the container type, such as "food waste container" or "recyclables container"), and the final 68 bits are the Serial Number (the container's unique serial number, such as the production batch and serial number).

[0038] The tag embeds coded information into the reflected radio waves using backscatter modulation (i.e., by varying the antenna impedance so that the amplitude of the reflected signal varies with the coded information), which is then fed back to the radio frequency unit (RFU). The RF unit's receiving module first uses a bandpass filter to remove electromagnetic interference outside the 860-960 MHz band (such as clutter from nearby electronic devices). It then uses amplitude-shift keying (ASK) demodulation to strip away the high-frequency carrier signal and extract the baseband digital signal containing the coded information (a binary sequence of 0s and 1s).

[0039] Decode the binary sequence: First, check the Header field to confirm whether the encoding version matches (if it does not match, trigger a re-read), then parse the EPCManager, ObjectClass, and SerialNumber fields in sequence, and combine these fields in the format of "organization code-container type-unique serial number" (such as "001-food waste-20230508001") to generate a unique container intrinsic identification code and store it in a temporary buffer.

[0040] Step 101: The container's intrinsic identification code, generated in step 100, is retrieved from a temporary buffer and accurately matched against a pre-set "device association list" (stored in the format of "container identification code - sensor array ID - communication port"). For example, if the identification code is "001-food waste-20230508001," the dynamic gravity sensor array with the ID "GSA-20230508001" is matched, and its corresponding communication port (e.g., COM3) is determined. An activation command (in the format of "activation + sensor ID + timestamp") is sent to the sensor array via the communication port, switching the array from a dormant state (power consumption <1mA) to an active state (power consumption 5-10mA). The array consists of four resistive strain gauge gravity sensors, mounted at the four corners of the container's bottom (5cm from the edge). Each element has a measurement range of 0-15kg and an accuracy of ±0.01kg. When garbage is placed in the container, its weight is transferred to the sensing element through the container shell. The strain gauge in the element deforms due to the force, causing a change in resistance (proportional to the weight). This change in resistance is converted into a weak voltage signal ranging from 0 to 10mV via a Wheatstone bridge circuit. The voltage signal is first amplified to 0-10V by an instrumentation amplifier (with an adjustable gain of 1000x) before being input into a 16-bit analog-to-digital converter (ADC), which converts the analog signal into a digital value ranging from 0 to 65535. The system then uses a preset calibration curve (generated through previous experiments: standard weights of 0kg, 5kg, 10kg, and 15kg were placed in the container, the corresponding digital values ​​were recorded, and a linear equation (e.g., "weight = digital value × 0.000238 - 0.005") is fitted to the four components' digital values, converting them to actual weights. The arithmetic mean of these four values ​​(to eliminate errors caused by uneven container placement) is then calculated to determine the current garbage weight. The above weight data are arranged in sequence according to the collection time (once every 0.5 seconds) to form a continuous garbage mass data stream (such as "1.2kg (10:00:00.000) → 1.5kg (10:00:00.500) → 1.5kg (10:00:01.000)...").

[0041] At the same time, after the container identification code is successfully matched, a startup signal is sent to the positioning module (e.g., Beidou BD-210). Upon startup, the positioning module first searches for and locks onto at least four Beidou satellites (if the satellite signal is weak, it supplements with GPS signals). It then calculates the distance between the module and each satellite by measuring the satellite signal propagation time (pseudorange). Combined with the precise satellite positions (obtained from the satellite ephemeris), it uses trilateration to calculate the module's three-dimensional coordinates (longitude, latitude, and altitude). The longitude and latitude in these three-dimensional coordinates are converted to degrees, minutes, and seconds (DMS) format in the WGS84 coordinate system: longitude values ​​range from -180° to 180°, and latitude values ​​range from -90° to 90°, with accuracy to six decimal places (approximately 0.1 meter). For example, 116.391° east longitude is converted to "116°23'27.6"; 39.904° north latitude is converted to "39°54'14.4"). The positioning module refreshes its coordinates every 30 seconds. If the coordinate deviation exceeds 0.5 meters for two consecutive times, differential positioning correction is initiated (calling the differential data of nearby base stations) to ensure position accuracy.

[0042] In addition, the system's built-in real-time clock (synchronized with Beidou satellite time, with an error of <1ms) records the time of each weight data collection and coordinate refresh, generates a timestamp (in the format of "YYYY-MM-DDHH:MM:SS.XXX", such as "2025-07-09 10:00:00.123"), and stores the timestamp in conjunction with the corresponding weight data and longitude and latitude data (such as "weight 1.2kg-longitude 116°23′27.6″-latitude 39°54′14.4″-time 2025-07-09 10:00:00.123").

[0043] Step 102: Create a temporary structured data framework for the container (using a two-dimensional table structure, similar to a "data table" in a relational database). The table has five preset fields, each with a clear format constraint:

[0044] "Container ID field": only accepts string type, with a fixed length of 20 characters (spaces are added if the length is less than 20 characters, and an error is reported if the length exceeds 20 characters). It stores the intrinsic identification code generated in step 100 (for example, "001-food waste-20230508001").

[0045] "Mass value field": accepts numeric values, retains 2 decimal places, and has a range of 0-50 kg (if exceeded, it will be marked as an "outlier"), and stores the weight data converted in step 101;

[0046] "Longitude field": accepts string type, format is "DD°MM′SS.SSSSSS″" (such as "116°23′27.600000″), range is -180° to 180°;

[0047] "Latitude field": the format is the same as longitude, ranging from -90° to 90°;

[0048] Timestamp field: Accepts a string type in the format of "YYYY-MM-DDHH:MM:SS.XXX". The time must be within ±10 minutes of the current system time. If it exceeds this limit, it will be considered a time synchronization error.

[0049] When filling in data, write the bound data in step 101 into the framework one by one in timestamp order: the "Container ID Field" of the first record is written with the identification code, the "Mass Value Field" is written with the first weight data (such as 1.2kg), the "Longitude / Latitude Field" is written with the corresponding coordinates, and the "Timestamp Field" is written with the corresponding time; the second record reuses the content of the "Container ID Field", and the remaining fields are written with the next set of data, and so on.

[0050] After filling is completed, the system performs multi-dimensional verification:

[0051] Format check: Check whether each field conforms to the preset format (e.g. whether the quality value contains non-numeric characters, whether the latitude and longitude contain the "°′" symbol);

[0052] Logical verification: Compare the timestamps of two adjacent records to ensure the interval is 0.5 seconds (weight collection interval) or 30 seconds (coordinate refresh interval). Any deviation exceeding 0.1 seconds is marked as a "time jump"; check whether the weight value increases reasonably with the amount of garbage input (if a negative number appears or a sudden decrease of more than 1kg occurs, and there is no garbage removal record, it is marked as "weight abnormal");

[0053] Range verification: Confirm that the longitude and latitude are within the city's administrative area (e.g., Beijing's area is approximately 115.7°-117.4° east longitude and 39.4°-41.6° north latitude). If they are outside the range, it will be considered an "abnormal location."

[0054] After verification, the system encapsulates the data in JSON format: "Container ID" is the root node, followed by "Observation Data List" sub-nodes. Each element in the list is a key-value pair for a record (for example, {"Container ID":"001-Kitchen Waste-20230508001", "Observation Data List":[{"Mass Value":"1.20","Longitude":"116°23′27.600000","Latitude":"39°54′14.400000","Timestamp":"2025-07-09 10:00:00.123"},...]}). During encapsulation, an "Anomaly Type" field is added to the anomaly-marked data (for example, {"Mass Value":"-0.50","Anomaly Type":"Weight Anomaly"...}). The resulting JSON data packet is the original spatiotemporal observation set, which is stored in the local cache and awaits upload to the edge computing node.

[0055] The combination of penetrating radio frequency technology and passive RFID tags enables contactless container identification reading, eliminating the tedious manual scanning process. It also overcomes the impact of environmental occlusion on identification and improves the stability and efficiency of parsing the container's intrinsic identification code. A dynamic gravity sensor array and linked positioning and timing modules enable real-time, continuous collection of garbage mass and simultaneous recording of spatiotemporal information.

[0056] The original spatiotemporal observation set is formed through structured encapsulation, and multi-dimensional data is integrated in an orderly manner to ensure the correlation and consistency between the data.

[0057] In a preferred embodiment of the present invention, the above step 2, inputting the original spatiotemporal observation set into the edge computing node, converting it into a triple structure according to the preset sanitation ontology rules, and outputting a normalized sanitation data body, may include:

[0058] Step 200: The original spatiotemporal observation set is transmitted to the edge computing node, and the following mapping operations are performed using the preset sanitation ontology rule base:

[0059] The container ID field is mapped to the entity identifier;

[0060] The quality numeric field is mapped to the quality attribute value;

[0061] The combination of longitude and latitude fields is mapped to spatial attribute values;

[0062] The timestamp field is mapped to the time attribute value;

[0063] Step 201 : Generate triple structure data based on the mapping result, including spatial relationship triples, quality status triples and time status triples, and output a normalized sanitation data body of triples.

[0064] In this embodiment of the present invention, the original spatiotemporal observation set (encapsulated in JSON format) is transmitted through the local communication interface of the edge computing node (such as LoRaWAN or Ethernet). Before transmission, it needs to undergo three layers of preprocessing:

[0065] Integrity check: After receiving data, the edge computing node first checks whether the JSON structure is complete (for example, whether it contains the five required key-value pairs such as the "container ID field" and the "quality value field"). If missing, it sends a retransmission instruction to the intelligent classification terminal (instruction format: "retransmit + missing field + timestamp");

[0066] Redundancy elimination: Delete records marked as "outliers" in the JSON (such as the "weight abnormality" and "location abnormality" data marked in step 102), and only retain valid records that have passed verification;

[0067] Protocol conversion: Convert the JSON format to the Protobuf format compatible with edge computing nodes (a lightweight binary format that reduces data volume).

[0068] The "Sanitation Ontology Rule Base" built into the edge computing node is a structured knowledge collection that contains four types of core rules (stored in XML format) that define field mapping standards:

[0069] Entity identifier rules: Specifies the conversion format from "container ID field → entity identifier", such as "entity identifier = 'entity' + container ID (special characters removed)". For example, if the container ID is "001-kitchen waste-20230508001", it will be converted to "entity001kitchen waste20230508001". It also includes entity type verification rules (for example, "the entity identifier of the kitchen waste container must contain the 'kitchen' prefix". If it does not match, it will be marked as "entity type exception").

[0070] Quality attribute value rules: These specify the standards for "quality value field → quality attribute value", including: unit uniformity (retaining two decimal places, with the unit fixed at "kg"), range verification (e.g., the single mass increment of a food waste container must not exceed 5kg; any excess will be marked as a "quality jump abnormality"), and anomaly correction (if the mass value remains unchanged for three consecutive records, add a "static quality" label).

[0071] Spatial attribute value rules: Specifies the integration standards of "longitude + latitude → spatial attribute value", including: coordinate format conversion (converting the "degrees, minutes, seconds" format to "decimal degrees", such as converting "116°23′27.600000″" to "116.391000", retaining 6 decimal places), regional association (using the preset "latitude and longitude-region code" comparison table to map coordinates to corresponding community / street codes, such as "116.391000,39.904000" corresponds to "Xicheng District-Financial Street Street-001"), and spatial accuracy verification (if the deviation between two coordinates exceeds 10 meters, it will be marked as "position drift anomaly").

[0072] Time attribute value rules: Specifies the unified standard for "timestamp field → time attribute value", including: time zone conversion (converting local time to UTC time, such as converting Beijing time "2025-07-09 10:00:00.123" to UTC time "2025-07-09 02:00:00.123"), format standardization (converting to the ISO8601 format "YYYY-MM-DDTHH:MM:SS.XXXZ", such as "2025-07-09T02:00:00.123Z"), and time continuity verification (if the time interval between adjacent records exceeds 1 minute, it is marked as "time interval anomaly").

[0073] The edge computing node calls the mapping engine in the rule library and executes mapping on each piece of preprocessed original data one by one:

[0074] Container ID field → Entity identifier: Extract the value of the "Container ID field" in the record (such as "001-kitchen waste-20230508001"), remove the special character "-" according to the "Entity identifier rule", and add the prefix "entitykitchen" (because the ID contains the "kitchen waste" identifier), generating "entitykitchen001 kitchen waste 20230508001"; at the same time, verify whether this identifier is in the "Valid entity list" in the rule library, and if it does not exist, mark it as "unknown entity".

[0075] Quality numerical field → Quality attribute value: Extract the value of the "Quality numerical field" (such as "1.20 kg"), confirm that the unit is "kg" according to the "Quality attribute value rule", and retain two decimal places; if the value is "1.2", it is supplemented to "1.20"; if the value is "15.30" (exceeding the rule of the single increment of 5 kg for the kitchen waste container), add the mark "[Increment anomaly]" after the attribute value, and finally generate "1.20 kg" or "15.30 kg[Increment anomaly]".

[0076] Longitude + Latitude → Spatial attribute value: Extract the "Longitude field" (such as "116°23′27.600000″") and the "Latitude field" (such as "39°54′14.400000″"), and convert them to decimal degrees "116.391000, 39.904000" according to the "Spatial attribute value rule"; match to "Xicheng District - Financial Street Sub-district - 001" through the "Longitude and Latitude - Region Coding" comparison table, and finally integrate it into "116.391000, 39.904000|Xicheng District - Financial Street Sub-district - 001".

[0077] Timestamp field → Time attribute value: Extract the "Timestamp field" (such as "2025-07-09 10:00:00.123"), and convert it to UTC time "2025-07-09T02:00:00.123Z" according to the "Time attribute value rule"; verify the time interval with the previous record (such as the previous one is "2025-07-09T02:00:00.123Z", and the current one is "2025-07-09T02:00:30.123Z", with an interval of 30 seconds, which conforms to the rule), and finally generate the standardized time attribute value.

[0078] Step 20, Based on the mapping results of Step 200, the edge computing node generates three types of triples in the triple format of "subject - predicate - object" (conforming to the RDF data model specification):

[0079] Spatial relation triple: with "entity identifier" as the subject, "spatial relation predicate" (preset as "locatedat" in the rule base) as the predicate, and "spatial attribute value" as the object. For example: "entitykitchen001 kitchen waste 20230508001 - locatedat - 116.391000, 39.904000|Xicheng District - Financial Street Sub - district - 001". When generating, it is necessary to verify the relevance between the subject and the object (such as "kitchen waste" in the entity identifier should match the type of garbage allowed to be placed in the area of the spatial attribute value. If only recyclables are allowed to be placed in the area, mark "spatial matching anomaly");

[0080] Quality status triple: with "entity identifier" as the subject, "quality relation predicate" (preset as "hasmass") as the predicate, and "quality attribute value" as the object. For example: "entitykitchen001 kitchen waste 20230508001 - hasmass - 1.20kg"; if the quality attribute value contains the mark "[increment anomaly]", then supplement "withanomaly" after the predicate, such as "hasmasswithanomaly";

[0081] Time status triple: with "entity identifier" as the subject, "time relation predicate" (preset as "datacollectedat") as the predicate, and "time attribute value" as the object. For example: "entitykitchen001 kitchen waste 20230508001 - datacollectedat - 2025 - 07 - 09T02:00:00.123Z". When generating, it is necessary to ensure that the time attribute value is consistent with the time order of the previous triple (time reversal is not allowed, otherwise mark "time series anomaly").

[0082] After the three types of triples are generated, the edge computing node performs integration and verification:

[0083] Consistency verification: Check whether the three types of triples with the same entity identifier are associated with the same record (such as whether the quality and time triples corresponding to the entity identifier of a certain spatial triple are the mapping results of the same time stamp. If not, trace the original data for correction);

[0084] Redundancy merging: If the spatial attribute values and entity identifiers of consecutive multiple records remain unchanged, and only the quality attribute values and time attribute values change, then retain the first spatial relation triple, and only update the quality and time triples subsequently (reduce data redundancy);

[0085] Anomaly summary: Store all triples marked as "anomaly" separately in the "anomaly data sub - module", and record the anomaly type, occurrence time, and possible reasons (such as the possible reason for "quality jump anomaly": sensor failure or a large - quantity one - time placement).

[0086] After the integration is completed, the edge computing node encapsulates the three types of triples in timestamp order into a standardized sanitation data body in JSON-LD format (including the "@context" context definition, which describes the semantics of the predicate in the triple, such as "locatedat" corresponds to "http: / / example.org / locatedat"), and sends it to the cloud data center via an encrypted transmission protocol (such as HTTPS), while retaining a local backup (retained for 72 hours).

[0087] Standardized field mapping is achieved through a pre-configured sanitation ontology rule base, unifying data formats and semantics (such as entity identifier naming conventions, coordinate and time formats), laying the foundation for cross-process data sharing. The triple structure (subject-verb-object) clearly associates the relationship between entities and attributes, making data scalable (attribute dimensions can be expanded by adding new predicates). Multiple rounds of verification (format, logic, and association verification) and anomaly marking mechanisms improve data accuracy and reduce decision bias caused by data errors. Separate storage of abnormal data facilitates problem tracing and equipment maintenance. Local processing at edge nodes reduces cloud computing pressure, and standardized encryption, transmission, and backup mechanisms for data bodies enhance data security and reliability.

[0088] In a preferred embodiment of the present invention, step 3 above performs spatiotemporal registration of the normalized sanitation data volume with two key facility status parameters: the real-time capacity saturation rate of regional transfer stations and the attenuation factor of the historical full load rate of mobile recycling vehicles. This forms a physically constrained sanitation status tensor, which may include:

[0089] Step 300: extracting spatial attribute values ​​and temporal attribute values ​​from the normalized sanitation data volume;

[0090] Step 301: Perform transfer station spatial matching based on spatial attribute values, i.e., obtain transfer station IoT sensor data in real time and calculate capacity saturation rate; perform topological overlay analysis on the container spatial attribute values ​​and transfer station geofences, and output spatial grid code;

[0091] Step 302: Perform time alignment of the collection and transportation plan based on the time attribute value. This involves analyzing the historical operation logs of the mobile recycling vehicle and fitting the decay coefficient of the full load rate with mileage. Align the container's time attribute value with the collection and transportation vehicle operation plan using a sliding time window, and output a time slice code.

[0092] Step 303: Integrate the spatial grid code, time slice code, quality attribute values ​​in the normalized sanitation data volume, capacity saturation rate, and fitted full load rate attenuation coefficient to generate a three-dimensional sanitation state tensor. The dimensions of the three-dimensional sanitation state tensor include:

[0093] X-axis: spatial grid encoding;

[0094] Y-axis: time slice encoding;

[0095] Z axis: state parameter vector.

[0096] In the embodiment of the present invention, the standardized sanitation data body is in JSON-LD format, which contains spatial relationship triples, quality status triples and time status triples. The extraction process is divided into two steps:

[0097] Spatial attribute value extraction:

[0098] Traverse the "spatial relationship triples" in the normalized sanitation data volume and extract the spatial attribute value from the "object" field of each triple (the format is "decimal coordinates | area code", such as "116.391000, 39.904000 | Xicheng District - Financial Street Subdistrict - 001");

[0099] Separate coordinates and area codes: Split them using the "|" symbol, retaining the first half of the pure coordinate information ("116.391000, 39.904000") as the core basis for subsequent spatial matching;

[0100] Coordinate format standardization: Convert coordinates to a floating-point format of "longitude, latitude" (e.g., 116.391000, 39.904000), retaining 6 decimal places, to ensure consistency with the transfer station geofence data format.

[0101] Time attribute value extraction:

[0102] Traverse the "time state triple" and extract the time attribute value from the "object" field (ISO8601 format, such as "2025-07-09T02:00:00.123Z");

[0103] Time format conversion: Convert UTC time to local time (e.g. Beijing time zone + 8 hours, converted to "2025-07-09 10:00:00.123");

[0104] Unified time granularity: retains to the minute level (such as "2025-07-09 10:00"), ignoring millisecond-level details, to facilitate matching with the time window of the removal plan.

[0105] In step 301, the edge computing node receives real-time IoT sensor data from all transfer stations in the area through the IoT gateway (using the MQTT protocol). The data includes: transfer station ID, current garbage storage capacity (unit: tons, uploaded every 5 minutes by the weighing sensor in the station), and designed maximum capacity (unit: tons, preset in the system database, such as "Transfer Station A has a maximum capacity of 50 tons").

[0106] Capacity saturation rate calculation: For each transfer station, calculate the capacity saturation rate using the formula "Capacity saturation rate = current storage capacity ÷ designed maximum capacity × 100%," rounding the result to two decimal places. For example, if transfer station A has a current storage capacity of 30 tons and a maximum capacity of 50 tons, the saturation rate = 30 ÷ 50 × 100% = 60.00%. If the current storage capacity exceeds the maximum capacity (e.g., 55 tons), the saturation rate is recorded as 100.00% and the station is marked "overloaded."

[0107] Topological overlay analysis of containers and transfer stations:

[0108] The system presets the geographic fence data of each transfer station (represented by a set of polygonal coordinates, such as the fence of transfer station A is "(116.380000, 39.890000), (116.390000, 39.890000), (116.390000, 39.900000), (116.380000, 39.900000)"), which represents the service area boundary of the transfer station. The "ray method" is used to determine whether the spatial attribute value (point coordinates) of the container is located within the geographic fence (polygon) of a transfer station: a virtual ray is drawn from the container coordinates in any direction (such as the east direction), and the number of intersections between the ray and the polygon boundary is counted - if it is an odd number, the container is within the fence (belongs to the service range of the transfer station); if it is an even number, it is outside the fence (needs to match other transfer stations). For example, the container coordinates (116.385000, 39.895000) are located within the fence of transfer station A, and it is determined to be the service object of the transfer station.

[0109] Spatial grid encoding output:

[0110] The city is divided into 100-meter x 100-meter square grids, with the southwest corner of the city as the origin (0, 0). The X-axis extends eastward, and the Y-axis extends northward. Each grid is coded as "area code + X-axis grid number + Y-axis grid number." For example, the area code for Xicheng District is "XC." A grid located at position 123 on the X-axis and position 456 on the Y-axis is coded as "XC-123-456." Based on the spatial coordinates of a container (e.g., 116.391000, 39.904000), the grid number is calculated as follows: X-axis number = (longitude - origin longitude) ÷ 0.001 (100 meters corresponds to approximately 0.001 degrees of longitude), Y-axis number = (latitude - origin latitude) ÷ 0.001. The result is rounded to the nearest integer. For example, if X=123 and Y=456, the spatial grid number for the container is "XC-123-456."

[0111] Step 302 extracts the operation logs of all mobile recycling vehicles in the past 30 days from the system database. The data includes: vehicle ID, mileage of each collection (unit: kilometers, the actual distance from the starting point to the end point is recorded by the vehicle's GPS), full load rate at departure (%, actual load capacity at departure ÷ maximum load capacity × 100%), and full load rate at arrival at the destination (%).

[0112] Attenuation factor calculation: For each vehicle, the full load rate attenuation value is calculated by segmented mileage: the mileage is divided into intervals such as 0-5 kilometers, 5-10 kilometers, ..., and above 50 kilometers. The "average attenuation rate" of each interval is calculated = (full load rate at departure - full load rate at arrival) ÷ ​​interval mileage.

[0113] Determine the global attenuation factor: Take the arithmetic mean of the average attenuation rates for all vehicles in the same mileage range and use this as the unified "full load rate attenuation factor" for the region. For example, if the average attenuation rate for all vehicles in the 5-10 km range is 3.8% / km, the attenuation factor for this range is set at 3.8% / km.

[0114] The system has preset operating time windows for mobile recycling vehicles (e.g., "6:00-8:00, 8:00-10:00, ..., 18:00-20:00 daily," with each window consisting of two-hour periods). Each window is assigned a unique time code (e.g., "20250709-06" represents 6:00-8:00 on July 9, 2025). The container's time attribute value (e.g., "2025-07-09 10:00") is compared with the time window, and the code corresponding to the window it falls into is assigned. For example, 10:00 falls within the "8:00-10:00" window and is coded "20250709-08." If the time falls outside the preset operating window (e.g., 21:00), it is matched to the first window of the next day (e.g., "20250710-06") and marked as "off-peak." Time slices are encoded in the format of "date + window number", with the window number incrementing every two hours starting at midnight (e.g., 0:00-2:00 is 01, 2:00-4:00 is 02, ..., 18:00-20:00 is 10). For example, the time slice encoding for July 9, 2025, from 8:00-10:00 is "20250709-05".

[0115] Step 303: The construction of the three-dimensional sanitation state tensor uses "spatial grid encoding (X-axis)", "time slice encoding (Y-axis)", and "state parameter vector (Z-axis)" as core dimensions. The specific integration steps are as follows:

[0116] Collect all spatial grid codes output in step 301 (after removing duplicates, form an X-axis index, such as "XC-123-456" and "XC-123-457");

[0117] Collect all time slice codes output in step 302 (after deduplication, form a Y-axis index, such as "20250709-05", "20250709-06", etc.).

[0118] For each intersection of the X and Y axes (i.e., a grid slice at a certain time), the Z-axis vector contains three types of parameters:

[0119] Garbage quality parameters: Extract the garbage quality attribute value (average value, unit: kg) within the grid and time slice from the normalized sanitation data volume;

[0120] Transfer station saturation parameter: the capacity saturation rate (%) of the transfer station to which the grid belongs in the time slice;

[0121] Vehicle attenuation parameter: Match the full load rate attenuation factor (% / km) corresponding to the time slice.

[0122] If a grid has no data in a certain time slice (such as no garbage disposal), the quality parameter is recorded as 0.00kg, and the saturation rate and attenuation factor use the values ​​of the adjacent time slices of the same grid; if the quality parameter is >0 but the saturation rate = 100%, it is marked as "the garbage in this grid may not be transported in time"; if the attenuation factor is negative (an unreasonable value), the historical average value is called instead. The final generated three-dimensional sanitation status tensor is a structured data set. For example: in the "XC-123-456" grid and the "20250709-05" time slice, the state parameter vector is [120.50kg, 60.00%, 3.8% / km], which intuitively reflects the status of the sanitation facilities in this space-time unit.

[0123] Capacity saturation rate calculation and topological overlay analysis achieve spatial binding between waste containers and transfer stations, enabling data to reflect the true regional service relationship. Spatial grid coding unifies dispersed spatial data to a grid scale, facilitating regional analysis. Attenuation factor fitting and time window alignment quantify the load variation of recycling trucks and link container data with the time series of removal plans, resolving the analytical bias caused by the "temporal misalignment" of traditional data. The three-dimensional sanitation state tensor, by integrating multi-dimensional parameters (quality, saturation rate, and attenuation factor) and assigning spatial and temporal constraints, upgrades data from "scattered records" to "structured state descriptions with physical meaning," improving the reliability of analytical results.

[0124] In a preferred embodiment of the present invention, step 4, inputting the sanitation status tensor into a gated loop architecture, combining regional population mobility characteristics and commercial activity intensity time series variables to generate a garbage load probability distribution matrix, may include:

[0125] Step 400: Input the spatial grid code, time slice code, and state parameter vector of the three-dimensional sanitation state tensor into the gated recurrent unit, and access the regional population flow characteristic time series variables and commercial activity intensity time series variables in real time, where:

[0126] The population mobility characteristic time series variable is generated based on the switching frequency of mobile terminal base stations in the target area;

[0127] The commercial activity intensity time series variable is generated based on the time series data of electronic payment transaction volume in the target area;

[0128] Step 401: Use a gated recurrent unit to extract spatiotemporal features from the sanitation status tensor, fuse the population flow feature time series variables with the commercial activity intensity time series variables, and output a garbage load probability distribution matrix corresponding to each spatial grid code within a preset time period in the future. The garbage load probability distribution matrix contains probability values ​​for multiple garbage load levels, specifically including:

[0129] Step 4010: The spatial grid encoding dimension of the sanitation state tensor is input as a spatial topological basis into the memory cell of the gated recurrent unit; the time slice encoding dimension is input as a timing control variable; and the state parameter vector dimension is input as a dynamic observation feature.

[0130] Step 4011: triggering a fusion operation based on the spatiotemporal dynamic features extracted from the memory cells:

[0131] The population mobility feature time series variables are converted into feature vectors with the same dimensions as the memory cell hidden layer through the first feature embedding layer;

[0132] The business activity intensity time series variable is converted into a feature vector with the same dimension as the memory cell hidden layer through the second feature embedding layer;

[0133] Step 4012: During the hidden state update phase of the gated recurrent unit, the population flow feature vector and the commercial activity feature vector are used as gated adjustment factors to dynamically weight and modify the output features of the memory cells to obtain modified fusion features.

[0134] In step 4013, the corrected fusion features are mapped into a garbage load probability distribution matrix corresponding to each spatial grid code in a future preset time period through a fully connected layer, where the output of each spatial grid code includes a low load level probability value, a medium load level probability value, a high load level probability value, and an overload load level probability value.

[0135] In this embodiment of the present invention, the three-dimensional sanitation state tensor (including the spatial grid code, time slice code, and state parameter vector) must first be converted into a structured input format that can be processed by the gated recurrent unit (GRU):

[0136] Spatial grid coding: Convert spatial grid codes (e.g., "XC-123-456") into numerical spatial indexes. This is done by using a preset "grid-index mapping table" (e.g., sorting city grids from west to east and from south to north, and assigning unique integer IDs). This converts the string code into an integer (e.g., "XC-123-456" corresponds to ID = 589), which serves as the identifier of the spatial dimension.

[0137] Time slice encoding processing: Convert the time slice code (such as "20250709-05") into a time series index - assign consecutive integers (such as "20250709-05" is the 12th slice, corresponding to index = 12) in chronological order (from the first time slice to the current slice) to mark the position of the time dimension.

[0138] State parameter vector standardization: The three parameters in the state parameter vector (garbage mass, saturation rate, and attenuation factor) need to be standardized:

[0139] Garbage mass (unit: kg): normalized by the "maximum mass in the history of the grid", the formula is "normalized mass = current mass ÷ maximum mass of the grid within 30 days" (result range 0-1);

[0140] Capacity saturation rate (%): directly converted to decimal form (e.g. 60% → 0.6);

[0141] Full load rate attenuation factor (% / km): Normalized according to the "regional maximum attenuation factor" (for example, if the regional maximum attenuation factor is 5% / km, the current value is 3.8% / km → 3.8÷5=0.76).

[0142] The standardized state parameter vector (such as [0.35, 0.6, 0.76]) is used as the basic input feature of GRU.

[0143] By interfacing with telecommunications operators, we obtain base station handoff records for all mobile terminals (mobile phones, IoT devices) within the target area. We count the number of handoffs within each spatial grid every five minutes (a higher handoff frequency indicates greater population mobility). For example, the "XC-123-456" grid had 120 handoffs between 8:00 and 8:05. We aggregate these handoff frequencies every five minutes by time window (aligned with the time slice encoding of the sanitation status tensor, such as a two-hour window) and calculate the "average handoff frequency within the window" as the population mobility feature value for that time slice. We then sort these population mobility feature values ​​by spatial grid and time slice to form a time series variable matrix that matches the dimensions of the sanitation status tensor (for example, the population variable value for the "XC-123-456" grid at "20250709-05" is 135). This is then fed into the GRU in real time through the data interface.

[0144] Connect to the e-payment platforms of major merchants in the region (such as Alipay and WeChat Pay) to obtain the total transaction volume (in RMB) per hour within each spatial grid. Transactions are mapped to the corresponding spatial grid based on their geographic location (based on the GPS location at the time of payment). For example, the total transaction volume for the "XC-123-456" grid between 9:00 and 10:00 is RMB 5,000. Hourly transaction volume is aggregated into two-hour time slices (aligned with the time slices of the sanitation status tensor). The "total transaction volume within the window" is calculated and normalized by the "maximum historical single-window transaction volume in the region" (e.g., if the maximum single-window transaction volume in the region is 20,000 RMB, the current value is 5,000 RMB → 5,000 ÷ 20,000 = 0.25). This serves as the commercial activity intensity value for that time slice. Similarly, sorting by spatial grid and time slice creates a commercial activity time series variable matrix (e.g., the commercial variable value for the "XC-123-456" grid at "20250709-05" is 0.25), which is then fed into the GRU in real time.

[0145] In step 4010, the gated recurrent unit (GRU) consists of three core modules: the "memory cell," the "update gate," and the "reset gate." The input allocation rules are as follows:

[0146] Spatial topological basis input memory cells: The integer ID (such as 589) encoded by the spatial grid is converted into a one-hot vector (dimension = total number of grids in the region, such as 10,000 dimensions, only the position corresponding to the ID is 1, and the rest are 0), which is used as the spatial topological basis input memory cell, so that the cell "remembers" the spatial location characteristics of the grid (such as whether it is close to a commercial area or a residential area).

[0147] Time series control variable input update gate: The integer index of the time slice encoding (such as 12) is converted into a time feature vector (containing derivative features such as "hour", "whether it is a weekend", and "whether it is a holiday". For example, 12 corresponds to "10:00-12:00", "non-weekend", and "non-holiday", encoded as [10, 0, 0]), and input into the update gate to adjust the retention ratio of historical information in the memory cell (for example, retaining more historical data during peak hours).

[0148] Dynamic observation feature input reset gate: The standardized state parameter vector (such as [0.35, 0.6, 0.76]) is input into the reset gate to "forget" irrelevant historical information (such as reducing the impact of historical load data when the attenuation factor is abnormal).

[0149] Step 4011, first feature embedding layer (population mobility): The population mobility time series variable (e.g., 135 times / 5 minutes) is first normalized (according to the maximum switching frequency of the region, e.g., 500 times / 5 minutes → 135 ÷ 500 = 0.27) and then input into the first embedding layer; the embedding layer is a fully connected network (input dimension = 1, output dimension = GRU memory cell hidden layer dimension, e.g., 64 dimensions). The scalar is converted into a 64-dimensional vector (e.g., [0.12, 0.35, ..., 0.21]) through a linear transformation (weight matrix × input + bias) to ensure consistency with the memory cell hidden layer dimension.

[0150] Second feature embedding layer (business activity): The time series variable of business activity intensity (such as 0.25) is also input into the second embedding layer (the structure is the same as the first embedding layer, and the weight matrix is ​​independent) and converted into a 64-dimensional vector (such as [0.08, 0.42, ..., 0.15]). The dimension matches the memory cell hidden layer.

[0151] Step 4012, during the GRU hidden state update phase, the demographic and commercial features are integrated using the following logic:

[0152] Dynamic adjustment of the update gate and reset gate: The update gate outputs a "retention weight" (such as 0.8, indicating that 80% of historical information is retained) based on the timing control variable (such as the time slice "10:00-12:00"), and the reset gate outputs a "reset weight" (such as 0.3, indicating that only 30% of irrelevant information is retained) based on the state parameter vector (such as high saturation rate).

[0153] Weighted fusion of population and commercial features: Multiply the population feature vector (64 dimensions) and the commercial feature vector (64 dimensions) by the "dynamic weight" respectively - the weight is calculated by the current hidden state (for example, the population vector weight is 0.6 when the population is dense, and the commercial vector weight is 0.5 when the business is active), and then add them together to obtain the fusion adjustment vector (for example, 0.6×population vector + 0.5×commercial vector).

[0154] Memory cell output correction: Use the fused adjustment vector to correct the output characteristics of the memory cell (original output + adjustment vector × correction coefficient) so that the output simultaneously reflects the combined impact of sanitation status, population mobility, and commercial activities (for example, when the weight of the commercial vector is high, the characteristics are more focused on the rule that "the amount of garbage increases with the growth of transaction volume").

[0155] In step 4013, the corrected fusion feature (64 dimensions) is input into the fully connected layer (input dimension = 64, output dimension = 4) and processed by linear transformation + Softmax activation function:

[0156] Linear transformation: convert the 64-dimensional features into 4 raw scores (e.g. [-1.2, 0.5, 2.1, -0.8]);

[0157] Softmax activation: Convert the original score into a probability value, the formula is "probability of a certain level = ",in, Indicates the The raw score of each level; Indicates the total number of levels (here ); is a natural constant; is the category index (from 1 to ); Indicates the The raw scores of the four categories are guaranteed to sum to 1 (e.g., [0.05, 0.25, 0.6, 0.1]).

[0158] Probability level definition: The four probability values ​​outputted correspond to:

[0159] Low load level: garbage volume ≤ 50% of the grid's daily average volume (e.g. ≤ 50kg);

[0160] Medium load level: 50%<waste volume≤100% (50-100kg);

[0161] High load level: 100%<waste volume≤150% (100-150kg);

[0162] Overload level: garbage volume>150% (>150kg).

[0163] Matrix assembly: The probability values ​​of all spatial grids in a preset future time period (e.g., 24 hours, including 12 time slices) are assembled according to the dimensions of "grid × time slice × level" to form a garbage load probability distribution matrix (e.g., a three-dimensional matrix of 10,000 grids × 12 slices × 4 levels).

[0164] Through standardization and access to multi-source time series variables, dispersed data such as sanitation status, population mobility, and commercial activity are integrated into a unified input, addressing the limitations of analysis that relies solely on a single data set. The input allocation mechanism enables the GRU to specifically "learn" spatial, temporal, and state features, avoiding prediction bias caused by feature confusion. Embedding layer transformations adapt low-dimensional time series variables to high-dimensional hidden layers, ensuring the effective integration of multiple factors. Gating adjustments dynamically weight memory cell outputs, enabling the model to flexibly capture complex relationships such as "population surge → increased waste volume" and "business activity → increased recyclables," thereby improving feature extraction accuracy. The garbage load probability distribution matrix probabilistically quantifies the possible range of future garbage volumes (rather than a single predicted value), addressing the problem of resource misallocation caused by "point predictions."

[0165] In a preferred embodiment of the present invention, step 5, solving the saddle point equilibrium of the removal path cost functional and the retention risk functional based on the garbage load probability distribution matrix and outputting resource scheduling instructions to the removal vehicle terminal, may include:

[0166] Step 500: Using the probability values ​​in the garbage load probability distribution matrix as input parameters, a cost functional for the removal path is constructed. The cost functional's calculation variables include the fuel consumption corresponding to the distance traveled by the removal vehicle, the mechanical loss coefficient associated with the vehicle's full load factor, and the labor cost required for the removal operation. Simultaneously, a retention risk functional is constructed based on the garbage load probability distribution matrix. The risk functional's calculation variables include the probability of excessive garbage accumulation in each spatial grid and the penalty cost resulting from excessive accumulation.

[0167] Step 501, optimizing the cost functional and the risk functional by a saddle point equilibrium algorithm to find the final Pareto solution that brings the cost and risk into equilibrium, specifically includes:

[0168] Step 5010: Calculate the removal path cost based on the four-level probability values ​​encoded in each spatial grid in the garbage load probability distribution matrix:

[0169] Calculate the fuel consumption factor of the basic driving route based on the low and medium probability values;

[0170] The mechanical wear adjustment factor is generated by combining the high and overload level probability values ​​and the full load rate attenuation factor;

[0171] Generate manpower time allocation factors based on the time attribute value of the sanitation data body;

[0172] The fuel consumption factor, mechanical wear adjustment factor and manpower time allocation factor are input into the path cost integration unit to generate a comprehensive path cost index;

[0173] Step 5011: Based on the overload level probability value in the garbage load probability distribution matrix, perform detention risk calculation:

[0174] Extracting spatial grid codes whose overload level probability values ​​are greater than a preset threshold;

[0175] Calculate the penalty coefficient based on the capacity saturation rate;

[0176] Generate emergency disposal cost coefficient based on environmental hazard level;

[0177] Input the penalty coefficient and emergency disposal cost coefficient into the risk integration unit to generate a comprehensive detention risk index;

[0178] Step 5012: Input the comprehensive path cost index and the comprehensive retention risk index into the saddle point equilibrium controller to perform dual-objective optimization:

[0179] The first optimization stage: fix the comprehensive detention risk index and reconfigure the transportation path sequence to make the comprehensive path cost index reach the final value;

[0180] The second optimization stage: fix the comprehensive path cost index and redistribute the removal priority to make the comprehensive detention risk index reach the final value;

[0181] Step 5013: Iterate until the change in the optimization results for two consecutive times is less than the set tolerance threshold, and then output a three-dimensional scheduling instruction set, including the final transportation path sequence encoded in the spatial grid, the transportation operation time window of the preset period, and the emergency disposal identifier of the overload probability grid;

[0182] Step 502: Generate resource scheduling instructions including a cleaning route sequence, an operation time window, and a priority identifier based on the Pareto final solution;

[0183] Step 503: Transmit the resource scheduling instruction to the transportation vehicle terminal of the corresponding spatial grid in real time.

[0184] In this embodiment of the present invention, the cost functional is used to quantify the comprehensive cost of garbage collection and transportation. It contains three core calculation variables and the construction logic is as follows:

[0185] Fuel consumption calculation: This is based on the "low load probability" and "medium load probability" (together the "normal load probability") in the garbage load probability distribution matrix. If the normal load probability is high (e.g., ≥70%), the garbage load in that grid is stable, and the shortest route (e.g., a straight-line distance from the starting point to the grid) can be planned, resulting in shorter driving distances and lower fuel consumption. If the normal load probability is low (e.g., <50%), multiple round trips may be required (e.g., if a sudden increase in garbage volume requires secondary collection), driving distance increases, and fuel consumption increases proportionally. Ultimately, driving distance is converted into fuel consumption (e.g., if fuel consumption is 0.15 liters per kilometer and the fuel price is 8 yuan / liter, a distance of 10 kilometers would cost 12 yuan).

[0186] Mechanical loss coefficient calculation: Combine the "high load probability" and "overload load probability" (together the "high load probability") with the "load rate attenuation factor" fitted in step 302. A high high load probability (e.g., ≥60%) indicates that the vehicle is likely to be highly loaded for extended periods. Combined with the attenuation factor (e.g., 3.8% / km), mechanical wear (e.g., tire wear and engine loss) is exacerbated. The loss coefficient is calculated as "high load probability x attenuation factor" (e.g., 60% x 3.8% / km = 2.28% / km), which is then converted to a cost (e.g., wear and tear cost per kilometer is 0.5 yuan, 2.28% x 0.5 = 0.0114 yuan / km).

[0187] Labor cost calculation: This is based on the "time attribute value" of the standardized sanitation data volume (e.g., "2025-07-09 08:00"). During peak hours (e.g., 7:00-9:00, 17:00-19:00), when labor demand is high, the hourly wage is calculated at 1.5 times the base hourly wage. During off-peak hours (e.g., 9:00-17:00), the base hourly wage is used; during nighttime hours (e.g., 19:00-7:00), the base hourly wage is used. Labor cost = duration of work (e.g., 2 hours) × corresponding hourly wage.

[0188] The cost functional integrates three variables through weighted summation (preset weights: fuel 40%, mechanical loss 20%, and manpower 40%) to form a quantitative cost indicator.

[0189] The risk functional is used to assess the potential losses from waste retention. It contains two core calculation variables and its construction logic is as follows:

[0190] Excess accumulation probability extraction: Extract the "overload load level probability" (for example, the overload probability of a grid is 40%) from the garbage load probability distribution matrix, and screen out grids with probability values ​​exceeding the preset threshold (for example, 30%) and mark them as "high-risk grids."

[0191] Penalty cost calculation: Combined with the "Transfer Station Capacity Saturation Rate" in step 301—if the saturation rate is ≤ 60%, the penalty coefficient is 0.3 (calculated as 30% of the basic fine of 1,000 yuan per trip); between 60% and 80%, the coefficient is 0.6; between 80% and 100%, the coefficient is 1.0; and if overloaded, the coefficient is 1.5. Penalty cost = overload probability × penalty coefficient × basic fine (e.g., 40% × 1.0 × 1,000 = 400 yuan).

[0192] The risk functional quantifies the detention risk by multiplying the probability of excess accumulation by the penalty cost (the higher the probability and the heavier the penalty, the greater the risk value).

[0193] In step 5010, for each spatial grid, calculate the mean of the "low load probability + medium load probability" (e.g., 30% low + 40% medium = 70%) and assign values ​​according to the rule that "the higher the normal load probability, the lower the fuel consumption factor": 70% → factor 0.7 (baseline value 1.0), 50% → factor 1.0, 30% → factor 1.3 (the smaller the value, the lower the fuel cost).

[0194] Mechanical wear adjustment factor generation: Calculate the average of "high load probability + overload probability" (e.g., 20% high + 10% overload = 30%), multiply by the full load rate attenuation factor (e.g., 3.8% / km) to obtain the wear adjustment factor. The larger the value, the higher the wear cost).

[0195] Generation of labor time allocation factor: Based on the "time period label" of the time attribute value (such as "morning rush hour"), the value is assigned according to the preset rules: morning rush hour → 1.5, off-peak → 1.0, night → 2.0 (the larger the value, the higher the labor cost).

[0196] Comprehensive path cost index integration: Input the three factors into the path cost integration unit and calculate according to the formula "fuel factor × 0.4 + wear factor × 0.2 + manpower factor × 0.4". The value is the comprehensive path cost index (the smaller the value, the lower the cost).

[0197] In step 5011, all spatial grids are traversed to select grids with an "overload probability > 30%" (e.g., grid A has an overload probability of 45% and grid B has an overload probability of 50%), forming a high-risk list. For high-risk grids, the capacity saturation rate of their associated transfer stations is matched: grid A corresponds to a transfer station saturation rate of 70% (a coefficient of 0.6); grid B corresponds to a transfer station saturation rate of 90% (a coefficient of 1.0). Combined with the grid's "environmental hazard level" (presumably, the kitchen waste grid is level 3, and the recyclables grid is level 1), level 3 is assigned a coefficient of 1.5, and level 1 is assigned a coefficient of 1.0. The emergency disposal cost coefficient is calculated as: penalty coefficient × environmental hazard coefficient (grid A: 0.6 × 1.5 = 0.9; grid B: 1.0 × 1.0 = 1.0).

[0198] The emergency response cost coefficient is input into the risk integration unit and calculated according to "grid overload probability × coefficient" (grid A: 45% × 0.9 = 0.405; grid B: 50% × 1.0 = 0.5). The results of all high-risk grids are then summed (0.405 + 0.5 = 0.905) to obtain the comprehensive detention risk index (the larger the value, the higher the risk).

[0199] Step 5012, first optimization stage (fixed risk, optimized cost): set the upper limit of the comprehensive detention risk index (e.g., ≤ 0.9), keep the removal priority of high-risk grids unchanged, and re-plan the removal route sequence:

[0200] Merge the removal tasks of adjacent grids (for example, grid A and grid C are 500 meters apart and are combined into the same route) to reduce the distance vehicles travel in idling;

[0201] Adjust the route sequence (e.g., change from "far to near" to "circular route") to shorten the total driving distance.

[0202] Each time an adjustment is made, a new comprehensive path cost index is calculated until the cost no longer decreases (e.g., from 1.108 to 0.95).

[0203] Second optimization stage (fixed cost, optimized risk): set the upper limit of the comprehensive path cost index (e.g. ≤0.95), keep the path sequence unchanged, and redistribute the removal priority:

[0204] High-risk grids (e.g., grid B with a 50% overload probability) are marked as "priority removal" (priority 1) and scheduled within the first 30 minutes of the operation window;

[0205] Low-risk grids (e.g., overload probability 20%) are marked as “routine removal” (priority 3) and scheduled in the second half of the time window.

[0206] After each adjustment, a new comprehensive detention risk index is calculated until the risk no longer decreases (e.g., from 0.905 to 0.75).

[0207] In step 5013, after two consecutive optimizations, if the change in the comprehensive path cost index (e.g., from 0.95 to 0.94, a change of 1.05%) and the change in the comprehensive detention risk index (e.g., from 0.75 to 0.74, a change of 1.33%) are both less than the set tolerance threshold (e.g., 2%), the iteration is stopped.

[0208] Three-dimensional scheduling instruction set generation:

[0209] Cleaning route sequence: According to the optimized route, record the order of the grids (such as grid B → grid A → grid C), and mark the driving distance of each grid (such as B to A: 1.2 kilometers);

[0210] Operation time window: Assign a specific time period to each grid (e.g., Grid B: 8:00-8:30; Grid A: 8:30-9:00) to match the vehicle's operating capacity (e.g., cleaning time per grid is approximately 30 minutes);

[0211] Emergency handling identifier: Mark “★” for grids with an overload probability >50% (such as grid B) to indicate that the vehicle needs priority handling.

[0212] Step 502: Generate structured resource scheduling instructions based on the three-dimensional scheduling instruction set, including:

[0213] Removal route sequence: Describe the route in words (e.g., "Start from the transfer station → take XX Road to Grid B → follow XX Street to Grid A → turn to XX Lane to Grid C → return to the transfer station") and attach simple map coordinates;

[0214] Job time window: clearly define the start and end time of each grid (e.g., "Grid B: 2025-07-10 08:00-08:30") and indicate the total job duration (e.g., 2 hours);

[0215] Priority identification: Use numbers 1-5 to indicate priority (1 is the highest), such as grid B is marked "Priority 1" and grid C is marked "Priority 3".

[0216] The instructions must be format-checked (to ensure there are no time conflicts and route continuity) and then converted into a format recognizable by the vehicle terminal (such as JSON).

[0217] Step 503: In real-time transmission via the Internet of Vehicles (IoV) (e.g., 4G / 5G), the command is encrypted before transmission (using the AES-128 encryption algorithm) to prevent information leakage. Based on the spatial grid code, the command is sent to the terminal of the cleaning vehicle responsible for that area (for example, if grid B falls within the responsibility of "cleaning vehicle 007," the command is sent directly to that vehicle). After receiving the command, the vehicle terminal automatically responds with a "received" confirmation message. If no confirmation is received within 10 seconds, the edge computing node automatically retransmits the command (up to three times) to ensure delivery.

[0218] By constructing cost and risk functionals, the system quantifies dispersed costs like fuel, machinery, and manpower, and risks like overloading and penalties, resolving the difficulty of balancing costs and risks and providing a calculable target for optimization. A dual-objective optimization approach (fixing risk to reduce cost, and vice versa) and iterative iteration achieves a dynamic balance between cost and risk, avoiding extreme scenarios where excessive pursuit of low costs leads to waste stagnation or excessive focus on risk leads to cost surges. The resulting dispatch instructions better align with actual operational needs. A three-dimensional dispatch instruction set (routes, time windows, and emergency markings) transforms waste removal tasks from a "fuzzy schedule" into structured instructions tailored to grid, time period, and priority, improving execution efficiency. Instruction generation and encrypted transmission ensure accuracy, security, and real-time delivery, reducing issues like empty vehicle runs and missed deliveries. Emergency markings prioritize high-risk areas, mitigating environmental issues caused by waste accumulation.

[0219] like Figure 2 As shown, an embodiment of the present invention also provides a smart sanitation waste recycling system, including:

[0220] The parsing module is used to parse the container's intrinsic identification code through the penetrating radio frequency unit of the intelligent classification terminal, obtain the garbage quality data stream and generate the original spatiotemporal observation set;

[0221] The conversion module is used to input the original spatiotemporal observation set into the edge computing node, convert it into a triple structure according to the preset sanitation ontology rules, and output the normalized sanitation data body;

[0222] The tensor construction module is used to perform spatiotemporal registration of the normalized sanitation data volume with two key facility status parameters: the real-time capacity saturation rate of regional transfer stations and the attenuation factor of the historical full load rate of mobile recycling vehicles, to form a sanitation status tensor with physical constraints;

[0223] The load probability module is used to input the sanitation status tensor into the gated loop architecture, combine it with the regional population flow characteristics and commercial activity intensity time series variables, and generate the garbage load probability distribution matrix;

[0224] The scheduling instruction module is used to solve the saddle point equilibrium of the transportation path cost functional and the detention risk functional based on the garbage load probability distribution matrix, and output resource scheduling instructions to the transportation vehicle terminal.

[0225] It should be noted that this system is a system corresponding to the above method, and all implementation methods in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0226] The embodiment of the present invention further provides a computer-readable storage medium storing instructions, which, when executed on a computer, causes the computer to execute the above-described method. All implementations in the above-described method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0227] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A smart sanitation waste recycling method, characterized in that: The method comprises: Step 1: The penetrating radio frequency unit of the intelligent classification terminal parses the container's intrinsic identification code and obtains the garbage quality data stream to generate the original spatiotemporal observation set; Step 2: Input the original spatiotemporal observation set into the edge computing node, convert it into a triple structure according to the preset sanitation ontology rules, and output the normalized sanitation data body; Step 3: Perform spatiotemporal registration between the normalized sanitation data volume and two key facility status parameters: the real-time capacity saturation rate of regional transfer stations and the attenuation factor of the historical full load rate of mobile recycling vehicles, to form a sanitation status tensor with physical constraints. Step 4: Input the sanitation status tensor into the gated loop architecture, combine it with the regional population flow characteristics and the commercial activity intensity time series variables, and generate the garbage load probability distribution matrix; Step 5: Solve the saddle point equilibrium of the transportation path cost functional and the detention risk functional based on the garbage load probability distribution matrix, and output resource scheduling instructions to the transportation vehicle terminal.

2. The smart sanitation waste recycling method according to claim 1 is characterized in that: The penetrating radio frequency unit of the intelligent classification terminal parses the container's intrinsic identification code and obtains the garbage quality data stream to generate the original spatiotemporal observation set, including: The passive RFID tag embedded in the garbage container is read contactlessly by a penetrating radio frequency unit to resolve the container's unique intrinsic identification code; Based on the container's intrinsic identification code, the dynamic gravity sensor array is activated to collect the continuous change data stream of the garbage mass bound to the container in real time, and the positioning module is triggered to obtain the container's real-time geographic coordinates and timestamp; The container intrinsic identification code, the bound garbage quality data stream, the geographic coordinates and the timestamp are encapsulated into a structured original spatiotemporal observation set, wherein the original spatiotemporal observation set includes a container ID field, a quality value field, a longitude field, a latitude field and a timestamp field.

3. The smart sanitation waste recycling method according to claim 2 is characterized in that: The original spatiotemporal observation set is input into the edge computing node, converted into a triple structure according to the preset sanitation ontology rules, and the normalized sanitation data volume is output, including: The original spatiotemporal observation set is transmitted to the edge computing node, and the following mapping operations are performed using the preset sanitation ontology rule base: The container ID field is mapped to the entity identifier; The quality numeric field is mapped to the quality attribute value; The combination of longitude and latitude fields is mapped to spatial attribute values; The timestamp field is mapped to the time attribute value; Based on the mapping results, triple structure data is generated, including spatial relationship triples, quality status triples and time status triples, and the normalized sanitation data body of triples is output.

4. The smart sanitation waste recycling method according to claim 3 is characterized in that: The normalized sanitation data volume is spatiotemporally aligned with two key facility status parameters: the real-time capacity saturation rate of regional transfer stations and the attenuation factor of the historical full load rate of mobile recycling vehicles. This forms a physically constrained sanitation status tensor, including: Extract spatial attribute values ​​and temporal attribute values ​​from the normalized sanitation data volume; Perform transfer station spatial matching based on spatial attribute values, i.e., obtain transfer station IoT sensor data in real time and calculate capacity saturation rate; perform topological overlay analysis on container spatial attribute values ​​and transfer station geofences, and output spatial grid code; Perform time alignment of the collection and transportation plan based on the time attribute value. This involves analyzing the historical operation logs of mobile recycling vehicles and fitting the decay coefficient of the full load rate with mileage. Align the container's time attribute value with the collection and transportation vehicle operation plan using a sliding time window and output a time slice code. The spatial grid coding, time slice coding, quality attribute values ​​in the normalized sanitation data volume, capacity saturation rate, and fitted full load rate attenuation coefficient with mileage are integrated to generate a three-dimensional sanitation state tensor. The dimensions of the three-dimensional sanitation state tensor include: X-axis: spatial grid encoding; Y-axis: time slice encoding; Z axis: state parameter vector.

5. The smart sanitation waste recycling method according to claim 4 is characterized in that: The sanitation status tensor is input into the gated loop architecture, combined with the regional population flow characteristics and commercial activity intensity time series variables to generate the garbage load probability distribution matrix, including: The spatial grid code, time slice code and state parameter vector of the three-dimensional sanitation state tensor are input into the gated recurrent unit, and the regional population flow characteristic time series variables and commercial activity intensity time series variables are accessed in real time, where: The population mobility characteristic time series variable is generated based on the switching frequency of mobile terminal base stations in the target area; The commercial activity intensity time series variable is generated based on the time series data of electronic payment transaction volume in the target area; The spatiotemporal features of the sanitation status tensor are extracted through a gated recurrent unit, and the time series variables of population flow characteristics and commercial activity intensity are integrated to output the garbage load probability distribution matrix corresponding to each spatial grid code in a preset time period in the future. The garbage load probability distribution matrix contains probability values ​​of multiple garbage quantity levels.

6. The smart sanitation waste recycling method according to claim 5, characterized in that: The spatiotemporal features of the sanitation status tensor are extracted through a gated recurrent unit, and the population flow characteristic time series variables and the commercial activity intensity time series variables are integrated to output the garbage load probability distribution matrix corresponding to each spatial grid code in the future preset time period. The garbage load probability distribution matrix contains the probability values ​​of multiple garbage volume levels, including: The spatial grid encoding dimension of the sanitation state tensor is used as the spatial topological basis to input the memory cells of the gated recurrent unit; the time slice encoding dimension is used as the timing control variable input; and the state parameter vector dimension is used as the dynamic observation feature input; Based on the spatiotemporal dynamic features extracted from memory cells, the fusion operation is triggered: The population mobility feature time series variables are converted into feature vectors with the same dimensions as the memory cell hidden layer through the first feature embedding layer; The business activity intensity time series variable is converted into a feature vector with the same dimension as the memory cell hidden layer through the second feature embedding layer; In the hidden state update phase of the gated recurrent unit, the population flow feature vector and the commercial activity feature vector are used as gate adjustment factors to dynamically weight and modify the output features of the memory cells to obtain the modified fusion features. The corrected fusion features are mapped into the garbage load probability distribution matrix corresponding to each spatial grid code in the future preset time period through the fully connected layer, where the output of each spatial grid code contains the probability value of low load level, medium load level, high load level and overload load level.

7. The smart sanitation waste recycling method according to claim 6 is characterized in that: Based on the probability distribution matrix of garbage load, the saddle point equilibrium of the transportation path cost functional and the detention risk functional is solved, and resource scheduling instructions are output to the transportation vehicle terminal, including: Using the probability values ​​in the garbage load probability distribution matrix as input parameters, a collection and transportation path cost functional is constructed. The calculation variables of this cost functional include the fuel consumption corresponding to the distance traveled by the collection and transportation vehicle, the mechanical loss coefficient related to the vehicle's full load factor, and the labor cost required for the collection and transportation operation. Simultaneously, a detention risk functional is constructed based on the garbage load probability distribution matrix. The calculation variables of this risk functional include the probability of excessive garbage accumulation in each spatial grid and the penalty cost caused by excessive accumulation. Optimize the cost functional and risk functional through the saddle point equilibrium algorithm to find the final Pareto solution that balances cost and risk. Generate resource scheduling instructions including the cleaning route sequence, operation time window and priority identification according to the Pareto final solution; The resource scheduling instructions are transmitted in real time to the transportation vehicle terminal of the corresponding spatial grid.

8. The smart sanitation waste recycling method according to claim 7 is characterized in that: The cost functional and risk functional are optimized through the saddle point equilibrium algorithm to find the final Pareto solution that balances cost and risk, including: Based on the four-level probability values ​​coded for each spatial grid in the garbage load probability distribution matrix, the removal path cost calculation is performed: Calculate the fuel consumption factor of the basic driving route based on the low and medium probability values; The mechanical wear adjustment factor is generated by combining the high and overload level probability values ​​and the full load rate attenuation factor; Generate manpower time allocation factors based on the time attribute value of the sanitation data body; The fuel consumption factor, mechanical wear adjustment factor and manpower time allocation factor are input into the path cost integration unit to generate a comprehensive path cost index; Based on the overload level probability values ​​in the garbage load probability distribution matrix, the detention risk calculation is performed: Extracting spatial grid codes whose overload level probability values ​​are greater than a preset threshold; Calculate the penalty coefficient based on the capacity saturation rate; Generate emergency disposal cost coefficient based on environmental hazard level; Input the penalty coefficient and emergency disposal cost coefficient into the risk integration unit to generate a comprehensive detention risk index; The comprehensive path cost index and the comprehensive detention risk index are input into the saddle point equilibrium controller to perform dual-objective optimization: The first optimization stage: fix the comprehensive detention risk index and reconfigure the transportation path sequence to make the comprehensive path cost index reach the final value; The second optimization stage: fix the comprehensive path cost index and redistribute the removal priority to make the comprehensive detention risk index reach the final value; The algorithm is iterated until the change in the optimization results for two consecutive times is less than the set tolerance threshold, and then a three-dimensional scheduling instruction set is output, including the final transportation path sequence encoded by the spatial grid, the transportation operation time window of the preset period, and the emergency disposal identifier marking the overload probability grid.

9. A smart sanitation waste recycling system, which implements the method according to any one of claims 1 to 8, characterized in that: include: The parsing module is used to parse the container's intrinsic identification code through the penetrating radio frequency unit of the intelligent classification terminal, obtain the garbage quality data stream and generate the original spatiotemporal observation set; The conversion module is used to input the original spatiotemporal observation set into the edge computing node, convert it into a triple structure according to the preset sanitation ontology rules, and output the normalized sanitation data body; The tensor construction module is used to perform spatiotemporal registration of the normalized sanitation data volume with two key facility status parameters: the real-time capacity saturation rate of regional transfer stations and the attenuation factor of the historical full load rate of mobile recycling vehicles, to form a sanitation status tensor with physical constraints; The load probability module is used to input the sanitation status tensor into the gated loop architecture, combine it with the regional population flow characteristics and commercial activity intensity time series variables, and generate the garbage load probability distribution matrix; The scheduling instruction module is used to solve the saddle point equilibrium of the transportation path cost functional and the detention risk functional based on the garbage load probability distribution matrix, and output resource scheduling instructions to the transportation vehicle terminal.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program, which implements the method according to any one of claims 1 to 8 when executed by a processor.

Images