An AI algorithm-based waste recycling vehicle dynamic scheduling system and method

Abstract

The application discloses a kind of based on AI algorithm's waste recycling vehicle dynamic scheduling system and method, belong to intelligent logistics field;System includes task analysis module, vehicle state sensing module, AI scheduling core module etc.;Task analysis module identifies waste article class by attention mechanism U-Net algorithm, generates the task parameter set containing priority;Vehicle state sensing module outputs accurate vehicle data by Kalman filtering algorithm;AI scheduling core module takes "task completion rate, energy consumption, carbon emission reduction" as target, outputs scheduling instruction by multi-objective optimization model and improved Dijkstra algorithm;Data feedback module calculates carbon emission reduction according to Sichuan local standard and updates model;The application solves the problem that prior art scene adaptation is poor, energy consumption is high, task completion rate is improved by 25%, empty running rate is reduced by 18%, and provides technical support for "double carbon" target.

Description

Technical Field

[0001] This invention belongs to the technical field of intelligent logistics and waste recycling, and relates to a dynamic scheduling system and method for waste recycling vehicles based on AI algorithms, which is applicable to waste recycling in various scenarios such as urban communities and industrial parks. Background Technology

[0002] The efficiency of waste recycling vehicle dispatching directly affects recycling costs and carbon emission reduction benefits. Traditional dispatching relies on manual experience, with an empty running rate as high as 25%-30%. Existing waste recycling vehicle dispatching mostly adopts a fixed route mode, which has problems such as delayed task response, low vehicle utilization, and high energy consumption and carbon emissions.

[0003] Traditional scheduling methods struggle to accurately match recycling tasks with vehicle status, lack the ability to adjust to sudden increases in tasks or road anomalies, and have limited accuracy in identifying waste types, leading to chaotic recycling priorities. Existing technologies have significant shortcomings:

[0004] 1. Patent application number CN202510563201.0 only focuses on optimizing energy consumption of logistics vehicles and does not take into account the differences in the types of waste materials when designing loading and priority mechanisms, resulting in a task delay rate of over 40%;

[0005] 2. The patent application number CN202510767429.1 focuses on event-driven scheduling, but does not design a dedicated image recognition algorithm for waste recycling, and cannot distinguish the transportation needs of industrial and domestic waste.

[0006] 3. The patent with application number CN202410620323.4 only realizes the estimated allocation of vehicle resources and does not involve the power management and carbon footprint accounting of new energy vehicles, which does not meet the needs of the "dual carbon" scenario.

[0007] Waste recycling scheduling technology has gone through three stages: "human experience → single-objective algorithm → multi-dimensional AI scheduling".

[0008] Phase 1 (before 2015): Relying on manual allocation, scheduling error > 30%;

[0009] The second phase (2016-2022) introduced simple path algorithms (such as the traditional Dijkstra's algorithm), but did not integrate the characteristics of the recycling scenario. For example, the patent CN202210179369.8 only targets cleaning vehicles and does not cover waste recycling.

[0010] The third phase (from 2023 to present) requires the integration of AI image recognition, multi-objective optimization, and carbon footprint accounting.

[0011] Therefore, in order to solve the above-mentioned technical problems, the technical solution of this application is proposed. Summary of the Invention

[0012] The purpose of this invention is to provide a dynamic scheduling system and method based on AI algorithms to solve the problems of "poor scenario adaptability, high energy consumption, and unquantifiable carbon emission reduction" in existing scheduling systems, and to achieve precise matching of recycling tasks and vehicle resources.

[0013] The technical solution adopted in this invention is as follows:

[0014] A dynamic scheduling system and method for waste recycling vehicles based on AI algorithms is disclosed. The system includes a task parsing module, a vehicle status perception module, an AI scheduling core module, a path output module, a data feedback module, and an anomaly handling module. The functions and technical features of each module are as follows:

[0015] (1) Task parsing module: The core adopts the U-Net image segmentation algorithm with attention mechanism. This algorithm adds a waste material feature extraction channel to the coding layer of the U-Net basic network, which can accurately identify the difference in reflectivity of materials such as plastic and metal. The recognition accuracy of common waste in Sichuan Province is no less than 97%. The task parsing module collects waste images through the camera of the recycling station, and identifies the item category after processing by the algorithm, generating a complete task parameter set including category label, loading coefficient, recycling priority and station coordinates. The recycling priority adopts a weighted calculation method, which is determined by combining the decay cycle (weight increase for perishable waste), the amount of backlog at the station (weight increase for backlog over 24 hours) and the recycling value (weight increase for high calorific value waste such as precious metals), to ensure that urgent and high-value tasks are executed first. The training sample library of this algorithm contains 120,000 images of common waste in Sichuan Province, covering different lighting scenarios such as sunny days and rainy days, and different placement states such as stacking and scattering. The sample labeling accuracy is ≥99%, which provides a guarantee for the recognition accuracy.

[0016] (2) Vehicle Status Perception Module: Composed of a BDS / GPS dual-mode positioning unit, a resistance strain gauge load cell, a battery status sensor, and a battery management system (BMS); wherein the positioning error of the dual-mode positioning unit is ≤3 meters, the measurement accuracy of the resistance strain gauge load cell is ±0.05 tons, the battery status sensor works in conjunction with the battery management system, and the battery management system (BMS) serves as the core processing unit for vehicle battery data, collecting and analyzing key operating data of the battery pack in real time, specifically including individual battery voltage (collection accuracy ±5mV), total battery pack voltage (collection accuracy ±0.1V), individual battery temperature (collection range -30℃~85℃, accuracy ±1℃), and charging / discharging current. (Data acquisition range -200A~200A, accuracy ±1A) and battery health status (SOH, calculation accuracy ±2%), while also having early warning data output functions for overvoltage, overcurrent, and overtemperature protection; the vehicle status perception module integrates three types of information, namely "BDS / GPS positioning data, load sensor data, and standardized data output by the battery management system", through the Kalman filter algorithm, filtering out positioning noise, load data fluctuations, and battery data interference in rural road sections or high-rise building areas, and finally outputting standardized data with an error ≤3%, specifically including the vehicle's current location, remaining power (directly mapped from the SOC value calculated by the BMS), remaining loading space, and real-time load, providing accurate vehicle status basis for scheduling and matching.

[0017] (3) AI scheduling core module: As the decision-making center of the system, it establishes communication with the task parsing module, vehicle status perception module, urban traffic cloud platform and anomaly handling module; the AI ​​scheduling core module has a built-in multi-objective optimization model, and the objective function is set as "0.4×task completion rate + 0.3×energy consumption reduction rate + 0.3×carbon emission reduction". The calculation of "energy consumption reduction rate" requires calling the "battery management system energy consumption data" (such as power consumption per unit mileage) output by the vehicle status perception module, which not only ensures the efficient completion of the recycling task, but also takes into account the goals of energy saving and environmental protection. Route planning is implemented using an improved Dijkstra algorithm, which adds three constraints: a weight coefficient of 0.8 for dedicated new energy lanes (to reduce travel time), a weight coefficient of 1.5 for construction sections (to reduce energy consumption during detours), and direct filtering of weight-restricted sections (to prevent vehicle violations). The route calculation time is ≤1 second, ensuring timely dispatch response. At the same time, the AI ​​dispatch core module communicates with the city's traffic cloud platform in real time, updating the road congestion index and temporary traffic control information every 5 seconds to dynamically optimize the route planning results. Furthermore, the remaining battery power data output by the BMS is incorporated during the planning process to avoid planning routes that exceed the vehicle's range.

[0018] (4) Path output module: It is dedicated to the transmission of dispatch instructions. After receiving the vehicle matching results and path planning instructions output by the AI ​​dispatch core module, it pushes them to the corresponding vehicle terminal through the wireless communication network. The terminal displays them in a combination of text and graphics, which is convenient for the driver to view and execute. If the instruction involves a long-distance task, the terminal will also display the "remaining battery power-mileage matching relationship" estimated by the BMS to help the driver judge the feasibility of the range. The path output module ensures the accuracy and timeliness of instruction transmission and is a key link connecting decision-making and execution.

[0019] (5) Data Feedback Module: This module includes a data acquisition unit and a carbon footprint calculation unit. The data acquisition unit collects task execution data (such as transportation mileage, actual energy consumption, calculated from the total power consumption recorded by the BMS, and task completion time), vehicle status change data (such as the SOC decay curve and SOH change value updated in real time by the BMS), and waste treatment data throughout the process. The carbon footprint calculation unit strictly adopts the Sichuan Provincial Local Standard "Specification for Carbon Footprint Calculation of Recyclable Resources" and calculates carbon emission reduction using the formula "transportation mileage × unit energy consumption carbon emission coefficient - waste recycling carbon emission reduction benchmark value". The "unit energy consumption" data is directly derived from the statistical results of the battery management system. All collected data is transmitted back to the AI ​​scheduling core module via the data bus to dynamically update the weight parameters of "energy consumption reduction rate" in the multi-objective optimization model, thereby continuously improving scheduling accuracy.

[0020] (6) Anomaly Handling Module: Three types of trigger thresholds are preset: vehicle fault, task increment, and road closure. The specific vehicle fault thresholds are "battery SOC ≤ 20% (output and determined by BMS in real time), load sensor data fluctuation > 10% and GPS positioning drift > 10 meters". The sub-condition "BMS warning data triggers fault" is added. When the BMS outputs overvoltage, overcurrent or overtemperature warnings, the vehicle fault response will be triggered even if the SOC threshold is not reached. When the monitored data reaches any threshold, the anomaly handling module immediately triggers the anomaly response and sends a signal to the AI ​​scheduling core module. The AI ​​scheduling core module recalculates the scheduling instructions (such as task transfer, path replanning) through the particle swarm algorithm. The anomaly response time is ≤ 2 seconds. At the same time, after the vehicle fault is triggered, an audible and visual alarm will be automatically sent to the scheduling terminal. The alarm information will include key abnormal data of BMS (such as "battery cell overtemperature: 45℃"), which is convenient for managers to accurately judge the fault type.

[0021] Dynamic scheduling method

[0022] The scheduling method of the above system forms a complete closed loop, realizing efficient linkage between various modules. The specific steps are as follows:

[0023] Step 1: The task parsing module starts image acquisition at the recycling site, completes waste item category identification through the U-Net image segmentation algorithm with attention mechanism, and generates a task parameter set containing category labels, loading coefficients, recycling priorities and site coordinates. Then, the parameter set is transmitted to the AI ​​scheduling core module through the data bus to complete the initial upload of task information.

[0024] Step 2: The vehicle status perception module synchronously starts the operation of each component. The BDS / GPS dual-mode positioning unit collects real-time location, the resistance strain gauge load sensor acquires load data, and the battery management system (BMS) collects battery pack voltage, temperature, current, and SOH data, and standardizes them into three core data categories: "remaining charge (SOC), estimated driving range, and health status". This data is then transmitted to the battery status sensor for secondary verification. After fusion and noise reduction using the Kalman filter algorithm, standardized data containing "current vehicle location, remaining charge (BMS calculated value), remaining loading space, real-time load, and BMS health status" is output and synchronized to the AI ​​scheduling core module via the data bus.

[0025] Step 3: After receiving the task parameter set and vehicle status data, the AI ​​scheduling core module first calls the multi-objective optimization model to calculate the vehicle matching result by combining "task priority, vehicle remaining loading space (matching loading coefficient), BMS output remaining power (matching transportation mileage requirements), and BMS health status (prioritizing scheduling vehicles with SOH≥80%)". Then, combined with the congestion index and temporary control information pushed in real time by the urban traffic cloud platform (updated every 5 seconds), the improved Dijkstra algorithm is used to complete the path planning and generate a scheduling instruction that includes "matching vehicle number, driving route, task order, and BMS range warning threshold (if it is below 25%, it needs to return)".

[0026] Step 4: The route output module receives the scheduling instructions from the AI ​​scheduling core module and accurately pushes them to the corresponding vehicle terminal through the wireless communication network. The terminal displays the route in the form of "route map + task list + BMS real-time data". The BMS data updates the remaining power, estimated range and fault warning status in real time, which makes it convenient for the driver to monitor the battery status during the task.

[0027] Step 5: During task execution, the data feedback module continuously collects data such as "transportation mileage, total power consumption recorded by the BMS (converted to actual energy consumption), task completion status, and the SOC decay trend of the BMS". After calculating carbon emission reduction through the carbon footprint accounting unit, the module transmits data such as "energy consumption data (from the BMS), carbon emission reduction, and changes in BMS health status" back to the AI ​​scheduling core module via the data bus. The AI ​​scheduling core module uses this data to update the weight parameters of "energy consumption reduction rate" in the multi-objective optimization model. At the same time, for vehicles with poor BMS health status, the module adjusts the subsequent matching task mileage (prioritizing short-distance tasks), realizing iterative optimization of the scheduling strategy and forming a closed loop of "data-decision-execution-feedback".

[0028] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0029] 1. A dynamic scheduling system and method for waste recycling vehicles based on AI algorithms, which achieves precise matching of tasks and vehicles through AI algorithms, improves Dijkstra's algorithm to ensure optimal path, shortens task response time by more than 30%, and reduces vehicle empty running rate by 25%.

[0030] 2. In this invention, the multi-objective optimization model takes into account the task completion rate, energy consumption, and carbon emission reduction targets. Combined with the planning of new energy dedicated channels and "precise calculation of BMS energy consumption data", the energy consumption per unit task is reduced by 18% and the carbon emission reduction is increased by 20%. At the same time, the scheduling strategy based on the health status of BMS extends the life of vehicle batteries by 15% to 20%.

[0031] 3. In this invention, the anomaly handling module, combined with the real-time early warning data of the BMS, achieves a rapid response within 2 seconds, avoiding safety risks such as battery overcharging and overheating, and improving fault handling efficiency by 40%; the BMS anomaly data attached to the audible and visual alarm reduces the fault investigation time of management personnel by 60%.

[0032] 4. In this invention, the U-Net attention mechanism algorithm has an accuracy of no less than 97%, providing a precise basis for task priority division and reducing the problem of mismatched recycling. Attached Figure Description

[0033] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments, experimental examples, and comparative examples will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort, wherein:

[0034] Figure 1 This is an architecture diagram of the system modules of the present invention;

[0035] Figure 2 This is a framework diagram of the U-Net image segmentation algorithm with attention mechanism in the system of this invention;

[0036] Figure 3 This is an architecture diagram of the vehicle status perception module in the system of the present invention;

[0037] Figure 4 This is an architecture diagram of the AI ​​scheduling core module in the system of this invention;

[0038] Figure 5 This is an architecture diagram of the exception handling module in the system of the present invention;

[0039] Figure 6 This is an architecture diagram of the data feedback module in the system of the present invention;

[0040] Figure 7 This is a diagram showing the connection between the AI ​​scheduling core module and the urban traffic cloud platform in the system of this invention.

[0041] Figure 8 This is a flowchart illustrating the steps of the method of the present invention;

[0042] Figure 9 This is a framework diagram for calculating recycling priority in the system of the present invention;

[0043] Figure 10 This is a screenshot of the high-frequency visualization of the system of this invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings, embodiments, experimental examples, and comparative examples. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described embodiments are merely some embodiments of the invention, and not all embodiments. The components of the embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0045] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0046] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0047] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0048] I. Implementation Examples

[0049] Example 1

[0050] This invention discloses a dynamic scheduling system and method for waste recycling vehicles based on AI algorithms, such as... Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 , Figure 7 , Figure 8 , Figure 9 , Figure 10 As shown, the specific implementation of this embodiment, combined with a waste recycling scheduling scenario in a district of Chengdu, Sichuan Province, further illustrates the present invention:

[0051] System Deployment: High-definition cameras are installed at 15 recycling stations in the area to connect to task analysis modules. All 30 new energy recycling vehicles are equipped with BDS / GPS dual-mode positioning units, resistance strain gauge load sensors, high-precision battery management systems (BMS, sampling frequency 1Hz, supporting CAN bus communication) and battery status sensors. The dispatch center is equipped with an AI dispatch core module and a communication interface with the urban traffic cloud platform. All modules achieve bidirectional interaction through a 5G data bus, and BMS data is transmitted to the vehicle status perception module in real time via the CAN bus.

[0052] Task execution: A camera at a recycling station captures images of a large amount of stacked waste. The task parsing module identifies the waste through the U-Net attention mechanism and generates a set of task parameters: "Category label: PET plastic, Loading coefficient: 0.8, Recycling priority: High (accumulation amount exceeds 24 hours), Station coordinates: XX Road XX No. (12 kilometers from the nearest vehicle)", which is then synchronized to the AI ​​scheduling core module. At this time, the status data of three nearby idle vehicles are uploaded synchronously. The BMS data of vehicle A shows "SOC: 80%, estimated range: 50 km, SOH: 92%", the BMS data of vehicle B shows "SOC: 35%, estimated range: 18 km, SOH: 85%", and the BMS data of vehicle C shows "SOC: 19%, estimated range: 8 km, SOH: 88%". The core module calculates through a multi-objective optimization model that the remaining BMS power of vehicle A can meet the requirement of "12 km round trip + 20% redundancy". Finally, vehicle A is matched, and after planning the route with information from the traffic cloud platform, the route output module pushes the instruction, and the vehicle terminal simultaneously displays "BMS range warning: below 25%, please return".

[0053] Anomaly Handling: While vehicle A was returning from its mission, the BMS detected in real time that "the temperature of a single battery cell has risen to 46℃ (over-temperature threshold 45℃)" and immediately output over-temperature warning data. After the vehicle status perception module captured the warning, it triggered the vehicle fault threshold of the anomaly handling module and transmitted the signal to the AI ​​scheduling core module within 1.2 seconds. The core module rescheduled the task using the particle swarm optimization algorithm, transferring the remaining short-distance task of vehicle A to the nearby vehicle D, and planned the path of vehicle A to the nearest charging point. At the same time, the scheduling terminal received an audible and visual alarm, and the alarm message displayed "Vehicle A: BMS single cell over-temperature (46℃), it is recommended to charge and cool down immediately". After confirmation by the management personnel, the task handover was completed.

[0054] Feedback Optimization: After the task is completed, the data feedback module collects data such as "transportation distance 26 km, total power consumption recorded by BMS: 8.5 kWh (energy consumption per unit mileage 0.33 kWh / km), carbon emission reduction 0.02 tons, and no change in vehicle A's BMS SOH", and sends it back to the core module. The core module combines this data and fine-tunes the weight of "energy consumption reduction rate" from 0.3 to 0.32. For vehicles with energy consumption per unit mileage higher than 0.35 kWh / km, subsequent short tasks within 5 km are prioritized, achieving precise optimization of the scheduling strategy.

[0055] In this embodiment, the system achieves rapid task response and efficient exception handling. Furthermore, through the precise application of BMS data, task interruptions caused by insufficient battery life are avoided. The energy consumption per unit task is reduced by 42% compared to traditional scheduling, and energy consumption is reduced by 21%.

[0056] Example 2

[0057] This invention discloses a dynamic scheduling system and method for waste recycling vehicles based on AI algorithms, such as... Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 , Figure 7 , Figure 8 , Figure 9 , Figure 10 As shown, the specific implementation method of this embodiment is: Chengdu Jinjiang District Community Recycling (Household Waste)

[0058] Scenario: 18 community stations, 8 new energy vehicles, 45 tasks per day (mainly waste paper and plastic);

[0059] Process: A site uploads an image of stacked waste. The task analysis module identifies that it contains 60% PET plastic and 40% waste paper, with a loading coefficient of 0.7 and a backlog of 2.5 tons, thus achieving a priority of Level 1 (time limit of 4 hours). The vehicle status perception module collects information on vehicle Sichuan A·XXXX5: location Chunxi Road, SOC 85%, and remaining space of 4 tons. The AI ​​scheduling core module calculates a matching degree of 93 points and improves the Dijkstra algorithm to plan the route "Chunxi Road → New Energy Dedicated Channel → Community Site → Longtansi Sorting Center".

[0060] Results: The task was completed in 3.2 hours, with an empty run rate of 7%, energy consumption of 5.1 kWh / ton, and carbon emission reduction of 0.09 tons of CO2 equivalent.

[0061] Example 3

[0062] This invention discloses a dynamic scheduling system and method for waste recycling vehicles based on AI algorithms, such as... Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 , Figure 7 , Figure 8 , Figure 9 , Figure 10 As shown, the specific implementation method of this embodiment is: recycling (industrial waste) in the Panzhihua Vanadium-Titanium Industrial Park.

[0063] Scenario: 6 industrial sites, 5 corrosion-resistant vehicles, recycling waste vanadium-titanium alloy (load capacity ≤ 5 tons);

[0064] Process: The station uploads the image of waste alloy. The task parsing module identifies the material (reflectivity 0.2), loading factor 0.6, and priority level two (time limit 8 hours); Vehicle status of Chuan D・XXXX8: SOC 78%, remaining space 5.5 tons, location at the park entrance; The AI dispatching core module combines the park traffic restriction information (Steel City Avenue is prohibited after 18:00) and plans to complete the transportation before 16:00.

[0065] Result: There is no traffic restriction delay, energy consumption is 8.0 kWh / ton, and carbon emission reduction is 0.16 tons of CO2 equivalent.

[0066] Example 4

[0067] A waste recycling vehicle dynamic scheduling system and method based on AI algorithm according to the present invention, as Figure 1 、 Figure 2 、 Figure 3 、 Figure 4 、 Figure 5 、 Figure 6 、 Figure 7 、 Figure 8 、 Figure 9 、 Figure 10 shown. The specific implementation method of this embodiment is: Scheduling during heavy rain in Yucheng District, Ya'an City (extreme scenario)

[0068] Scenario: Heavy rain causes partial blockade of G318, 12 urban and rural stations, 7 vehicles;

[0069] Process: A certain rural station uploads a waste paper task (backlog 3 tons). The exception handling module monitors the blockade of G318 → triggers recalculation; The AI dispatching core module screens the vehicle Chuan T・XXXX3 closest to the station (SOC 65%) and plans the route "Rural Road Y002 → detour route → station";

[0070] Result: Response time is 1.8 seconds, the task is completed within 4 hours, and there is only a 20-minute delay compared to the conventional route.

[0071] Example 5

[0072] A waste recycling vehicle dynamic scheduling system and method based on AI algorithm according to the present invention, as Figure 1 、 Figure 2 、 Figure 3 、 Figure 4 、 Figure 5 、 Figure 6 、 Figure 7 、 Figure 8 、 Figure 9 、 Figure 10 shown. The specific implementation method of this embodiment is: Centralized recycling during the Chengdu Universiade (emergency scenario)

[0073] Scenario: Ten stations around the Universiade Village have been temporarily assigned a task to dispose of 15 tons of waste, with five vehicles en route;

[0074] Process: The exception handling module monitors task increments → splits the task into 3 vehicles with remaining space ≥ 3 tons; the AI ​​scheduling core module plans a composite path of "Universiade Village → Temporary Transfer Point → Sorting Center";

[0075] Results: Recovery was completed in 2.5 hours, with no additional vehicles added, and the empty-running rate was reduced by 5% compared to regular dispatching.

[0076] II. Core Algorithm Technical Details

[0077] 1. Attention Mechanism U-Net Algorithm: The encoding layer adds 3 material feature convolutional kernels (3×3), and the decoding layer focuses on the waste area through attention gating units. The training uses the Adam optimizer with a learning rate of 0.001. After 50 iterations, the recognition accuracy reaches 97%.

[0078] 2. Multi-objective optimization model: Objective function Z = 0.4 × (number of completed tasks / total number of tasks) + 0.3 × (1 - actual energy consumption / baseline energy consumption) + 0.3 × (carbon emission reduction / baseline value), with constraints of "remaining power ≥ transportation mileage × unit energy consumption" and "remaining space ≥ task weight × loading coefficient".

[0079] 3. Improved Dijkstra algorithm: Path cost = distance × (1 + congestion index) + road segment weight coefficient, new energy dedicated lane weight 0.8, construction road segment weight 1.5, weight limit road segment cost set to infinity (automatic filtering).

[0080] III. Experimental Design

[0081] Controlled variables principle: All experimental cases and comparative cases use the same scenario parameters (number of recycling stations, vehicle configuration, total task volume), and the experimental period is 30 days. The interference of holidays and extreme weather (except for special experiments) is excluded. The core test indicators are unified as follows: task completion rate (percentage of completion within the time limit), vehicle empty running rate (empty running mileage / total mileage), energy consumption per unit of cargo (kWh / ton), carbon emission reduction (ton CO2 equivalent, only calculated in this patent), and abnormal response time (seconds).

[0082] Scenario selection logic: Covering four typical recycling scenarios in Sichuan and surrounding areas: densely populated urban communities (household waste), urban and rural areas at the junction of Sichuan and Chongqing (mixed waste), chemical industrial parks (hazardous waste), and extreme weather (heavy rain), matching the core advantage of the patented "multi-scenario adaptation".

[0083] Comparative selection: Four patents that are closest to existing patents were selected, representing “energy consumption optimization type”, “event-driven type”, “resource allocation type” and “cleaning-specific type” respectively, to ensure comprehensive comparison dimensions.

[0084] IV. Experimental Examples (Technical Solution of this Patent)

[0085] Experimental Example 1: Recycling in Urban Dense Communities in Wuhou District, Chengdu (mainly domestic waste)

[0086] Scenario Parameters: 22 community recycling stations (average spacing of 1.2 km), equipped with 10 new energy recycling vehicles (endurance of 350 km, cargo hold volume of 8 m³), 58 daily tasks on average, the waste is mainly composed of waste paper (40%), PET plastic (35%), and aluminum alloy (25%), and the proportion of first-priority tasks (backlog ≥ 2 tons) is 30%.

[0087] Application of Core Technologies:

[0088] Task Analysis Module: Adopt the attention mechanism U-Net algorithm, collect images through the Hikvision DS-2CD3T46WD-I5 cameras deployed at the stations, identify the reflectance characteristics of PET plastic (0.32 - 0.48), match the loading coefficient of 0.8, and calculate the priority according to "0.5 × backlog + 0.3 × decay period + 0.2 × recycling value" (for example, the decay period of the leftover food bucket is 1 day, and the priority is automatically upgraded to the first level);

[0089] Vehicle Status Sensing Module: The positioning error of the BDS / GPS dual-mode positioning unit (Huawei HiLink A200) is 2.1 meters, and the accuracy of the resistive strain type load sensor (AVIC Electro-Measurement YZC-320) is ±0.03 tons. The SOC value of vehicle Chuan A・XXXX9 is 82% and the remaining space is 5.2 m³ are collected in real time;

[0090] AI Scheduling Core Module: The matching degree calculated by the multi-objective optimization model is 94 points (adaptability 40 points + power matching degree 35 points + distance coefficient 19 points). The improved Dijkstra algorithm preferentially selects the new energy special lane on the Second Ring Road of Chengdu to avoid the congested section of Kehua Middle Road (congestion index 72%).

[0091] Experimental Data: Task completion rate of 98.3%, vehicle empty running rate of 7.2%, unit cargo energy consumption of 5.1 kWh / ton, daily average carbon emission reduction of 1.4 tons of CO2 equivalent, and abnormal response time of 1.6 seconds (recalculation triggered by a flat tire of a vehicle).

[0092] Experimental Example 2: Recycling in Urban-Rural Areas at the Sichuan-Chongqing Border in Longchang City, Neijiang (mixed waste)

[0093] Scenario Parameters: 25 recycling stations (including 16 rural stations, and the proportion of rural roads is 60%), equipped with 8 new energy vehicles, 42 daily tasks on average, the waste includes domestic waste paper (30%), crop straws (25%), waste furniture (20%), and small waste metals (25%), and the tasks at rural stations are scattered (the farthest spacing is 8 km).

[0094] Core technology applications:

[0095] Task parsing module: The attention mechanism U-Net algorithm distinguishes straw from waste furniture through image segmentation and matches a comprehensive loading coefficient of 0.7 (20% space needs to be reserved for the fluffy straw).

[0096] Vehicle status perception module: The Kalman filter algorithm filters the positioning noise on rural roads, corrects the positioning error of vehicle Chuan K·XXXX2 from the original 12 meters to 2.8 meters, and real-time feedbacks the remaining battery power of 68% and the load of 2.3 tons.

[0097] Path optimization: The improved Dijkstra algorithm avoids narrow bridges (weight limit of 5 tons), plans the route of "urban station → rural road Y008 → rural station → Longchang sorting center", and avoids the congested section of trucks on G321 national highway.

[0098] Experimental data: The task completion rate is 95.2%, the vehicle empty running rate is 9.8%, the unit cargo energy consumption is 6.2 kWh / ton, the daily average carbon emission reduction is 1.6 tons of CO2 equivalent (the carbon emission reduction of straw recycling is calculated according to the Sichuan agricultural carbon sink standard), and the abnormal response time is 1.9 seconds (temporarily adding the task of demolishing construction waste).

[0099] Experimental example 3: Recycling in the chemical industrial park of Naxi District, Luzhou City (hazardous waste)

[0100] Scenario parameters: 7 industrial recycling points (all are chemical enterprises), equipped with 5 anti-corrosion new energy vehicles (the cargo hold has corrosion-resistant coating, load capacity of 5 tons), 15 daily tasks, the waste is waste chemical barrels (60%, flammable and explosive), waste circuit boards (30%), waste steel (10%), the proportion of first-priority tasks (waste chemical barrels) is 70%, and the freight channels in the park are restricted to pass (8:00 - 18:00).

[0101] Core technology applications:

[0102] Task parsing module: The attention mechanism U-Net algorithm identifies the identification of waste chemical barrels (such as the words "flammable"), automatically sets the priority to the first level (execution time limit of 2 hours), and matches the loading coefficient of 0.6 (reserving a safety distance).

[0103] Vehicle status perception module: The battery status sensor monitors the charge and discharge efficiency in real time. When the SOC of vehicle Chuan E·XXXX5 drops to 28%, it automatically triggers a charging reminder.

[0104] Scheduling constraint: The AI scheduling core module combines the restricted travel information in the park, centrally assigns 3 waste chemical barrel tasks to be executed from 9:00 to 11:00, and the path avoids the living area of the park.

[0105] Experimental data: Task completion rate is 100%, vehicle empty running rate is 5.9%, energy consumption per ton of goods is 8.0 kWh / ton, daily average carbon emission reduction is 0.8 tons of CO2 equivalent, and abnormal response time is 1.5 seconds (2 tons of waste chemical drums are temporarily added at a recycling point).

[0106] Experimental Example 4: Recycling in Heavy Rain Weather in Yucheng District, Ya'an City (Extreme Scenario)

[0107] Scene parameters: 12 urban and rural sites, 6 new energy vehicles, 2 sections of road on G318 are waterlogged and closed due to heavy rain, and among the original 18 single tasks, 3 are of first-level priority (waste in waterlogged areas of communities, preventing secondary pollution).

[0108] Application of core technologies:

[0109] Abnormal trigger: The abnormal handling module monitors the "G318 closed" information pushed by the urban traffic cloud platform and triggers recalculation within 1.2 seconds;

[0110] Task splitting: The AI scheduling core module uses the particle swarm algorithm to split the task of vehicle Chuan T・XXXX3 originally assigned to the direction of the closed road section to the 2 idle vehicles (Chuan T・XXXX6, Chuan T・XXXX8) closest to the waterlogged community;

[0111] Route planning: Improve the Dijkstra algorithm to automatically match the detour route (Rural Road Y003 → Provincial Road S104) to avoid waterlogging points.

[0112] Experimental data: All 3 first-level tasks are completed within 1.5 hours, the overall task completion rate is 94.4%, the vehicle empty running rate only increases by 1.2% compared with normal weather, the energy consumption per ton of goods is 6.8 kWh / ton, the single carbon emission reduction is 0.5 tons of CO2 equivalent, and the abnormal response time is 1.7 seconds.

[0113] V. Comparative Example (Existing Patent Technical Solution)

[0114] Comparative Example 1: Based on Patent CN202510563201.0 (Energy Consumption Fitting Type)

[0115] Technical basis: The core is "screening vehicles by energy consumption fitting", without waste identification and priority scheduling functions, and only aiming at the lowest energy consumption.

[0116] Experimental scenario: Exactly the same as Experimental Example 1 (Community in Wuhou District, Chengdu).

[0117] Experimental data: Task completion rate is 72.4% (30% of first-level tasks are delayed, and the waste paper backlog at 2 sites exceeds 3 tons), vehicle empty running rate is 25.6% (due to the lack of matching loading factor, 2 vehicles return with overload for adjustment), energy consumption per ton of goods is 6.9 kWh / ton, there is no carbon emission reduction accounting, and abnormal response time is 14.8 seconds.

[0118] Difference analysis: Failure to identify waste items led to loading imbalance; failure to distinguish priorities led to delays in emergency tasks; empty run rate was 18.4 percentage points higher than this patent; and energy consumption was 35.3% higher.

[0119] Comparative Example 2: Based on patent CN202510767429.1 (Event-Driven)

[0120] Technical basis: The core is "natural language processing to extract event information", there is no dedicated waste identification algorithm, and it is unable to distinguish between the transportation requirements of industrial and domestic waste.

[0121] Experimental setting: exactly the same as Experiment Example 3 (Luzhou Naxi Chemical Industrial Park).

[0122] Experimental data: Task completion rate 81.3% (2 waste chemical drum tasks were delayed outside the restricted hours), vehicle empty running rate 18.7% (1 vehicle entered the restricted area of ​​the park and detoured 12 kilometers), unit cargo energy consumption 10.6 kWh / ton, no carbon emission reduction accounting, abnormal response time 7.9 seconds.

[0123] Difference analysis: The hazardous properties of waste chemical drums could not be identified, and the park's traffic restriction rules were not taken into account. The task delay rate was 18.7 percentage points higher than that of this patent, and the energy consumption was 32.5% higher.

[0124] Comparative Example 3: Based on patent CN202410620323.4 (Resource Allocation Type)

[0125] Technical basis: The core is "configuring vehicles according to the carrying capacity", without new energy vehicle power management or rural road condition adaptation functions.

[0126] Experimental scenario: completely consistent with Experiment Example 2 (Longchang urban and rural areas in Neijiang).

[0127] Experimental data: Task completion rate 77.1% (task delay rate of 58% at rural stations), vehicle empty run rate 28.3% (one vehicle detoured 20 kilometers due to failure to recognize the weight limit on a narrow bridge), energy consumption per unit of cargo 9.1 kWh / ton, no carbon emission reduction accounting, and abnormal response time 11.5 seconds.

[0128] Difference analysis: The location noise of rural roads was not filtered, the battery power was not monitored (one vehicle ran out of power in a rural area), the empty driving rate was 18.5 percentage points higher than that of this patent, and the energy consumption was 46.8% higher.

[0129] Comparative Example 4: Based on patent CN202210179369.8 (Cleaning-specific type)

[0130] Technical basis: The core is "road consumption registration triggers cleaning tasks", which is only applicable to road cleaning vehicles and has no design for waste recycling scenarios.

[0131] Experimental scenario: exactly the same as Experiment 4 (Ya'an rain city rainstorm weather).

[0132] Experimental data: Task completion rate 72.2% (5 tasks were delayed due to route planning not avoiding water accumulation points), vehicle empty run rate 26.4% (3 vehicles repeatedly went to the same area), energy consumption per unit of cargo 9.7 kWh / ton, no carbon emission reduction accounting, and abnormal response time 13.2 seconds.

[0133] Difference Analysis: It cannot cope with sudden road conditions such as rainstorms and flooding, the route planning has no detour dynamic adjustment function, the task completion rate is 22.2 percentage points lower than that of this patent, and the energy consumption is 42.6% higher.

[0134] VI. Summary Table of Core Performance Comparison

[0135] Table 1: Comparison of core indicators between experimental and comparative examples

[0136]

[0137] VII. Experimental Conclusions

[0138] Creative Support: This patent solves the core defect of existing patents' "generalized application of general algorithms" by combining the technical features of "attention mechanism U-Net waste identification + multi-objective optimization scheduling + carbon footprint accounting", achieving breakthrough improvements in task completion rate (increased by 15%-28%), empty running rate (reduced by 8%-22%), and energy consumption (reduced by 31%-47%).

[0139] Scenario adaptability: Special optimizations for specific scenarios such as Sichuan chemical industrial parks and urban and rural areas in Sichuan and Chongqing have improved the adaptability of the scheduling scheme by more than 40% compared with existing patents.

[0140] Sufficiency of disclosure: The sensor model (Huawei HiLink A200 positioning unit), algorithm parameters (U-Net attention mechanism iterated for 50 rounds), and local standards (Sichuan renewable resource carbon footprint specification) were clearly specified in the experiment.

[0141] Policy alignment: The accurate carbon emission reduction calculation function (error ≤ 5%) is adapted to the "dual carbon" target.

[0142] The above description is only a preferred embodiment, experimental example, and comparative example of the present invention, and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A dynamic scheduling system for waste recycling vehicles based on AI algorithms, characterized in that, It includes a task parsing module, a vehicle status perception module, an AI scheduling core module, a path output module, and a data feedback module. Each module achieves bidirectional data interaction through a data bus. The task parsing module identifies waste item categories using the U-Net image segmentation algorithm with an attention mechanism, and generates a task parameter set containing category labels, loading coefficients, recycling priorities, and site coordinates. The recycling priority is calculated by weighting the decay cycle, backlog amount, and recycling value. The vehicle status perception module uses a Kalman filter algorithm to fuse GPS positioning data, load sensor data and battery management system data to output the vehicle's current location, remaining battery power, remaining loading space and real-time load. The AI ​​scheduling core module communicates with the first two modules respectively. It has a built-in multi-objective optimization model with an objective function of "0.4×task completion rate + 0.3×energy consumption reduction rate + 0.3×carbon emission reduction". It outputs vehicle matching results and route planning instructions. The path output module pushes the path planning instructions and vehicle matching results generated by the AI ​​scheduling core module to the corresponding vehicle terminal. The data feedback module collects all data during the task execution process and uses it to dynamically update the parameters of the multi-objective optimization model.

2. The dynamic scheduling system for waste recycling vehicles based on AI algorithms according to claim 1, characterized in that, The attention mechanism of the task parsing module, U-Net image segmentation algorithm, adds a waste material feature extraction channel to the coding layer of the U-Net basic network. This channel identifies the difference in reflectivity between plastic and metal materials to optimize the classification results. The algorithm has an accuracy of no less than 97% in identifying common waste items.

3. The dynamic scheduling system for waste recycling vehicles based on AI algorithms according to claim 1, characterized in that, The vehicle status perception module consists of a BDS / GPS dual-mode positioning unit, a resistance strain gauge load cell, and a battery status sensor; the positioning error of the dual-mode positioning unit is ≤3 meters, the measurement accuracy of the resistance strain gauge load cell is ±0.05 tons, and the battery status sensor collects battery SOC value and charge / discharge efficiency data in real time.

4. The dynamic scheduling system for waste recycling vehicles based on AI algorithms according to claim 1, characterized in that, The path planning function of the AI ​​scheduling core module is implemented using an improved Dijkstra algorithm. This algorithm adds three constraints: the weight coefficient of the new energy dedicated channel is set to 0.8, the weight coefficient of the construction section is set to 1.5, and the weight-restricted section is directly filtered. The overall path calculation time is ≤1 second.

5. A dynamic scheduling system for waste recycling vehicles based on AI algorithms according to claim 1, characterized in that, It also includes an anomaly handling module, which communicates with the AI ​​scheduling core module and sets thresholds for three types of triggers: vehicle malfunction, task increment, and road closure. When the monitored data reaches any of the thresholds, an anomaly response is triggered. The AI ​​scheduling core module recalculates the scheduling instructions using the particle swarm optimization algorithm, and the anomaly response time is ≤2 seconds.

6. The dynamic scheduling system for waste recycling vehicles based on AI algorithms according to claim 1, characterized in that, The data feedback module integrates a carbon footprint accounting unit, which adopts the Sichuan Provincial Local Standard "Specification for Carbon Footprint Accounting of Recyclable Resources". The carbon emission reduction data is obtained through the calculation formula "transportation mileage × unit energy consumption carbon emission coefficient - waste recycling carbon emission reduction benchmark value".

7. A dynamic scheduling system for waste recycling vehicles based on AI algorithms according to claim 1, characterized in that, The training sample library of the U-Net image segmentation algorithm with the attention mechanism contains 120,000 images of common waste in Sichuan Province. The samples cover scenes with different lighting intensities and different stacking states, and the sample annotation accuracy is ≥99%.

8. A dynamic scheduling system for waste recycling vehicles based on AI algorithms according to claim 5, characterized in that, The specific thresholds for triggering vehicle malfunctions include battery SOC ≤ 20%, load sensor data fluctuation > 10%, and GPS positioning drift distance > 10 meters. After triggering the thresholds for vehicle malfunctions, the anomaly handling module automatically sends an audible and visual alarm to the dispatch terminal.

9. A dynamic scheduling system for waste recycling vehicles based on AI algorithms according to claim 1, characterized in that, The AI ​​scheduling core module establishes real-time communication with the city traffic cloud platform to continuously acquire road congestion index and temporary traffic control information, with a data update frequency of 5 seconds per update.

10. A dynamic scheduling method for waste recycling vehicles based on AI algorithms, characterized in that, Applied to the system described in any one of claims 1-9, the method achieves closed-loop scheduling through the linkage of various modules, and specifically includes the following steps: Step 1: The task parsing module collects images of waste at the recycling station through the camera, identifies the item type through the U-Net image segmentation algorithm with attention mechanism, and generates a task parameter set containing category tags, loading coefficients, recycling priorities and station coordinates. This set is then transmitted to the AI ​​scheduling core module via the data bus. Step 2: The vehicle status perception module collects positioning, load, and battery data through various sensors. After fusion processing by the Kalman filter algorithm, it outputs standardized vehicle current location and remaining battery status data, which are synchronized to the AI ​​scheduling core module through the data bus. Step 3: The AI ​​scheduling core module combines the received task parameter set and vehicle status data, calls the built-in multi-objective optimization model to calculate the vehicle matching result, completes path planning through the improved Dijkstra algorithm, and generates scheduling instructions; Step 4: The path output module receives the scheduling instructions from the AI ​​scheduling core module and pushes them to the corresponding vehicle terminal. The vehicle then executes the retrieval task according to the instructions. Step 5: The data feedback module collects task execution data and vehicle status change data throughout the process, calculates carbon emission reduction, and then sends the data back to the AI ​​scheduling core module via the data bus to update the parameters of the multi-objective optimization model and form a scheduling closed loop.

Images