CN-121981359-A - Intelligent garbage dynamic clearing and dispatching method and system based on recycle stream optimization

CN121981359ACN 121981359 ACN121981359 ACN 121981359ACN-121981359-A

Abstract

The invention discloses an intelligent garbage dynamic clearing and dispatching method and system based on recycling logistics optimization, which belong to the technical field of garbage transportation and dispatching, and specifically comprise the steps of characteristic splicing full-load rate data, opening and closing frequency data and people stream density distribution data, constructing a node characteristic sequence with a time stamp, constructing a space topology adjacency matrix based on the space geographic position of each garbage delivery terminal, and fusing to generate space-time diagram data; the method comprises the steps of inputting space-time diagram data into a space-time diagram convolution network, extracting area topology association features and time sequence dependence features, outputting a capacity evolution prediction curve of a rubbish delivery terminal, calculating overflow risk indexes of the corresponding rubbish delivery terminal according to the capacity evolution prediction curve, generating a clearing task node set, combining the clearing task node set, current position coordinates of a clearing vehicle and dynamic residual load of the clearing vehicle, and generating a multi-vehicle collaborative dynamic clearing route map through a deep reinforcement learning model.

Inventors

JI RAN
ZhuGe Yingbei
HONG XINYI
QIU YI
LIAO ZIHAN
LIN GUANGYU
LIU GUANRUI
LIU FANGFEI
Shen Zining

Assignees

厦门大学嘉庚学院

Dates

Publication Date: 20260505
Application Date: 20260408

Claims (9)

1. The intelligent garbage dynamic clearance scheduling method based on recycle stream optimization is characterized by comprising the following steps of: s1, collecting full rate data, opening and closing frequency data and people flow density distribution data of an area where the full rate data, the opening and closing frequency data and the people flow density distribution data of a garbage delivery terminal are located; S2, carrying out characteristic splicing on full-load rate data, switching frequency data and people stream density distribution data, constructing a node characteristic sequence with a time stamp, constructing a space topology adjacency matrix based on the space geographic position of each rubbish delivery terminal, and fusing to generate space-time diagram data; S3, inputting the space-time diagram data into a space-time diagram convolution network to extract area topology association features and time sequence dependency features, and outputting a capacity evolution prediction curve of the rubbish delivery terminal; s4, calculating a full risk index of the corresponding rubbish delivery terminal according to the capacity evolution prediction curve, and extracting nodes exceeding a preset risk threshold according to the full risk index to generate a clearing task node set; S5, combining the clearing task node set, the current position coordinates of the clearing vehicles and the dynamic residual load of the clearing vehicles, and generating a multi-vehicle collaborative dynamic clearing route map through a deep reinforcement learning model; S6, converting the dynamic clearing route map into a navigation instruction, and sending the navigation instruction to a vehicle-mounted operation terminal of the clearing vehicle, acquiring feedback task state data of the vehicle-mounted operation terminal to update strategy parameters of the deep reinforcement learning model, and simultaneously collecting actual capacity data of the rubbish delivery terminal to update network parameters of the space-time diagram convolutional network.
2. The method for dynamically clearing and dispatching intelligent garbage based on recycle stream optimization according to claim 1, wherein in the step S2, the characteristic splicing of full-load rate data, switching frequency data and people stream density distribution data is performed, and the process of constructing a node characteristic sequence with a time stamp is as follows: time alignment is carried out on full rate data, switching frequency data and people stream density distribution data of the garbage delivery terminal according to the unified sampling moment; Performing characteristic splicing on the time-aligned full-load rate data, the switching frequency data and the people stream density distribution data to generate a characteristic vector corresponding to the sampling moment; And adding the time stamp corresponding to the sampling time to the feature vector, arranging the feature vectors in time sequence, and constructing a node feature sequence with the time stamp.
3. The intelligent garbage dynamic clearance scheduling method based on recycle stream optimization according to claim 1, wherein the process of constructing a space topology adjacency matrix based on the space geographic position of each garbage delivery terminal and fusing to generate space-time diagram data is as follows: The method comprises the steps of obtaining the spatial geographic position of each rubbish delivery terminal, and calculating the geographic distance between the rubbish delivery terminals through a coordinate analysis algorithm; calculating the space weight of each rubbish delivery terminal according to the geographic distance, and arranging the space weights according to the node numbers to construct a space topology adjacency matrix; And mapping the node characteristic sequence with the time stamp to the nodes of the space topology adjacency matrix, and fusing the node characteristics and the topology structure to generate space-time diagram data.
4. The method for dynamically clearing and scheduling intelligent garbage based on recycle stream optimization according to claim 1, wherein in the step S3, the process of inputting space-time diagram data into a space-time diagram convolution network to extract area topology association features and time sequence dependency features and outputting a capacity evolution prediction curve of a garbage delivery terminal is as follows: Inputting the space-time diagram data into a space diagram convolution layer provided with a learning weight matrix, aggregating the characteristic data of adjacent rubbish delivery terminals according to the learning weight matrix, and extracting the topological association characteristics of the region; inputting the region topology association features into a time convolution layer provided with a time self-attention mechanism, executing expansion convolution calculation along the direction of a time stamp sequence, and extracting time sequence dependency features; and inputting the time sequence dependent characteristics into a full-connection output layer, performing dimension reduction mapping on the time sequence dependent characteristics by using a linear transformation matrix of the full-connection output layer, and outputting a capacity evolution prediction curve of the garbage delivery terminal.
5. The method for intelligent garbage dynamic clearance scheduling based on recycle stream optimization according to claim 1, wherein in the step S4, the process of calculating the overflow risk index of the corresponding garbage delivery terminal according to the capacity evolution prediction curve, extracting the nodes exceeding the preset risk threshold according to the overflow risk index, and generating the clearance task node set is as follows: Calculating a first derivative of the capacity evolution prediction curve in a preset future time window to obtain a capacity increase rate, and executing a constant integral operation on the capacity evolution prediction curve in the preset future time window to obtain an accumulated capacity load; Inputting the capacity growth rate and the accumulated capacity load into a preset risk assessment function, performing exponential mapping on the capacity growth rate by using the risk assessment function, carrying out weighted summation on the capacity growth rate and the accumulated capacity load, and outputting a full risk overflow index corresponding to the garbage delivery terminal; comparing the overflow risk index of the corresponding rubbish delivery terminal with a preset risk threshold value in a numerical value mode, judging the rubbish delivery terminal with the overflow risk index larger than the preset risk threshold value, and extracting the rubbish delivery terminal as a node exceeding the preset risk threshold value; And arranging the extracted nodes exceeding the preset risk threshold in a descending order according to the corresponding overflow risk indexes, and combining the nodes exceeding the preset risk threshold according to the descending order to generate a clearing task node set.
6. The method for scheduling dynamic garbage collection and transportation based on recycle stream optimization according to claim 1, wherein in step S5, the process of generating a multi-vehicle collaborative dynamic transportation route map by combining the transportation task node set, the current position coordinates of the transportation vehicle and the dynamic residual load of the transportation vehicle through a deep reinforcement learning model is as follows: Splicing the clearing task node set, the current position coordinates of the clearing vehicle and the dynamic residual load of the clearing vehicle to construct an environment state matrix, and inputting the environment state matrix into a deep reinforcement learning model; performing mapping calculation on the environment state matrix by using the deep reinforcement learning model, and outputting an execution action vector of an unassigned target node in the collection of the clearing task nodes of each clearing vehicle; updating the current position coordinates and dynamic residual load of the clearing vehicle according to the execution motion vector, adding the target node to a path sequence corresponding to the clearing vehicle, and eliminating the allocated node; And repeatedly updating the environment state matrix and inputting the environment state matrix into the deep reinforcement learning model until the clearing task node set is empty, and summarizing the path sequence of each clearing vehicle to generate a multi-vehicle collaborative dynamic clearing route map.
7. The method for dynamically clearing and dispatching intelligent garbage based on recycle stream optimization according to claim 6, wherein the process of repeatedly updating the environment state matrix and inputting the environment state matrix into the deep reinforcement learning model until clearing task node sets are empty and summarizing path sequences of all clearing vehicles to generate a multi-vehicle collaborative dynamic clearing route map is as follows: Judging that unassigned target nodes exist in the clearing task node set, and recombining the updated current position coordinates and dynamic residual load to construct an environment state matrix, and inputting the environment state matrix into a deep reinforcement learning model; stopping updating the environment state matrix when the unassigned target nodes of the clearing task node set are completely removed, and calling the complete path sequences corresponding to all clearing vehicles; And carrying out space mapping on the fetched complete path sequence of each clearing vehicle, and generating a multi-vehicle collaborative dynamic clearing route map according to the track coordinate combination of the space mapping.
8. The method for intelligent garbage dynamic clearance scheduling based on recycle stream optimization according to claim 1, wherein in the step S6, the process of acquiring feedback task state data of a vehicle-mounted operation terminal to update policy parameters of the deep reinforcement learning model and simultaneously collecting actual capacity data of a garbage delivery terminal to update network parameters of the space-time diagram convolutional network is as follows: Receiving feedback task state data uploaded by a vehicle-mounted operation terminal, acquiring task time consumption, calculating a rewarding value, and synchronously collecting actual capacity data acquired by a rubbish delivery terminal at a corresponding time node; Inputting the rewarding numerical value into the deep reinforcement learning model to calculate a strategy gradient matrix, and executing back propagation by using the strategy gradient matrix to update strategy parameters of the deep reinforcement learning model; And calculating the deviation of the actual capacity data and the predicted value of the capacity evolution prediction curve to construct a loss function, and executing the network parameters of the back propagation update time-space diagram convolution network according to the loss function.
9. An intelligent garbage dynamic clearance scheduling system based on recycle stream optimization, which is characterized in that the intelligent garbage dynamic clearance scheduling method based on recycle stream optimization is implemented in any one of claims 1-8, and is characterized by comprising the following steps: The data acquisition module is used for acquiring full rate data, switching frequency data and people flow density distribution data of the area where the garbage delivery terminal is positioned; the data fusion module is used for carrying out characteristic splicing on full-load rate data, switching frequency data and people stream density distribution data, constructing a node characteristic sequence with a time stamp, constructing a space topology adjacency matrix based on the space geographic position of each rubbish delivery terminal, and fusing to generate space-time diagram data; The capacity prediction module is used for inputting the space-time diagram data into a space-time diagram convolution network to extract the region topology association characteristic and the time sequence dependence characteristic and outputting a capacity evolution prediction curve of the rubbish delivery terminal; The task generation module is used for calculating the overflow risk index of the corresponding rubbish delivery terminal according to the capacity evolution prediction curve, extracting nodes exceeding a preset risk threshold according to the overflow risk index, and generating a clearing task node set; the route planning module is used for generating a multi-vehicle collaborative dynamic clearing route map through the deep reinforcement learning model by combining the clearing task node set, the current position coordinates of the clearing vehicles and the dynamic residual load of the clearing vehicles; the closed loop feedback module is used for converting the dynamic clearing route map into a navigation instruction and sending the navigation instruction to the vehicle-mounted operation terminal of the clearing vehicle, acquiring feedback task state data of the vehicle-mounted operation terminal to update the strategy parameters of the deep reinforcement learning model, and simultaneously collecting actual capacity data of the garbage delivery terminal to update the network parameters of the space-time diagram convolutional network.

Description

Intelligent garbage dynamic clearing and dispatching method and system based on recycle stream optimization Technical Field The invention relates to the technical field of garbage transportation scheduling, in particular to an intelligent garbage dynamic clearing and scheduling method and system based on recycle stream optimization. Background The continuous deepening of smart city construction and the wide popularization of the Internet of things technology gradually introduce informationized management means in the fields of urban public health and environmental infrastructures, and especially in the field of urban garbage recycling and clearing logistics, the informationized technology becomes an important support for optimizing transportation capacity configuration and improving scheduling efficiency. In the conventional urban solid waste management and garbage collection and clearance process, the management department usually relies on a pre-planned regional grid, a fixed clearance logistics route and a clearance fleet with regular shifts to execute daily garbage collection and clearance logistics operation. In order to assist the clearing logistics process, the partial area is provided with basic Internet of things sensing equipment at the garbage collection terminal, the basic physical state of the container is monitored in real time, and the state signals are summarized to the dispatching center through the communication network, so that basic data support is provided for clearing logistics dispatching. The operation mode based on regular inspection, fixed-point clearing, fixed-route scheduling and basic data acquisition forms a basic framework of the current urban area garbage recycling and clearing logistics system, maintains basic operation of urban environmental sanitation to a certain extent, but is increasingly prominent in terms of logistics capacity optimization and path dynamic adjustment. However, the existing garbage collection and clearance logistics scheduling mechanism generally adopts a fixed clearance route and cycle duration, and cannot deeply analyze dynamic nonlinear characteristics of garbage generation rates under different physical space and time dimensions. The static logistics scheduling management mode causes serious unbalance of the supply and demand of the transportation capacity, the collection terminal in the high-frequency delivery area is easy to overflow and stay, and the low-frequency delivery area frequently generates ineffective transportation, so that ineffective consumption of transportation resources of the transportation logistics and rapid increase of operation and maintenance cost are caused. In addition, the current system usually only depends on a single threshold trigger sensor in the link of bottom data acquisition and analysis, lacks the deep fusion processing capability of multidimensional heterogeneous data, and fails to construct a capacity evolution prediction model with foresight. The system cannot conduct dynamic self-adaptive dispatch and logistics path real-time planning of the clearance tasks based on the global logistics view angle, and efficient circulation and refined logistics scheduling management of the garbage recycling full link under the urban complex scene are difficult to support. Disclosure of Invention The invention aims to provide an intelligent garbage dynamic clearing and dispatching method and system based on recycle stream optimization, which solve the problems in the background technology: The aim of the invention can be achieved by the following technical scheme: An intelligent garbage dynamic clearance scheduling method based on recycle stream optimization comprises the following steps: s1, collecting full rate data, opening and closing frequency data and people flow density distribution data of an area where the full rate data, the opening and closing frequency data and the people flow density distribution data of a garbage delivery terminal are located; S2, carrying out characteristic splicing on full-load rate data, switching frequency data and people stream density distribution data, constructing a node characteristic sequence with a time stamp, constructing a space topology adjacency matrix based on the space geographic position of each rubbish delivery terminal, and fusing to generate space-time diagram data; S3, inputting the space-time diagram data into a space-time diagram convolution network to extract area topology association features and time sequence dependency features, and outputting a capacity evolution prediction curve of the rubbish delivery terminal; s4, calculating a full risk index of the corresponding rubbish delivery terminal according to the capacity evolution prediction curve, and extracting nodes exceeding a preset risk threshold according to the full risk index to generate a clearing task node set; S5, combining the clearing task node set, the current position coordinates of the clearing vehicles