Logistics truck multi-mode transportation scheduling method and system based on internet of things

Abstract

This invention provides a multi-modal transportation scheduling method and system for logistics freight vehicles based on the Internet of Things (IoT), belonging to the field of transportation scheduling technology. The method includes: obtaining multiple multi-source real-time datasets through IoT terminals; calculating the dynamic health of each dump truck; performing state analysis of multiple dump trucks; performing group fairness enhancement training on a pre-trained scheduling model; inputting multiple health labels, a resource state mapping table, and the enhanced scheduling model into a multi-objective optimizer to generate a truck-cargo pairing scheme and an unloading point allocation scheme, and then executing transportation scheduling. This invention solves the technical problems of existing technologies where logistics freight vehicle transportation scheduling relies on static rules and lacks real-time perception and fairness constraints, leading to low resource utilization and uneven load distribution among vehicle groups. It achieves the technical effect of improving resource utilization and realizing fair group scheduling of logistics freight vehicles through real-time perception via IoT and the introduction of fairness enhancement.

Description

Technical Field

[0001] This invention relates to the field of transportation scheduling technology, specifically to a method and system for multi-modal transportation scheduling of logistics freight vehicles based on the Internet of Things. Background Technology

[0002] Multimodal transportation scheduling for logistics freight vehicles is widely used in scenarios such as mining transportation, construction waste transportation, and port collection and distribution. Its core task is to coordinate and allocate multiple dump trucks, cargo boxes, and loading and unloading equipment to achieve efficient transfer of goods from loading points to unloading points. In actual operation, vehicles need to frequently switch between different loading and unloading points according to task requirements and coordinate scheduling among multiple transportation modes to ensure transportation continuity and operational efficiency. In existing technologies, scheduling management is usually carried out using manual planning or rule-driven methods. By pre-setting transportation routes, fixed vehicle allocation strategies, or simple priority rules, the matching of trucks and cargo boxes and task allocation are achieved. At the same time, some technologies introduce basic optimization algorithms to perform static optimization calculations based on vehicle location, transportation distance, and task urgency to improve overall transportation efficiency.

[0003] However, the existing scheduling methods mentioned above lack real-time perception of vehicle operating status, cargo box operation status, and loading and unloading equipment occupancy, and cannot reflect the dynamically changing operating environment. This results in scheduling decisions lagging behind the actual operating status. Furthermore, the lack of constraints on load differences among different vehicle groups and long-term fairness of use leads to some vehicles operating at high loads for extended periods while others have low utilization rates. Consequently, overall transportation resources are not fully utilized, scheduling strategies are difficult to dynamically adjust with the actual operating status, and overall scheduling efficiency is limited.

[0004] In summary, existing technologies suffer from the technical problems of relying on static rules for logistics freight vehicle transportation scheduling and lacking real-time perception and fair constraints, resulting in low resource utilization and uneven load distribution among vehicle groups. Summary of the Invention

[0005] The purpose of this application is to provide a multi-modal transportation scheduling method and system for logistics freight vehicles based on the Internet of Things, in order to solve the technical problems of existing technologies, such as logistics freight vehicle transportation scheduling relying on static rules and lacking real-time perception and fair constraints, resulting in low resource utilization and uneven load distribution among vehicle groups.

[0006] In view of the above problems, this application provides a method and system for multi-modal transportation scheduling of logistics freight vehicles based on the Internet of Things.

[0007] The first aspect of this application provides a multi-modal transportation scheduling method for logistics freight vehicles based on the Internet of Things (IoT). The method includes: collecting real-time data on the unloading load, cab status, cargo box status, and loading / unloading equipment status of multiple dump trucks via IoT terminals to obtain multiple multi-source real-time datasets; calculating the dynamic health of each dump truck based on the multiple multi-source real-time datasets to obtain multiple health labels; performing state analysis on the multiple dump trucks based on the multiple multi-source real-time datasets to construct a resource state mapping table; dividing the multiple dump trucks into multiple vehicle groups based on their cumulative unloading load, and performing group fairness enhancement training on a pre-trained scheduling model based on the multiple multi-source real-time datasets and the resource state mapping table to obtain an enhanced scheduling model; inputting the multiple health labels, the resource state mapping table, and the enhanced scheduling model into a multi-objective optimizer to generate a cab-cargo box pairing scheme and an unloading point allocation scheme, and executing transportation scheduling.

[0008] Optionally, GPS positioning modules, speed sensors, and CAN bus readers are installed on the tractor units of multiple dump trucks to collect real-time location, speed, cumulative running time, and fault codes of the corresponding vehicles, obtaining multiple tractor unit status data. Load sensors, accelerometers, and hydraulic cylinder pressure and temperature sensors are installed on the dump trailers of multiple dump trucks to collect real-time load, number of dump lifting operations, cylinder pressure, and cylinder temperature of the corresponding vehicles, obtaining multiple dump load data and multiple cargo box status data. Equipment controller interface readers are installed on the loading and unloading equipment corresponding to multiple dump trucks to collect the busy / idle status and fault status of each loading and unloading equipment, obtaining multiple loading and unloading equipment status data. The multiple tractor unit status data, multiple dump load data, multiple cargo box status data, and multiple loading and unloading equipment status data are aggregated with the dump trucks to obtain multiple multi-source real-time datasets.

[0009] Optionally, the number of self-unloading lifts is read from multiple self-unloading load data, and the wear increment is calculated by reading the load mass lifted in each self-unloading lift action. After accumulation, multiple wear increments are obtained. The cumulative running time is read from multiple vehicle head status data from multiple multi-source real-time datasets. For every hour of running time read, the running wear is increased by a fixed step size to obtain multiple running wear amounts. The multiple wear increments and the multiple running wear amounts are weighted to obtain multiple health decay degrees. The initial health degree is subtracted from the multiple health decay degrees to obtain multiple dynamic health degrees. The dynamic health degree is used as the vehicle's health label to obtain multiple health labels.

[0010] Optionally, based on multiple vehicle head status data collected from multiple multi-source real-time datasets, status labels are established for the vehicle heads of multiple dump trucks, resulting in multiple vehicle head status labels; based on multiple cargo box status data, status labels are established for the cargo boxes of multiple dump trucks, resulting in multiple cargo box status labels; based on multiple loading and unloading equipment status data, status labels are established for each loading and unloading equipment, resulting in multiple loading and unloading equipment status labels; and a resource status mapping table is constructed based on the multiple vehicle head status labels, multiple cargo box status labels, and multiple loading and unloading equipment status labels.

[0011] Optionally, the vehicle head status label includes three states: idle, in operation, and under maintenance; the cargo box status label includes four states: idle, loading, fully loaded and awaiting towing, and unloading; and the loading and unloading equipment status label includes three states: idle, occupied, and faulty.

[0012] Optionally, extract the locomotives with the idle label from the multiple locomotive status labels to obtain the left node set; extract the cargo boxes with the idle label or fully loaded and waiting to be towed from the multiple locomotive status labels to obtain the right node set; if the unloading directions of the locomotive and the cargo box are compatible and the traction capacity of the locomotive is greater than or equal to the load capacity of the cargo box, then establish an edge between the left and right nodes, use the Hungarian algorithm or greedy algorithm to solve the minimum cost matching, and store the matching result together with the status label of each resource in the resource status mapping table.

[0013] Optionally, multiple cumulative dump loads are extracted from multiple dump load data; the multiple dump trucks are clustered according to their magnitude to divide them into multiple vehicle groups; scheduling efficiency loss analysis is performed on the multiple vehicle groups based on the multiple multi-source real-time datasets to obtain the average scheduling loss of the multiple vehicle groups; the vehicle group corresponding to the minimum average scheduling loss of the multiple vehicle groups is taken as the optimal group, and the difference between the average scheduling loss of the multiple vehicle groups and the minimum average scheduling loss of the multiple vehicle groups is calculated as the fairness gap of the multiple groups; based on the fairness gap of the multiple groups and the resource state mapping table, the pre-trained scheduling model is trained to enhance group fairness to obtain an enhanced scheduling model.

[0014] Optionally, based on the multiple group fairness gap and resource status mapping table, difficult negative samples and simple negative samples are constructed; training data of the pre-trained scheduling model is obtained, and difficult negative samples and simple negative samples are merged to obtain an updated training dataset; the pre-trained scheduling model is trained to enhance group fairness using the updated training dataset to obtain an enhanced scheduling model.

[0015] Optionally, all vehicle heads, cargo boxes, and loading / unloading equipment with idle status labels are read from the resource status mapping table to obtain a candidate resource set. For vehicle groups with a group fairness gap greater than or equal to a preset gap threshold among multiple group fairness gaps, resource samples with a matching cost greater than or equal to a preset cost threshold that are expected to match the current scheduling strategy are selected from the candidate resource set, and the selected resource samples are used as hard negative samples. For vehicle groups with a group fairness gap less than a preset gap threshold among multiple group fairness gaps, resource samples with a matching cost less than a preset cost threshold are selected from the candidate resource set, and the selected resource samples are used as simple negative samples.

[0016] The second aspect of this application provides an IoT-based multi-mode transportation scheduling system for logistics freight vehicles. The system includes: a data acquisition module for real-time acquisition of unloading load data, truck head status data, cargo box status data, and loading / unloading equipment status data from multiple dump trucks via IoT terminals, obtaining multiple multi-source real-time datasets; a health calculation module for calculating the dynamic health of each dump truck based on the multiple multi-source real-time datasets, obtaining multiple health labels; a status parsing module for performing status parsing of multiple dump trucks based on the multiple multi-source real-time datasets, constructing a resource status mapping table; a scheduling model acquisition module for dividing the multiple dump trucks into multiple vehicle groups based on their cumulative unloading load, and performing group fairness enhancement training on a pre-trained scheduling model based on the multiple multi-source real-time datasets and the resource status mapping table, obtaining an enhanced scheduling model; and a scheme generation module for inputting the multiple health labels, the resource status mapping table, and the enhanced scheduling model into a multi-objective optimizer to generate a truck head-cargo box pairing scheme and an unloading point allocation scheme, and executing transportation scheduling.

[0017] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0018] The method provided in this application collects real-time data on the unloading load, cab status, cargo box status, and loading / unloading equipment status of multiple dump trucks via an IoT terminal, obtaining multiple multi-source real-time datasets. Based on these datasets, the dynamic health of each dump truck is calculated, resulting in multiple health labels. State analysis of the multiple dump trucks is performed, constructing a resource status mapping table. The trucks are divided into multiple groups based on their cumulative unloading load, and a pre-trained scheduling model is trained to enhance group fairness using the datasets and resource status mapping table, resulting in an enhanced scheduling model. The health labels, resource status mapping table, and enhanced scheduling model are input into a multi-objective optimizer to generate cab-cargo box pairing schemes and unloading point allocation schemes, and transportation scheduling is executed. This achieves the technical effect of improving the utilization rate of logistics freight vehicle scheduling resources and realizing fair group scheduling through real-time IoT sensing and the introduction of fairness enhancement.

[0019] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0021] Figure 1 A flowchart illustrating the IoT-based multi-mode transportation scheduling method for logistics freight vehicles provided in this application.

[0022] Figure 2 A schematic diagram of the structure of the IoT-based multi-modal transportation scheduling system for logistics freight vehicles provided in this application.

[0023] Figure labeling: Data acquisition module 11, health calculation module 12, status parsing module 13, scheduling model acquisition module 14, scheme generation module 15. Detailed Implementation

[0024] This application provides a multi-modal transportation scheduling method and system for logistics freight vehicles based on the Internet of Things (IoT). It addresses the technical problems of existing technologies where logistics freight vehicle transportation scheduling relies on static rules and lacks real-time perception and fairness constraints, leading to low resource utilization and uneven load distribution among vehicle groups. The method achieves the technical effect of improving resource utilization and realizing fair scheduling of logistics freight vehicles through real-time perception via IoT and the introduction of fairness enhancement.

[0025] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0026] Example 1, as Figure 1 As shown, this application provides a multi-modal transportation scheduling method for logistics freight vehicles based on the Internet of Things (IoT). The IoT-based multi-modal transportation scheduling method for logistics freight vehicles includes: By collecting real-time data on the unloading load, cab status, cargo box status, and loading / unloading equipment status of multiple dump trucks through IoT terminals, multiple multi-source real-time datasets are obtained.

[0027] Furthermore, by using IoT terminals to collect real-time data on the unloading load, cab status, cargo box status, and loading / unloading equipment status of multiple dump trucks, multiple multi-source real-time datasets are obtained. These include: installing GPS positioning modules, speed sensors, and CAN bus readers on the tractor units of multiple dump trucks to collect real-time location, speed, cumulative running time, and fault codes, thus obtaining multiple cab status data; installing load sensors, accelerometers, and hydraulic cylinder pressure and temperature sensors on the unloading trailers of multiple dump trucks to collect real-time load, unloading lifting times, cylinder pressure, and cylinder temperature, thus obtaining multiple unloading load data and multiple cargo box status data; installing equipment controller interface readers on the loading / unloading equipment corresponding to multiple dump trucks to collect the busy / idle status and fault status of each loading / unloading device, thus obtaining multiple loading / unloading equipment status data; and summarizing the multiple cab status data, multiple unloading load data, multiple cargo box status data, and multiple loading / unloading equipment status data across the dump trucks to obtain multiple multi-source real-time datasets.

[0028] Specifically, an onboard IoT data acquisition terminal is deployed on each dump truck. This terminal includes multiple sensor interfaces for synchronously collecting and time-aligning the status of different physical objects during vehicle operation. Specifically, a GPS positioning module, speed sensor, and CAN bus reader are installed on the tractor unit. The GPS positioning module utilizes Global Positioning System (GPS) technology to accurately acquire the vehicle's real-time location coordinates, including longitude and latitude, and records them as continuous trajectory data using timestamps. The speed sensor collects the vehicle's instantaneous speed, and the CAN bus reader reads data from the vehicle's internal CAN bus. The CAN bus is a communication network within the vehicle, through which the vehicle's cumulative operating time (total driving time since the vehicle was put into use) and fault codes are obtained. Fault codes are specific codes generated by the vehicle's electronic control system when a fault is detected, indicating the type and location of the fault. Based on the real-time location, speed, cumulative operating time, and fault codes collected by the GPS positioning module, speed sensor, and CAN bus reader, multiple tractor unit status data sets are constructed.

[0029] Load sensors, accelerometers, and hydraulic cylinder pressure and temperature sensors are installed on the dump trailers of multiple dump trucks to collect data. The load sensors, based on strain gauge or piezoelectric measurement principles, deform when the trailer bears the weight of cargo, generating an electrical signal. This signal is then converted into readable real-time load data, expressed in kilograms or tons, by a signal processing circuit. The accelerometers measure the changes in the trailer's motion during the dumping process. By identifying the number of vertical acceleration changes, the number of dumping operations—the number of complete lifting and lowering cycles—is determined. The hydraulic cylinder pressure and temperature sensors collect data on the pressure and temperature within the hydraulic cylinders during the dumping process. Pressure reflects the cylinder's workload, while temperature helps monitor its operational status and prevent damage due to overheating. Based on data collected from load sensors, accelerometers, and hydraulic cylinder pressure and temperature sensors, multiple self-unloading load data and multiple cargo box status data are obtained. The self-unloading load data includes real-time load, self-unloading lifting times, cylinder pressure, and cylinder temperature, while the cargo box status data refers to real-time load.

[0030] Simultaneously, equipment controller interface readers are installed on the loading and unloading equipment corresponding to the vehicles to collect data from the loading and unloading equipment of multiple dump trucks. These interface readers connect to the loading and unloading equipment controller via an industrial communication interface, such as Modbus or OPC-UA protocol, to collect status information for each piece of equipment, including busy / idle status and fault status. The busy / idle status indicates whether the equipment is currently performing loading and unloading operations, while the fault status indicates whether the equipment has malfunctioned and cannot operate normally. The busy / idle status is a binary description of the resource occupancy of the loading and unloading equipment, and the fault status is marked based on the equipment's self-checking system or alarm signals.

[0031] The collected data on the status of multiple truck heads, multiple dump loads, multiple cargo boxes, and multiple loading and unloading equipment are summarized according to the self-dumping freight vehicle. That is, the status data of the truck head, dump load, cargo box, and its associated loading and unloading equipment for each vehicle are integrated together to form multiple multi-source real-time datasets. Each multi-source real-time dataset contains comprehensive status information of a certain self-dumping freight vehicle and its related loading and unloading equipment.

[0032] For example, there are three dump trucks: vehicle A, vehicle B, and vehicle C. For the tractor unit of vehicle A, the real-time location collected by the GPS positioning module is (116.404, 39.915), the speed sensor collects a speed of 50 km / h, the CAN bus reader collects a cumulative operating time of 1000 hours, and the fault code is "no fault," represented by code 0. On the dump trailer of vehicle A, the load sensor collects a real-time load of 20 tons, the accelerometer records 5 dump lifting operations, and the hydraulic cylinder pressure and temperature sensor collects a cylinder pressure of 10 MPa and a cylinder temperature of 40℃. For the loading and unloading equipment related to vehicle A, the equipment controller interface reader collects an idle status, represented by code 1, and a fault status, represented by code 0, indicating "no fault." Similarly, the above data collection is performed on vehicles B and C. Finally, the cab status data, dump load data, cargo box status data, and loading and unloading equipment status data corresponding to vehicles A, B, and C are summarized to form three multi-source real-time datasets. Each dataset completely records the real-time status information of a dump truck and its related loading and unloading equipment.

[0033] By uniformly collecting and summarizing the operating status of the truck head, the load of the self-unloading operation, the cargo box load, and the occupancy status of the loading and unloading equipment, the operating status of vehicles and loading and unloading equipment can be comprehensively reflected, thereby improving the accuracy and efficiency of logistics freight vehicle transportation scheduling.

[0034] The dynamic health status of each dump truck is calculated based on the multiple multi-source real-time datasets, and multiple health status labels are obtained.

[0035] Furthermore, the dynamic health of each dump truck is calculated based on the multiple multi-source real-time datasets to obtain multiple health labels. This includes: reading the number of dump lifting operations from multiple dump load data sets, and calculating the wear increment based on the load mass lifted in each dump lifting operation, accumulating these increments to obtain multiple wear increments; reading the cumulative running time from multiple truck head status data sets from multiple multi-source real-time datasets, adding a fixed step size of running wear for every hour of running time read, obtaining multiple running wear amounts; weighting the multiple wear increments and the multiple running wear amounts to obtain multiple health decay degrees; subtracting the multiple health decay degrees from the initial health to obtain multiple dynamic health values, and using the dynamic health values ​​as the vehicle's health labels to obtain multiple health labels.

[0036] Specifically, the dump load data of each dump truck is extracted from multiple multi-source real-time datasets, and the number of dump lifting operations and the corresponding load mass for each lifting operation are read from multiple dump load data. The number of dump lifting operations is identified by detecting the characteristic acceleration changes generated by the hydraulic lifting action using an accelerometer. Each obvious lifting action corresponds to one count. The lifting load mass is obtained through timestamp alignment, that is, based on the time point of each lifting action, the corresponding real-time load value is extracted from the load sensor data at the same time point as the actual load mass of that lifting operation.

[0037] When calculating wear increments, since the greater the load mass, the higher the wear on the cargo box and related components, the wear increment for each lift is calculated by multiplying the load mass by the wear coefficient. The wear coefficient can be set according to the vehicle's design parameters and actual operating experience. For example, the wear coefficient can be set as a piecewise function: when the load is ≤20 tons, k=0.015; when it is 20~30 tons, k=0.02; and when it exceeds 30 tons, i.e., in overload conditions, k=0.03. If a certain lift load is 25 tons, corresponding to k=0.02, then the wear increment for that lift is 0.5. If the load is 32 tons, k=0.03, then the wear increment is 0.96, thus reflecting the amplifying effect of high load on wear. By accumulating the wear increments for each lift, multiple wear increments can be obtained. These wear increments reflect the cumulative wear of the cargo box caused by frequent lifting and carrying of goods.

[0038] Simultaneously, the cumulative running time is extracted from the vehicle's status data from multi-source real-time datasets. For every hour of running time read, the running wear is increased by a preset fixed step size. By performing differential calculations on the running time, the current moment is subtracted from the last recorded moment to obtain the time increment, which is then accumulated to form the running wear amount. The fixed step size is determined comprehensively based on factors such as vehicle type, operating environment, and life parameters of key components, reflecting the basic mechanical wear of the vehicle in non-operating conditions. For example, based on the manufacturer's lifespan indicators, such as an engine design life of 20,000 hours, a basic wear rate is set, and the total lifespan is normalized to 100, resulting in a basic wear of 0.005 per hour. Then, adjustments are made based on actual operating conditions, introducing an environmental coefficient of 1.2 to 1.5 in high-load construction environments to obtain the final step size value. If a vehicle has a base step size of 0.005 and an environmental correction factor of 1.2, then the fixed step size is 0.006. If the current cumulative operating time increases from 1200 hours to 1203.5 hours, the time increment is 3.5 hours, corresponding to an operating wear amount of 3.5 × 0.006 = 0.021. After obtaining the operational wear amount and the running wear amount, the two are weighted and fused. Weighting refers to assigning different weight coefficients based on the differences in the degree of impact of different wear sources on vehicle lifespan. For example, operational wear weight α and running wear weight β. The calculation formula can be expressed as: Health Decay = α × Operational Wear Amount + β × Running Wear Amount. The weight coefficients can be determined through expert evaluation, historical data analysis, or experimental testing.

[0039] The weighted wear increment and operational wear are summed to obtain multiple health decay values, which represent the degree of decline in vehicle health due to wear. Multiple health decay values ​​are then subtracted from the initial health value to obtain multiple dynamic health values. The initial health value is the baseline health status of the vehicle when it is put into use, typically set to 100% or 1. The dynamic health value reflects the actual health status of the vehicle at the current moment; the lower the value, the more severe the wear and the worse the health status. The calculated dynamic health value for each vehicle is tagged and output as a health label, which is then bound to the vehicle ID.

[0040] By accurately calculating the dynamic health status of each dump truck and generating health status tags, the health status of vehicles can be monitored in real time. This enables logistics truck scheduling to not only make decisions based on transportation efficiency, but also to comprehensively consider vehicle lifespan and reliability, achieving intelligent scheduling with health constraints. This effectively avoids continuous overloading of vehicles under high intensity, improving the success rate and safety of transportation tasks.

[0041] Based on the multiple multi-source real-time datasets, perform status analysis on multiple dump trucks and construct a resource status mapping table.

[0042] Furthermore, based on the multiple multi-source real-time datasets, state analysis of multiple dump trucks is performed to construct a resource state mapping table, including: establishing state labels for the truck heads of multiple dump trucks based on the state data of multiple truck heads in the multiple multi-source real-time datasets, obtaining multiple truck head state labels; establishing state labels for the cargo boxes of multiple dump trucks based on the state data of multiple cargo boxes, obtaining multiple cargo box state labels; establishing state labels for each loading and unloading device based on the state data of multiple loading and unloading devices, obtaining multiple loading and unloading device state labels; and constructing a resource state mapping table based on the multiple truck head state labels, multiple cargo box state labels, and multiple loading and unloading device state labels.

[0043] Furthermore, the vehicle head status label includes three states: idle, in operation, and under maintenance; the cargo box status label includes four states: idle, loading, fully loaded and awaiting towing, and unloading; and the loading and unloading equipment status label includes three states: idle, occupied, and faulty.

[0044] Specifically, based on multiple real-time datasets from multiple sources, status labels are established for the cabs of multiple dump trucks, resulting in multiple cab status labels. These labels include three states: idle, in task, and under maintenance. The idle state indicates that the cab is not currently assigned a transportation task, determined by: zero speed and no task execution signal detected on the CAN bus. In the idle state, the current position coordinates of the cab are recorded for subsequent path planning and nearest-neighbor matching. The in-task state indicates that the cab has been assigned a transportation task and is currently executing it, determined by: the vehicle being in operation. The identifier of the cargo box currently being towed by the cab and the remaining task time are also obtained. The remaining task time is calculated by subtracting the executed time from the total task duration. The cargo box identifier is a unique number for each cargo box, used to distinguish different cargo boxes. The under maintenance state is triggered by CAN bus fault code marking. When a critical fault code is detected, the cab status is set to under maintenance, and it is not involved in scheduling.

[0045] Then, based on multiple cargo compartment status data, status tags are established for the cargo compartments of multiple dump trucks, resulting in multiple cargo compartment status tags. These tags include four states: idle, loading, fully loaded and awaiting towing, and unloading. The idle state indicates that the cargo compartment is not loaded with goods and is not participating in a scheduling task. This is determined by the load being below a set threshold, such as <5% of the rated load, and no loading / unloading operation records. In the idle state, the cargo compartment position and unloading direction compatibility type are recorded. The unloading direction compatibility type refers to the unloading direction allowed by the cargo compartment structure, such as rear tipping or side tipping. This parameter is derived from the vehicle configuration database and is used for subsequent matching of unloading points. The loading state is determined jointly by the loading / unloading equipment status and the load change trend. That is, when the equipment is occupied and the cargo compartment load continues to increase, it is determined to be loading. Combined with historical loading efficiency, the estimated loading completion time is predicted based on the loading volume per unit time. The fully loaded, awaiting-towing status refers to a cargo box load reaching the rated load threshold, such as ≥90% of the rated value, and not currently being towed by a tractor. This location is typically a loading station, where cargo information, such as ore type and weight, is recorded. This information comes from the loading system or is manually entered and used for subsequent transport matching. The unloading status is determined by both hydraulic lifting actions and equipment occupancy status.

[0046] Furthermore, based on the status data of multiple loading and unloading equipment, a status tag is established for each piece of equipment, resulting in multiple status tags for loading and unloading equipment. These status tags include three states: idle, occupied, and faulty. The idle state indicates that the equipment is not performing a loading or unloading task. This is determined by the busy / idle status collected by the interface reader of the installed equipment controller. In the idle state, the equipment location and type are recorded, with the equipment type sourced from the equipment asset management system. The occupied state indicates that the equipment is serving a specific cargo compartment. The identifier of the currently served cargo compartment and the estimated release time are obtained through the interface. The estimated release time can be predicted using standard operation duration or real-time operation progress. The faulty state is directly determined by reading the fault status data. Once a loading or unloading equipment enters a faulty state, it is removed from the schedulable resources.

[0047] Then, based on multiple tractor status labels, multiple cargo box status labels, and multiple loading and unloading equipment status labels, a resource status mapping table is constructed. The resource status mapping table is a structured data table that integrates the status information of tractors, cargo boxes, and loading and unloading equipment to display the current status of all transportation resources in a clear and intuitive way. Different columns can be set in the table to correspond to the identifiers of tractors, cargo boxes, and loading and unloading equipment, as well as their respective status labels and related detailed information, such as the location coordinates of the tractor, the loading status of the cargo box, and the expected release time of the loading and unloading equipment.

[0048] By accurately analyzing the status of the tractor, cargo box, and loading / unloading equipment and constructing a resource status mapping table, we can grasp the distribution and usage of transportation resources in real time and comprehensively. This provides an accurate data foundation for subsequent tractor-cargo matching, route planning, and multi-objective optimization, enabling scheduling decisions to be more scientific and reasonable, avoiding resource conflicts and waste, and improving transportation efficiency.

[0049] Furthermore, based on the multiple tractor status labels, multiple cargo box status labels, and multiple loading and unloading equipment status labels, a resource status mapping table is constructed, including: extracting tractors with idle labels from the multiple tractor status labels to obtain a set of left nodes; extracting cargo boxes with idle labels or fully loaded cargo boxes waiting to be towed from the multiple tractor status labels to obtain a set of right nodes; if the unloading directions of the tractor and cargo box are compatible and the tractor's traction capacity is greater than or equal to the cargo box's load capacity, then an edge is established between the left and right nodes, and the minimum cost matching is solved using the Hungarian algorithm or a greedy algorithm, and the matching result is stored together with the status labels of each resource in the resource status mapping table.

[0050] Specifically, from multiple locomotive status labels, locomotives with the "idle" label are extracted, and their relevant information, such as locomotive number and current coordinates, is extracted and used as elements in the left node set to construct the left node set. For example, if an idle locomotive has the locomotive number T001 and its current coordinates are (116.301, 39.956), then the left node set records: T001, (116.301, 39.956). From multiple cargo box status labels, cargo boxes with the "idle" label or fully loaded and awaiting towing are extracted, and the cargo box information that meets the conditions, such as cargo box number, location, unloading direction compatibility type, and cargo box load capacity, is extracted and used as elements in the right node set to form the right node set. Then, a matching judgment is performed between locomotives and cargo boxes. If the unloading directions of the locomotive and cargo box are compatible, that is, the unloading direction of the cargo box can meet the unloading requirements after the locomotive towing, and the towing capacity of the locomotive is greater than or equal to the load capacity of the cargo box, then an edge is established between the two nodes. The traction capacity data of the tractor unit is stored in the basic information database of the tractor unit, while the load capacity of the cargo box is obtained in real time through load sensors. For example, if the tractor unit T001 has a traction capacity of 30 tons and the cargo box C001 has a load capacity of 20 tons, and the unloading directions are compatible, then an edge is established between the nodes corresponding to T001 and C001.

[0051] Then, the Hungarian algorithm or greedy algorithm is used to solve the minimum cost matching problem. First, a cost matrix is ​​constructed based on distance cost, waiting time cost, and load balancing cost. The cost matrix is ​​used to represent the comprehensive scheduling cost of each tractor-cargo combination. Each matrix element represents the total cost of performing the matching task between the tractor and the cargo, which can be expressed as: Cij=w1Dij+w2Tij+w3Lij, where w1 is the distance cost weight, w2 is the time cost weight, and w3 is the load balancing cost weight, and satisfies w1+w2+w3=1. The weights are set according to actual needs, such as w1 is 0.4, w2 and w3 are 0.3 respectively. The distance cost Dij is used to reflect the empty driving cost from the tractor to the cargo, which is obtained by calculating the shortest path through the electronic map API or the pre-set road network map. The time cost Tij is used to reflect the scheduling response efficiency, which is calculated as cargo waiting time for traction + distance cost / average speed of the fleet. The load balancing cost is used to avoid overuse of some tractors, which is obtained by dividing the historical task load of the tractor by the maximum task load of the fleet. The cost matrix is ​​initialized, and for tractor-cargo combinations that do not meet traction capacity constraints or have incompatible unloading directions, the corresponding cost is set to positive infinity or a maximum value to indicate an unmatchable relationship.

[0052] Then, the Hungarian algorithm steps are executed: First, perform row subtraction on each row of the cost matrix, i.e., subtract the minimum value of each row, so that each row has at least one zero element; Second, perform column subtraction on each column, so that each column also has at least one zero element, thus forming a normalized matrix; Third, find independent zero elements in the matrix, i.e., select only one zero in any row and column to construct an initial match; If the current number of independent zeros is insufficient to complete the matching of all nodes, then cover all zero elements with the minimum coverage line, adjust the minimum value of the uncovered elements, and iteratively update the matrix; repeat this process until a zero-matching structure covering all locomotive nodes can be obtained. The index pairs corresponding to the final zero elements are the optimal matching relationship between the locomotive and the cargo box, which minimizes the total matching cost globally, thus outputting the optimal scheduling scheme.

[0053] The specific implementation of the greedy algorithm to solve the minimum cost matching problem is as follows: After constructing the set of tractor units and cargo boxes and their cost matrices, firstly, a candidate edge set is generated for all feasible tractor-cargo box combinations, and the matching cost of each edge is calculated. Then, all candidate edges are sorted in ascending order of cost to form an ordered edge list. During initialization, all tractor units and cargo boxes are marked as unmatched. Then, the sorted edge set is traversed sequentially. For the edge with the minimum cost, if tractor unit i is not matched and cargo box is not matched, its constraints are checked, including traction capacity constraints and unloading direction compatibility. If the constraints are still satisfied, the tractor unit and cargo box are immediately matched, and their states are marked as occupied or matched. If either end is already matched, the edge is skipped and the next candidate edge is traversed. This process continues until all tractor units are matched or there are no feasible edges. Choose the appropriate algorithm based on the specific scenario and requirements. For example, if you value the accuracy and global optimality of the matching, you can choose the Hungarian algorithm. If you have high requirements for computational efficiency and allow for a certain degree of approximate optimal solution, you can choose the greedy algorithm.

[0054] The matching results and the status labels of each resource are stored together in the resource status mapping table. This makes the resource status mapping table not only contain static status information, such as idle or full load, but also dynamic scheduling relationship information, such as matched traction relationships and corresponding costs, thus forming a complete basic data structure for scheduling decisions.

[0055] By optimally matching idle locomotives with cargo compartments that meet the constraints, and by performing a global solution with the goal of minimizing costs, the resource state mapping table can reflect the optimal scheduling relationship between locomotives and cargo compartments in real time. This reduces the idle waiting time and empty driving distance of vehicles and cargo compartments, achieves overall optimization of transportation routes and resource allocation, improves resource utilization, and reduces overall transportation costs.

[0056] The multiple dump trucks are divided into multiple vehicle groups based on their cumulative dumping load. The pre-trained scheduling model is then trained to enhance group fairness based on the multiple multi-source real-time datasets and resource status mapping tables to obtain an enhanced scheduling model.

[0057] Furthermore, based on the cumulative dumping load of the vehicles, the multiple dump trucks are divided into multiple vehicle groups. The pre-trained scheduling model is then subjected to group fairness enhancement training based on the multiple multi-source real-time datasets and a resource state mapping table to obtain an enhanced scheduling model. This includes: extracting multiple cumulative dumping loads from the multiple dumping load data; clustering the multiple cumulative dumping loads to divide the multiple dump trucks into multiple vehicle groups; performing scheduling efficiency loss analysis on the multiple vehicle groups based on the multiple multi-source real-time datasets to obtain the average scheduling loss of the multiple vehicle groups; selecting the vehicle group corresponding to the minimum average scheduling loss among the multiple vehicle groups as the optimal group, and calculating the difference between the average scheduling loss of the multiple vehicle groups and the minimum average scheduling loss among the multiple vehicle groups as the group fairness gap; and then, based on the group fairness gap and the resource state mapping table, performing group fairness enhancement training on the pre-trained scheduling model to obtain the enhanced scheduling model.

[0058] Specifically, multiple dumping load data points for each dump truck are extracted from a multi-source real-time dataset. Multiple cumulative dumping loads are then calculated based on these data points. The cumulative dumping load refers to the total sum of the loads carried and unloaded by the vehicle during all dumping operations within a preset statistical period. For example, if a dump truck unloads 10 tons on the first day, 15 tons on the second day, and 8 tons on the third day, then the cumulative dumping load of that vehicle up to the third day is 10 + 15 + 8 = 33 tons.

[0059] After obtaining the cumulative dump load of all vehicles, clustering is performed based on the magnitude of multiple cumulative dump loads. K-means clustering or density-based clustering algorithms, such as DBSCAN, are used to group the vehicles, thus dividing multiple dump trucks into multiple vehicle groups. A vehicle group refers to a set of vehicles with similar long-term operating load levels, such as a high-load group, a medium-load group, and a low-load group. This division reflects the differences in vehicle usage intensity.

[0060] Based on multiple multi-source real-time datasets, a scheduling efficiency loss analysis is performed on various vehicle groups. This scheduling efficiency loss reflects the performance degradation of each group of vehicles relative to the theoretically optimal scheduling state under the current scheduling strategy. The calculation is based on indicators including, but not limited to, average waiting time, empty driving distance, task completion delay, and resource matching failure rate. These indicators are derived from scheduling result records in the resource status mapping table and running trajectory data in the multi-source real-time datasets. The above four types of indicators are normalized according to a unified dimension, such as Min-Max normalization, and then weighted and fused according to preset weight coefficients to obtain the single-vehicle scheduling loss value. The weight coefficients can be set based on actual needs and expert experience. By statistically averaging the scheduling losses of all vehicles within each vehicle group, the mean scheduling loss of each group is obtained, reflecting the overall efficiency level of different load groups under the current scheduling.

[0061] The vehicle group with the minimum mean scheduling loss among multiple vehicle groups is selected as the optimal group. The difference between the mean scheduling loss of multiple vehicle groups and the minimum mean scheduling loss of multiple vehicle groups is calculated as the fairness gap among the multiple groups. The optimal group represents the vehicle group with the highest scheduling efficiency under the current scheduling strategy. The group fairness gap reflects the degree of difference in scheduling efficiency between other vehicle groups and the optimal group. For example, if the mean scheduling loss of group 1 is 2 hours, the mean scheduling loss of group 2 is 1.5 hours, and the mean scheduling loss of group 3 is 3 hours, then the optimal group is group 2. The group fairness gap of group 1 is 2 - 1.5 = 0.5 hours, and the group fairness gap of group 3 is 3 - 1.5 = 1.5 hours.

[0062] The fairness gap between groups and the resource status mapping table are used as constraints and feedback signals to input into the pre-trained scheduling model. By introducing fairness constraint terms, the pre-trained scheduling model is optimized to enhance group fairness. This allows the scheduling model to reduce the performance differences between different vehicle groups while optimizing the global scheduling cost, thereby obtaining a scheduling model with enhanced group fairness and achieving a more balanced task allocation among vehicle groups with different loads.

[0063] By constructing a vehicle group segmentation mechanism and analyzing group scheduling losses, the scheduling model not only pursues the optimal global efficiency, but also takes into account the fairness of resource allocation among different vehicle groups, thereby avoiding the problem of high-load vehicles being continuously overloaded and low-load vehicles being idle for a long time, making the scheduling strategy more reasonable and balanced.

[0064] Furthermore, based on the multiple group fairness gaps and resource state mapping tables, the pre-trained scheduling model is subjected to group fairness enhancement training to obtain an enhanced scheduling model. This includes: constructing hard negative samples and simple negative samples based on the multiple group fairness gaps and resource state mapping tables; obtaining the training data of the pre-trained scheduling model, merging the hard negative samples and simple negative samples to obtain an updated training dataset; and using the updated training dataset to perform group fairness enhancement training on the pre-trained scheduling model to obtain the enhanced scheduling model.

[0065] Furthermore, constructing difficult negative samples and simple negative samples includes: reading all vehicle heads, cargo boxes, and loading / unloading equipment with idle status labels from the resource status mapping table to obtain a candidate resource set; for vehicle groups with a group fairness gap greater than or equal to a preset gap threshold among multiple group fairness gaps, selecting resource samples from the candidate resource set whose expected matching cost with the current scheduling strategy is greater than or equal to a preset cost threshold, and using the selected resource samples as difficult negative samples; for vehicle groups with a group fairness gap less than a preset gap threshold among multiple group fairness gaps, selecting resource samples from the candidate resource set whose matching cost is less than a preset cost threshold, and using the selected resource samples as simple negative samples.

[0066] Specifically, all idle tractor units, idle cargo compartments, and idle loading / unloading equipment are read in real-time from the resource status mapping table to form a candidate resource set. Fairness differences are screened among vehicle groups. When the fairness difference of a vehicle group is greater than or equal to a preset difference threshold, the group is identified as a highly unfair group. The preset difference threshold refers to the boundary for judging the scheduling fairness difference between vehicle groups. It is set as follows: based on the average scheduling loss of each vehicle group over multiple historical scheduling cycles, the standard deviation σ and mean μ of the average scheduling loss of all groups are calculated. A statistical interval is used as the basis for setting the threshold. For example, the preset difference threshold θ = μ + λσ can be set, where λ is an empirical coefficient, such as between 0.5 and 1.5, to adjust the fairness sensitivity. Alternatively, a quantile method can be used, such as taking the 75th or 80th percentile of the historical fairness difference distribution as the preset difference threshold, thus ensuring that only groups significantly deviating from the average scheduling level are identified as highly unfair groups. For this type of highly unfair group, resource samples with a predicted matching cost greater than or equal to a preset cost threshold under the current scheduling strategy are further screened from the candidate resource set and defined as difficult negative samples. The preset cost threshold is used to distinguish the scheduling difficulty level. Its setting method is based on the historical distribution characteristics of the cost matrix. The cost value of all successfully matched locomotive-cargo combinations is extracted from the historical scheduling records. Similarly, the mean plus 0.5 to 1.5 times the standard deviation is used as the preset cost threshold to control the intensity of difficult sample screening. Alternatively, the quantile method can be directly used, such as taking the 90th percentile value as the difficult sample threshold, so that only combinations with costs significantly higher than the normal scheduling level are defined as difficult negative samples, thereby strengthening the model's learning ability for extreme scheduling scenarios.

[0067] Conversely, vehicle groups with a fairness gap of less than a preset threshold are classified as high-fairness or low-bias groups. For such groups, resource combinations with matching costs less than a preset cost threshold are selected from the candidate resource set and defined as simple negative samples. These samples typically correspond to situations with relatively balanced resource distribution, short scheduling paths, or low waiting times, and are used to enhance the model's stable learning ability in typical scheduling scenarios. After constructing the negative samples, a basic training dataset for the pre-trained scheduling model is introduced. This basic training dataset comes from historical scheduling task records, specifically tractor-cargo pairing samples whose actual execution efficiency during historical operations exceeds a preset efficiency standard. This preset efficiency standard can be set based on average task completion time, empty mileage rate, or comprehensive cost indicators.

[0068] Historical high-efficiency positive samples are fused with newly constructed difficult and simple negative samples to form an updated training dataset, thereby constructing a multi-level training dataset containing high-efficiency decision samples, difficult constraint samples, and regular samples. The updated training dataset is used to perform group fairness enhancement training on the pre-trained scheduling model to obtain an enhanced scheduling model. The pre-trained scheduling model is preferably constructed as a multi-layer graph matching model based on a graph neural network to achieve the initial optimal matching decision between the tractor and the cargo compartment. Specifically, the resource state mapping table at each scheduling time is constructed as a bipartite graph G=(V,E), where the node set V includes the tractor node set H, the cargo compartment node set T, and the loading / unloading equipment node set D, and E represents the edge set in the graph structure. Each node type shares a message passing framework within the same graph structure, and type identifier vectors are used for differentiation during the input feature encoding stage to achieve heterogeneous information expression rather than heterogeneous parameter structures.

[0069] During graph construction, candidate matching edges are built between tractor nodes and cargo box nodes by default. Valid edge relationships are established only if basic scheduling constraints are met. These basic scheduling constraints include tractor traction capacity being greater than or equal to cargo box load capacity, and compatibility in unloading directions. For node pairs that do not meet the constraints, their corresponding edges are assigned a value of 0 or an unreachable marker in the adjacency matrix, thus forming a constrained sparse bipartite adjacency matrix. Auxiliary association edges are built between tractor nodes and loading / unloading equipment nodes, and between cargo box nodes and loading / unloading equipment nodes, respectively, based on geographical distance thresholds and task dependencies, to reflect the impact of loading / unloading resource occupancy on scheduling feasibility.

[0070] In the feature encoding stage, an initial feature vector is constructed for each node, including location coordinates, current state encoding, load capacity or current load, waiting time, and historical task frequency. This vector is uniformly mapped to a 128-dimensional embedding space through a fully connected layer, with ReLU as the activation function. This mapping layer shares parameters across all nodes but concatenates them using node type embedding vectors, enabling the scheduling model to distinguish between three semantic categories: tractor, cargo compartment, and equipment. In the graph encoding stage, a three-layer graph convolutional network (GCN) is used for message passing updates. In the matching calculation stage, a matching score matrix is ​​constructed for any tractor node and cargo compartment node, and softmax normalization is performed on each tractor dimension to obtain a probability distribution Pij, which represents the probability of tractor i selecting cargo compartment j. In the output layer, the final tractor-cargo compartment allocation result is output through a Hungarian matching layer or a greedy algorithm approximation matching layer. In the model training stage, supervised learning is used, with the tractor-cargo compartment pairing result with the best actual execution efficiency in historical scheduling records as the label to construct a cross-entropy loss function.

[0071] In the group fairness enhancement training phase based on the updated training dataset, the input data includes training data, merged hard and easy negative samples, and attached vehicle group labels. After the model forward propagates to obtain the predicted matching results, a comprehensive cost function is calculated to form the cost loss, which is defined as the mean squared error between the predicted matching cost and the historical best cost. The group fairness loss is constructed using a group mean difference constraint method, that is, calculating the average scheduling loss of each vehicle group under the current model, and using the optimal group loss as a benchmark, calculating the sum of the absolute differences between other groups and the benchmark group, thereby quantifying the scheduling performance differences between groups. The cross-entropy loss function, cost loss, and group fairness loss are weighted and combined to form the total loss, and the model parameters are updated through a gradient descent algorithm, such as the Adam optimizer, so that the scheduling model gradually reduces the performance differences between different vehicle groups while reducing the overall scheduling cost. The initial learning rate of the graph neural network is set to 1e. -3 And a cosine decay strategy is used to gradually decrease it to 1e. -5 The batch size was set to 64, and the training epochs ranged from 50 to 200. To avoid overfitting, an early stopping mechanism was introduced: training was terminated when the validation set loss no longer decreased within 10 consecutive epochs. The convergence criterion for the augmented scheduling model was defined as: the total loss decreased by less than 1e... -4 Alternatively, the validation set loss may tend to stabilize within a certain fluctuation range. After multiple rounds of iterative training, the scheduling model not only maintains the original scheduling efficiency but also achieves a more balanced resource allocation capability among different load groups, ultimately resulting in an enhanced scheduling model with improved group fairness.

[0072] By selectively filtering resource samples from vehicle groups with varying fairness levels, constructing and updating the training dataset, and enhancing the pre-trained scheduling model, the shortcomings of traditional scheduling models in terms of group fairness can be effectively addressed. This makes scheduling strategies more reasonable and fair, improves the satisfaction of vehicle groups with different workloads, and ultimately enhances the stability and efficiency of the entire logistics scheduling process.

[0073] The multiple health labels, resource status mapping table, and enhanced scheduling model are input into the multi-objective optimizer to generate a truck head-cargo box pairing scheme and an unloading point allocation scheme, and then the transportation scheduling is executed.

[0074] Specifically, multiple health labels, resource status mapping tables, and a scheduling model enhanced with group fairness are uniformly input into a multi-objective optimizer. The multi-objective optimizer constructs a unified optimization objective function, which includes three objectives: minimizing scheduling costs, balancing resource health, and ensuring scheduling fairness. The minimizing scheduling costs objective consists of transportation distance costs, task waiting time costs, and loading / unloading equipment occupancy time costs. The balancing resource health objective avoids overuse of a single vehicle by constraining the task allocation ratio between high-load vehicles and low-health vehicles. The scheduling fairness objective constrains the differences in scheduling losses between different vehicle groups, ensuring that the overall scheduling result meets the requirements of group balance.

[0075] The three objectives mentioned above are unified into a single optimization function through a weighted approach. The weight coefficients are dynamically adjusted according to the operating strategy. During the solution process, the multi-objective optimizer employs a combination of constraint-based screening and iterative search: First, the matching probability output by the enhanced scheduling model is used to initially screen candidate pairings, eliminating infeasible solutions that violate traction capacity constraints, unloading direction constraints, and health constraints. Then, an iterative optimization strategy is used to gradually update the matching scheme in the feasible solution space. In each iteration, the total objective function value is recalculated, and the current optimal solution structure is retained until the objective function converges or the maximum number of iterations is reached. The final output includes two parts: the first part is the optimal matching relationship between the truck head and the cargo compartment, used to determine the main body responsible for transport task execution; the second part is the unloading point allocation scheme, which maps each cargo compartment task to a specific unloading station or loading / unloading equipment, satisfying spatial optimality and equipment availability constraints, thus forming a complete transport scheduling execution scheme for execution.

[0076] By integrating vehicle health status, real-time resource status, and learned scheduling strategies, and through a multi-objective optimization mechanism, the final transformation from matching decisions to executable scheduling schemes is achieved. This enables logistics freight vehicle transportation scheduling to not only optimize transportation efficiency but also take into account equipment health lifespan and group fairness, thereby achieving comprehensive scheduling control that is safe, economical, and balanced, and improving the level of intelligence in logistics transportation scheduling.

[0077] Example 2 is based on the same inventive concept as the IoT-based multi-modal transportation scheduling method for logistics freight vehicles in the previous examples, such as... Figure 2 As shown, this application provides an IoT-based multi-modal transportation scheduling system for logistics freight vehicles, wherein the IoT-based multi-modal transportation scheduling system for logistics freight vehicles includes: The data acquisition module 11 is used to collect real-time data on the unloading load, cab status, cargo box status, and loading / unloading equipment status of multiple dump trucks through IoT terminals, obtaining multiple multi-source real-time datasets; the health calculation module 12 is used to calculate the dynamic health of each dump truck based on the multiple multi-source real-time datasets, obtaining multiple health labels; the state parsing module 13 is used to perform state parsing of multiple dump trucks based on the multiple multi-source real-time datasets, and construct a resource state mapping table; the scheduling model acquisition module 14 is used to divide the multiple dump trucks into multiple vehicle groups based on the cumulative unloading load of the vehicles, and perform group fairness enhancement training on the pre-trained scheduling model based on the multiple multi-source real-time datasets and the resource state mapping table, obtaining an enhanced scheduling model; the scheme generation module 15 is used to input the multiple health labels, the resource state mapping table, and the enhanced scheduling model into a multi-objective optimizer, generate a cab-cargo box pairing scheme and an unloading point allocation scheme, and execute transportation scheduling.

[0078] Furthermore, the data acquisition module 11 is also used to: install GPS positioning modules, speed sensors, and CAN bus readers on the tractor heads of multiple dump trucks respectively, collect the real-time position, speed, cumulative running time, and fault codes of the corresponding vehicles, and obtain multiple tractor head status data; install load sensors, accelerometers, and hydraulic cylinder pressure and temperature sensors on the dump trailers of multiple dump trucks respectively, collect the real-time load, number of dump lifting operations, cylinder pressure, and cylinder temperature of the corresponding vehicles, and obtain multiple dump load data and multiple cargo box status data; install equipment controller interface readers on the loading and unloading equipment corresponding to multiple dump trucks, collect the busy / idle status and fault status of each loading and unloading equipment, and obtain multiple loading and unloading equipment status data; and summarize the multiple tractor head status data, multiple dump load data, multiple cargo box status data, and multiple loading and unloading equipment status data with the dump trucks to obtain multiple multi-source real-time datasets.

[0079] Furthermore, the health calculation module 12 is also used to: read the number of self-unloading lifts from multiple self-unloading load data, and calculate the wear increment by reading the load mass lifted in each self-unloading lift action, and obtain multiple wear increments after accumulation; read the cumulative running time from multiple vehicle head status data from multiple multi-source real-time datasets, and increase the running wear by a fixed step size for every hour of running time read, to obtain multiple running wear amounts; weight the multiple wear increments and the multiple running wear amounts to obtain multiple health decay degrees, subtract the multiple health decay degrees from the initial health to obtain multiple dynamic health degrees, and use the dynamic health degree as the vehicle's health label to obtain multiple health labels.

[0080] Furthermore, the state parsing module 13 is also used to: establish state labels for the heads of multiple dump trucks based on multiple head status data from multiple multi-source real-time datasets, and obtain multiple head status labels; establish state labels for the cargo boxes of multiple dump trucks based on multiple cargo box status data, and obtain multiple cargo box status labels; establish state labels for each loading and unloading device based on multiple loading and unloading device status data, and obtain multiple loading and unloading device status labels; and construct a resource state mapping table based on the multiple head status labels, multiple cargo box status labels, and multiple loading and unloading device status labels.

[0081] Furthermore, the vehicle head status label includes three states: idle, in operation, and under maintenance; the cargo box status label includes four states: idle, loading, fully loaded and awaiting towing, and unloading; and the loading and unloading equipment status label includes three states: idle, occupied, and faulty.

[0082] Furthermore, the state parsing module 13 is also used to: extract the locomotives with idle labels from the multiple locomotive state labels to obtain a set of left nodes; extract the cargo boxes with idle labels or fully loaded and waiting to be towed from the multiple locomotive state labels to obtain a set of right nodes; if the unloading directions of the locomotive and the cargo box are compatible and the traction capacity of the locomotive is greater than or equal to the load capacity of the cargo box, then establish an edge between the left and right nodes, use the Hungarian algorithm or greedy algorithm to solve the minimum cost matching, and store the matching result and the state labels of each resource in the resource state mapping table.

[0083] Furthermore, the scheduling model acquisition module 14 is also used to: extract multiple cumulative dump loads from multiple dump load data; cluster the multiple cumulative dump loads according to their magnitudes to divide the multiple dump trucks into multiple vehicle groups; perform scheduling efficiency loss analysis on the multiple vehicle groups based on the multiple multi-source real-time datasets to obtain the average scheduling loss of the multiple vehicle groups; take the vehicle group corresponding to the minimum value among the average scheduling losses of the multiple vehicle groups as the optimal group, and calculate the difference between the average scheduling loss of the multiple vehicle groups and the minimum value among the average scheduling losses of the multiple vehicle groups as the fairness gap among the multiple groups; and perform group fairness enhancement training on the pre-trained scheduling model according to the fairness gap among the multiple groups and the resource state mapping table to obtain an enhanced scheduling model.

[0084] Furthermore, the scheduling model acquisition module 14 is also used to: construct difficult negative samples and simple negative samples according to the multiple group fairness gap and resource status mapping table; acquire the training data of the pre-trained scheduling model, merge the difficult negative samples and simple negative samples to obtain an updated training dataset; and use the updated training dataset to perform group fairness enhancement training on the pre-trained scheduling model to obtain an enhanced scheduling model.

[0085] Furthermore, the scheduling model acquisition module 14 is also used to: read all vehicle heads, cargo boxes, and loading / unloading equipment with idle status labels from the resource status mapping table to obtain a candidate resource set; for vehicle groups with a group fairness gap greater than or equal to a preset gap threshold among multiple group fairness gaps, select resource samples from the candidate resource set whose expected matching cost with the current scheduling strategy is greater than or equal to a preset cost threshold, and use the selected resource samples as hard negative samples; for vehicle groups with a group fairness gap less than a preset gap threshold among multiple group fairness gaps, select resource samples from the candidate resource set whose matching cost is less than a preset cost threshold, and use the selected resource samples as simple negative samples.

[0086] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The IoT-based multi-mode transportation scheduling method and specific examples in the aforementioned Embodiment 1 are also applicable to the IoT-based multi-mode transportation scheduling system for logistics freight vehicles in this embodiment. Through the foregoing detailed description of the IoT-based multi-mode transportation scheduling method for logistics freight vehicles, those skilled in the art can clearly understand the IoT-based multi-mode transportation scheduling system for logistics freight vehicles in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0087] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0088] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A multi-modal transportation scheduling method for logistics freight vehicles based on the Internet of Things, characterized in that, The method includes: The system collects real-time data on the unloading load, cab status, cargo box status, and loading / unloading equipment status of multiple dump trucks through IoT terminals, thereby obtaining multiple multi-source real-time datasets. The dynamic health status of each dump truck is calculated based on the multiple multi-source real-time datasets, and multiple health status labels are obtained. Based on the multiple multi-source real-time datasets, perform status analysis on multiple dump trucks and construct a resource status mapping table. The multiple dump trucks are divided into multiple vehicle groups based on their cumulative dumping load, and the pre-trained scheduling model is trained to enhance group fairness based on the multiple multi-source real-time datasets and resource status mapping table to obtain an enhanced scheduling model. The multiple health labels, resource status mapping table, and enhanced scheduling model are input into the multi-objective optimizer to generate a truck head-cargo box pairing scheme and an unloading point allocation scheme, and then the transportation scheduling is executed.

2. The IoT-based multi-mode transportation scheduling method for logistics freight vehicles as described in claim 1, characterized in that, By collecting real-time data on the unloading load, cab status, cargo box status, and loading / unloading equipment status of multiple dump trucks through IoT terminals, multiple multi-source real-time datasets are obtained, including: GPS positioning modules, speed sensors, and CAN bus readers are installed on the tractor heads of multiple dump trucks to collect the real-time location, speed, cumulative running time, and fault codes of the corresponding vehicles, thereby obtaining multiple tractor head status data. Load sensors, accelerometers, and hydraulic cylinder pressure and temperature sensors were installed on the dump trailers of multiple dump trucks to collect real-time load, number of dump lifting operations, cylinder pressure, and cylinder temperature of the corresponding vehicles, thereby obtaining multiple dump load data and multiple cargo box status data. Install equipment controller interface readers on the loading and unloading equipment corresponding to multiple dump trucks to collect the busy / idle status and fault status of each loading and unloading equipment and obtain status data of multiple loading and unloading equipment. Multiple data points on the status of truck heads, dump loads, cargo boxes, and loading / unloading equipment are aggregated together with the data of dump trucks to obtain multiple multi-source real-time datasets.

3. The IoT-based multi-mode transportation scheduling method for logistics freight vehicles as described in claim 1, characterized in that, The dynamic health status of each dump truck is calculated based on the multiple multi-source real-time datasets, resulting in multiple health labels, including: The number of self-unloading lifts is read from multiple self-unloading load data, and the wear increment is calculated by reading the load mass lifted in each self-unloading lift action. After accumulation, multiple wear increments are obtained. The cumulative running time is read from multiple locomotive status data from multiple multi-source real-time datasets. For every hour of running time read, the running wear is increased by a fixed step size to obtain multiple running wear values. The multiple wear increments and multiple operational wear amounts are weighted to obtain multiple health decay values. The initial health value is then subtracted from each of the multiple health decay values ​​to obtain multiple dynamic health values. The dynamic health values ​​are then used as the vehicle's health labels to obtain multiple health labels.

4. The IoT-based multi-mode transportation scheduling method for logistics freight vehicles as described in claim 1, characterized in that, Based on the aforementioned multiple multi-source real-time datasets, the status analysis of multiple dump trucks is performed, and a resource status mapping table is constructed, including: Based on multiple vehicle head status data from multiple multi-source real-time datasets, status labels are established for the vehicle heads of multiple dump trucks, resulting in multiple vehicle head status labels. Based on multiple cargo compartment status data, status labels are established for the cargo compartments of multiple dump trucks, resulting in multiple cargo compartment status labels. Based on the status data of multiple loading and unloading equipment, a status label is established for each loading and unloading equipment to obtain multiple status labels for loading and unloading equipment. A resource status mapping table is constructed based on the multiple vehicle head status labels, multiple cargo box status labels, and multiple loading and unloading equipment status labels.

5. The IoT-based multi-mode transportation scheduling method for logistics freight vehicles as described in claim 4, characterized in that, The vehicle front status label includes three states: idle, in operation, and under maintenance. The cargo compartment status label includes four states: idle, loading, fully loaded and awaiting towing, and unloading. The status labels of the loading and unloading equipment include three states: idle, occupied, and faulty.

6. The IoT-based multi-mode transportation scheduling method for logistics freight vehicles as described in claim 4, characterized in that, Based on the multiple vehicle head status labels, multiple cargo box status labels, and multiple loading / unloading equipment status labels, a resource status mapping table is constructed, including: Extract the vehicle heads with idle labels from the multiple vehicle head status labels to obtain the left node set; Extract cargo boxes with idle or fully loaded cargo boxes awaiting towing from multiple vehicle head status labels to obtain the right node set; If the unloading directions of the tractor and the cargo box are compatible and the tractor's traction capacity is greater than or equal to the cargo box's load capacity, then establish an edge between the left and right nodes, use the Hungarian algorithm or a greedy algorithm to solve the minimum cost matching problem, and store the matching results and the state labels of each resource in the resource state mapping table.

7. The IoT-based multi-mode transportation scheduling method for logistics freight vehicles as described in claim 1, characterized in that, The multiple dump trucks are divided into multiple vehicle groups based on their cumulative dumping load. The pre-trained scheduling model is then trained to enhance group fairness using the multiple multi-source real-time datasets and resource state mapping tables, resulting in an enhanced scheduling model, including: Extract multiple cumulative self-unloading loads from multiple self-unloading load data; Clustering is performed based on the magnitude of the cumulative dump loads to divide the multiple dump trucks into multiple vehicle groups; Based on the multiple multi-source real-time datasets, the scheduling efficiency loss of the multiple vehicle groups is analyzed to obtain the average scheduling loss of the multiple vehicle groups. The vehicle group corresponding to the minimum mean of scheduling loss among multiple vehicle groups is taken as the optimal group, and the difference between the mean of scheduling loss among multiple vehicle groups and the minimum mean of scheduling loss among multiple vehicle groups is calculated as the fairness gap among multiple groups. Based on the multiple group fairness gaps and resource status mapping tables, the pre-trained scheduling model is trained to enhance group fairness, thereby obtaining an enhanced scheduling model.

8. The IoT-based multi-mode transportation scheduling method for logistics freight vehicles as described in claim 7, characterized in that, Based on the multiple group fairness gaps and resource state mapping tables, the pre-trained scheduling model is subjected to group fairness enhancement training to obtain an enhanced scheduling model, including: Based on the fairness gap and resource status mapping table of the multiple groups, construct difficult negative samples and simple negative samples; Obtain the training data of the pre-trained scheduling model, merge hard negative samples and easy negative samples, and obtain an updated training dataset; By updating the training dataset, the pre-trained scheduling model is trained to enhance group fairness, thus obtaining an enhanced scheduling model.

9. The IoT-based multi-mode transportation scheduling method for logistics freight vehicles as described in claim 8, characterized in that, include: Read all locomotives, cargo boxes, and loading / unloading equipment with the status label of "idle" from the resource status mapping table to obtain a candidate resource set; For a group of vehicles whose group fairness gap is greater than or equal to a preset gap threshold among multiple groups, select resource samples from the candidate resource set whose expected matching cost with the current scheduling strategy is greater than or equal to a preset cost threshold, and use the selected resource samples as difficult negative samples. For vehicle groups with a group fairness gap less than a preset gap threshold among multiple groups, resource samples with a matching cost less than a preset cost threshold are selected from the candidate resource set, and the selected resource samples are used as simple negative samples.

10. A multi-modal transportation dispatching system for logistics freight vehicles based on the Internet of Things, characterized in that: The steps for implementing the IoT-based multi-modal transportation scheduling method for logistics freight vehicles according to any one of claims 1 to 9 include: The data acquisition module is used to collect real-time data on the unloading load, cab status, cargo box status, and loading and unloading equipment status of multiple dump trucks through IoT terminals, and obtain multiple multi-source real-time datasets. The health calculation module is used to calculate the dynamic health of each dump truck based on the multiple multi-source real-time datasets and obtain multiple health labels. The status resolution module is used to perform status resolution of multiple dump trucks based on the multiple multi-source real-time datasets and construct a resource status mapping table. The scheduling model acquisition module is used to divide the multiple dump trucks into multiple vehicle groups based on the cumulative dump load of the vehicles, and to perform group fairness enhancement training on the pre-trained scheduling model based on the multiple multi-source real-time datasets and resource status mapping table to obtain an enhanced scheduling model. The scheme generation module is used to input the multiple health labels, resource status mapping table and enhanced scheduling model into the multi-objective optimizer to generate a truck head-cargo box pairing scheme and an unloading point allocation scheme, and to execute transportation scheduling.

Images