A transport intelligent scheduling method and device based on trajectory data

Abstract

This invention discloses a transportation intelligent scheduling method and device based on trajectory data. The method includes designing a dynamic multi-layered electronic fence, extracting transport vehicle trajectories to identify the operation stages of transport vehicles, identifying virtual excavators to determine excavation characteristics, identifying spatiotemporal distribution to determine queuing characteristics in the work area, determining the queuing status, constructing a road network map of the work area, collecting real-time transport vehicle trajectories to update the road network map, constructing an energy consumption and emission estimation model, determining the transportation scheduling optimization objective function, adjusting the real-time paths and excavation characteristics of transport vehicles according to the road network map of the work area, recalculating the transportation scheduling optimization objective function, iteratively adjusting until the scheduling objective is met, and then outputting the transportation scheduling plan. This method can effectively improve the intelligent management level of earthwork transportation, reduce unnecessary waste in the transportation process, improve transportation efficiency and environmental friendliness, and has good interpretability, making it directly applicable to intelligent transportation scheduling devices.

Description

Technical Field

[0001] This invention relates to the fields of traffic engineering and earthwork transportation management, and in particular to a method and apparatus for intelligent transportation scheduling based on trajectory data. Background Technology

[0002] With the acceleration of urbanization and the rapid development of the construction industry, the earthwork transportation industry is facing increasingly severe challenges. Traditional earthwork transportation scheduling mainly relies on manual management and simple planning, which is inefficient and prone to problems such as traffic congestion, vehicle idling, and energy waste. Especially in large-scale earthwork projects, with the increase in the number of vehicles, how to rationally allocate transportation vehicles, avoid congestion, optimize transportation routes, and reduce energy consumption and emissions has become a key issue.

[0003] Existing earthwork transportation management systems mostly rely on static planning and simple timetables for scheduling, lacking real-time data collection and analysis. This leads to vehicles potentially stalling during transport due to traffic congestion, road problems, or equipment malfunctions, resulting in unnecessary time and energy waste. Furthermore, existing scheduling systems typically fail to provide real-time feedback on transportation status and lack intelligent energy consumption and emission monitoring functions, hindering their ability to effectively assist transportation managers in adjusting routes or optimizing transportation strategies. Therefore, this invention proposes a transportation intelligent scheduling method and device based on trajectory data. By collecting and analyzing vehicle trajectory data in real time, it dynamically adjusts transportation routes and scheduling schemes, achieving comprehensive optimization of the earthwork transportation process. Through modules for work stage identification, work area feature identification, work area road network updates, energy consumption and emission estimation, and transportation scheduling, it effectively improves transportation efficiency, reduces energy consumption and costs, and can respond to emergencies in real time, improving the safety and flexibility of the transportation process. This presents a more intelligent and precise earthwork transportation scheduling solution, providing support for the green development of the earthwork transportation industry. Summary of the Invention

[0004] The purpose of this invention is to provide a transportation intelligent scheduling method and device based on trajectory data.

[0005] To achieve the above objectives, the present invention is implemented according to the following technical solution: This invention includes the following steps: A dynamic multi-layer electronic fence is designed based on construction characteristics. The trajectory of transport vehicles is extracted, and the operation stage of transport vehicles is identified based on the dynamic multi-layer electronic fence. The dynamic multi-layer electronic fence includes a digging operation area fence and an outer monitoring fence. The operation stages of transport vehicles include the soil loading stage, soil transportation stage, soil unloading stage, and return stage. Virtual excavator identification is performed on transport vehicles in similar areas to determine excavation characteristics. Based on the spatiotemporal distribution of these vehicles, queuing characteristics in the work area are determined. Spatial density clustering analysis is used to analyze these queuing characteristics to determine the queuing status. The excavation characteristics include excavator location, number of excavators, loading time, active excavation area, excavator workload, and soil removal rate. The queuing characteristics in the work area include vehicle entry / exit frequency, average dwell time, average queuing time, average passing time, and average vehicle spacing. The work area includes an excavation area and an unloading area. A road network map of the work area is constructed based on the historical transport vehicle trajectories. The road network map of the work area is updated by collecting real-time transport vehicle trajectories. The energy consumption characteristics of historical transport vehicles are extracted to calculate fuel energy consumption and combustion emissions. An energy consumption and emission estimation model is constructed by combining the corresponding historical transport vehicle trajectories, excavation characteristics, queuing characteristics of the work area, and weather characteristics. The objective function for optimizing transportation scheduling is determined. A formulaic search is used to adjust the real-time routes of transport vehicles and excavation characteristics based on the road network map of the work area. The objective function for optimizing transportation scheduling is recalculated and iteratively adjusted until the scheduling objective is met, at which point the transportation scheduling plan is output.

[0006] Furthermore, the method for identifying the operational stage of a transport vehicle includes: The original trajectory point set of vehicles in the core coordinate point of the work area and the surrounding area in the most recent time is extracted. Using the core coordinate point of the work area as the seed, a density-based spatial clustering algorithm is used to perform cluster analysis on the surrounding trajectory points to determine the internal work vehicle point cloud cluster and the external vehicle point cloud cluster. The internal work vehicle point cloud cluster is characterized by being dense, disordered and frequently turning. The external vehicle point cloud cluster is characterized by being linear, orderly and having the same direction. Alpha Shape is used to extract the largest irregularly shaped working area that can encompass all internal point cloud clusters of working vehicles. The boundary of the largest working area is used as the working fence. A fixed range is extended outward from the largest working area to generate an outer monitoring fence. The point cloud cluster area outside the working fence is set as the transportation area channel. The working fence includes a digging area fence and a unloading area fence. The outer monitoring fence is used to lock the transportation vehicles entering the working area in advance, including the digging area outer monitoring fence and the unloading area outer monitoring fence. The system filters the trajectories of transport vehicles entering the outer monitoring fence. When a transport vehicle's trajectory enters the work fence beyond the first time threshold and its average speed is lower than the speed threshold, the system determines whether the transport vehicle is in the loading or unloading stage based on the work area. When a transport vehicle's trajectory moves from the work fence into the transport area channel and is continuously updated, the system determines whether the transport vehicle is in the soil transport stage or the return stage based on the work area. When a transport vehicle's trajectory moves outside the work fence and is not continuously updated, the system determines that it is in a non-operation stage.

[0007] Furthermore, the method for identifying and determining excavation characteristics using a virtual excavator includes: Extract the trajectory and associated information of transport vehicles within the fenced area of ​​the excavation zone during a fixed time period; the associated information includes timestamp, acceleration, orientation angle, load, fuel consumption, coordinate sequence, and vehicle identification; the trajectory of the same transport vehicle may appear multiple times at different points in time within the fixed time period; The locations and dwell times of vehicles whose dwell time exceeds a second time threshold are extracted from the trajectories of each transport vehicle within a fixed time period. A density-based spatial clustering algorithm is used to analyze the corresponding vehicle locations and determine spatial point clusters. The center point of the cluster is taken as the excavator location, the number of clusters is taken as the number of excavators, the cluster area is taken as the active excavation area, the average dwell time of a single transport vehicle is taken as the loading time, and the average cumulative loading time of the cluster area is taken as the excavator's working intensity. The cumulative load of all vehicle trajectories leaving the excavation area fence within a fixed time period is calculated, and the soil removal rate is calculated to obtain the excavation characteristics.

[0008] 5. Further, the method for determining the queuing status includes: The frequency of vehicle entry and exit and the average dwell time can be extracted directly from the trajectory of transport vehicles and the location of the fence in the excavation area within a fixed time period. The trajectories of all vehicles entering the excavation area within a fixed time period are time-aligned. Hierarchical clustering based on Euclidean distance is used to cluster all trajectory points according to their spatial location, forming multiple interaction space clusters. Each interaction space cluster represents a local area where vehicle interaction may occur. The local area includes loading points and narrow intersections. Analyze the status of all vehicles in each spatial cluster to determine the average queuing time, average passing time, and average vehicle spacing. The queuing feature vector is composed of vehicle entry and exit frequency, average dwell time, average queuing time, average passing time, and average vehicle spacing. Evaluation feature vectors are obtained by assigning evaluation weights to different queuing features. Calculate the weighted Euclidean distance between the evaluation feature vector and each historical evaluation feature vector, and select historical evaluation feature vectors whose Euclidean distance is less than the dynamic distance threshold to form a neighborhood point set; The congestion influence of each historical evaluation feature vector within the neighborhood point set on the evaluation feature vector is calculated, and the baseline congestion index for the corresponding time period of each historical evaluation feature vector is extracted to calculate the comprehensive queuing status evaluation value; the baseline congestion index is obtained through long-term statistics. The expression for the comprehensive queuing status evaluation value is as follows: ; ; in This is a comprehensive queuing status assessment value. To evaluate the feature vector, For historical evaluation feature vectors, For the neighborhood point set, Historical evaluation feature vector For evaluating feature vectors The impact of congestion Historical evaluation feature vector The benchmark congestion index, Historical evaluation feature vector The local density of the corresponding region, i.e., the number of historical evaluation feature vectors. It is a smoothing constant; The queuing status is determined by comparing the comprehensive queuing status assessment value with the preset assessment threshold.

[0009] Furthermore, the method for updating the road network map of the work area includes: The Douglas-Puk algorithm is used to simplify the historical transport vehicle trajectory, identify significant turning points as intersections, and designate the excavator location as the loading point. The intersection, loading point, and entrance / exit of the excavation area fence are used as graph nodes. The average travel time of corresponding historical transport vehicle trajectories between two adjacent graph nodes is taken as the initial travel time of the corresponding adjacent graph node edge. The average passing distance and average passing time of corresponding historical transport vehicle trajectories between two adjacent graph nodes with passing interactions are extracted and used as the initial passing distance and initial passing time of the corresponding adjacent graph node edge. The edge weights are calculated based on the initial travel time, initial passing distance, and initial passing time of the adjacent graph node edge. The initial road network graph model of the work area is constructed, and real-time trajectories are continuously collected. The trajectory similarity function is used to identify new paths and calculate the edge weights of each segment of the new path. The expression is: ; ; in For the adjacent graph nodes of the new path, edges The edge weights of the corresponding road segments, For road section The initial passage time, For road section The length of the road segment Standard road speed limit Weighting of passing time As the passing distance weight, For road section The initial passing time, Standard passing time, For road section The initial passing distance, Standard passing distance, For a new path With historical road network trajectory similarity, This is the distance attenuation coefficient. For new trajectory points With historical road network The minimum distance, For a new path The number of tracking points between China and Singapore; The product of the mean edge weight and the trajectory similarity is calculated as the update score. When the update score is less than the update score threshold, a new path is used to update the road network model.

[0010] Furthermore, the method for constructing the energy consumption and emission estimation model includes: Extracting historical energy consumption characteristics of transport vehicles to calculate fuel consumption and combustion emissions, and extracting historical operating characteristics of each transport vehicle within the excavation area fence based on historical transport vehicle trajectories; the operating characteristics include the number of turns, the cumulative absolute value of turns, the travel distance, and the number of intersections; A comprehensive estimation dataset is composed of fuel energy consumption, combustion emissions, and corresponding historical operating characteristics, excavation characteristics, work area queuing characteristics, and weather characteristics. The comprehensive estimation dataset is divided into a training set and a test set according to the plan in a 6:4 ratio. The energy consumption and emission estimation model is trained using the training set, and the performance of the energy consumption and emission estimation model is tested using the test set. The energy consumption and emission estimation model includes an input layer, a feature fusion layer, an energy consumption estimation layer, an emission estimation layer, and an output layer. The feature fusion layer uses a self-attention mechanism to weightedly fuse operational features, excavation features, work area queuing features, and weather features to obtain fused estimation features; the energy consumption estimation layer uses a BP neural network to capture the nonlinear relationship between fuel energy consumption and fused estimation features to predict fuel energy consumption; the emission estimation layer embeds an emission factor mapping model to predict combustion emissions based on the predicted fuel energy consumption; the emission factor mapping model is fitted according to vehicle type and emission standards; the output layer is connected to the emission estimation layer through a fully connected layer to output the predicted fuel energy consumption and predicted combustion emissions. The energy consumption and emission estimation model uses a composite weighted mean square error loss function to improve the model's prediction accuracy and an adaptive moment estimator optimizes the model parameters; the composite weighted mean square error loss function includes fuel energy consumption prediction loss and combustion emission prediction loss.

[0011] Furthermore, the method for determining the objective function for transportation scheduling optimization includes: Based on queuing state characteristics, energy consumption costs, and pollution emission penalties, a multi-objective dynamic scheduling optimization objective function is constructed, expressed as follows: ; in To optimize the objective function for transportation scheduling, , To avoid delays and reduce costs, , For energy consumption cost weighting and energy consumption cost, , To determine the pollution weight and pollution emission costs, This is the time penalty coefficient. This is the energy consumption penalty coefficient. To account for delays, including average queuing time and average passing time, Standard delay time, For target fuel energy consumption, pollutants A set of.

[0012] Secondly, a transportation intelligent scheduling device based on trajectory data includes: Operation phase identification module: used to design dynamic multi-layer electronic fences according to construction characteristics, extract the trajectory of transport vehicles and identify the operation phase of transport vehicles based on the dynamic multi-layer electronic fences; Work area feature recognition module: used to identify excavators from similar areas to determine excavation features, identify the spatiotemporal distribution of similar areas to determine queuing features in the work area, and use spatial density clustering analysis to determine queuing status in the work area queuing features. Work area road network module: used to build a work area road network map based on historical transport vehicle trajectories, and collect real-time transport vehicle trajectories to update the work area road network map; Energy consumption and emission estimation module: used to extract the energy consumption characteristics of historical transport vehicles to calculate fuel energy consumption and combustion emissions, and to build an energy consumption and emission estimation model by combining the corresponding historical transport vehicle trajectory, excavation characteristics, work area queuing characteristics and weather characteristics. Transportation scheduling module: Used to determine the objective function for transportation scheduling optimization. It uses a formulaic search to adjust the real-time routes of transport vehicles and excavation characteristics based on the road network map of the work area, recalculates the objective function for transportation scheduling optimization, and iterates until the scheduling objective is met, then outputs the transportation scheduling plan.

[0013] The beneficial effects of this invention are: This invention relates to a transportation intelligent scheduling method and device based on trajectory data. Compared with the prior art, this invention has the following technical advantages: The intelligent transportation scheduling method and device based on trajectory data provided by this invention can accurately identify the operating status of muck transport vehicles under various complex construction conditions, and complete the global optimization of transportation routes and vehicle assignments based on optimization algorithms. Through real-time trajectory acquisition, electronic fence determination, automatic status recognition, and ant colony algorithm scheduling, this invention effectively improves the real-time performance and rationality of scheduling decisions. Field test results show that this invention has high recognition accuracy and scheduling stability, can significantly improve the efficiency of earthwork transportation, and has good engineering application value and promotion potential. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the steps of a transportation intelligent scheduling method based on trajectory data according to the present invention. Detailed Implementation

[0015] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0016] The present invention provides a transportation intelligent scheduling method and apparatus based on trajectory data, comprising the following steps: like Figure 1 As shown, this embodiment includes the following steps: A dynamic multi-layer electronic fence is designed based on construction characteristics. The trajectory of transport vehicles is extracted, and the operation stage of transport vehicles is identified based on the dynamic multi-layer electronic fence. The dynamic multi-layer electronic fence includes a digging operation area fence and an outer monitoring fence. The operation stages of transport vehicles include the soil loading stage, soil transportation stage, soil unloading stage, and return stage. Virtual excavator identification is performed on transport vehicles in similar areas to determine excavation characteristics. Based on the spatiotemporal distribution of these vehicles, queuing characteristics in the work area are determined. Spatial density clustering analysis is used to analyze these queuing characteristics to determine the queuing status. The excavation characteristics include excavator location, number of excavators, loading time, active excavation area, excavator workload, and soil removal rate. The queuing characteristics in the work area include vehicle entry / exit frequency, average dwell time, average queuing time, average passing time, and average vehicle spacing. The work area includes an excavation area and an unloading area. A road network map of the work area is constructed based on the historical transport vehicle trajectories. The road network map of the work area is updated by collecting real-time transport vehicle trajectories. The energy consumption characteristics of historical transport vehicles are extracted to calculate fuel energy consumption and combustion emissions. An energy consumption and emission estimation model is constructed by combining the corresponding historical transport vehicle trajectories, excavation characteristics, queuing characteristics of the work area, and weather characteristics. The objective function for optimizing transportation scheduling is determined. A formulaic search is used to adjust the real-time routes of transport vehicles and excavation characteristics based on the road network map of the work area. The objective function for optimizing transportation scheduling is recalculated and iteratively adjusted until the scheduling objective is met, at which point the transportation scheduling plan is output.

[0017] In this embodiment, the method for identifying the operational stage of a transport vehicle includes: The original trajectory point set of vehicles in the core coordinate point of the work area and the surrounding area in the most recent time is extracted. Using the core coordinate point of the work area as the seed, a density-based spatial clustering algorithm is used to perform cluster analysis on the surrounding trajectory points to determine the internal work vehicle point cloud cluster and the external vehicle point cloud cluster. The internal work vehicle point cloud cluster is characterized by being dense, disordered and frequently turning. The external vehicle point cloud cluster is characterized by being linear, orderly and having the same direction. Alpha Shape is used to extract the largest irregularly shaped working area that can encompass all internal point cloud clusters of working vehicles. The boundary of the largest working area is used as the working fence. A fixed range is extended outward from the largest working area to generate an outer monitoring fence. The point cloud cluster area outside the working fence is set as the transportation area channel. The working fence includes a digging area fence and a unloading area fence. The outer monitoring fence is used to lock the transportation vehicles entering the working area in advance, including the digging area outer monitoring fence and the unloading area outer monitoring fence. The system filters the trajectories of transport vehicles entering the outer monitoring fence. When the trajectory of a transport vehicle enters the work fence beyond the first time threshold and the average speed is lower than the speed threshold, the transport vehicle is determined to be in the loading or unloading stage based on the work area. When the trajectory of a transport vehicle moves from the work fence into the transport area channel and is continuously updated, the transport vehicle is determined to be in the soil transport stage or the return stage based on the work area. When the trajectory of a transport vehicle moves out of the work fence and is not continuously updated, it is determined to be in a non-work stage. In the actual assessment, the earthwork transportation scheduling of the new project X is carried out by the scheduling system. The system obtains the original trajectory point set of all transportation vehicles that have been active in the project area in the past 24 hours. Each point contains a timestamp and latitude and longitude coordinates. At the same time, the core coordinate points of the excavation area and the unloading area (excavation area center point W0, unloading area center point X0) are obtained from the engineering drawings. To ensure data continuity and high accuracy, this invention installs a mobile data acquisition device on each transport vehicle. This device integrates a high-precision GPS module, accelerometer, gyroscope, weighing and communication module, and can continuously record information such as the vehicle's latitude and longitude, speed, acceleration, heading angle, timestamp, fuel consumption, and load at a frequency of 1Hz to 5Hz. All data is uploaded to the cloud dispatch server in real time via cellular network or Wi-Fi. The raw vehicle trajectory data collected is subject to limitations such as GPS positioning accuracy, signal obstruction, differences in device sampling rates, and the randomness of driving behavior, resulting in a certain degree of noise and deviation. To ensure the accuracy of subsequent analysis and scheduling algorithms, this invention designs a trajectory standardization process that integrates data cleaning, trajectory smoothing, and geometric projection correction. Data cleaning phase: The system uses velocity thresholding and acceleration constraint methods to remove abnormal jump points and drift points. When the velocity change of consecutive sampling points exceeds the vehicle dynamics limit (e.g., instantaneous acceleration exceeds 3 m / s²), the system removes these points. 2 Or instantaneous deceleration exceeding -4 m / s 2 If the signal is unstable, the point is identified as an abnormal location point and removed. To correct timestamp drift caused by signal instability, the system uses a time window-based interpolation algorithm to ensure the temporal continuity of trajectory data. Trajectory smoothing stage: Kalman filtering and moving average algorithms are introduced to dynamically smooth the trajectory coordinate sequence, eliminating random jitter and preserving the true driving trend; Geometric Projection Correction Stage: A trajectory geometric projection correction method based on the road centerline is proposed. This method uses the high-frequency trajectory of long-term vehicle operation as a basis, extracts multiple typical driving paths, and generates the corresponding road centerline model through curve fitting algorithms (such as spline interpolation or least squares fitting). This centerline represents the geometric reference of the actual road where the vehicle is driving and is used to correct the offset error of a single trajectory. For a single trajectory to be corrected, the system performs geometric projection according to the following rules: When the vehicle is driving on a straight section of the road, the vertical projection method is used, that is, the trajectory points are orthogonally projected to the road centerline according to the shortest distance; when the vehicle is driving on a turning or bend section of the road, the curve tangent projection method is used. The system first calculates the tangential direction of the centerline corresponding to the trajectory point, and adjusts the position of the projection point according to the angle between the vehicle's driving direction and the tangent direction to eliminate the curvature effect, thereby avoiding the spatial error caused by simple vertical projection at curves. The corrected trajectory not only fits the road geometry better in space, but also significantly improves the consistency between multi-vehicle trajectory data, which is convenient for subsequent trajectory-based clustering analysis, energy consumption estimation, and path recognition. Standardization processing: Unifying the timestamp format, coordinate system (WGS84 or GCJ02) and status labels (excavation, earthmoving, unloading, return). The average spatial error of the trajectory data after this processing is controlled within 3 meters, which improves the accuracy by about 40% compared with the original data, providing a highly reliable input basis for subsequent virtual excavator identification, queuing early warning and energy consumption analysis. Taking the excavation area as an example, the system uses the core coordinate point W0 as the initial seed and analyzes all trajectory points within a 500-meter radius around it using a density-based spatial clustering algorithm. The analysis found that the trajectory points are clearly divided into two types of clusters: (1) Internal operation vehicle point cloud cluster (Cluster_A): This cluster has a high point density, a relatively concentrated distribution range but an irregular shape, and the connection directions between points are chaotic, showing frequent turning and stopping characteristics, which is consistent with the behavior pattern of vehicles loading, moving and waiting in the pit; (2) External vehicle point cloud cluster (Cluster_B): This cluster has a low point density, and the point sequence can be connected into a clear, consistent linear path, which is consistent with the characteristics of vehicles driving normally on roads outside the construction area; The same analysis was performed on the core coordinate point X0 of the unloading area, and similar internal operation point cloud clusters (Cluster_C) and external path point cloud clusters (Cluster_D) were obtained. For the identified internal work point cluster Cluster_A, the Alpha Shape algorithm is used to process and generate an irregular polygonal boundary that is close to all external points and may contain depressions, resulting in an irregular polygonal region R_w with an area of ​​approximately 8500 square meters. This region accurately encompasses all trajectory points related to loading operations and is defined as the maximum work area of ​​the excavation zone. The boundary of the maximum work area is directly set as the excavation zone fence. Similarly, the internal work point cluster Cluster_C is processed to obtain the unloading zone fence. The main coherent path area outside the work fence and represented by the external vehicle point clusters (Cluster_B, Cluster_D) is defined as the transportation zone channel. The system continuously monitors vehicle trajectories. When a vehicle's trajectory point enters a certain outer monitoring fence (taking the outer monitoring fence of the excavation area as an example, the fixed range extending outward is 15m), the system marks the vehicle as "approaching the work area". If the vehicle's subsequent trajectory enters the corresponding work fence (excavation area fence), and the transport vehicle's trajectory enters the work fence for more than the first time threshold (30s) and the average speed is lower than the speed threshold (5km / h), the transport vehicle is determined to be in the soil loading stage. When the transport vehicle's trajectory moves from the excavation area fence into the transport area channel and is continuously updated, the transport vehicle is determined to be in the soil transport stage.

[0018] In this embodiment, the method for identifying and determining excavation characteristics using a virtual excavator includes: Extract the trajectory and associated information of transport vehicles within the fenced area of ​​the excavation zone during a fixed time period; the associated information includes timestamp, acceleration, orientation angle, load, fuel consumption, coordinate sequence, and vehicle identification; the trajectory of the same transport vehicle may appear multiple times at different points in time within the fixed time period; The locations and dwell times of vehicles whose dwell time exceeds a second time threshold are extracted from the trajectories of each transport vehicle within a fixed time period. A density-based spatial clustering algorithm is used to analyze the corresponding vehicle locations and determine spatial point clusters. The center point of the cluster is taken as the excavator location, the number of clusters is taken as the number of excavators, the cluster area is taken as the active excavation area, the average dwell time of a single transport vehicle is taken as the loading time, and the average cumulative loading time of the cluster area is taken as the working intensity of the excavator. The cumulative load of all vehicle trajectories that leave the excavation area fence within a fixed time period is calculated and the soil removal rate is calculated to obtain the excavation characteristics. In actual evaluation, taking the earthwork transportation scheduling of the X new construction project as an example, all 30 muck trucks were equipped with mobile data collection devices to collect transportation trajectories and corresponding information in real time. All data were organized and stored using vehicle identifiers (such as license plate numbers or on-board device IDs) as indexes. Taking a fixed time period from 8:00 to 10:00 as an example, the system queries the database for the trajectory data and related information of all vehicles that have entered the "excavation area A fence". For the same vehicle, its trajectory may have multiple discontinuous trajectory segments due to multiple round trips during this time period. For example, vehicle V001 entered the excavation area to complete loading at 8:15-8:30 and 9:30-9:45 respectively. The system analyzes each extracted trajectory segment and identifies time periods when the vehicle stays within the excavation area fence for more than a second time threshold (set to 120 seconds in this embodiment). This threshold is used to filter out brief stops caused by yielding, U-turns, etc., and focuses on high-probability loading operations. For example, the system identifies that vehicle V001 was basically stationary near the coordinate point (E116.403, N39.904) from 8:17:30 to 8:27:30 for 600 seconds, which exceeds the threshold. Therefore, the location (taking the average coordinates during the stay) and the stay time (600 seconds) are recorded as a valid loading event. The system performs the same processing on all vehicles within this time period, resulting in a total of 45 valid loading events. Each event includes its geographical location and duration. The system takes the geographical locations of the 45 loading events as input, uses a density-based spatial clustering algorithm to analyze the corresponding vehicle locations, and determines 5 spatial point groups (the corresponding spatial point group area is the active excavation area, and the number of excavators is 5). The center point of the spatial point group is taken as the excavator location. The loading time is taken as the average single dwell time of the transport vehicle, which is 580s. The excavator's working intensity is taken as the average cumulative loading time of each spatial point group, which is 0.80 (the average cumulative loading time within 2 hours is 96min). The earthwork discharge rate is taken as the ratio of the cumulative load of all vehicles when they leave the excavation area fence to the fixed time period, which is 540 cubic meters / hour (corresponding to an earthwork discharge rate of 108 cubic meters / hour for each excavator).

[0019] 6. In this embodiment, the method for determining the queuing status includes: The frequency of vehicle entry and exit and the average dwell time can be extracted directly from the trajectory of transport vehicles and the location of the fence in the excavation area within a fixed time period. The trajectories of all vehicles entering the excavation area within a fixed time period are time-aligned. Hierarchical clustering based on Euclidean distance is used to cluster all trajectory points according to their spatial location, forming multiple interaction space clusters. Each interaction space cluster represents a local area where vehicle interaction may occur. The local area includes loading points and narrow intersections. Analyze the status of all vehicles in each spatial cluster to determine the average queuing time, average passing time, and average vehicle spacing. The queuing feature vector is composed of vehicle entry and exit frequency, average dwell time, average queuing time, average passing time, and average vehicle spacing. Evaluation feature vectors are obtained by assigning evaluation weights to different queuing features. Calculate the weighted Euclidean distance between the evaluation feature vector and each historical evaluation feature vector, and select historical evaluation feature vectors whose Euclidean distance is less than the dynamic distance threshold to form a neighborhood point set; The congestion influence of each historical evaluation feature vector within the neighborhood point set on the evaluation feature vector is calculated, and the baseline congestion index for the corresponding time period of each historical evaluation feature vector is extracted to calculate the comprehensive queuing status evaluation value; the baseline congestion index is obtained through long-term statistics. The expression for the comprehensive queuing status evaluation value is as follows: ; ; in This is a comprehensive queuing status assessment value. To evaluate the feature vector, For historical evaluation feature vectors, For the neighborhood point set, Historical evaluation feature vector For evaluating feature vectors The impact of congestion Historical evaluation feature vector The benchmark congestion index, Historical evaluation feature vector The local density of the corresponding region, i.e., the number of historical evaluation feature vectors. It is a smoothing constant; The queuing status is determined by comparing the comprehensive queuing status assessment value with the preset assessment threshold. In actual assessment, the number of times the transport vehicle trajectory intersects with the excavation area fence within a fixed time period is directly extracted. The frequency of vehicle entry and exit per unit time is calculated based on the number of intersections. The time length between the intersection points of the same transport vehicle trajectory and the excavation area fence is counted, and the average time length between all intersection points is taken as the average dwell time. The time length between the intersection points is taken as the starting point of the time of entering the excavation area fence and the ending point of the time of leaving the excavation area fence. Analyze the state of all vehicles within each spatial cluster: When the interaction spatial cluster is located within a group of spatial points, there are at least two transport vehicles within the interaction spatial cluster, and the corresponding vehicles are simultaneously stationary, it is determined that a queuing interaction has occurred within the interaction spatial cluster. The continuous time during which each pair of vehicles is simultaneously stationary is extracted as the duration of a single queuing event. When there are two transport vehicles within the interaction spatial cluster, and the instantaneous spatial distance between them is less than the unobstructed distance threshold and their speeds are both greater than the movement threshold, it is determined that a passing interaction has occurred within the interaction spatial cluster. The continuous time during which the corresponding vehicles satisfy the passing interaction is extracted as the duration of a single passing event, and the corresponding instantaneous spatial distance is extracted as the passing interval. The average queuing time is the average of the single queuing events of all transport vehicles; the sum of the times of all passing events of the same transport vehicle during the soil loading stage is the passing time of a single transport time, and the average passing time of a single transport time of transport vehicles is the average passing time; the average passing distance of all transport vehicles is the average vehicle spacing. Taking the earthwork transportation scheduling of the X new construction project as an example, during the fixed time period from 8:00 to 9:00, the vehicle entry and exit frequency of 28 vehicles / hour and the average dwell time of 580 seconds were directly extracted. The status of all vehicles in each spatial cluster was analyzed to determine the average queuing time of 105 seconds, the average passing time of 30 seconds, and the average vehicle spacing of 5.2 meters. The queuing feature vector was determined to be [28, 580, 105, 30, 5.2]. The queuing feature vector is adjusted based on the evaluation weight vector [0.3, 0.25, 0.3, 0.1, 0.05] to obtain the evaluation feature vector [8.4, 145, 31.5, 3, 0.26]. After standardization, it becomes [0.36, 0.375, 0.54, 0.008, -0.025]. The weighted Euclidean distance between the evaluation feature vector and each historical evaluation feature vector is calculated. Historical evaluation feature vectors with a weighted Euclidean distance less than the dynamic distance threshold of 0.15 are selected to form a neighborhood point set, and the comprehensive queuing state evaluation value is calculated to be 1.58 (local density). Represents the historical evaluation feature vector The number of historical evaluation feature vectors within a certain radius neighborhood in the historical database, if If there are 3 other historical vectors within a radius of 0.1, then Take 3; Baseline Congestion Index From 1 / 2 / 3, they are respectively labeled "Smooth Traffic" / "Light Congestion" / "Congestion"), according to the queuing status mapping ( - Smooth flow, - Mild congestion -Congestion), which determines the queuing status in the excavation area during the fixed time period from 8:00 to 9:00 as congestion.

[0020] In this embodiment, the method for updating the road network map of the work area includes: The Douglas-Puk algorithm is used to simplify the historical transport vehicle trajectory, identify significant turning points as intersections, and designate the excavator location as the loading point. The intersection, loading point, and entrance / exit of the excavation area fence are used as graph nodes. The average travel time of corresponding historical transport vehicle trajectories between two adjacent graph nodes is taken as the initial travel time of the corresponding adjacent graph node edge. The average passing distance and average passing time of corresponding historical transport vehicle trajectories between two adjacent graph nodes with passing interactions are extracted and used as the initial passing distance and initial passing time of the corresponding adjacent graph node edge. The edge weights are calculated based on the initial travel time, initial passing distance, and initial passing time of the adjacent graph node edge. The initial road network graph model of the work area is constructed, and real-time trajectories are continuously collected. The trajectory similarity function is used to identify new paths and calculate the edge weights of each segment of the new path. The expression is: ; ; in For the adjacent graph nodes of the new path, edges The edge weights of the corresponding road segments, For road section The initial passage time, For road section The length of the road segment Standard road speed limit Weighting of passing time As the passing distance weight, For road section The initial passing time, Standard passing time, For road section The initial passing distance, Standard passing distance, For a new path With historical road network trajectory similarity, This is the distance attenuation coefficient. For new trajectory points With historical road network The minimum distance, For a new path The number of tracking points between China and Singapore; The product of the mean edge weight and the trajectory similarity is calculated as the update score. When the update score is less than the update score threshold, a new path is used to update the road network model. In the actual assessment, taking the X new construction project on the 10th day as an example, due to the advancement of the work surface, a new temporary access road was opened on site, which leads directly from the intersection J5 to the soil loading point E2. The system continuously monitors the real-time trajectory. In one hour in the morning, the system detected that the trajectories of multiple vehicles traveling from J5 to E2 (25 new trajectory points and 2 new road sections) were significantly different from any path in the historical road network (such as detouring via J8). Using a standard passing distance of 5m, a standard passing time of 6s, a standard road segment speed of 8m / s, a passing time weight of 0.1, a passing distance weight of 0.15, and a distance decay coefficient of 0.8, the road segment edge weights and trajectory similarity are calculated to be 0.31 / 0.4 and 0.45, respectively. The update score is calculated to be 0.16 (less than the update score threshold of 0.2). The new path is used to update the road network graph model, and the edges of each adjacent graph node are associated with the initial travel time, initial passing distance, and initial passing time of the corresponding road segment.

[0021] In this embodiment, the method for constructing the energy consumption and emission estimation model includes: Extracting historical energy consumption characteristics of transport vehicles to calculate fuel consumption and combustion emissions, and extracting historical operating characteristics of each transport vehicle within the excavation area fence based on historical transport vehicle trajectories; the operating characteristics include the number of turns, the cumulative absolute value of turns, the travel distance, and the number of intersections; A comprehensive estimation dataset is composed of fuel energy consumption, combustion emissions, and corresponding historical operating characteristics, excavation characteristics, work area queuing characteristics, and weather characteristics. The comprehensive estimation dataset is divided into a training set and a test set according to the plan in a 6:4 ratio. The energy consumption and emission estimation model is trained using the training set, and the performance of the energy consumption and emission estimation model is tested using the test set. The energy consumption and emission estimation model includes an input layer, a feature fusion layer, an energy consumption estimation layer, an emission estimation layer, and an output layer. The feature fusion layer uses a self-attention mechanism to weightedly fuse operational features, excavation features, work area queuing features, and weather features to obtain fused estimation features; the energy consumption estimation layer uses a BP neural network to capture the nonlinear relationship between fuel energy consumption and fused estimation features to predict fuel energy consumption; the emission estimation layer embeds an emission factor mapping model to predict combustion emissions based on the predicted fuel energy consumption; the emission factor mapping model is fitted according to vehicle type and emission standards; the output layer is connected to the emission estimation layer through a fully connected layer to output the predicted fuel energy consumption and predicted combustion emissions. The energy consumption and emission estimation model uses a composite weighted mean square error loss function to improve the model's prediction accuracy and an adaptive moment estimator optimizes the model parameters; the composite weighted mean square error loss function includes fuel energy consumption prediction loss and combustion emission prediction loss; In actual assessments, the method for calculating fuel consumption and combustion emissions based on the historical energy consumption characteristics of transport vehicles is as follows: The instantaneous power model (IPM model) of vehicle dynamics is used to calculate the fuel consumption of each transport vehicle. The IPM model uses the speed and acceleration of each transport vehicle in the transport trajectory as input variables and calculates the instantaneous power consumption based on the principle of vehicle force balance. ; ; ; ; in For instantaneous power consumption, To determine the effective payload mass, the virtual excavator identification module automatically switches between loaded and empty states. To accelerate the transport vehicles, For the speed of transport vehicles, For rolling resistance, This is the roller resistance coefficient. It is the acceleration due to gravity. For air resistance, air density, The air drag coefficient, The vehicle's frontal area is determined by extracting the corresponding vehicle model using the vehicle identifier. For slope resistance, The road slope angle; Instantaneous fuel consumption rate is calculated based on engine combustion efficiency model and lower heating value of fuel, and then fuel energy consumption during the vehicle's entry and exit from the excavation zone enclosure is obtained through numerical integration: ; ; in Instantaneous fuel consumption rate For engine combustion efficiency, The lower heating value of fuel oil. For vehicles Fuel consumption during the period of entering and exiting the excavation area fence. The time it takes for transport vehicles to enter the fenced area of ​​the excavation zone. The time it takes for transport vehicles to leave the fenced area of ​​the excavation zone; Calculate the fuel consumption of all transport vehicles within the excavation zone enclosure within a unit time period, and use this as the unit time transportation fuel consumption of the excavation zone. Input the unit time transportation fuel consumption of the excavation zone into the emission factor mapping model to calculate the combustion emissions of the excavation zone per unit time: ; in pollutants Combustion emissions, pollutants The fuel emission factor is automatically matched with the factor value based on the vehicle type and emission standard (such as China V and China VI).

[0022] In this embodiment, the method for determining the objective function for transportation scheduling optimization includes: Based on queuing state characteristics, energy consumption costs, and pollution emission penalties, a multi-objective dynamic scheduling optimization objective function is constructed, expressed as follows: ; in To optimize the objective function for transportation scheduling, , To avoid delays and reduce costs, , For energy consumption cost weighting and energy consumption cost, , To determine the pollution weight and pollution emission costs, This is the time penalty coefficient. This is the energy consumption penalty coefficient. To account for delays, including average queuing time and average passing time, Standard delay time, For target fuel energy consumption, pollutants A set; In the actual assessment, the Analytic Hierarchy Process (AHP) was used to comprehensively determine the weights of delay time, energy cost, and emissions pollution by combining the opinions of project management, scheduling experts, and environmental protection departments. A three-tiered judgment matrix of "efficiency-cost-environmental protection" was constructed, and consistency checks were performed to ultimately determine the relative weights of each objective to reflect the project's strategic priority (e.g., during the project's expedited construction phase, the time weight may be increased; in areas with strict environmental controls, emission rights may be increased). In the X new construction project (during the project's expedited construction phase), the weights of delay time, energy cost, and emissions pollution were set to 0.5, 0.3, and 0.2, respectively, with time penalty coefficients and energy penalty coefficients set to 1.5 and 0.8, respectively (when the actual delay time exceeds the standard delay time, the cost will be significantly amplified; when the actual fuel consumption exceeds the target fuel consumption, additional cost penalties will be incurred). The pollutant set includes carbon dioxide (CO2) and nitrogen oxides (NOx). xand fine particulate matter (PM) 2.5 ; Taking the earthwork transportation scheduling of the new X project as an example, optimize the transportation scheduling plan for the fixed time period from 9:00 to 10:00. Take the transportation scheduling plan from 9:00 to 10:00 as the initial plan, and use A* heuristic search to perform heuristic search within the excavator feature range (adjust the excavator position in the excavation area, adjust the number of working excavators according to the number of excavators on the construction site, and adjust the excavator excavation speed) to update the excavation area features. Extract the new excavator locations and the entrances / exits of the excavation area fence. Generate new transport vehicle trajectories based on the road network map of the work area (calculate the Euclidean distance between the existing excavator locations and the historical loading points, and take the transport vehicle trajectory corresponding to the historical loading point with the smallest Euclidean distance as the new transport vehicle trajectory; the entry time of each transport vehicle is determined according to the transport vehicle operation phase time before the update; after the transport vehicles enter the site, allocate the excavator that starts loading the fastest and is closest to the site for loading; the loading time of each excavator is determined by the excavator's digging speed). Set up the new transport vehicle... The vehicle trajectories are overlaid, and the road segment information (initial passage time, initial passing distance, and initial passing time) of the graph node edges contained in the transport vehicle trajectory is calculated. If the road segment information exceeds the corresponding threshold, the transport vehicle trajectory of the corresponding road segment is finely adjusted (if there are multiple transport vehicle trajectories on the same road segment (including different vehicles from the same loading point passing through the same road segment, or different vehicles from different loading points passing through the same road segment), the transport vehicle trajectory of that road segment is diverted to an adjacent road segment), new transport vehicle trajectories are obtained, and spatiotemporal distribution identification is performed to determine the queuing characteristics of the work area, and the operation characteristics are calculated; The new work area queuing characteristics, excavation characteristics, operation characteristics, and corresponding weather characteristics (wind speed, rainfall, temperature, visibility) are input into the energy consumption and emission estimation model to obtain predicted fuel energy consumption and predicted fuel emissions. The transportation scheduling optimization objective function is then recalculated. The iteration is repeated until the maximum number of iterations is reached or the rate of change of the transportation scheduling optimization objective function is less than 0.5% for 5 consecutive iterations. The latest excavation characteristics (excavator location, number of excavators, excavator workload) and the transportation trajectory of the corresponding transport vehicles for each excavator (only the corresponding two-dimensional driving path is output, without the corresponding time information) are used as the optimal transportation scheduling scheme.

[0023] Secondly, a transportation intelligent scheduling device based on trajectory data includes: Operation phase identification module: used to design dynamic multi-layer electronic fences according to construction characteristics, extract the trajectory of transport vehicles and identify the operation phase of transport vehicles based on the dynamic multi-layer electronic fences; Work area feature recognition module: used to identify excavators from similar areas to determine excavation features, identify the spatiotemporal distribution of similar areas to determine queuing features in the work area, and use spatial density clustering analysis to determine queuing status in the work area queuing features. Work area road network module: used to build a work area road network map based on historical transport vehicle trajectories, and collect real-time transport vehicle trajectories to update the work area road network map; Energy consumption and emission estimation module: used to extract the energy consumption characteristics of historical transport vehicles to calculate fuel energy consumption and combustion emissions, and to build an energy consumption and emission estimation model by combining the corresponding historical transport vehicle trajectory, excavation characteristics, work area queuing characteristics and weather characteristics. Transportation scheduling module: Used to determine the objective function for transportation scheduling optimization. It uses a formulaic search to adjust the real-time routes of transport vehicles and excavation characteristics based on the road network map of the work area, recalculates the objective function for transportation scheduling optimization, and iterates until the scheduling objective is met, then outputs the transportation scheduling plan.

[0024] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A transportation intelligent scheduling method based on trajectory data, characterized in that, Includes the following steps: S1. Design a dynamic multi-layer electronic fence based on construction characteristics, extract the trajectory of transport vehicles, and identify the operation stage of transport vehicles based on the dynamic multi-layer electronic fence; the dynamic multi-layer electronic fence includes a digging operation area fence and an outer monitoring fence. The operation phases of the transport vehicles include the loading phase, the transport phase, the unloading phase, and the return phase. S2. Virtual excavator identification is performed on transport vehicles in similar areas to determine excavation characteristics. Based on the spatiotemporal distribution of transport vehicles in similar areas, queuing characteristics of the work area are determined. Spatial density clustering analysis is used to analyze the queuing characteristics of the work area to determine the queuing status. The excavation characteristics include excavator location, number of excavators, loading time, active excavation area, excavator workload, and soil removal rate. The queuing characteristics of the work area include vehicle entry and exit frequency, average dwell time, average queuing time, average passing time, and average vehicle spacing. The work area includes an excavation area and an unloading area. S3. Construct a road network map of the work area based on the historical transport vehicle trajectories, collect real-time transport vehicle trajectories to update the road network map of the work area, extract the energy consumption characteristics of historical transport vehicles to calculate fuel energy consumption and combustion emissions, and construct an energy consumption and emission estimation model by combining the corresponding historical transport vehicle trajectories, excavation characteristics, work area queuing characteristics and weather characteristics. S4. Determine the objective function for transportation scheduling optimization. Use formulaic search to adjust the real-time routes of transport vehicles and excavation characteristics based on the road network map of the work area. Recalculate the objective function for transportation scheduling optimization and iterate until the scheduling objective is met, then output the transportation scheduling plan.

2. The intelligent transportation scheduling method based on trajectory data according to claim 1, characterized in that, The method for identifying the operational stage of a transport vehicle includes: The original trajectory point set of vehicles in the core coordinate point of the work area and the surrounding area in the most recent time is extracted. Using the core coordinate point of the work area as the seed, a density-based spatial clustering algorithm is used to perform cluster analysis on the surrounding trajectory points to determine the internal work vehicle point cloud cluster and the external vehicle point cloud cluster. The internal work vehicle point cloud cluster is characterized by being dense, disordered and frequently turning. The external vehicle point cloud cluster is characterized by being linear, orderly and having the same direction. Alpha Shape is used to extract the largest irregularly shaped working area that can encompass all internal point cloud clusters of working vehicles. The boundary of the largest working area is used as the working fence. A fixed range is extended outward from the largest working area to generate an outer monitoring fence. The point cloud cluster area outside the working fence is set as the transportation area channel. The working fence includes a digging area fence and a unloading area fence. The outer monitoring fence is used to lock the transportation vehicles entering the working area in advance, including the digging area outer monitoring fence and the unloading area outer monitoring fence. The system filters the trajectories of transport vehicles entering the outer monitoring fence. When a transport vehicle's trajectory enters the work fence beyond the first time threshold and its average speed is lower than the speed threshold, the system determines whether the transport vehicle is in the loading or unloading stage based on the work area. When a transport vehicle's trajectory moves from the work fence into the transport area channel and is continuously updated, the system determines whether the transport vehicle is in the soil transport stage or the return stage based on the work area. When a transport vehicle's trajectory moves outside the work fence and is not continuously updated, the system determines that it is in a non-operation stage.

3. The intelligent transportation scheduling method based on trajectory data according to claim 1, characterized in that, The method for identifying and determining excavation characteristics using virtual excavators includes: Extract the trajectory and associated information of transport vehicles within the fenced area of ​​the excavation zone during a fixed time period; the associated information includes timestamp, acceleration, orientation angle, load, fuel consumption, coordinate sequence, and vehicle identification; the trajectory of the same transport vehicle may appear multiple times at different points in time within the fixed time period; The locations and dwell times of vehicles whose dwell time exceeds a second time threshold are extracted from the trajectories of each transport vehicle within a fixed time period. A density-based spatial clustering algorithm is used to analyze the corresponding vehicle locations and determine spatial point clusters. The center point of the cluster is taken as the excavator location, the number of clusters is taken as the number of excavators, the cluster area is taken as the active excavation area, the average dwell time of a single transport vehicle is taken as the loading time, and the average cumulative loading time of the cluster area is taken as the excavator's working intensity. The cumulative load of all vehicle trajectories leaving the excavation area fence within a fixed time period is calculated, and the soil removal rate is calculated to obtain the excavation characteristics.

4. The intelligent transportation scheduling method based on trajectory data according to claim 1, characterized in that, The method for determining the queuing status includes: The frequency of vehicle entry and exit and the average dwell time can be extracted directly from the trajectory of transport vehicles and the location of the fence in the excavation area within a fixed time period. The trajectories of all vehicles entering the excavation area within a fixed time period are time-aligned. Hierarchical clustering based on Euclidean distance is used to cluster all trajectory points according to their spatial location, forming multiple interaction space clusters. Each interaction space cluster represents a local area where vehicle interaction may occur. The local area includes loading points and narrow intersections. Analyze the status of all vehicles in each spatial cluster to determine the average queuing time, average passing time, and average vehicle spacing. The queuing feature vector is composed of vehicle entry and exit frequency, average dwell time, average queuing time, average passing time, and average vehicle spacing. Evaluation feature vectors are obtained by assigning evaluation weights to different queuing features. Calculate the weighted Euclidean distance between the evaluation feature vector and each historical evaluation feature vector, and select historical evaluation feature vectors whose Euclidean distance is less than the dynamic distance threshold to form a neighborhood point set; The congestion influence of each historical evaluation feature vector within the neighborhood point set on the evaluation feature vector is calculated, and the baseline congestion index for the corresponding time period of each historical evaluation feature vector is extracted to calculate the comprehensive queuing status evaluation value; the baseline congestion index is obtained through long-term statistics. The expression for the comprehensive queuing status evaluation value is as follows: ； ； in This is a comprehensive queuing status assessment value. To evaluate the feature vector, For historical evaluation feature vectors, For the neighborhood point set, Historical evaluation feature vector For evaluating feature vectors The impact of congestion Historical evaluation feature vector The benchmark congestion index, Historical evaluation feature vector The local density of the corresponding region, i.e., the number of historical evaluation feature vectors. It is a smoothing constant; The queuing status is determined by comparing the comprehensive queuing status assessment value with the preset assessment threshold.

5. The intelligent transportation scheduling method based on trajectory data according to claim 1, characterized in that, The method for updating the road network map of the work area includes: The Douglas-Puk algorithm is used to simplify the historical transport vehicle trajectory, identify significant turning points as intersections, and designate the excavator location as the loading point. The intersection, loading point, and entrance / exit of the excavation area fence are used as graph nodes. The average travel time of corresponding historical transport vehicle trajectories between two adjacent graph nodes is taken as the initial travel time of the corresponding adjacent graph node edge. The average passing distance and average passing time of corresponding historical transport vehicle trajectories between two adjacent graph nodes with passing interactions are extracted and used as the initial passing distance and initial passing time of the corresponding adjacent graph node edge. The edge weights are calculated based on the initial travel time, initial passing distance, and initial passing time of the adjacent graph node edge. The initial road network graph model of the work area is constructed, and real-time trajectories are continuously collected. The trajectory similarity function is used to identify new paths and calculate the edge weights of each segment of the new path. The expression is: ； ； in For the adjacent graph nodes of the new path, edges The edge weights of the corresponding road segments, For road section The initial passage time, For road section The length of the road segment Standard road speed limit Weighting of passing time As the passing distance weight, For road section The initial passing time, Standard passing time, For road section The initial passing distance, Standard passing distance, For a new path With historical road network trajectory similarity, This is the distance attenuation coefficient. For new trajectory points With historical road network The minimum distance, For a new path The number of tracking points between China and Singapore; The product of the mean edge weight and the trajectory similarity is calculated as the update score. When the update score is less than the update score threshold, a new path is used to update the road network model.

6. The intelligent transportation scheduling method based on trajectory data according to claim 1, characterized in that, The method for constructing the energy consumption and emission estimation model includes: Extracting historical energy consumption characteristics of transport vehicles to calculate fuel consumption and combustion emissions, and extracting historical operating characteristics of each transport vehicle within the excavation area fence based on historical transport vehicle trajectories; the operating characteristics include the number of turns, the cumulative absolute value of turns, the travel distance, and the number of intersections; A comprehensive estimation dataset is composed of fuel energy consumption, combustion emissions, and corresponding historical operating characteristics, excavation characteristics, work area queuing characteristics, and weather characteristics. The comprehensive estimation dataset is divided into a training set and a test set according to the plan in a 6:4 ratio. The energy consumption and emission estimation model is trained using the training set, and the performance of the energy consumption and emission estimation model is tested using the test set. The energy consumption and emission estimation model includes an input layer, a feature fusion layer, an energy consumption estimation layer, an emission estimation layer, and an output layer. The feature fusion layer uses a self-attention mechanism to weightedly fuse operational features, excavation features, work area queuing features, and weather features to obtain fused estimation features; the energy consumption estimation layer uses a BP neural network to capture the nonlinear relationship between fuel energy consumption and fused estimation features to predict fuel energy consumption; the emission estimation layer embeds an emission factor mapping model to predict combustion emissions based on the predicted fuel energy consumption; the emission factor mapping model is fitted according to vehicle type and emission standards; the output layer is connected to the emission estimation layer through a fully connected layer to output the predicted fuel energy consumption and predicted combustion emissions. The energy consumption and emission estimation model uses a composite weighted mean square error loss function to improve the model's prediction accuracy and an adaptive moment estimator optimizes the model parameters; the composite weighted mean square error loss function includes fuel energy consumption prediction loss and combustion emission prediction loss.

7. The intelligent transportation scheduling method based on trajectory data according to claim 1, characterized in that, The method for determining the objective function for transportation scheduling optimization includes: Based on queuing state characteristics, energy consumption costs, and pollution emission penalties, a multi-objective dynamic scheduling optimization objective function is constructed, expressed as follows: ； in To optimize the objective function for transportation scheduling, , To avoid delays and reduce costs, , For energy consumption cost weighting and energy consumption cost, , To determine the pollution weight and pollution emission costs, This is the time penalty coefficient. This is the energy consumption penalty coefficient. To account for delays, including average queuing time and average passing time, Standard delay time, For target fuel energy consumption, pollutants A set of.

8. A transportation intelligent scheduling device based on trajectory data, used to execute the method according to any one of claims 1-7, characterized in that, include: Operation phase identification module: used to design dynamic multi-layer electronic fences according to construction characteristics, extract the trajectory of transport vehicles and identify the operation phase of transport vehicles based on the dynamic multi-layer electronic fences; Work area feature recognition module: used to identify excavators from similar areas to determine excavation features, identify the spatiotemporal distribution of similar areas to determine queuing features in the work area, and use spatial density clustering analysis to determine queuing status in the work area queuing features. Work area road network module: used to build a work area road network map based on historical transport vehicle trajectories, and collect real-time transport vehicle trajectories to update the work area road network map; Energy consumption and emission estimation module: used to extract the energy consumption characteristics of historical transport vehicles to calculate fuel energy consumption and combustion emissions, and to build an energy consumption and emission estimation model by combining the corresponding historical transport vehicle trajectory, excavation characteristics, work area queuing characteristics and weather characteristics. Transportation scheduling module: Used to determine the objective function for transportation scheduling optimization. It uses a formulaic search to adjust the real-time routes of transport vehicles and excavation characteristics based on the road network map of the work area, recalculates the objective function for transportation scheduling optimization, and iterates until the scheduling objective is met, then outputs the transportation scheduling plan.

Images