Big data-based intelligent scheduling method and system for logistics transport vehicles

Abstract

This invention discloses a method and system for intelligent scheduling of logistics vehicles based on big data, belonging to the field of logistics big data scheduling technology. It includes establishing a logistics transportation scheduling big data warehouse, collecting and storing raw data such as historical orders, vehicle files, real-time traffic, weather, and warehouse operation status; standardizing, cleaning, and integrating the data to form a multi-dimensional scheduling decision feature set; identifying high-frequency delivery areas and frequently congested road sections through spatiotemporal clustering algorithms; generating regional heat maps and road network risk layers; dynamically correcting the road network risk layers by combining real-time traffic and weather information to obtain a real-time road network status layer; and performing multi-objective programming operations by integrating orders to be scheduled, available vehicle pools, regional heat maps, and real-time road network status layers to generate a preliminary scheduling scheme. This invention can accurately present the spatiotemporal distribution characteristics of orders and road networks, dynamically adapt to the real-time transportation environment, and improve the multi-dimensional data support and computational adaptability of scheduling decisions.

Description

Technical Field

[0001] This invention belongs to the field of logistics big data scheduling technology, specifically a method and system for intelligent scheduling of logistics transportation vehicles based on big data. Background Technology

[0002] Traditional logistics vehicle dispatching relies heavily on manual experience to formulate plans, collecting only limited data such as historical orders and basic vehicle files. It lacks a comprehensive big data warehouse for logistics dispatching that integrates historical orders, vehicle files, real-time traffic conditions, weather, and warehouse operational status. Raw data is merely processed without standardized cleaning and correlation integration, failing to generate a multi-dimensional set of dispatching decision features reflecting the spatiotemporal coupling relationships between orders, vehicles, road networks, weather, and warehouses. Conventional dispatching techniques often employ simple path planning logic, failing to utilize spatiotemporal clustering algorithms to mine historical data features, generate regional heatmaps indicating historical order density, or create road network risk layers indicating congestion frequency and average delay time.

[0003] Existing scheduling schemes directly call isolated real-time traffic and weather data for scheduling calculations, without dynamically correcting them based on historical road network risk layers, thus failing to generate a realistic real-time road network status layer. Multi-objective programming operations only use pending orders and available vehicle status as inputs, without combining regional heat maps and real-time road network status layers. This invention aims to implement spatiotemporal clustering processing based on spatiotemporally coupled multidimensional features to generate corresponding regional heat maps and road network risk layers. It then corrects the road network risk layers using real-time traffic and weather information, using all four types of data as inputs to generate a scheduling scheme for multi-objective programming. Summary of the Invention

[0004] This invention aims to solve at least one of the technical problems existing in the prior art;

[0005] To this end, the present invention proposes an intelligent scheduling method for logistics transportation vehicles based on big data, including:

[0006] Establish a big data warehouse for logistics transportation scheduling, collect and store raw scheduling data, which includes historical order information, vehicle file information, real-time traffic information, weather information, and warehouse operation status information;

[0007] The raw scheduling data in the logistics transportation scheduling big data warehouse is standardized, cleaned, and integrated to form a multi-dimensional scheduling decision feature set. The multi-dimensional scheduling decision feature set is used to reflect the coupling relationship between orders, vehicles, road networks, weather, and warehouses in the time and space dimensions.

[0008] Based on the multidimensional scheduling decision feature set, a spatiotemporal clustering algorithm is used to identify high-frequency delivery areas and frequently congested road sections, generating regional heat maps and road network risk layers. The regional heat maps are used to mark the order density of different geographical areas in a historical period, and the road network risk layers are used to mark the congestion frequency and average delay time of each road section in a historical period.

[0009] Based on the real-time traffic information and weather information, the road network risk layer is dynamically corrected to generate a real-time road network status layer.

[0010] By combining the details of the orders to be dispatched, the status of the available vehicle pool, the regional heat map, and the real-time road network status layer, a multi-objective programming operation is performed to generate a preliminary vehicle dispatching plan.

[0011] Furthermore, the raw scheduling data in the logistics transportation scheduling big data warehouse is subjected to standardized cleaning and correlation integration to form a multi-dimensional scheduling decision feature set, including:

[0012] The historical order information is parsed to extract the cargo volume, weight, geographical coordinates of the delivery start and end points, promised delivery time window, and actual completion timestamp for each order;

[0013] The vehicle file information is parsed to extract the vehicle's identification, load capacity, cargo volume, current location coordinates, current cargo status, and maintenance records for each vehicle.

[0014] The real-time traffic information is analyzed to extract the traffic speed, congestion level, and event announcements of each road segment in the entire road network at continuous time points;

[0015] The weather status information is analyzed to extract temperature, precipitation, wind speed, visibility, and disaster warning signals at each meteorological monitoring point at continuous time points;

[0016] By associating historical order information, vehicle file information, real-time traffic information, weather information, and warehouse operation status information under the same time and space dimensions, a multi-dimensional scheduling decision feature set indexed by vehicle, order, road segment, and time point is constructed.

[0017] Furthermore, based on the aforementioned multi-dimensional scheduling decision feature set, a spatiotemporal clustering algorithm is used to identify high-frequency delivery areas and frequently congested road sections, generating regional heat maps and road network risk layers, including:

[0018] Extract the geographical coordinates of the delivery destinations of all historical orders from the multidimensional scheduling decision feature set, and form a set of coordinate points on the electronic map;

[0019] The coordinate point set is analyzed using a density-based spatial clustering algorithm to identify the spatial range where the order point density exceeds a set threshold. The spatial range is defined as the high-frequency delivery area, and the order concentration index is calculated for each high-frequency delivery area.

[0020] Extract the traffic speed sequence of all road segments in the historical period from the multidimensional scheduling decision feature set, and calculate the average daily congestion duration and speed fluctuation variance of each road segment;

[0021] Road segments whose average daily congestion duration exceeds a duration threshold or whose speed fluctuation variance exceeds a variance threshold are defined as the frequently congested road segments.

[0022] On the electronic map, the order concentration index of the high-frequency delivery area is rendered with different color depths to form the area heat map, and the frequently congested road sections and their historical average delay time are identified with different risk levels to form the road network risk layer.

[0023] Furthermore, based on the real-time traffic information and weather information obtained in real time, the road network risk layer is dynamically corrected to generate a real-time road network status layer, including:

[0024] Obtain the real-time traffic information at the current moment, and identify the road segments marked as congested or slow-moving and their estimated duration;

[0025] Obtain the current weather status information and identify the areas where warnings have been issued regarding road icing, heavy rain, and dense fog that may affect driving;

[0026] The identified currently congested road segments are compared with historically frequently congested road segments. For road segments that are both currently congested and historically frequently congested, their risk levels are weighted and increased based on their historical risk levels.

[0027] The warning areas affecting driving are mapped onto the road network, thereby increasing the basic traffic resistance coefficient of all road sections within the warning areas;

[0028] Information on road segments with weighted risk levels and road segments with increased basic traffic resistance coefficients is integrated and overlaid on the original road network risk layer to generate a real-time road network status layer that reflects instantaneous road conditions and weather effects. The real-time road network status layer integrates historical risk patterns with the instantaneous impact of current traffic and weather.

[0029] Furthermore, the process of combining the details of the orders to be dispatched, the status of the available vehicle pool, the regional heatmap, and the real-time road network status layer to perform multi-objective programming operations and generate a preliminary vehicle dispatching plan includes:

[0030] Parse the details of the orders to be dispatched to obtain the cargo attributes, delivery address coordinates and time requirements of all orders to be delivered;

[0031] Scan the status of the available vehicle pool to obtain the identifiers, current locations, load capacities, and available cargo compartment volumes of all vehicles that are idle or about to be idle.

[0032] Based on the coordinates of the delivery address, the heat map of the region is queried to obtain the order concentration index of the order destination region, and the orders are pre-grouped based on the order concentration index;

[0033] Based on the real-time road network status layer, calculate the multi-segment path travel time from the current location of each available vehicle to the pickup and delivery points of all orders in its assigned order group;

[0034] With the objectives of minimizing total transportation cost, maximizing vehicle occupancy rate, and minimizing average order delay time as the objective functions, and with vehicle capacity, order time window, and route connectivity as constraints, a multi-objective optimization model is constructed.

[0035] Solve the multi-objective optimization model to output the preliminary vehicle scheduling scheme, which includes the matching relationship between vehicles and orders, the order of pickup, the delivery route, and the estimated timetable for each node.

[0036] The preliminary vehicle dispatching plan is used to indicate the order allocation, pickup sequence, delivery route, and expected time nodes for each dispatched vehicle.

[0037] Furthermore, it also includes a step of dynamically rescheduling the preliminary vehicle scheduling plan:

[0038] After the initial vehicle dispatching plan is implemented, the real-time location, speed and remaining capacity of each dispatched vehicle are continuously monitored.

[0039] The system receives real-time updates of the real-time traffic and weather information and dynamically refreshes the real-time road network status layer.

[0040] The real-time location of the vehicle is compared with the planned path in the preliminary vehicle scheduling scheme to calculate the deviation between the actual progress and the planned progress. When the deviation exceeds the tolerance threshold, a rescheduling evaluation is triggered.

[0041] When a rescheduling assessment is triggered, all currently en route orders, orders that have not yet started execution, real-time vehicle status, the refreshed real-time road network status layer, and the regional heat map are used as inputs to re-execute the multi-objective planning operation;

[0042] The new scheduling scheme obtained by recalculation is compared with the original scheme. If the new scheme is better than the original scheme in terms of the objective function by more than the improvement threshold, the scheduling instructions of the unexecuted part are replaced by the new scheme.

[0043] Furthermore, when triggering a rescheduling evaluation, the multi-objective programming operation is re-executed using all currently en route orders, orders not yet started, real-time vehicle status, the refreshed real-time road network status layer, and the regional heatmap as input, including:

[0044] Completed order tasks and ongoing tasks that cannot be changed are locked. The remaining variable tasks are defined as a set of tasks to be rescheduled. The set of tasks to be rescheduled includes the remaining delivery points of orders in transit and all orders that have not yet started execution.

[0045] The status definition of the available vehicle pool is updated based on the real-time location of the vehicles, the remaining capacity of the carriages, and the current task status.

[0046] Based on the refreshed real-time road network status layer, the estimated travel time from the real-time location of each available vehicle to all task points in the set of tasks to be rescheduled is recalculated.

[0047] With the goal of minimizing new costs and delays, a local rescheduling optimization model is constructed based on the set of tasks to be rescheduled, the updated status of the available vehicle pool, and the recalculated estimated travel time.

[0048] The local rescheduling optimization model is solved quickly, and the adjusted scheduling instruction sequence is output for the set of tasks to be rescheduled.

[0049] Furthermore, it also includes a self-optimization step for scheduling strategy parameters based on historical data feedback:

[0050] Periodically extract complete execution data of completed scheduling cycles from the logistics transportation scheduling big data warehouse. The complete execution data includes scheduling plan, actual execution trajectory, actual time consumption and various event records.

[0051] By comparing the estimated time nodes in the scheduling plan with the actual time nodes in the actual execution trajectory, the travel time prediction error of each segment is calculated.

[0052] Analyze the scheduling anomalies caused by vehicle malfunctions, traffic accidents, and order changes in the actual execution data and their handling results;

[0053] Using machine learning algorithms, with the multidimensional scheduling decision feature set as input and minimizing the trip time prediction error and optimizing the abnormal event handling results as training objectives, the internal parameters in the path planning model and time prediction model used in scheduling decisions are iteratively adjusted.

[0054] The iteratively adjusted model parameters are updated in the model used for the multi-objective programming operation to achieve adaptive optimization of the scheduling strategy.

[0055] Furthermore, the step of utilizing machine learning algorithms, taking the multi-dimensional scheduling decision feature set as input and minimizing the travel time prediction error and optimizing the handling of abnormal events as training objectives, iteratively adjusts the internal parameters of the path planning model and time estimation model used in scheduling decisions, including:

[0056] Construct a training sample set, each sample containing a snapshot of the multidimensional scheduling decision feature set at a historical scheduling moment, the estimated travel time series generated by the corresponding scheduling decision, and the actual travel time series and abnormal event records that subsequently occurred;

[0057] Define a loss function, which is a weighted sum of a trip time estimation error term and an abnormal event handling effect evaluation term.

[0058] The training sample set is input into a path planning model and a time prediction model containing internal parameters to be adjusted, and forward propagation calculation is performed to obtain the prediction output;

[0059] Calculate the loss function value between the predicted output and the true value, and calculate the gradient of the loss function with respect to the model's internal parameters using the backpropagation algorithm;

[0060] Based on the calculated gradient, the internal parameters of the path planning model and the time prediction model are updated using an optimization algorithm.

[0061] The process of inputting training samples and updating parameters is repeated until the loss function value converges or reaches the predetermined number of training rounds, thus completing the iterative adjustment of the model's internal parameters.

[0062] Furthermore, the present invention also includes a big data-based intelligent scheduling system for logistics and transportation vehicles, the system comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the big data-based intelligent scheduling method for logistics and transportation vehicles described above.

[0063] Compared with the prior art, the beneficial effects of the present invention are:

[0064] A spatiotemporal clustering algorithm is used to process the multidimensional scheduling decision feature set reflecting the coupling relationship between orders, vehicles, road networks, weather, and warehouses in the temporal and spatial dimensions. This accurately identifies high-frequency delivery areas and frequently congested road sections, generating regional heat maps that label the order density of different geographical areas within historical periods. At the same time, a road network risk layer is generated that labels the congestion frequency and average delay time of each road section within historical periods. The order distribution characteristics and road network operation characteristics in historical scheduling data are transformed into visualized and quantifiable layer data, fully presenting the historical distribution patterns of regional order aggregation and road congestion. This refines the spatiotemporal dimension feature parameters required for scheduling decisions and solidifies the inherent attributes of regional delivery and road network traffic implicit in historical data.

[0065] Based on real-time traffic and weather information, dynamic correction operations are performed on the road network risk layer to form a real-time road network status layer that fits the current transportation environment. The details of orders to be dispatched, the status of the available vehicle pool, regional heat maps, and the real-time road network status layer are simultaneously incorporated into the input system of multi-objective programming operations. By correcting historical road network risk parameters through real-time data, the parameter differences between historical layers and actual road network operating states are reduced, expanding the input dimensions and data coverage of multi-objective programming operations. This allows the reference data on which dispatch operations rely to simultaneously integrate historical patterns and real-time environmental characteristics, ensuring that the generation process of vehicle dispatching schemes takes into account both historical data patterns and real-time operating conditions. This matches the dynamically changing road network and order status in logistics and transportation scenarios, optimizing the data source structure and parameter adaptability of dispatching operations. Attached Figure Description

[0066] Figure 1 This is a flowchart illustrating the steps of the intelligent scheduling method for logistics transportation vehicles based on big data as described in this invention.

[0067] Figure 2 A flowchart for spatiotemporal clustering identification and layer generation;

[0068] Figure 3 This is a dynamic traffic delay coefficient distribution map for the road segment;

[0069] Figure 4 A bar chart comparing the status of logistics vehicles;

[0070] Figure 5 A bar chart comparing the statistics and handling effectiveness of abnormal logistics scheduling events. Detailed Implementation

[0071] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0072] See Figure 1 The overall implementation scheme of the intelligent scheduling method for logistics transportation vehicles based on big data provided by this invention is as follows: A logistics transportation scheduling big data warehouse is established, collecting and storing raw scheduling data including historical order information, vehicle file information, real-time traffic information, weather information, and warehouse operation status information. The raw scheduling data in the logistics transportation scheduling big data warehouse is standardized, cleaned, and integrated to form a multi-dimensional scheduling decision feature set that reflects the coupling relationship between orders, vehicles, road networks, weather, and warehouses in the temporal and spatial dimensions. Based on this multi-dimensional scheduling decision feature set, a spatiotemporal clustering algorithm is used to identify high-frequency delivery areas and frequently congested road sections, generating regional heat maps for labeling historical order density and road network risk layers for labeling historical congestion frequency and average delay time. Based on real-time traffic information and weather information, the road network risk layer is dynamically corrected to generate a real-time road network status layer that integrates historical patterns and instantaneous impacts. Combining the details of orders to be scheduled, the status of the available vehicle pool, the regional heat map, and the real-time road network status layer, multi-objective programming operations are performed to generate a preliminary vehicle scheduling plan.

[0073] In one embodiment of the present invention, the process of performing standardized cleaning and correlation integration on the raw scheduling data in the logistics transportation scheduling big data warehouse to form a multi-dimensional scheduling decision feature set includes the following specific operations: Parsing historical order information to extract the cargo volume, weight, geographical coordinates of delivery origin and destination, promised delivery time window, and actual completion timestamp for each order. Parsing vehicle file information to extract the identification, load capacity, cargo volume, current location coordinates, current cargo status, and maintenance records for each vehicle. Parsing real-time traffic information to extract the traffic speed, congestion level, and event announcements for each road segment across the entire road network at continuous time points. Parsing weather information to extract the temperature, precipitation, wind speed, visibility, and disaster warning signals for each meteorological monitoring point at continuous time points. Correlating historical order information, vehicle file information, real-time traffic information, weather information, and warehouse operation status information in the same time and space dimensions to construct a multi-dimensional scheduling decision feature set indexed by vehicle, order, road segment, and time point.

[0074] In practical implementation, the raw scheduling data in the logistics transportation scheduling big data warehouse undergoes standardized cleaning and correlation integration to form a multi-dimensional scheduling decision feature set. This operation involves multiple ordered processing stages. Specifically, the parsing of historical order information requires extracting from the raw scheduling data the cargo volume, weight, origin and destination geographic coordinates, promised delivery time window, and timestamp of the actual order completion for each order. In some embodiments, the parsing of vehicle profile information requires extracting from the raw scheduling data the vehicle identification code, marked load capacity, actual cargo volume, reported current location coordinates, current cargo status, and historical maintenance records for each transport vehicle. Optionally, the parsing of real-time traffic information requires extracting from the raw data stream the average traffic speed, real-time congestion level, and event announcement text issued by the traffic management department for each road segment within the entire urban road network at consecutive time points.

[0075] In practical implementation, the analysis of weather status information requires extracting temperature, precipitation, wind speed, visibility, and disaster warning signal levels recorded by various meteorological monitoring points at continuous time points from meteorological data sources. It can be understood that the core step in constructing a multi-dimensional scheduling decision feature set is to correlate and integrate historical order information, vehicle file information, real-time traffic information, weather status information, and warehouse operation status information under the same time and spatial dimensions. The correlation operation aligns and matches scattered data entries based on a unified timestamp and geospatial grid. In some embodiments, the correlation result is constructed as a normalized data set with vehicle identifier, order number, road segment number, and time point as a composite index; this data set is the multi-dimensional scheduling decision feature set. Each record in the multi-dimensional scheduling decision feature set integrates multi-source feature values ​​related to scheduling decisions at a specific time and location.

[0076] In practice, the geographical coordinates of delivery origin and destination points parsed from historical order information are converted into a standard latitude and longitude format, while the current location coordinates parsed from vehicle file information also need to undergo geocoding standardization. Traffic speeds extracted from real-time traffic information need to be normalized according to road grade to eliminate the impact of differences in baseline speeds across different road grades. Optionally, continuous numerical features such as precipitation and wind speed extracted from weather information are discretized into categorical features such as "light rain" and "moderate wind" for subsequent model processing. Warehouse operation status information, including the busyness of warehouse gates and the occupancy rate of loading and unloading platforms, also needs to be synchronized with external orders and vehicle information over time. The multi-dimensional scheduling decision feature set formed after the above standardization, cleaning, and integration operations provides a unified and multi-dimensional input data foundation for subsequent spatiotemporal pattern mining and intelligent decision-making calculations. During feature construction, a derived feature for quantifying the degree of regional order aggregation, the "order concentration index," can be calculated using the following formula:

[0077]

[0078] Where: symbol This represents the calculated order concentration index, with the symbol... This represents the total number of historical orders within a high-frequency delivery area obtained from a clustering, with the symbol... Indicates the geographical area of ​​the high-frequency delivery region, symbol This represents the weight factor of the goods for the i-th order within the region, denoted by [symbol]. This represents the total duration of the historical period for which statistics are collected. This index comprehensively reflects the number of orders and the weight load per unit area and per unit time, and is an important derived feature in the multidimensional scheduling decision feature set.

[0079] In one embodiment of the present invention, see [reference] Figure 2Based on a multidimensional scheduling decision feature set, the steps for identifying high-frequency delivery areas and frequently congested road segments using a spatiotemporal clustering algorithm, and generating regional heat maps and road network risk layers, are as follows: First, extract the geographical coordinates of the delivery destinations of all historical orders from the multidimensional scheduling decision feature set, forming a coordinate point set on an electronic map. Then, analyze the coordinate point set using a density-based spatial clustering algorithm to identify spatial ranges where order point density exceeds a set threshold. Define these spatial ranges as high-frequency delivery areas and calculate the order concentration index for each high-frequency delivery area. Next, extract the traffic speed sequences of all road segments within historical periods from the multidimensional scheduling decision feature set, calculating the average daily congestion duration and speed fluctuation variance for each road segment. Road segments with average daily congestion duration exceeding a duration threshold or speed fluctuation variance exceeding a variance threshold are selected and defined as frequently congested road segments. Finally, on the electronic map, render the order concentration index of high-frequency delivery areas with different color depths to form a regional heat map, and identify frequently congested road segments and their historical average delay times with different risk levels to form a road network risk layer.

[0080] In practical implementation, the process of identifying high-frequency delivery areas and frequently congested road sections based on a multi-dimensional scheduling decision feature set and generating corresponding visualization layers involves in-depth mining and feature calculation of historical spatiotemporal data. Specifically, from the constructed multi-dimensional scheduling decision feature set, the geographical coordinates of the delivery destination are extracted from all historical order records. These coordinates are cleaned and standardized to form a set of coordinate points usable for spatial analysis. This set of coordinate points is projected onto the coordinate system of an electronic map, with each point representing the destination location of a historical delivery. In some embodiments, a density-based spatial clustering algorithm, such as a noisy density-based spatial clustering method, is used to analyze the set of coordinate points on the electronic map. By setting a neighborhood radius and a minimum point threshold, the algorithm identifies dense spatial ranges where the order point density exceeds the set threshold. Each identified dense spatial range is defined as a high-frequency delivery area in the system.

[0081] In practical implementation, for each identified high-frequency delivery area, a quantified order concentration index needs to be calculated to characterize the concentration of its delivery demand. This calculation process involves counting the total number of historical orders falling within the high-frequency delivery area and measuring the geographical area of ​​that area. Optionally, the calculation of the order concentration index may consider not only the number of orders but also weight or volume weighting factors. On the other hand, identifying frequently congested road segments requires consideration from another dimension: a multi-dimensional set of scheduling decision features. In practical implementation, the time series of traffic speeds for each road segment within a set historical period is extracted from the set. This time series records the average vehicle speed of the road segment at different time points. Based on the traffic speed time series, two core indicators are calculated for each road segment: average daily congestion duration and speed fluctuation variance.

[0082] In some embodiments, calculating the average daily congestion duration requires first defining a congestion speed threshold, then calculating the cumulative duration of time the speed on that road segment was below the threshold each day within a historical period, and finally obtaining the daily average over the historical period. The speed fluctuation variance is calculated directly based on the numerical values ​​of the traffic speed time series, reflecting the stability of vehicle speed. In specific implementation, all road segments in the entire road network are screened according to preset discrimination rules. Road segments with an average daily congestion duration exceeding a preset duration threshold, or a speed fluctuation variance exceeding a preset variance threshold, are marked in the system and defined as frequently congested road segments. It is understood that the preset duration threshold and variance threshold can be adjusted according to the traffic characteristics of different cities. After screening, each frequently congested road segment is also associated with and its average delay duration calculated over the entire historical period. The average delay duration is obtained from the average difference between the actual travel time and the free-flow travel time.

[0083] In practical implementation, the process of generating regional heatmaps involves visualizing the high-frequency delivery areas and their order concentration indices calculated above. On the electronic map base map, the system maps the order concentration index value of each high-frequency delivery area onto a predefined color gradient. Areas with higher index values ​​are rendered with deeper color depths, thus forming a regional heatmap that intuitively displays the spatial distribution of order density. The process of generating road network risk layers involves marking frequently congested road segments and their attributes on the map. Based on the historical average delay time or other comprehensive risk scores, the system classifies each frequently congested road segment into different risk levels, for example, using red to represent high risk and yellow to represent medium risk. On the electronic map, these road segments are drawn using line segments of the corresponding risk level color and thickness, and the specific values ​​of the average delay time are marked on the line segments, ultimately forming a risk layer covering the entire road network. A comprehensive score is used to quantify the historical congestion risk of road segments. It can be calculated using the following formula:

[0084]

[0085] Where: symbol This indicates the comprehensive score of historical congestion risk for a road segment, represented by the symbol. This indicates the average daily congestion duration of this road segment, indicated by the symbol. The total duration of a day is used for normalization, and the symbol is... This represents the variance of speed fluctuations on that road segment, with the sign... The theoretical maximum value of the variance of velocity fluctuations is used for normalization, with the sign... Indicates the average delay time for this road segment, symbol The baseline value representing the average delay time is used for normalization. Coefficients , and These are preset weighting parameters used to weight different influencing factors, and they satisfy... This rating is the core basis for defining the risk level in the road network risk layer.

[0086] In one embodiment of the present invention, the operation of dynamically correcting the road network risk layer and generating a real-time road network status layer based on real-time traffic and weather information includes the following steps: Acquiring real-time traffic information at the current moment and identifying road segments marked as congested or slow-moving and their expected duration. Acquiring current weather information and identifying warning areas affecting driving, such as road icing, heavy rain, and dense fog. Comparing the identified currently congested road segments with historically frequently congested road segments, and weighting up the risk level of road segments that are simultaneously classified as both currently congested and historically frequently congested. Mapping the warning areas affecting driving onto the road network and increasing the basic drag coefficient of all road segments within the warning areas. Integrating the information of road segments with weighted risk level increases and road segments with increased basic drag coefficients, and overlaying this information onto the original road network risk layer to generate a real-time road network status layer reflecting the impact of instantaneous traffic conditions and weather. This real-time road network status layer integrates historical risk patterns with the instantaneous impact of current traffic and weather.

[0087] The process of generating a preliminary vehicle scheduling plan by combining the details of orders to be scheduled, the status of the available vehicle pool, regional heatmaps, and real-time road network status layers, and performing multi-objective programming operations, is as follows: First, the details of orders to be scheduled are parsed to obtain the cargo attributes, delivery address coordinates, and time requirements for all orders to be delivered. Second, the status of the available vehicle pool is scanned to obtain the identifiers, current locations, load capacities, and available cargo compartment volumes of all vehicles that are idle or about to become idle. Third, the regional heatmap is queried based on the delivery address coordinates to obtain the order concentration index of the order destination area, and orders are pre-grouped based on the order concentration index. Fourth, based on the real-time road network status layer, the multi-segment travel time from the current location of each available vehicle to the pickup and delivery points of all orders within its assigned order group is calculated. Fifth, a multi-objective optimization model is constructed with the objective functions of minimizing total transportation cost, maximizing vehicle occupancy rate, and minimizing average order delay time, and with constraints of vehicle capacity, order time windows, and path connectivity. Sixth, the multi-objective optimization model is solved to output a preliminary vehicle scheduling plan that includes vehicle-order matching relationships, pickup order, delivery routes, and estimated timetables for each node. The preliminary vehicle dispatch plan is used to indicate the order allocation, pickup sequence, delivery route, and expected time nodes for each dispatched vehicle.

[0088] In practical implementation, dynamically correcting the road network risk layer and generating a real-time road network status layer based on real-time traffic and weather information is a crucial step in connecting historical patterns with the current situation. In practice, the system continuously acquires real-time traffic information streams via an application programming interface (API). These streams contain speed, traffic flow, and event data for all road segments across the entire network. The parsing module identifies road segments marked as "congested" or "slow-moving" from the information stream and extracts the estimated duration of these abnormal conditions. Simultaneously, the system acquires current weather information via the API. This weather information comes from meteorological data services. The parsing module identifies the polygonal coordinate ranges of potentially hazardous weather warning areas, such as road icing warnings, heavy rain warnings, and fog warnings, from the weather information.

[0089] In practice, comparing the identified set of currently congested road segments with the set of historically frequently congested road segments is an automated process. The system performs the matching operation based on a unified road segment code. For road segments that match successfully—that is, road segments that exist in both the current and historical sets of frequently congested road segments—the system applies a weighted increase to their original historical risk level. The magnitude of this weighted increase is determined by a preset dynamic weighting coefficient, which takes into account the severity of the current congestion and its expected duration. In practice, the warning areas affecting driving are mapped to the road network for spatial calculations. The system overlays the polygon coordinates of the weather warning area with the digital road network layer, filtering out all road segments whose geometric location falls within the polygon of the warning area; these road segments are defined as affected road segments. For all affected road segments, the system increases their basic drag coefficient based on the warning type and level. For example, under a red warning for road icing, the drag coefficient is increased to a relatively high fixed value.

[0090] In practical implementation, generating a real-time road network status layer requires integrating the aforementioned corrected road segment information. The system creates a real-time road network status layer as a new digital layer, based on the original road network risk layer. All road segment information with weighted risk level enhancement, as well as all road segment information with increased basic traffic resistance coefficients, will be updated and overlaid on this base layer. It can be understood that each road segment in the real-time road network status layer possesses a comprehensive risk attribute value, which integrates its historical risk patterns with the instantaneous impact of current traffic and weather. In some embodiments, the system creates a version snapshot of the real-time road network status layer and records a timestamp for subsequent querying and analysis. Refer to Table 1, which shows a simplified example of the dynamic correction logic for road segment risk levels.

[0091] Table 1: Logical Adjustment Table for Road Section Risk Level

[0092]

[0093] In practice, multi-objective programming calculations are performed by combining details of orders to be dispatched, the status of the available vehicle pool, regional heat maps, and real-time road network status layers. The input and calculation process are clearly defined. The parsing module parses the details of orders to be dispatched, obtaining the volume and weight of goods, the latitude and longitude coordinates of the delivery address, and the customer's required final delivery time window for all orders to be delivered. The scanning module scans the status of the available vehicle pool, obtaining the vehicle identifiers of all vehicles with a status of "idle" or "soon to be idle," the real-time location coordinates reported by the vehicle via the GPS, the vehicle's maximum load capacity, and the vehicle's remaining available cargo space. Based on the delivery address coordinates, the system queries the regional heat map to obtain the order concentration index for each order's destination region. The dispatching logic pre-groups orders that are geographically adjacent and have similar indices based on the order concentration index for subsequent batch delivery.

[0094] In practical implementation, the system calculates the travel time of multiple path segments based on the real-time road network status layer. The system's path planning engine uses the latest traffic resistance coefficient and risk level of each road segment in the real-time road network status layer as weights, and employs a graph search algorithm that considers dynamic weights to calculate the optimal path and estimated travel time from the current location of each available vehicle to all order pickup and delivery points within its possible order group. When constructing the multi-objective optimization model, the objective function is typically set to minimize the total mileage cost, maximize the average load rate of all dispatched vehicles, and minimize the average delay time of all orders. The average order delay time is calculated by comparing the estimated delivery time of the order determined based on the travel time of multiple path segments with the promised delivery time window of the order. Constraints include that the total volume of orders allocated to each vehicle does not exceed the vehicle's cargo volume, the total weight does not exceed the load capacity, each order must be delivered within its time window, and the travel path must be connected within the road network topology. It is understood that solving the multi-objective optimization model requires mathematical programming methods or metaheuristic algorithms. In some embodiments, a non-dominated sorting genetic algorithm with an elitist strategy is used to solve the model. This algorithm can generate a solution set containing multiple non-dominated solutions. The dispatcher can select a solution from the solution set, or the system can automatically output an optimal solution as a preliminary vehicle dispatching plan based on preset rules. The preliminary vehicle dispatching plan is output in a structured data format, indicating the vehicle identifier of each dispatched vehicle, the list of order numbers assigned to that vehicle, the order in which these order pickup and delivery points are visited, the sequence of road segments traversed by the travel route, and the planned timestamp for arrival at each node. This is used to calculate the dynamic travel time of road segments. The formula can be expressed as:

[0095]

[0096] Where: symbol Represents the dynamic travel time calculated based on the real-time road network status layer, symbol This indicates the baseline travel time for this road segment under free-flow conditions, denoted by [symbol]. This represents the delay coefficient determined based on historically frequent congestion risk levels, with the symbol... This represents the delay coefficient determined based on the current congestion status in real-time traffic information, with the symbol... This represents the delay coefficient determined based on the warning level in the current weather status information. These three delay coefficients are directly read from the attribute fields of the corresponding road segments in the real-time road network status layer and together constitute the dynamic traffic delay coefficient. Based on the dynamic traffic time of the aforementioned road segments... The system's path planning engine employs a graph search algorithm that considers dynamic weights for each road segment. As the weight of this road segment in the graph, the optimal path is found from the current location of each available vehicle to all task points within its order group. The travel time of the multiple road segments from the origin to the destination is the total travel time of all road segments constituting this optimal path. sum.

[0097] See Figure 3 This is a dynamic traffic delay coefficient distribution map for road segments, used to quantify the impact of historical congestion, real-time traffic conditions, and weather warnings on road segment efficiency. Higher values ​​indicate slower traffic. R005 has all three coefficients at 1.0, representing the most efficient road segment in the current network. R002, influenced by both real-time traffic conditions and weather warnings, has a significantly higher delay coefficient than other segments and requires priority scheduling and avoidance. R003 / R004 / R008 have at least one coefficient of 1.0, primarily influenced by a single factor, and their traffic pressure is relatively controllable. When calculating dynamic travel time, the three delay coefficients are multiplied by the baseline travel time to obtain a more accurate prediction of road segment travel time. High-risk road segments are identified, and detours are proactively implemented when generating vehicle scheduling plans to reduce delay risks. The combined effect of the three coefficients is used to classify road segments into real-time risk levels, assisting in the generation of a dynamic road network status layer.

[0098] In one embodiment of the present invention, the method further includes a step of dynamically rescheduling and evaluating the preliminary vehicle scheduling plan. After the preliminary vehicle scheduling plan begins execution, the real-time location, speed, and remaining capacity of each scheduled vehicle are continuously monitored. Real-time updated traffic and weather information is received, and the real-time road network status layer is dynamically refreshed. The real-time location of the vehicles is compared with the planned path in the preliminary vehicle scheduling plan, and the deviation between the actual progress and the planned progress is calculated. When the deviation exceeds a tolerance threshold, a rescheduling evaluation is triggered. When a rescheduling evaluation is triggered, all currently en route orders, orders not yet executed, real-time vehicle status, the refreshed real-time road network status layer, and the regional heatmap are used as inputs to re-execute the multi-objective programming operation. The re-calculated new scheduling plan is compared with the original plan. If the new plan is superior to the original plan in terms of the objective function to an improvement threshold exceeding a certain level, the new plan replaces the subsequent unexecuted scheduling instructions.

[0099] When a rescheduling assessment is triggered, all currently en route orders, orders not yet executed, real-time vehicle status, the updated real-time road network status layer, and the regional heatmap are used as inputs to re-execute a multi-objective programming operation. The internal process is as follows: Completed order tasks and currently executing tasks that cannot be changed are locked. The remaining variable tasks are defined as the set of tasks to be rescheduled. This set includes the remaining delivery points of en route orders and all orders not yet executed. The status definition of the available vehicle pool is updated based on the real-time vehicle location, remaining vehicle capacity, and current task status. Based on the updated real-time road network status layer, the estimated travel time from the real-time location of each available vehicle to all task points in the set of tasks to be rescheduled is recalculated. With the goal of minimizing additional costs and delays, a local rescheduling optimization model is constructed based on the set of tasks to be rescheduled, the updated status of the available vehicle pool, and the recalculated estimated travel times. The local rescheduling optimization model is solved quickly, outputting an adjusted scheduling instruction sequence for the set of tasks to be rescheduled.

[0100] In practical implementation, the step of dynamically rescheduling and evaluating the initial vehicle dispatching plan is the core mechanism for addressing changes in the external environment and internal execution deviations during plan execution. In practice, after the initial vehicle dispatching plan begins execution, the system continuously monitors the real-time location (latitude and longitude coordinates), speed, and remaining cargo capacity of each dispatched vehicle via vehicle IoT terminals. The system receives real-time traffic and weather information updated from external data sources and dynamically refreshes the real-time road network status layer based on this latest information to ensure the timeliness of traffic condition assessments. In practice, comparing the real-time location of a vehicle with the planned path in the initial vehicle dispatching plan is a continuous calculation process. The system periodically calculates the deviation between the actual progress and the planned progress of the vehicles. The deviation calculation comprehensively considers time and spatial distance differences. In some embodiments, the deviation is defined as the absolute difference between the actual time and the planned time for a vehicle to reach the current planned node, divided by the total planned duration of that task segment to obtain a standardized ratio.

[0101] In practice, when the calculated deviation exceeds the system's preset tolerance threshold, the system automatically triggers a rescheduling evaluation process. When triggering the rescheduling evaluation, the system uses all currently en route orders, all orders yet to be executed, the real-time status of all relevant vehicles, the updated real-time road network status layer, and the regional heatmap as new input datasets to re-execute the multi-objective programming operation. This re-calculation considers all changes that have occurred since the initial vehicle scheduling plan was generated. The system compares the re-calculated new scheduling plan with the original plan, comparing objective function values ​​such as total mileage, estimated total delay time, and total vehicle utilization. If the new scheduling plan's overall evaluation value on the objective function is better than the original plan by more than a preset improvement threshold, the system automatically replaces the remaining unexecuted scheduling instructions with the new plan and sends adjustment instructions to the corresponding vehicle terminals and dispatcher interfaces. Table 2 shows a simplified example of vehicle plan deviation calculation and rescheduling trigger judgment.

[0102] Table 2: Vehicle Dispatch Progress Deviation Calculation and Rescheduling Trigger Table

[0103]

[0104] In practice, the internal process of re-executing multi-objective programming calculations when triggering a rescheduling evaluation is a local optimization process focused on the remaining variable tasks. Completed order tasks and ongoing tasks that cannot be changed are locked. The system makes judgments based on order status and vehicle location; for example, order tasks where vehicles have arrived and begun loading / unloading are considered immutable. The remaining variable tasks are defined as a set of tasks to be rescheduled, which includes all orders en route with remaining undelivered delivery points and all orders that have not yet been started by any vehicle. In practice, based on the real-time location coordinates of vehicles, remaining cargo capacity, and the status of currently executing tasks, the system updates the status definition of the available vehicle pool, updating vehicles participating in rescheduling from their original idle or soon-to-be-idle status to a status including specific location, remaining cargo capacity, and estimated next idle time. Based on the refreshed real-time road network status layer, the route planning engine recalculates the estimated travel time from the real-time location of each available vehicle to all task points in the set of tasks to be rescheduled, using the latest dynamic road segment travel times.

[0105] In practical implementation, aiming to minimize additional costs and delays, a local rescheduling optimization model is constructed based on the set of tasks to be rescheduled, the updated state of the available vehicle pool, and the recalculated estimated travel time. It can be understood that the objective function of the local rescheduling optimization model focuses on the additional costs incurred due to the adjustment plan, such as the additional detour distance for vehicles and the potential additional delays to existing orders caused by rescheduling. In some embodiments, the constraints of the local rescheduling optimization model include ensuring that tasks already locked in the original plan are not affected, vehicle capacity constraints, and time window constraints for newly inserted tasks. To quickly solve the local rescheduling optimization model, a computationally efficient heuristic algorithm is needed, such as a large-scale neighborhood search algorithm, to output an adjusted scheduling instruction sequence for the set of tasks to be rescheduled within a short time. The adjusted scheduling instruction sequence clarifies the new task sequence, new path, and new estimated time nodes that each vehicle should execute starting from the current moment. This is used to calculate the scheduling progress deviation value. Here is an example:

[0106]

[0107] Where: symbol This represents the overall scheduling progress deviation value for a single vehicle, with the symbol... This represents the total number of planned nodes that have been completed and are currently being executed for the vehicle within the current cycle, indicated by the symbol. Represents the actual timestamp of the vehicle arriving at the i-th node, symbol This represents the timestamp of the planned arrival at the i-th node in the preliminary vehicle dispatching plan, with the symbol... This represents the planned total travel time from the starting point to the i-th node. The system calculates the total travel time for each dispatched vehicle. The value is then compared with a preset global tolerance threshold.

[0108] See Figure 4 This is a bar chart comparing the status of five logistics vehicles, visually displaying their speed and remaining cargo space. It's used for dynamic rescheduling and load factor optimization analysis. Vehicles with high load factors (V1002 (85%) and V1004 (80%)) have low remaining cargo space but high resource utilization efficiency, making them suitable to continue their current delivery tasks. Vehicles with low load factors (V1003 (50%) and V1005 (55%) have sufficient remaining space and can be merged with new orders during rescheduling to improve load factor. Vehicles traveling at high speeds (V1002 (60km / h) and V1004 (55km / h) have high traffic efficiency and are likely traveling on smooth roads. Vehicle traveling at low speeds (V1003 (30km / h)) may be affected by congestion / road conditions and requires close monitoring of progress deviations. For V1003 (low speed + low load factor), if the progress deviation exceeds a threshold, rescheduling can be triggered, allocating new orders to this vehicle to improve load factor.

[0109] In one embodiment of the present invention, the method further includes a self-optimization step for scheduling strategy parameters based on historical data feedback. Complete execution data for completed scheduling cycles is periodically extracted from a logistics transportation scheduling big data warehouse. This complete execution data includes scheduling plans, actual execution trajectories, actual time consumption, and various event records. The estimated time nodes in the scheduling plan are compared with the actual time nodes in the actual execution trajectory to calculate the travel time prediction error for each route segment. The scheduling anomalies caused by vehicle malfunctions, traffic accidents, and order changes, and their handling results, are analyzed in the actual execution data. Using a machine learning algorithm, with a multi-dimensional scheduling decision feature set as input and minimizing travel time prediction errors and optimizing anomaly event handling results as training objectives, the internal parameters in the path planning model and time prediction model used in scheduling decisions are iteratively adjusted. The iteratively adjusted model parameters are then updated in the model used for multi-objective programming operations to achieve adaptive optimization of the scheduling strategy.

[0110] Using machine learning algorithms, and taking a multi-dimensional scheduling decision feature set as input, with the training objectives of minimizing travel time prediction error and optimizing the handling of abnormal events, the internal parameters of the path planning model and time prediction model used in scheduling decisions are iteratively adjusted. The specific training process is as follows: A training sample set is constructed, with each sample containing a snapshot of the multi-dimensional scheduling decision feature set at a historical scheduling moment, the predicted travel time series generated by the corresponding scheduling decision, and the subsequent actual travel time series and abnormal event records. A loss function is defined, which is a weighted sum of the travel time prediction error term and the abnormal event handling effect evaluation term. The training sample set is input into the path planning model and time prediction model containing the internal parameters to be adjusted, and forward propagation is performed to obtain the predicted output. The loss function value between the predicted output and the true value is calculated, and the gradient of the loss function with respect to the model's internal parameters is calculated using the backpropagation algorithm. Based on the calculated gradient, the internal parameters of the path planning model and time prediction model are updated using an optimization algorithm. The process from training sample input to parameter update is repeated until the loss function value converges or a predetermined number of training rounds are reached, completing the iterative adjustment of the model's internal parameters.

[0111] In practical implementation, a closed-loop learning mechanism was established based on the self-optimization steps of scheduling strategy parameters using historical data feedback, enabling the scheduling decision model to continuously improve itself. The system periodically extracts complete execution data from the logistics and transportation scheduling big data warehouse, covering a single scheduling cycle. A complete execution data cycle includes the initially generated scheduling plan, the sequence of actual execution trajectory points recorded by the vehicle via GPS, the actual time timestamps for each trajectory point and task node, and various event records such as vehicle malfunctions, traffic accidents, and customer order changes. In practical implementation, by comparing the estimated time nodes in the scheduling plan with the actual time nodes in the actual execution trajectory, the system aligns and compares the planned and actual timelines on the same route segments, calculating the travel time prediction error for each segment. This travel time prediction error is the difference between the actual travel time and the model's estimated travel time. The system analyzes the scheduling anomalies and their handling results recorded in the actual execution data. The analysis includes identifying delivery interruptions caused by vehicle mechanical failures, route closures caused by sudden traffic accidents, and unplanned operations caused by customer requests to change delivery addresses or times. The system also records the final handling methods and additional time consumption caused by these events.

[0112] In practical implementation, machine learning algorithms are used to iteratively adjust the internal parameters of the path planning model and time estimation model used in scheduling decisions. The machine learning algorithms take a multi-dimensional set of scheduling decision features from historical data as input features, and aim to minimize travel time prediction errors and optimize the handling of abnormal events as training and optimization objectives. It can be understood that the path planning model and time estimation model are core computational components integrated into multi-objective programming operations, and their internal parameters include various weight coefficients in the road resistance function and regression coefficients in the time estimation model. In some embodiments, at the beginning of each model training cycle, the system extracts a large number of recent historical scheduling records and their corresponding complete execution data from the logistics transportation scheduling big data warehouse to construct training samples. Optionally, the operation of constructing the training sample set is completed by a dedicated data preprocessing module. Each training sample contains a snapshot of the multi-dimensional scheduling decision feature set at a historical scheduling decision moment, a series of estimated travel times for each road segment generated by that scheduling decision, and subsequent actual travel time sequences and abnormal event records.

[0113] In practical implementation, a loss function is defined to guide model training. This loss function is a weighted average of a travel time prediction error term and an anomaly handling effectiveness evaluation term. The travel time prediction error term measures the overall difference between the model-predicted travel time series and the actual travel time series, while the anomaly handling effectiveness evaluation term quantifies the gap between the robustness of the model's suggested path or solution and the actual handling result when similar anomalies occur. The constructed training sample set is input into the path planning model and the time prediction model, which contain the internal parameters to be adjusted. Forward propagation is performed to obtain the predicted output. Forward propagation simulates the complete reasoning process from a multi-dimensional scheduling decision feature set to generating predicted travel times and path solutions. It can be understood that calculating the loss function value between the predicted output and the true value is accomplished through a pre-defined loss function formula. The gradient of the loss function value with respect to each internal parameter in the path planning model and the time prediction model is calculated using the backpropagation algorithm. The backpropagation algorithm propagates the error layer by layer from the output layer to the input layer along the model's computational graph and calculates the partial derivatives of each parameter.

[0114] In practical implementation, based on the calculated gradient values, optimization algorithms are used to update the internal parameters of the path planning model and the time prediction model. Commonly used optimization algorithms include stochastic gradient descent and adaptive moment estimation. The system repeatedly executes the complete process from training sample input, forward propagation calculation, loss calculation, backpropagation calculation to parameter update, until the loss function value converges to a stable state on the validation set or reaches the predetermined upper limit of training epochs, completing one iterative adjustment of the model's internal parameters. In some embodiments, the iterative adjustment process is performed offline on a dedicated machine learning training server to reduce the performance impact on the online scheduling system. After the iterative adjustment is completed, the latest model parameters obtained after training convergence are updated in the path planning model and time prediction model used by the online scheduling system for multi-objective planning operations, replacing the old model parameters, thereby achieving adaptive optimization of the scheduling strategy without manual intervention. Optionally, the loss function used for model training... One specific form can be expressed as:

[0115]

[0116] Where: symbol This represents the total loss value during model training, measured in units of time. (Symbol) and These are hyperparameter weights used to balance the two losses; both are dimensionless coefficients. (Symbol) Represents the total number of road segments participating in the comparison in a training sample, with the symbol... The symbol represents the model's estimated travel time for the j-th road segment. This represents the actual travel time for the j-th road segment. This represents absolute error, with the dimension of time. (Symbol) This represents the total number of anomalous event instances identified in a training sample, with the symbol […]. Represents the k-th exception event instance, symbol It is a mapping function whose input is an instance of an exception event. The function takes the characteristic data as input and outputs the additional time delay caused by the abnormal event. The output dimension is also time. Both terms on the right-hand side of the formula have the dimension of time. After being weighted and summed using dimensionless weights, the total loss is... The dimensions are consistent, both being time.

[0117] See Figure 5This is a bar chart comparing the statistics and handling effectiveness of logistics scheduling anomalies. It is used to analyze the frequency of occurrence of different types of anomalies and the system's handling capabilities, serving as core data for scheduling strategy self-optimization. Traffic accidents (120 times), vehicle malfunctions (85 times), and order changes (75 times) are the three main factors affecting scheduling, accounting for approximately 88% in total. Weather impacts (30 times) and other anomalies (10 times) occur less frequently, but still need to be included in model training to cover edge scenarios. The handling effectiveness of all types of anomalies is at an extremely low level (<0.05), indicating that the current scheduling system has significant shortcomings in handling anomalies, which is the core breakthrough point for strategy optimization. The handling effectiveness of high-frequency anomalies is slightly higher than that of low-frequency anomalies, reflecting that the system has some adaptability to high-frequency scenarios, but overall, it still needs significant improvement. For traffic accidents and vehicle malfunctions, the focus should be on optimizing route replanning, vehicle replacement, and order transfer logic, as these two types of anomalies have the greatest impact on scheduling stability.

[0118] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for intelligent scheduling of logistics transportation vehicles based on big data, characterized in that, include: Establish a big data warehouse for logistics transportation scheduling, collect and store raw scheduling data, which includes historical order information, vehicle file information, real-time traffic information, weather information, and warehouse operation status information; The raw scheduling data in the logistics transportation scheduling big data warehouse is standardized, cleaned, and integrated to form a multi-dimensional scheduling decision feature set. The multi-dimensional scheduling decision feature set is used to reflect the coupling relationship between orders, vehicles, road networks, weather, and warehouses in the time and space dimensions. Based on the multidimensional scheduling decision feature set, a spatiotemporal clustering algorithm is used to identify high-frequency delivery areas and frequently congested road sections, generating regional heat maps and road network risk layers. The regional heat maps are used to mark the order density of different geographical areas in a historical period, and the road network risk layers are used to mark the congestion frequency and average delay time of each road section in a historical period. Based on the real-time traffic information and weather information, the road network risk layer is dynamically corrected to generate a real-time road network status layer. By combining the details of the orders to be dispatched, the status of the available vehicle pool, the regional heat map, and the real-time road network status layer, a multi-objective programming operation is performed to generate a preliminary vehicle dispatching plan.

2. The intelligent scheduling method for logistics transportation vehicles based on big data according to claim 1, characterized in that, The raw scheduling data in the logistics transportation scheduling big data warehouse is standardized, cleaned, and integrated to form a multi-dimensional scheduling decision feature set, including: The historical order information is parsed to extract the cargo volume, weight, geographical coordinates of the delivery start and end points, promised delivery time window, and actual completion timestamp for each order; The vehicle file information is parsed to extract the vehicle's identification, load capacity, cargo volume, current location coordinates, current cargo status, and maintenance records for each vehicle. The real-time traffic information is analyzed to extract the traffic speed, congestion level, and event announcements of each road segment in the entire road network at continuous time points; The weather status information is analyzed to extract temperature, precipitation, wind speed, visibility, and disaster warning signals at each meteorological monitoring point at continuous time points; By associating historical order information, vehicle file information, real-time traffic information, weather information, and warehouse operation status information under the same time and space dimensions, a multi-dimensional scheduling decision feature set indexed by vehicle, order, road segment, and time point is constructed.

3. The intelligent scheduling method for logistics transportation vehicles based on big data according to claim 2, characterized in that, Based on the aforementioned multidimensional scheduling decision feature set, a spatiotemporal clustering algorithm is used to identify high-frequency delivery areas and frequently congested road sections, generating regional heat maps and road network risk layers, including: Extract the geographical coordinates of the delivery destinations of all historical orders from the multidimensional scheduling decision feature set, and form a set of coordinate points on the electronic map; The coordinate point set is analyzed using a density-based spatial clustering algorithm to identify the spatial range where the order point density exceeds a set threshold. The spatial range is defined as the high-frequency delivery area, and the order concentration index is calculated for each high-frequency delivery area. Extract the traffic speed sequence of all road segments in the historical period from the multidimensional scheduling decision feature set, and calculate the average daily congestion duration and speed fluctuation variance of each road segment; Road segments whose average daily congestion duration exceeds a duration threshold or whose speed fluctuation variance exceeds a variance threshold are defined as the frequently congested road segments. On the electronic map, the order concentration index of the high-frequency delivery area is rendered with different color depths to form the area heat map, and the frequently congested road sections and their historical average delay time are identified with different risk levels to form the road network risk layer.

4. The intelligent scheduling method for logistics transportation vehicles based on big data according to claim 3, characterized in that, Based on the real-time traffic information and weather information, the road network risk layer is dynamically corrected to generate a real-time road network status layer, including: Obtain the real-time traffic information at the current moment, and identify the road segments marked as congested or slow-moving and their estimated duration; Obtain the current weather status information and identify the areas where warnings have been issued regarding road icing, heavy rain, and dense fog that may affect driving; The identified currently congested road segments are compared with historically frequently congested road segments. For road segments that are both currently congested and historically frequently congested, their risk levels are weighted and increased based on their historical risk levels. The warning areas affecting driving are mapped onto the road network, thereby increasing the basic traffic resistance coefficient of all road sections within the warning areas; Information on road segments with weighted risk levels and road segments with increased basic traffic resistance coefficients is integrated and overlaid on the original road network risk layer to generate a real-time road network status layer that reflects instantaneous road conditions and weather effects. The real-time road network status layer integrates historical risk patterns with the instantaneous impact of current traffic and weather.

5. The intelligent scheduling method for logistics transportation vehicles based on big data according to claim 4, characterized in that, The process involves combining the details of the orders to be dispatched, the status of the available vehicle pool, the regional heatmap, and the real-time road network status layer to perform multi-objective programming operations and generate a preliminary vehicle dispatching plan, including: Parse the details of the orders to be dispatched to obtain the cargo attributes, delivery address coordinates and time requirements of all orders to be delivered; Scan the status of the available vehicle pool to obtain the identifiers, current locations, load capacities, and available cargo compartment volumes of all vehicles that are idle or about to be idle. Based on the coordinates of the delivery address, the heat map of the region is queried to obtain the order concentration index of the order destination region, and the orders are pre-grouped based on the order concentration index; Based on the real-time road network status layer, calculate the multi-segment path travel time from the current location of each available vehicle to the pickup and delivery points of all orders in its assigned order group; The calculation process for the multi-segment path travel time is as follows: For each road segment constituting the path, the dynamic travel time of that road segment is calculated based on its free-flow time and the dynamic travel delay coefficient read from the real-time road network status layer. The dynamic travel delay coefficient integrates a delay coefficient determined based on historical frequent congestion risk levels, a delay coefficient determined based on the current congestion status in real-time traffic information, and a delay coefficient determined based on the warning level in the current weather information. The multi-segment path travel time is obtained by summing the dynamic travel times of all road segments constituting the path. With the objectives of minimizing total transportation cost, maximizing vehicle occupancy rate, and minimizing average order delay time as the objective functions, and with vehicle capacity, order time window, and route connectivity as constraints, a multi-objective optimization model is constructed. The average order delay time is calculated by comparing the estimated delivery time of the order, which is determined by the travel time of the multi-segment route, with the promised delivery time window of the order. Solve the multi-objective optimization model to output the preliminary vehicle scheduling scheme, which includes the matching relationship between vehicles and orders, the order of pickup, the delivery route, and the estimated timetable for each node. The preliminary vehicle dispatching plan is used to indicate the order allocation, pickup sequence, delivery route, and expected time nodes for each dispatched vehicle.

6. The intelligent scheduling method for logistics transportation vehicles based on big data according to claim 5, characterized in that, It also includes a step of dynamically rescheduling the preliminary vehicle scheduling plan: After the initial vehicle dispatching plan is implemented, the real-time location, speed and remaining capacity of each dispatched vehicle are continuously monitored. The system receives real-time updates of the real-time traffic and weather information and dynamically refreshes the real-time road network status layer. The real-time location of the vehicle is compared with the planned path in the preliminary vehicle scheduling scheme to calculate the deviation between the actual progress and the planned progress. When the deviation exceeds the tolerance threshold, a rescheduling evaluation is triggered. When a rescheduling assessment is triggered, all currently en route orders, orders that have not yet started execution, real-time vehicle status, the refreshed real-time road network status layer, and the regional heat map are used as inputs to re-execute the multi-objective planning operation; The new scheduling scheme obtained by recalculation is compared with the original scheme. If the new scheme is better than the original scheme in terms of the objective function by more than the improvement threshold, the scheduling instructions of the unexecuted part are replaced by the new scheme.

7. The intelligent scheduling method for logistics transportation vehicles based on big data according to claim 6, characterized in that, When triggering a rescheduling evaluation, the multi-objective programming operation is re-executed using all currently en route orders, orders not yet started, real-time vehicle status, the updated real-time road network status layer, and the regional heatmap as input, including: Completed order tasks and ongoing tasks that cannot be changed are locked. The remaining variable tasks are defined as a set of tasks to be rescheduled. The set of tasks to be rescheduled includes the remaining delivery points of orders in transit and all orders that have not yet started execution. The status definition of the available vehicle pool is updated based on the real-time location of the vehicles, the remaining capacity of the carriages, and the current task status. Based on the refreshed real-time road network status layer, the estimated travel time from the real-time location of each available vehicle to all task points in the set of tasks to be rescheduled is recalculated. With the goal of minimizing new costs and delays, a local rescheduling optimization model is constructed based on the set of tasks to be rescheduled, the updated status of the available vehicle pool, and the recalculated estimated travel time. The local rescheduling optimization model is solved quickly, and the adjusted scheduling instruction sequence is output for the set of tasks to be rescheduled.

8. The intelligent scheduling method for logistics transportation vehicles based on big data according to claim 7, characterized in that, It also includes a self-optimization step for scheduling strategy parameters based on historical data feedback: Periodically extract complete execution data of completed scheduling cycles from the logistics transportation scheduling big data warehouse. The complete execution data includes scheduling plan, actual execution trajectory, actual time consumption and various event records. By comparing the estimated time nodes in the scheduling plan with the actual time nodes in the actual execution trajectory, the travel time prediction error of each segment is calculated. Analyze the scheduling anomalies caused by vehicle malfunctions, traffic accidents, and order changes in the actual execution data and their handling results; Using machine learning algorithms, with the multidimensional scheduling decision feature set as input and minimizing the trip time prediction error and optimizing the abnormal event handling results as training objectives, the internal parameters in the path planning model and time prediction model used in scheduling decisions are iteratively adjusted. The iteratively adjusted model parameters are updated in the model used for the multi-objective programming operation to achieve adaptive optimization of the scheduling strategy.

9. The intelligent scheduling method for logistics transportation vehicles based on big data according to claim 8, characterized in that, The method utilizes machine learning algorithms, taking the multi-dimensional scheduling decision feature set as input and minimizing the travel time prediction error and optimizing the handling of abnormal events as training objectives, to iteratively adjust the internal parameters of the path planning model and time estimation model used in scheduling decisions, including: Construct a training sample set, each sample containing a snapshot of the multidimensional scheduling decision feature set at a historical scheduling moment, the estimated travel time series generated by the corresponding scheduling decision, and the actual travel time series and abnormal event records that subsequently occurred; Define a loss function, which is a weighted sum of a trip time estimation error term and an abnormal event handling effect evaluation term. The training sample set is input into a path planning model and a time prediction model containing internal parameters to be adjusted, and forward propagation calculation is performed to obtain the prediction output; Calculate the loss function value between the predicted output and the true value, and calculate the gradient of the loss function with respect to the model's internal parameters using the backpropagation algorithm; Based on the calculated gradient, the internal parameters of the path planning model and the time prediction model are updated using an optimization algorithm. The process of inputting training samples and updating parameters is repeated until the loss function value converges or reaches the predetermined number of training rounds, thus completing the iterative adjustment of the model's internal parameters.

10. A big data-based intelligent dispatching system for logistics vehicles, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent scheduling method for logistics transportation vehicles based on big data as described in any one of claims 1 to 9.

Images