Digitized shared heavy truck transportation management system

Abstract

The invention discloses a digital shared heavy truck transportation management system. The system comprises an original data acquisition module, a transportation management feature acquisition module, a transportation supply and demand prediction module, a transportation demand scheduling module and a transportation demand intelligent management module. The invention relates to the technical field of information data processing, in particular to a digital shared heavy truck transportation management system, which innovatively adopts an improved feature clustering processing method, combines a multi-level feature selection mechanism fusing physical distribution and statistical strength, and improves the transportation efficiency. A more accurate, stable and input feature set is provided for the prediction model; a feature structural hierarchical modeling strategy and a gating fusion feature generation mechanism are innovatively proposed, and the precision and stability of regional-level transportation task prediction are improved; a regional supply and demand balanced multi-target transport capacity allocation optimization strategy is innovatively proposed, the response efficiency and the operation stability of shared scheduling are effectively improved, and intelligent management of new energy heavy truck transportation is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of information data processing, and in particular to a digital shared heavy truck transportation management system. Background Art

[0002] A digital shared heavy-duty truck transport management system refers to a management system that uses digital technology and big data analysis technology to intelligently integrate and optimize the entire process of new energy heavy-duty truck transportation. Its core is to achieve shared allocation of truck capacity resources, accurate matching of transportation needs and dynamic monitoring of the transportation process through real-time data collection, transmission and analysis, so as to improve the efficiency of heavy-duty truck transportation, reduce empty driving rate and cost, and support the intelligent and green development of logistics and transportation.

[0003] However, the raw data in traditional shared heavy-duty truck transport management systems generally have technical problems such as strong multi-source heterogeneity, severe time series fluctuations, and serious interference between positive and negative feature values, resulting in insufficient precision in transport task modeling and low prediction accuracy, which in turn affects the stability of the scheduling algorithm and the reasonable allocation of vehicle capacity; in traditional shared heavy-duty truck transport management systems, there are technical problems such as the lack of spatial structure modeling and the difficulty in unified modeling of continuous and discontinuous features in the regional-level transport task prediction process, resulting in obvious bottlenecks in the system in terms of prediction accuracy, scenario adaptability, and scheduling response efficiency; in traditional shared heavy-duty truck transport management systems, scheduling has technical problems such as rigid regional attribution, lack of steady-state compensation mechanism, and single scheduling optimization goal, which leads to low capacity matching accuracy. Summary of the Invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a digital shared heavy-duty truck transportation management system. In view of the technical problems that the original data of the traditional shared heavy-duty truck transportation management system generally have strong multi-source heterogeneity, severe time series fluctuations and serious interference of positive and negative feature values, which lead to insufficient accuracy of transportation task modeling and low prediction accuracy, thereby affecting the stability of the scheduling algorithm and the reasonable allocation of vehicle capacity, this solution innovatively adopts an improved feature clustering processing method, combines a multi-level feature selection mechanism that integrates physical distribution and statistical intensity, introduces the median vector as the initial clustering center, avoids the sensitivity of the mean to extreme data structures, improves the stability of clustering initialization and clustering quality, and The feature deviation index is used to measure the degree of dispersion of each feature within the cluster, and the heterogeneous changes of transportation characteristics are captured from the physical attribute level. At the same time, combined with the statistical indicators of the number of abnormal samples, the disturbance sensitivity of each feature in the real scene is quantified, and a multi-level feature selection mechanism that integrates physical distribution and statistical intensity is constructed to provide a more accurate, stable and interpretable input feature set for the prediction model; in the traditional shared heavy truck transportation management system, there are technical problems such as the lack of spatial structure modeling and the difficulty in unified modeling of continuous and discontinuous features in the regional transportation task prediction process, which leads to obvious bottlenecks in the system in terms of prediction accuracy, scene adaptability and scheduling response efficiency. This solution innovatively proposes A feature structural hierarchical modeling strategy and gated fusion feature generation mechanism are proposed. By structurally decoupling the feature set of transport task volume prediction, dynamic features with spatial correlation and temporal continuity and discontinuous features are modeled separately, avoiding information interference and modeling deviation caused by feature mixing, and significantly improving the clarity of feature expression and structural matching. After extracting the main path features and auxiliary path features, a dual-gating mechanism is introduced to calculate the gating vectors of the main path and auxiliary path respectively, which can dynamically perceive the importance of features from different sources, automatically highlight the task-dominant features and suppress redundant interference features, improve the accuracy and stability of regional transport task prediction, and provide a more reliable data basis for capacity scheduling; To address the technical issues of rigid regional attribution, lack of steady-state compensation mechanism, and single scheduling optimization objective in traditional shared heavy-duty truck transport management systems, which lead to low capacity matching accuracy, this solution innovatively proposes a hierarchical scheduling attribution mechanism based on vehicle regional existence probability. Combined with probability thresholds, it divides vehicle dynamics into three categories to construct a flexible soft-attribution scheduling mechanism, significantly improving the scheduling system's ability to discern vehicle availability and the stability of task allocation under uncertain environments. A multi-objective capacity allocation optimization strategy for regional supply and demand balance is proposed, and a regional-level multi-objective optimization function is constructed to generate the optimal allocation relationship between regions. This solves the problem that traditional scheduling cannot simultaneously take into account supply and demand balance, path cost, and energy efficiency constraints.A regional priority selection strategy was designed to prioritize vehicles with the highest scores within the region, reducing idle driving rates and dispatching costs, improving the region's internal capacity absorption capacity, and achieving flexible resource allocation, intelligent supply and demand matching, and multi-objective allocation coordination in the dispatch of new energy heavy-duty trucks. This effectively improved the response efficiency and operational stability of shared dispatch, providing key support for the refined regulation of new energy shared heavy-duty truck capacity and low-carbon and efficient transportation.

[0005] The technical solution adopted by the present invention is as follows: the present invention provides a digital shared heavy truck transportation management system, including a raw data acquisition module, a transportation management feature acquisition module, a transportation supply and demand forecasting module, a transportation demand scheduling module and a transportation demand intelligent management module;

[0006] The raw data collection module specifically collects raw data of truck transportation management through the truck transportation management platform;

[0007] The transport management feature acquisition module specifically performs data optimization processing and transport management feature processing on the original truck transport management data to obtain a transport feature data set;

[0008] The transportation supply and demand forecasting module specifically constructs a regional transportation task volume forecasting model through a feature structured hierarchical modeling strategy and a gated fusion feature generation mechanism for forecasting, adopts a long short-term memory neural network to predict the transport capacity at the vehicle level, and calculates the probability of each vehicle in each target area through a classification prediction model. The transportation supply and demand results are obtained by combining the regional transportation task volume forecast results, the vehicle capacity forecast results at the vehicle level, and the predicted probability of the vehicle in each forecast area.

[0009] The transport demand scheduling module specifically divides vehicles into core vehicles, alternative vehicles and shadow vehicles based on the probability of their presence in each area; calculates the difference between transport capacity supply and demand in each area, constructs a regional allocation relationship matrix, and uses a multi-objective optimization algorithm to generate the optimal inter-regional transport capacity allocation plan; according to the optimized allocation relationship matrix, core vehicles are allocated in the region first; if the resources in the region are insufficient, alternative vehicles are allocated from the supply region for cross-regional supplementation based on the allocation quota in the allocation matrix; if neither the core nor the alternative vehicles can meet the task requirements, the shadow vehicles in the region are called for backup scheduling; the allocation results of various types of vehicles are summarized to obtain the new energy shared heavy-duty truck vehicle scheduling strategy;

[0010] The intelligent transportation demand management module specifically implements intelligent management of new energy shared heavy-duty trucks based on the new energy shared heavy-duty truck scheduling strategy.

[0011] Furthermore, the raw data acquisition module specifically collects the raw data required for truck transport management through the truck transport management platform to obtain truck transport management raw data; the truck transport management raw data includes transport demand data and single-vehicle transport capacity data; the transport demand data includes transport order data, operation demand spatiotemporal dimension data, external influencing factor data and task statistics data; the single-vehicle transport capacity data includes vehicle status data, vehicle energy consumption data, vehicle transport order data and single-vehicle transport capacity statistics.

[0012] Furthermore, the transportation management feature acquisition module is used to perform data optimization processing and transportation management feature processing on the truck transportation management raw data to obtain a transportation feature data set; the transportation feature data set includes vehicle transportation management optimization data, a transportation task volume prediction feature set, and a vehicle capacity prediction feature set; specifically, the following steps are included:

[0013] Data optimization processing, specifically including data cleaning, data standardization and data label definition, to obtain truck transportation management optimization data;

[0014] The data cleaning specifically includes removing duplicate data, processing missing values, identifying outliers and checking logical consistency;

[0015] The data standardization processing includes time format standardization processing, unit standardization processing and classification field numerical processing;

[0016] The data tag definitions specifically include transportation demand tag definitions, shared vehicle capacity tag definitions, time window tag definitions, and area tag definitions;

[0017] Transport management feature processing involves constructing a transport management feature model and using truck transport management optimization data as input to obtain a transport task volume prediction feature set and a vehicle capacity prediction feature set. This process includes the following steps:

[0018] The transportation management feature model is constructed by using an improved feature clustering method combined with a multi-level feature selection mechanism that integrates physical distribution and statistical intensity. The model includes the following steps:

[0019] Obtain the feature sample matrix, specifically by extracting the candidate feature set from the data through statistical analysis methods and constructing the feature sample matrix , where m represents the number of samples and n represents the number of candidate features;

[0020] The initial cluster center is constructed by calculating the median of each column of the feature sample matrix D to form a median vector M as the initial cluster center, performing preliminary clustering operations, and obtaining the initial clustering results. ,like If the number of cluster samples is less than the minimum cluster volume threshold t or the clustering result is empty, the feature sample matrix D is divided into positive sample group and negative sample group according to the main direction of the feature;

[0021] Cluster center difference determination, specifically, calculating the mean P of the positive sample group and the mean Q of the negative sample group as candidates for alternative cluster centers; if , then choose one from P or Q as a single cluster center to generate clustering results, generate clustering results ,like , then P and Q are respectively used as cluster centers, and two sub-clusters are generated at the same time and ;in, Indicates the critical value for judging the degree of difference between P and Q;

[0022] Small sample fallback clustering processing, specifically if the number of samples in any cluster in any clustering result is , then call the partial priority clustering algorithm to perform backfill processing and generate supplementary clusters ;

[0023] Clustering termination condition determination: if the number of remaining unclustered samples is less than the preset lower limit threshold τ, the clustering operation is terminated to obtain the valid cluster set C;

[0024] Feature cluster formation, specifically in each effective cluster For each feature j, the deviation degree of the feature value within the cluster relative to the mean of the dimension is calculated to obtain the deviation index , and according to the deviation index , sort all features, group features with similar deviations into the same group, and regard each group as a feature cluster , the formula used is as follows:

[0025] ;

[0026] Where, Indicates that feature j is effectively clustered The degree of deviation in Indicates effective clustering The set of sample values of the jth feature in , Indicates that feature j is effectively clustered The mean of

[0027] Key feature selection involves applying the boxplot method to identify outliers for each candidate feature in the feature cluster, counting the number of abnormal samples contained in each feature, and selecting the feature with the largest number of abnormal samples in each feature cluster as the representative feature of the cluster. All selected features are aggregated to form the key feature set F.

[0028] Obtaining a transport task volume prediction feature set, specifically by inputting transport demand data in the truck transport management optimization data into a transport management feature model to obtain a transport task volume prediction feature set;

[0029] A vehicle capacity prediction feature set is obtained, specifically by inputting the single vehicle transport capacity data in the truck transport management optimization data into the transport management feature model to obtain the vehicle capacity prediction feature set.

[0030] Furthermore, the transportation supply and demand forecasting module includes transportation task volume forecasting, vehicle capacity forecasting, vehicle regional location forecasting, and obtaining transportation supply and demand results, specifically including the following steps:

[0031] Transport task volume forecasting is used to predict future transport demand within a region. It includes the following steps:

[0032] Constructing a transportation task volume prediction model includes the following steps:

[0033] The regional graph structure is constructed by first dividing the transportation space into multiple prediction areas according to the regional labels and then constructing the regional graph based on the adjacency relationship between the areas, the cargo flow path and the road connectivity. , generate the adjacency matrix A according to the connection relationship between regions; where R represents the regional node set, E represents the edge set, and the elements in the adjacency matrix are Indicates area and The connection relationship between them;

[0034] Acquire multi-source features of transportation demand. Specifically, divide the transportation task volume prediction feature set into continuous feature subsets and discontinuous feature subsets based on the temporal continuity and business attributes of the features.

[0035] Continuous feature processing, specifically organizing continuous feature subsets into three-dimensional input tensors by region, time step, and feature dimension, and inputting the tensors into the graph convolutional network GCN and the gated recurrent unit network GRU for joint modeling to obtain the main path features at the regional level ;

[0036] Auxiliary state feature discontinuous processing, specifically constructing the discontinuous feature subset into a two-dimensional input matrix, and then inputting it into the multi-layer perceptron network structure, sequentially performing nonlinear mapping through multiple fully connected layers and GELU activation functions to obtain the auxiliary area business scenario features ;

[0037] Generate fusion features, specifically obtain the main path gating vector through the gating vector generation function and auxiliary path gating vector , main path gating vector Main path feature is the input, auxiliary path gating vector Splicing vectors with main path and auxiliary path features The gating vector generation function includes two layers of fully connected neural networks, which use GELU and Sigmoid activation functions for nonlinear transformation to obtain the path gating vector; finally, the fusion feature is obtained by weighting the gating vectors of the main path and auxiliary path features and splicing them together. ;

[0038] Multi-objective regional transport demand forecasting, specifically by integrating the feature vector Input to the multi-target regression output layer and output for each region and ;in, It indicates the total cargo tonnage value predicted for the area during the target time period. Indicates the predicted total transport mileage value corresponding to the area;

[0039] The transport task volume prediction model training specifically comprises extracting transport demand data within a historical period to construct a training data set, and generating a historical transport task volume prediction feature set through the transport management feature acquisition module. The transport demand label of the corresponding period is used as the supervision output field to train the transport task volume prediction model to obtain a trained transport task volume prediction model;

[0040] Regional transport task volume forecasting is used to achieve advance forecasting of the transport task volume of each region within the target forecast period. Specifically, the transport demand data of N consecutive time periods before the forecast period are used to form a dataset to be forecasted, and the real-time transport task volume forecast feature set is obtained through the transport management feature acquisition module. The real-time transport task volume forecast feature set is used as the input data of the trained transport task volume forecast model to obtain the regional transport task volume forecast result, which includes the total cargo tonnage value of each region. and total transport mileage ;

[0041] Vehicle capacity forecasting specifically includes the following steps:

[0042] A single-vehicle capacity prediction model is constructed and trained. Specifically, a long short-term memory neural network is used to construct the single-vehicle capacity prediction model. The single-vehicle transport capacity data within the historical period is used as the training dataset. The transportation management feature acquisition module generates a historical vehicle capacity prediction feature set. The vehicle capacity label of the corresponding period is used as the supervision output field to train the single-vehicle capacity prediction and obtain the trained single-vehicle capacity prediction model.

[0043] Single-vehicle capacity prediction is specifically to form a dataset to be predicted using the single-vehicle transport capacity data of N consecutive time periods before the prediction period, and obtain a real-time vehicle capacity prediction feature set through the transportation management feature acquisition module, and use the real-time vehicle capacity prediction feature set as the input data of the trained single-vehicle capacity prediction model to obtain a single-vehicle capacity prediction result, which includes the load capacity of each vehicle. , Remaining mileage and service time factor ;

[0044] Vehicle regional location prediction, specifically, using a vehicle location prediction model based on a long short-term memory neural network and a Softmax classification layer, using the single-vehicle transport capacity data of N consecutive time periods before the prediction period to form a vehicle regional location prediction dataset, and obtaining a real-time location prediction feature set through the transportation management feature acquisition module, and using the real-time location prediction feature set as the input data of the vehicle location prediction model to obtain the vehicle's presence probability in the area ;

[0045] Obtaining the transport supply and demand results, specifically by calculating the transport intensity value, calculating the single vehicle capacity value, and summarizing the regional total capacity value, to obtain the transport supply and demand results, the transport supply and demand results including the transport intensity value, the regional total available capacity, and the vehicle comprehensive capacity value, including the following steps:

[0046] Transport intensity value calculation, specifically based on the regional transport task volume forecast results, calculate each region Transport intensity value , defined as the product of total cargo tonnage and total transport mileage;

[0047] Single vehicle capacity calculation, specifically the calculation of each shared new energy heavy truck based on the single vehicle capacity prediction results Comprehensive vehicle capacity value within the target time period , defined as the product of load capacity, remaining endurance and service time coefficient;

[0048] The total capacity value of the region is summarized, specifically the comprehensive capacity value of the bicycle The probability of the vehicle being in the area Multiply them together to get the weighted capacity value of the vehicle in the area. , then for all the people in the target time period, in the area The predicted probability of existence The corresponding weighted capacity value Summarize and get the total available capacity of the region .

[0049] Furthermore, the transportation demand scheduling module specifically includes the following steps:

[0050] Vehicle stratification, specifically based on the probability of a vehicle's presence in an area , and combined with the preset core attribution probability threshold and the minimum scheduling trust threshold , all vehicles are divided into the following three categories, divided into the core vehicle set , Alternative Vehicle Set and Shadow Vehicle Set ;

[0051] Calculate the regional capacity supply and demand difference, specifically based on the transport intensity value of each region and regional total available capacity , calculate the regional capacity supply and demand difference ;

[0052] Regional multi-objective capacity balance optimization, specifically based on the regional capacity supply and demand gap , build a regional allocation relationship matrix ; The regional allocation relationship matrix Each element in represents the demand supply area Towards unmet need areas The number of vehicles to be dispatched; and a regional multi-objective optimization function is constructed. The regional multi-objective optimization function is optimized using a non-dominated sorting genetic algorithm to obtain a set of Pareto optimal solutions. From the optimal solution set, the solution that minimizes the sum of the absolute values of the supply and demand differences in all regions is preferentially selected. The solution is taken as the final regional scheduling result, and the optimized regional allocation relationship matrix is obtained. The regional multi-objective optimization function includes minimizing the total demand difference of insufficient capacity in each region , minimize the total idle mileage caused by regional allocation and minimize the number of vehicles that are below the battery threshold after allocation;

[0053] Core vehicles are allocated first, specifically according to the optimized regional allocation relationship matrix and gather the core vehicles in the area The vehicles in the area are assigned to the transport task pool of the area first and marked as assigned, and the remaining transport intensity value of the area is updated. If the core vehicle concentrates the comprehensive transport capacity value of the vehicle When the transport intensity value of the region is fully covered and there are some redundant core vehicles, these redundant core vehicles are included in the candidate vehicle set to obtain the regional core vehicle allocation result;

[0054] The candidate vehicle selection is specifically to perform scheduling scoring on all candidate vehicles through the vehicle-level scheduling scoring function to obtain the candidate vehicle scheduling score value, sort the scheduling score values of all candidate vehicles from high to low, and allocate them according to the following local priority selection strategy and cross-regional allocation selection strategy to obtain the regional candidate vehicle allocation result. The local priority selection strategy is specifically to give priority to selecting resources from the candidate vehicles with higher scores in the local area, and supplement the transportation task requirements of the local area in turn until the task of the area is fully met or the candidate vehicle pool is exhausted; the cross-regional allocation selection strategy is only when the task requirements of the local area have been met, according to Based on the allocation quota demand of the target area in the target area, the candidate vehicle set that can be allocated is screened from the corresponding supply area and sorted by the score value until the allocation quota of the target area is met or the candidate resources are exhausted. The formula used is as follows:

[0055] ;

[0056] Where, represents the vehicle-level scheduling scoring function, 、 and are respectively expressed as the weighted coefficients of each scoring item, Indicates the path distance from the current location of the vehicle to the target area. Indicates the penalty term brought by the vehicle's current battery status, which is used to restrict vehicles with low battery status from participating in scheduling;

[0057] Shadow vehicle backup scheduling: Specifically, in areas where there is a gap in task demand, the scheduling score of each shadow vehicle in the area is calculated according to the vehicle-level scheduling scoring function, and the scores are allocated in descending order until the task demand is met; if the demand supply area The number of vehicles that should be dispatched has not been reached If the shadow vehicle resources in this area cannot meet the required quota for all tasks, the system will mark the unassigned tasks as failed and will no longer call other regional vehicles to participate in the backup scheduling, thus obtaining the regional shadow vehicle allocation result;

[0058] The dispatching strategy is generated by integrating the regional core vehicle allocation results, regional alternative vehicle allocation results, and regional shadow vehicle allocation results to obtain the new energy shared heavy-duty truck vehicle dispatching strategy.

[0059] Furthermore, the intelligent transportation demand management module specifically allocates vehicle resources and issues tasks in each target area according to the new energy shared heavy-duty truck vehicle scheduling strategy, executes scheduling instructions, dynamically manages vehicle operating status, and realizes intelligent management of shared transportation capacity of new energy heavy-duty trucks.

[0060] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0061] (1) In view of the technical problems that the raw data in the traditional shared heavy truck transportation management system generally have strong multi-source heterogeneity, severe time series fluctuations and serious interference between positive and negative values of features, which lead to insufficient accuracy in transportation task modeling and low prediction accuracy, thus affecting the stability of the scheduling algorithm and the reasonable allocation of vehicle capacity, this scheme innovatively adopts an improved feature clustering processing method, combined with a multi-level feature selection mechanism that integrates physical distribution and statistical intensity, introduces the median vector as the initial cluster center, avoids the sensitivity of the mean to extreme data structures, improves the stability of cluster initialization and clustering quality, measures the degree of dispersion of each feature within the cluster based on the feature deviation index, captures the heterogeneous changes of transportation characteristics from the physical attribute level, and combines the statistical indicator of the number of abnormal samples to quantify the disturbance sensitivity of each feature in real scenarios, and constructs a multi-level feature selection mechanism that integrates physical distribution and statistical intensity to provide a more accurate, stable and interpretable input feature set for the prediction model.

[0062] (2) In view of the technical problems of the lack of spatial structure modeling and the difficulty in unified modeling of continuous and discontinuous features in the traditional shared heavy truck transportation management system in the process of regional transportation task prediction, which leads to obvious bottlenecks in the system in terms of prediction accuracy, scene adaptability and scheduling response efficiency, this scheme innovatively proposes a feature structure hierarchical modeling strategy and a gated fusion feature generation mechanism. By structurally decoupling the transportation task volume prediction feature set, dynamic features with spatial correlation and temporal continuity and discontinuous features are modeled separately, avoiding information interference and modeling deviation caused by feature mixing, and significantly improving the clarity of feature expression and structural matching. After extracting the main path features and auxiliary path features, a dual gating mechanism is introduced to calculate the gating vectors of the main path and auxiliary path respectively. It can dynamically perceive the importance of features from different sources, automatically highlight the task-dominant features and suppress redundant interference features, improve the accuracy and stability of regional transportation task prediction, and provide a more reliable data basis for capacity scheduling.

[0063] (3) In response to the technical problems of rigid regional attribution, lack of steady-state compensation mechanism and single scheduling optimization target in the traditional shared heavy-duty truck transportation management system, which leads to low capacity matching accuracy, this scheme innovatively proposes to introduce a vehicle regional probability hierarchical scheduling attribution mechanism, and combines the probability threshold to divide the vehicle dynamics into three categories to construct a flexible soft attribution scheduling mechanism, which significantly improves the scheduling system's ability to judge vehicle availability and task allocation stability under uncertain environments; proposes a multi-objective capacity allocation optimization strategy for regional supply and demand balance, constructs a regional multi-objective optimization function, and generates the optimal allocation relationship between regions, solving the problem that traditional scheduling cannot simultaneously take into account supply and demand balance, path cost and energy efficiency constraints; designs a local priority selection strategy to give priority to matching vehicle resources with high scores in the local area, reduces the empty driving rate and scheduling cost, improves the internal digestion capacity of the region, realizes flexible resource attribution, intelligent supply and demand matching and multi-objective allocation coordination in the new energy heavy-duty truck scheduling process, improves the response efficiency and operation stability of shared scheduling, and provides key support for the refined regulation of new energy shared heavy-duty truck capacity and low-carbon and efficient transportation. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 A module diagram of a digital shared heavy truck transportation management system provided by the present invention;

[0065] Figure 2 A flow chart of building a transportation management feature model in the transportation management feature acquisition module;

[0066] Figure 3 This is a flow chart of the transportation supply and demand forecasting module;

[0067] Figure 4 A flow chart for building a transportation task volume forecasting model in the transportation supply and demand forecasting module;

[0068] Figure 5 This is a flow chart of the transportation demand scheduling module;

[0069] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0070] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0071] In the description of the present invention, it should be understood that terms such as "up", "down", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0072] Example 1, see Figure 1 The present invention provides a digital shared heavy truck transportation management system, which includes a raw data acquisition module, a transportation management feature acquisition module, a transportation supply and demand forecasting module, a transportation demand scheduling module and a transportation demand intelligent management module;

[0073] The raw data acquisition module specifically collects raw truck transport management data through the truck transport management platform and sends the data to the transport management feature acquisition module;

[0074] The transport management feature acquisition module receives the data sent by the original data acquisition module, performs data optimization processing and transport management feature processing on the original truck transport management data, obtains a transport feature data set, and sends the data to the transport supply and demand forecast module;

[0075] The transport supply and demand forecasting module receives data sent by the transport management feature acquisition module, specifically constructs a regional transport task volume forecasting model through a feature structured hierarchical modeling strategy and a gated fusion feature generation mechanism for forecasting, adopts a long short-term memory neural network to forecast the transport capacity at the vehicle level, calculates the probability of each vehicle existing in each target area through a classification forecasting model, combines the regional transport task volume forecast results, the vehicle capacity forecast results at the vehicle level, and the forecast probability of the vehicle in each forecast area to calculate the transport supply and demand results, and sends the data to the transport demand scheduling module;

[0076] The transport demand scheduling module receives data sent by the transport supply and demand forecasting module, specifically stratifies and assigns vehicles based on the probability of their presence in each area, and divides the vehicles into core vehicles, alternative vehicles, and shadow vehicles; calculates the difference between transport supply and demand in each area, and constructs a regional allocation relationship matrix, and uses a multi-objective optimization algorithm to generate the optimal inter-regional transport capacity allocation plan; according to the optimized allocation relationship matrix, core vehicles are allocated in the area first; if the resources in the area are insufficient, alternative vehicles are allocated from the supply area for cross-regional supplementation based on the allocation quota in the allocation matrix; if neither the core nor the alternative vehicles can meet the task requirements, the shadow vehicles in the area are called for backup scheduling; the allocation results of various types of vehicles are summarized to obtain the new energy shared heavy-duty truck vehicle scheduling strategy, and the data is sent to the transport demand intelligent management module;

[0077] The intelligent transportation demand management module specifically implements intelligent management of new energy shared heavy-duty trucks based on the new energy shared heavy-duty truck scheduling strategy.

[0078] Example 2, see Figure 1 , this embodiment is based on the above embodiment, the raw data acquisition module specifically collects the raw data required for truck transportation management through the truck transportation management platform to obtain the truck transportation management raw data; the truck transportation management raw data includes transportation demand data and single-vehicle transportation capacity data; the transportation demand data includes transportation order data, operation demand spatiotemporal dimension data, external influencing factor data and task statistics data; the single-vehicle transportation capacity data includes vehicle status data, vehicle energy consumption data, vehicle transportation order data and single-vehicle transportation capacity statistics; the transportation order data includes delivery time, arrival time, delivery place, delivery place, cargo type, cargo weight and transportation distance; the operation demand spatiotemporal dimension data includes Time window, regional location, whether it is a holiday and whether it is a weekend; the external influencing factor data includes weather type, e-commerce activity identification and traffic control identification; the task statistics data includes total cargo tonnage and total transportation mileage; the vehicle status data includes current battery power, loadable weight, remaining cruising range, geographic location, vehicle status, service time coefficient and time window; the vehicle energy consumption data includes power consumption rate and average power consumption of the task; the vehicle transportation order data includes current order delivery time, arrival time, shipping place, receiving place, cargo type, cargo weight, transportation distance and vehicle order planning route; the service time coefficient represents the proportion of time that the vehicle is in a dispatchable state within the time period, reflecting the vehicle scheduling availability.

[0079] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. The transportation management feature acquisition module is used to perform data optimization processing and transportation management feature processing on the truck transportation management raw data to obtain a transportation feature data set; the transportation feature data set includes vehicle transportation management optimization data, a transportation task volume prediction feature set, and a vehicle capacity prediction feature set. The specific steps include:

[0080] Data optimization processing is used to improve the quality, consistency, and structural standardization of raw data. Specifically, it includes data cleaning, data standardization, and data label definition to obtain optimized data for truck transportation management;

[0081] The data cleaning is used to improve the accuracy, completeness and logical consistency of the original data, specifically by removing duplicate data, processing missing values, identifying outliers and checking logical consistency;

[0082] The duplicate data removal is specifically to process duplicate data in the original data through a data deduplication algorithm;

[0083] The missing value processing is specifically processing the missing data of the fields in the original data by linear interpolation;

[0084] The outlier identification is specifically to process the physical unreasonable and logical abnormal values appearing in the original data through rule threshold detection;

[0085] The logical consistency check specifically processes the problems of time inconsistency, spatial conflict and business rule violation in the original data through the logic relationship check rules between fields;

[0086] The data standardization process is used to unify raw data of different sources, formats, units and structures, specifically time format standardization, unit standardization and classification field numerical processing;

[0087] The time format standardization process converts the time fields in the original data into a standard timestamp format through a time format standardization algorithm;

[0088] The unit standard processing unifies the measurement units of physical quantities in the original data through unit conversion processing;

[0089] The numerical processing of the categorical fields is performed by encoding the categorical fields, and the non-numerical fields are numerically processed using the one-hot encoding method;

[0090] The data tag definition is used to clarify the field functions in the original data, specifically the transportation demand tag definition, shared vehicle capacity tag definition, time window tag definition and area tag definition;

[0091] The transport demand tag definition is used to construct the supervisory output fields in the transport task prediction model, specifically including the total cargo tonnage and the total transport mileage;

[0092] The shared vehicle capacity tag definition is used to construct the supervision output fields in the vehicle capacity prediction model, specifically including the load capacity, remaining range and service time coefficient;

[0093] The time window label definition is used to build a time series prediction structure, clarifying the time period corresponding to each set of training data and prediction results. Specifically, the time window length is set to 6 hours and the sliding interval is set to 1 hour;

[0094] The regional label definition is used for geographic space division to support regional modeling, capacity aggregation and task scheduling. Specifically, the core demand of heavy trucks is concentrated on cargo distribution and transportation nodes, and the functional blocks within the city are used as the basis for prediction areas.

[0095] Transport management feature processing involves constructing a transport management feature model and using truck transport management optimization data as input to obtain a transport task volume prediction feature set and a vehicle capacity prediction feature set. This process includes the following steps:

[0096] Specifically, an improved feature clustering processing method is used, combined with a multi-level feature selection mechanism that integrates physical distribution and statistical intensity. The specific steps include:

[0097] Obtain the feature sample matrix, specifically by extracting the candidate feature set from the data through statistical analysis methods and constructing the feature sample matrix , where m represents the number of samples and n represents the number of candidate features;

[0098] The initial cluster center is constructed by calculating the median of each column of the feature sample matrix D to form a median vector M as the initial cluster center, performing preliminary clustering operations, and obtaining the initial clustering results. ,like If the number of cluster samples is less than the minimum cluster volume threshold t or the clustering result is empty, the feature sample matrix D is divided into a positive sample group and a negative sample group according to the main feature direction; the preliminary clustering operation means taking M as the center and assigning all samples to the center according to the Euclidean distance to form a cluster;

[0099] Cluster center difference determination is used to enhance the adaptability to asymmetric data structures. Specifically, the mean P of the positive sample group and the mean Q of the negative sample group are calculated as candidates for alternative cluster centers. , then choose one from P or Q as a single cluster center to generate clustering results, generate clustering results ,like , then P and Q are respectively used as cluster centers, and two sub-clusters are generated at the same time and ;in, Indicates the critical value for judging the degree of difference between P and Q;

[0100] Small sample fallback clustering processing is used to ensure that the final clustering results meet the clustering quality and sample representativeness requirements. Specifically, if the number of samples in any cluster in any clustering result is less than , then call the partial priority clustering algorithm to perform backfill processing and generate supplementary clusters ;

[0101] Clustering termination condition determination: if the number of remaining unclustered samples is less than the preset lower limit threshold τ, the clustering operation is terminated to obtain the valid cluster set C;

[0102] Feature cluster formation is used to identify the degree of deviation of feature dimensions in each valid cluster subset and group them, and to mine the heterogeneous differences of transportation characteristics from the physical attribute level to obtain feature clusters. Specifically, in each valid cluster For each feature j, the deviation degree of the feature value within the cluster relative to the mean of the dimension is calculated to obtain the deviation index , and according to the deviation index , sort all features, group features with similar deviations into the same group, and regard each group as a feature cluster , the formula used is as follows:

[0103] ;

[0104] Where, Indicates that feature j is effectively clustered The degree of deviation in Indicates effective clustering The set of sample values of the jth feature in , Indicates that feature j is effectively clustered The mean of

[0105] Key feature selection is used to select the most representative core features in each feature cluster, perform feature selection based on statistical characteristics, and construct the final feature set. Specifically, the box plot method is applied to each candidate feature in the feature cluster to identify outliers, and abnormal samples are marked based on the upper and lower quartiles and 1.5 times the interquartile range principle. The number of abnormal samples contained in each feature is counted. In each feature cluster, the feature with the largest number of abnormal samples is selected as the representative feature of the cluster, and all selected features are aggregated to form the key feature set F;

[0106] Obtaining a transport task volume prediction feature set, specifically by inputting transport demand data in the truck transport management optimization data into a transport management feature model to obtain a transport task volume prediction feature set;

[0107] A vehicle capacity prediction feature set is obtained, specifically by inputting the single vehicle transport capacity data in the truck transport management optimization data into the transport management feature model to obtain the vehicle capacity prediction feature set.

[0108] By performing the above operations, we address the technical problems in traditional shared heavy-duty truck transportation management systems, such as the common presence of strong multi-source heterogeneity, severe time series fluctuations, and severe interference between positive and negative feature values in raw data. These problems lead to insufficient precision in transportation task modeling and low prediction accuracy, which in turn affects the stability of the scheduling algorithm and the rational allocation of vehicle capacity. This solution innovatively adopts an improved feature clustering processing method, combined with a multi-level feature selection mechanism that integrates physical distribution and statistical intensity. The median vector is introduced as the initial cluster center to avoid the mean's sensitivity to extreme data structures, improve the stability of cluster initialization and clustering quality, measure the degree of dispersion of each feature within the cluster based on the feature deviation index, capture the heterogeneous changes in transportation characteristics from the physical attribute level, and combine the statistical indicator of the number of abnormal samples to quantify the perturbation sensitivity of each feature in real scenarios. This constructs a multi-level feature selection mechanism that integrates physical distribution and statistical intensity, providing a more accurate, stable, and interpretable input feature set for the prediction model.

[0109] Example 4, see Figure 1 、 Figure 3 and Figure 4 This embodiment is based on the above embodiment. The transportation supply and demand forecasting module is used to forecast the transportation task demand and available vehicle capacity within a target time period, including transportation task quantity forecasting, vehicle capacity forecasting, vehicle regional location forecasting, and obtaining transportation supply and demand results. Specifically, the module includes the following steps:

[0110] Transport task volume forecasting is used to predict future transport demand within a region. It includes the following steps:

[0111] Constructing a transportation task volume prediction model includes the following steps:

[0112] The regional graph structure is constructed to express the spatial connection relationship between each prediction area. Specifically, the transportation space is divided into multiple prediction areas according to the regional labels and the regional graph is constructed based on the adjacency relationship between the areas, the cargo flow path and the road connectivity. , generate the adjacency matrix A according to the connection relationship between regions; where R represents the regional node set, E represents the edge set, and the elements in the adjacency matrix are Indicates area and The connection relationship between them; the adjacency matrix is used as the input structure information of the graph convolutional network GCN; the connection relationship is defined as when the region and Adjacent in space, set Otherwise, set ;

[0113] Acquire multi-source features of transportation demand. Specifically, the transportation task volume forecast feature set is structurally divided into continuous feature subsets and discontinuous feature subsets based on the temporal continuity and business attributes of the features. The continuous feature subset mainly includes feature variables with spatial dependence and temporal continuous variation characteristics between prediction areas, including multi-time transportation order statistical features and regional meteorological characteristics within continuous time periods. The discontinuous feature subset mainly includes label-type, periodic, and discontinuous variation features.

[0114] Continuous feature processing is used to model regional transportation task features with temporal continuity and spatial correlation. Through graph structure propagation and dynamic trend extraction, the potential evolution law of regional transportation tasks is learned. Specifically, the continuous feature subset is organized into a three-dimensional input tensor according to region, time step and feature dimension, and the tensor is sequentially input into the graph convolutional network GCN and the gated recurrent unit network GRU for joint modeling to obtain the regional-level main path spatiotemporal joint features. ;

[0115] The graph convolutional network (GCN) uses the adjacency matrix to perform spatial aggregation and graph structure propagation on the features of each region, and extracts the spatial coupling features between regions.

[0116] The gated recurrent unit network GRU performs trend modeling and temporal pattern extraction on the time series input of each region;

[0117] Auxiliary state feature discontinuous processing is used to deeply model auxiliary features that do not have a continuous time structure but have a regulatory effect on the change of transportation tasks. Specifically, the discontinuous feature subset is constructed into a two-dimensional input matrix, which is then input into the multi-layer perceptron network structure. It is sequentially mapped through multiple fully connected layers and GELU activation functions for nonlinear mapping to obtain the auxiliary regional business scenario features. ;

[0118] Generate fusion features to perform weighted fusion of the feature representations of the main path and the auxiliary path to construct fusion features that have both regional structural characteristics and business status expression capabilities. Specifically, the main path gating vector is obtained through the gating vector generation function. and auxiliary path gating vector , where the main path gating vector Represented by main path features is the input, auxiliary path gating vector Splicing vectors with main path and auxiliary path features The gating vector generation function includes two layers of fully connected neural networks, which use GELU and Sigmoid activation functions for nonlinear transformation to obtain the path gating vector; finally, the fusion feature is obtained by weighting the gating vectors of the main path and auxiliary path features and splicing them together. ; The formula used is as follows:

[0119] ;

[0120] ;

[0121] ;

[0122] Where, and They represent the weight parameters and bias parameters of the first layer of the fully connected network in the main path, and Respectively represent the weight parameters and bias parameters of the second layer of the fully connected network of the main path, and They represent the weight parameters and bias parameters of the first layer of the fully connected network of the auxiliary path, and They represent the weight parameters and bias parameters of the second layer of the fully connected network of the auxiliary path, represents element-wise multiplication, Represents vector concatenation;

[0123] Multi-objective regional transport demand forecasting, specifically by integrating the feature vector Input to the multi-target regression output layer and output for each region and ;in, It indicates the total cargo tonnage value predicted for the area during the target time period. Indicates the predicted total transport mileage value corresponding to the area;

[0124] The transport task volume prediction model training specifically comprises extracting transport demand data within a historical period to construct a training data set, and generating a historical transport task volume prediction feature set through the transport management feature acquisition module. The transport demand label of the corresponding period is used as the supervision output field to train the transport task volume prediction model to obtain a trained transport task volume prediction model;

[0125] Regional transport task volume forecasting is used to achieve advance forecasting of the transport task volume of each region within the target forecast period. Specifically, the transport demand data of N consecutive time periods before the forecast period are used to form a dataset to be forecasted, and the real-time transport task volume forecast feature set is obtained through the transport management feature acquisition module. The real-time transport task volume forecast feature set is used as the input data of the trained transport task volume forecast model to obtain the regional transport task volume forecast result, which includes the total cargo tonnage value of each region. and total transport mileage ;

[0126] Vehicle capacity forecasting is used to predict the available capacity of each shared heavy-duty truck in the target area within a specified forecast time window at a vehicle-by-vehicle level. Specifically, it includes the following steps:

[0127] A single-vehicle capacity prediction model is constructed and trained. Specifically, a long short-term memory neural network is used to construct the single-vehicle capacity prediction model. The single-vehicle transport capacity data within the historical period is used as the training dataset. The transportation management feature acquisition module generates a historical vehicle capacity prediction feature set. The vehicle capacity label of the corresponding period is used as the supervision output field to train the single-vehicle capacity prediction and obtain the trained single-vehicle capacity prediction model.

[0128] Single-vehicle capacity prediction is specifically to form a dataset to be predicted using the single-vehicle transport capacity data of N consecutive time periods before the prediction period, and obtain a real-time vehicle capacity prediction feature set through the transportation management feature acquisition module, and use the real-time vehicle capacity prediction feature set as the input data of the trained single-vehicle capacity prediction model to obtain a single-vehicle capacity prediction result, which includes the load capacity of each vehicle. , Remaining mileage and service time factor ;

[0129] Vehicle regional location prediction is used to independently predict the regional location of each shared new energy heavy-duty truck within the target prediction time period, thereby providing a spatial attribution basis for the regional level available transport capacity aggregation. Specifically, a vehicle location prediction model is constructed based on a long short-term memory neural network and a Softmax classification layer. The single-vehicle transport capacity data of N consecutive time periods before the prediction period constitutes a vehicle regional location prediction dataset, and the real-time location prediction feature set is obtained through the transportation management feature acquisition module. The real-time location prediction feature set is used as the input data of the vehicle location prediction model to obtain the vehicle's presence probability in the region. ;in, Indicates vehicle In the area The probability of existence, and satisfy ;

[0130] Obtaining transport supply and demand results, which are used to quantitatively evaluate the supply and demand relationship between transport task requirements and dispatchable transport capacity in the target area within the forecast period, is done by calculating transport intensity values, calculating single vehicle capacity values, and summarizing regional total transport capacity values. The transport supply and demand results include transport intensity values, regional total available transport capacity, and vehicle comprehensive transport capacity values, and include the following steps:

[0131] The transport intensity value is calculated to measure the overall load level of regional transport tasks. Specifically, the transport intensity value is calculated for each region based on the regional transport task volume forecast results. Transport intensity value , defined as the product of total cargo tonnage and total transport mileage;

[0132] Single vehicle capacity calculation is used to evaluate the dispatchable transport capacity of a single vehicle level. Specifically, the capacity of each shared new energy heavy truck is calculated based on the single vehicle capacity prediction results. Comprehensive vehicle capacity value within the target time period , defined as the product of load capacity, remaining endurance and service time coefficient;

[0133] The regional total capacity value is summarized to obtain the overall dispatchable capacity level of each forecast area within the target time period, specifically the comprehensive capacity value of each vehicle The probability of the vehicle being in the area Multiply them together to get the weighted capacity value of the vehicle in the area. , then for all the people in the target time period, in the area The predicted probability of existence The corresponding weighted capacity value Summarize and get the total available capacity of the region , the formula used is as follows:

[0134] ;

[0135] Where, Indicates that the vehicle Will appear in the area .

[0136] By performing the above operations, the traditional shared heavy truck transport management system solves the technical problems of lack of spatial structure modeling and difficulty in unified modeling of continuous and discontinuous features in the regional transport task prediction process, which leads to obvious bottlenecks in the system in terms of prediction accuracy, scenario adaptability and scheduling response efficiency. This solution innovatively proposes a feature structured hierarchical modeling strategy and a gated fusion feature generation mechanism. By structurally decoupling the transport task volume prediction feature set, dynamic features with spatial correlation and temporal continuity and discontinuous features are modeled separately, avoiding information interference and modeling bias caused by feature mixing, and significantly improving the clarity of feature expression and structural matching. After extracting the main path features and auxiliary path features, a dual gating mechanism is introduced to calculate the gating vectors of the main path and auxiliary path respectively. This can dynamically perceive the importance of features from different sources, automatically highlight the task-dominant features and suppress redundant interference features, improve the accuracy and stability of regional transport task prediction, and provide a more reliable data foundation for capacity scheduling.

[0137] Example 5, see Figure 1 and Figure 5 This embodiment is based on the above embodiment, and the transportation demand scheduling module specifically includes the following steps:

[0138] Vehicle stratification, specifically based on the probability of a vehicle's presence in an area , and combined with the preset core attribution probability threshold and the minimum scheduling trust threshold , all vehicles are divided into the following three categories, divided into the core vehicle set , Alternative Vehicle Set and Shadow Vehicle Set ; The formula used is as follows:

[0139] ;

[0140] Calculate the regional capacity supply and demand difference to identify whether there is excess or shortage of capacity in each region, specifically based on the transport intensity value of each region and regional total available capacity , calculate the regional capacity supply and demand difference ;

[0141] Regional multi-objective capacity balance optimization is used to achieve dynamic allocation of inter-regional transport capacity at the macro level, specifically based on the regional capacity supply and demand difference , build a regional allocation relationship matrix ; The regional allocation relationship matrix Each element in represents the demand supply area Towards unmet need areas The number of vehicles to be dispatched; and a regional multi-objective optimization function is constructed. The regional multi-objective optimization function is optimized using a non-dominated sorting genetic algorithm to obtain a set of Pareto optimal solutions. From the optimal solution set, the solution that minimizes the sum of the absolute values of the supply and demand differences in all regions is preferentially selected. The solution is taken as the final regional scheduling result, and the optimized regional allocation relationship matrix is obtained. The regional multi-objective optimization function includes minimizing the total demand difference of insufficient capacity in each region , minimize the total idle mileage caused by regional allocation and minimize the number of vehicles that are below the battery threshold after allocation;

[0142] Prioritize core vehicles, which is used to prioritize the dispatch of core vehicles and improve dispatch stability. Specifically, it is based on the optimized regional dispatch relationship matrix. and gather the core vehicles in the area The vehicles in the area are assigned to the transport task pool of the area first and marked as assigned, and the remaining transport intensity value of the area is updated. If the core vehicle concentrates the comprehensive transport capacity value of the vehicle When the transport intensity value of the region is fully covered and there are some redundant core vehicles, these redundant core vehicles are included in the candidate vehicle set to obtain the regional core vehicle allocation result;

[0143] Candidate vehicle selection is used to screen vehicle resources with high scheduling priority from the alternative vehicle pool to supplement transportation capacity in areas where there is a gap in task demand. Specifically, all candidate vehicles are scored by the vehicle-level scheduling scoring function to obtain the candidate vehicle scheduling score value, and the scheduling score values of all candidate vehicles are sorted from high to low. They are allocated according to the following local priority selection strategy and cross-regional allocation selection strategy to obtain the regional candidate vehicle allocation result. The local priority selection strategy is to give priority to selecting resources from the candidate vehicles with higher scores in the local area, and supplement the transportation task demand of the local area in turn until the task of the area is fully met or the candidate vehicle pool is exhausted. The cross-regional allocation selection strategy is used only when the task demand of the local area has been met. Based on the allocation quota demand of the target area in the target area, the candidate vehicle set that can be allocated is screened from the corresponding supply area and sorted by the score value until the allocation quota of the target area is met or the candidate resources are exhausted. The formula used is as follows:

[0144] ;

[0145] Where, represents the vehicle-level scheduling scoring function, 、 and are respectively expressed as the weighted coefficients of each scoring item, Indicates the path distance from the current location of the vehicle to the target area. Indicates the penalty term brought by the vehicle's current battery status, which is used to restrict vehicles with low battery status from participating in scheduling;

[0146] Shadow vehicle backup scheduling: Specifically, in areas where there is a gap in task demand, the scheduling score of each shadow vehicle in the area is calculated according to the vehicle-level scheduling scoring function, and the scores are allocated in descending order until the task demand is met; if the demand supply area The number of vehicles to be transferred does not reach the transfer matrix If the shadow vehicle resources in this area cannot meet the required quota for all tasks, the system will mark the unassigned tasks as failed and will no longer call other regional vehicles to participate in the backup scheduling, thus obtaining the regional shadow vehicle allocation result;

[0147] The dispatching strategy is generated by integrating the regional core vehicle allocation results, regional alternative vehicle allocation results, and regional shadow vehicle allocation results to obtain the new energy shared heavy-duty truck vehicle dispatching strategy.

[0148] By implementing the above operations, this solution addresses the technical issues of rigid regional attribution, lack of steady-state compensation mechanisms, and single scheduling optimization objectives in traditional shared heavy-duty truck transportation management systems, which lead to low capacity matching accuracy. This solution innovatively introduces a hierarchical scheduling attribution mechanism based on vehicle regional existence probability. Combined with probability thresholds, it dynamically classifies vehicles into three categories, constructing a flexible soft-attribution scheduling mechanism. This significantly improves the scheduling system's ability to discern vehicle availability and the stability of task allocation under uncertainty. A multi-objective capacity allocation optimization strategy for regional supply and demand balance is proposed, and a regional-level multi-objective optimization function is constructed to generate optimal allocation relationships between regions. This addresses the problem that traditional scheduling cannot simultaneously balance supply and demand, path costs, and energy efficiency constraints. A local priority selection strategy is designed to prioritize vehicles with the highest scores within the region, reducing idle driving rates and scheduling costs, and improving the internal absorption capacity of regional capacity. This achieves flexible resource attribution, intelligent supply and demand matching, and multi-objective allocation coordination in the new energy heavy-duty truck scheduling process, effectively improving the response efficiency and operational stability of shared scheduling, and providing key support for the refined regulation of new energy shared heavy-duty truck capacity and low-carbon and efficient transportation.

[0149] Example 6, see Figure 1 This embodiment is based on the above embodiment. The intelligent transportation demand management module specifically allocates vehicle resources and issues tasks in each target area according to the new energy shared heavy-duty truck vehicle scheduling strategy, executes scheduling instructions, dynamically manages vehicle operating status, and realizes intelligent management of shared transportation capacity of new energy heavy-duty trucks.

[0150] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0151] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0152] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A digital shared heavy truck transport management system, characterized by: It includes original data acquisition module, transportation management feature acquisition module, transportation supply and demand forecasting module, transportation demand scheduling module and transportation demand intelligent management module; The raw data acquisition module specifically obtains raw data of truck transportation management through data acquisition; The transport management feature acquisition module specifically performs data optimization processing and transport management feature processing on the original data to obtain a transport feature data set; the transport management feature processing uses an improved feature clustering method using the median vector as the initial cluster center, combined with a multi-level feature selection mechanism that integrates physical distribution and statistical intensity to construct a transport management feature model, inputs the data into the model, and generates a key feature set; The transport supply and demand forecasting module specifically divides the transport task volume prediction feature set into continuous features and discontinuous features, processes the continuous features and discontinuous features separately, and obtains the main path and auxiliary path features; introduces a dual-gating mechanism to generate fusion features, constructs a transport task volume prediction model, obtains regional transport task volume prediction results, and completes single-vehicle level transport capacity prediction and vehicle regional position prediction respectively, and obtains transport supply and demand results to obtain transport supply and demand results; The transportation demand scheduling module specifically divides vehicles into three types of vehicle sets, constructs a regional allocation relationship matrix by calculating the difference between regional transportation capacity supply and demand, and designs a multi-objective optimization function, using a non-dominated sorting genetic algorithm to generate the optimized regional allocation relationship matrix; A local priority scheduling strategy is used to allocate core vehicle resources, a comprehensive scoring function is constructed for candidate vehicles and they are ranked and scheduled, a shadow vehicle backup mechanism is established, and a scheduling strategy for new energy shared heavy-duty trucks is obtained; The intelligent transportation demand management module specifically implements intelligent management of new energy shared heavy-duty trucks based on the new energy shared heavy-duty truck scheduling strategy.

2. A digital shared heavy truck transportation management system according to claim 1, characterized in that: The transport management feature acquisition module is used to perform data optimization processing and transport management feature processing on the original truck transport management data to obtain a transport feature data set; specifically, it includes the following steps: Data optimization processing, specifically including data cleaning, data standardization and data label definition, to obtain truck transportation management optimization data; The data cleaning specifically includes removing duplicate data, processing missing values, identifying outliers and checking logical consistency; The data standardization processing includes time format standardization processing, unit standardization processing and classification field numerical processing; The transportation management feature processing specifically includes the following steps: Construct a transportation management feature model using an improved feature clustering method combined with a multi-level feature selection mechanism that integrates physical distribution and statistical intensity. Obtaining a transport task volume prediction feature set, specifically by inputting transport demand data in the truck transport management optimization data into a transport management feature model to obtain a transport task volume prediction feature set; A vehicle capacity prediction feature set is obtained, specifically by inputting the single vehicle transport capacity data in the truck transport management optimization data into the transport management feature model to obtain the vehicle capacity prediction feature set.

3. A digital shared heavy truck transportation management system according to claim 1, characterized in that: The construction of the transport management feature model specifically includes the following steps: Obtain the feature sample matrix, specifically by extracting the candidate feature set from the data through statistical analysis methods and constructing the feature sample matrix , where m represents the number of samples and n represents the number of candidate features; Initial cluster center construction, specifically calculating the median of each column of the feature sample matrix D to form a median vector M as the initial cluster center, performing preliminary clustering operations, and obtaining the initial clustering results ,like If the number of cluster samples is less than the minimum cluster volume threshold t or the clustering result is empty, the feature sample matrix D is divided into positive sample group and negative sample group according to the main direction of the feature; Cluster center difference determination, specifically, calculating the mean P of the positive sample group and the mean Q of the negative sample group as candidates for alternative cluster centers; if , then choose one from P or Q as a single cluster center to generate clustering results, generate clustering results ,like , then P and Q are respectively used as cluster centers, and two sub-clusters are generated at the same time and ;in, Indicates the critical value for judging the degree of difference between P and Q; Small sample fallback clustering processing, specifically if the number of samples in any cluster in any clustering result is , then call the partial priority clustering algorithm to perform backfill processing and generate supplementary clusters ; Clustering termination condition determination: if the number of remaining unclustered samples is less than the preset lower limit threshold τ, the clustering operation is terminated to obtain the valid cluster set C; Feature cluster formation, specifically in each effective cluster For each feature j, the deviation degree of the feature value within the cluster relative to the mean of the dimension is calculated to obtain the deviation index , and according to the deviation index , sort all features, group features with similar deviations into the same group, and regard each group as a feature cluster , the formula used is as follows: ； Where, Indicates that feature j is effectively clustered The degree of deviation in Indicates effective clustering The set of sample values of the jth feature in , Indicates that feature j is effectively clustered The mean of Key feature selection involves applying the box plot method to each candidate feature in the feature cluster to identify outliers, counting the number of abnormal samples contained in each feature, and selecting the feature with the largest number of abnormal samples in each feature cluster as the representative feature of the cluster. All selected features are aggregated to form a key feature set F.

4. A digital shared heavy truck transportation management system according to claim 1, characterized in that: The transportation supply and demand forecasting module specifically includes the following steps: The transportation task volume forecast includes the following steps: Build a transportation task volume prediction model; The transport task volume prediction model training specifically comprises extracting transport demand data within a historical period to construct a training data set, and generating a historical transport task volume prediction feature set through the transport management feature acquisition module. The transport demand label of the corresponding period is used as the supervision output field to train the transport task volume prediction model to obtain a trained transport task volume prediction model; Regional transport task volume forecasting, specifically, the transport demand data of N consecutive time periods before the forecast period constitute the data set to be forecasted, and the real-time transport task volume forecast feature set is obtained through the transport management feature acquisition module, and the real-time transport task volume forecast feature set is used as the input data of the trained transport task volume forecast model to obtain the regional transport task volume forecast result, which includes the total cargo tonnage value of each region and total transport mileage ; Vehicle capacity forecasting specifically includes the following steps: A single-vehicle capacity prediction model is constructed and trained. Specifically, a long short-term memory neural network is used to construct the single-vehicle capacity prediction model. The single-vehicle transport capacity data within the historical period is used as the training dataset. The transportation management feature acquisition module generates a historical vehicle capacity prediction feature set. The vehicle capacity label of the corresponding period is used as the supervision output field to train the single-vehicle capacity prediction and obtain the trained single-vehicle capacity prediction model. Single-vehicle capacity prediction is specifically to form a dataset to be predicted using the single-vehicle transport capacity data of N consecutive time periods before the prediction period, and obtain a real-time vehicle capacity prediction feature set through the transportation management feature acquisition module, and use the real-time vehicle capacity prediction feature set as the input data of the trained single-vehicle capacity prediction model to obtain a single-vehicle capacity prediction result, which includes the load capacity of each vehicle. , Remaining mileage and service time factor ; Vehicle regional location prediction, specifically, using a vehicle location prediction model based on a long short-term memory neural network and a Softmax classification layer, using the single-vehicle transport capacity data of N consecutive time periods before the prediction period to form a vehicle regional location prediction dataset, and obtaining a real-time location prediction feature set through the transportation management feature acquisition module, and using the real-time location prediction feature set as the input data of the vehicle location prediction model to obtain the vehicle's presence probability in the area ; Obtaining the transport supply and demand results, specifically by calculating the transport intensity value, calculating the single vehicle capacity value, and summarizing the regional total capacity value, to obtain the transport supply and demand results, the transport supply and demand results including the transport intensity value, the regional total available capacity, and the vehicle comprehensive capacity value, including the following steps: Transport intensity value calculation, specifically based on the regional transport task volume forecast results, calculate each region Transport intensity value , defined as the product of total cargo tonnage and total transport mileage; Single vehicle capacity calculation, specifically the calculation of each shared new energy heavy truck based on the single vehicle capacity prediction results Comprehensive vehicle capacity value within the target time period , defined as the product of load capacity, remaining endurance and service time coefficient; The total capacity value of the region is summarized, specifically the comprehensive capacity value of the bicycle The probability of the vehicle being in the area Multiply them together to get the weighted capacity value of the vehicle in the area. , then for all the people in the target time period, in the area The predicted probability of existence The corresponding weighted capacity value Summarize and get the total available capacity of the region .

5. The digital shared heavy truck transportation management system according to claim 1 is characterized in that: The construction of the transport task volume prediction model specifically includes the following steps: The regional graph structure is constructed by first dividing the transportation space into multiple prediction areas according to the regional labels and then constructing the regional graph based on the adjacency relationship between the areas, the cargo flow path and the road connectivity. , generate the adjacency matrix A according to the connection relationship between regions; where R represents the regional node set, E represents the edge set, and the elements in the adjacency matrix are Indicates area and The connection relationship between them; Acquire multi-source features of transportation demand. Specifically, divide the transportation task volume prediction feature set into continuous feature subsets and discontinuous feature subsets based on the temporal continuity and business attributes of the features. Continuous feature processing, specifically organizing continuous feature subsets into three-dimensional input tensors by region, time step, and feature dimension, and inputting the tensors into the graph convolutional network GCN and the gated recurrent unit network GRU for joint modeling to obtain the main path features at the regional level ; Auxiliary state feature discontinuous processing, specifically constructing the discontinuous feature subset into a two-dimensional input matrix, and then inputting it into the multi-layer perceptron network structure, sequentially performing nonlinear mapping through multiple fully connected layers and GELU activation functions to obtain the auxiliary area business scenario features ; Generate fusion features, specifically obtain the main path gating vector through the gating vector generation function and auxiliary path gating vector , main path gating vector Main path feature is the input, auxiliary path gating vector Splicing vectors with main path and auxiliary path features The gating vector generation function includes two layers of fully connected neural networks, which use GELU and Sigmoid activation functions for nonlinear transformation to obtain the path gating vector; finally, the fusion feature is obtained by weighting the gating vectors of the main path and auxiliary path features and splicing them together. ; Multi-objective regional transport demand forecasting, specifically by integrating the feature vector Input to the multi-target regression output layer and output for each region and ;in, It indicates the total cargo tonnage value predicted for the area during the target time period. Indicates the predicted total transportation mileage value corresponding to the area.

6. A digital shared heavy truck transportation management system according to claim 1, characterized in that: The transportation demand scheduling module specifically includes the following steps: Vehicle stratification, specifically based on the probability of a vehicle's presence in an area , and combined with the preset core attribution probability threshold and the minimum scheduling trust threshold , all vehicles are divided into the following three categories, divided into the core vehicle set , Alternative Vehicle Set and Shadow Vehicle Set ; Calculate the regional capacity supply and demand difference, specifically based on the transport intensity value of each region and regional total available capacity , calculate the regional capacity supply and demand difference ; Regional multi-objective capacity balance optimization, specifically based on the regional capacity supply and demand gap , build a regional allocation relationship matrix ; The regional allocation relationship matrix Each element in represents the demand supply area Towards unmet need areas The number of vehicles to be dispatched; and a regional multi-objective optimization function is constructed. The regional multi-objective optimization function is optimized using a non-dominated sorting genetic algorithm to obtain a set of Pareto optimal solutions. From the optimal solution set, the solution that minimizes the sum of the absolute values of the supply and demand differences in all regions is preferentially selected. The solution is taken as the final regional scheduling result, and the optimized regional allocation relationship matrix is obtained. The regional multi-objective optimization function includes minimizing the total demand difference of insufficient capacity in each region , minimize the total idle mileage caused by regional allocation and minimize the number of vehicles that are below the battery threshold after allocation; Core vehicles are allocated first, specifically according to the optimized regional allocation relationship matrix and gather the core vehicles in the area The vehicles in the area are assigned to the transport task pool of the area first and marked as assigned, and the remaining transport intensity value of the area is updated. If the core vehicle concentrates the comprehensive transport capacity value of the vehicle When the transport intensity value of the region is fully covered and there are some redundant core vehicles, these redundant core vehicles are included in the candidate vehicle set to obtain the regional core vehicle allocation result; The candidate vehicle selection is specifically to perform scheduling scoring on all candidate vehicles through the vehicle-level scheduling scoring function to obtain the candidate vehicle scheduling score value, sort the scheduling score values of all candidate vehicles from high to low, and allocate them according to the following local priority selection strategy and cross-regional allocation selection strategy to obtain the regional candidate vehicle allocation result. The local priority selection strategy is specifically to give priority to selecting resources from the candidate vehicles with higher scores in the local area, and supplement the transportation task requirements of the local area in turn until the task of the area is fully met or the candidate vehicle pool is exhausted; the cross-regional allocation selection strategy is only when the task requirements of the local area have been met, according to Based on the allocation quota demand of the target area in the target area, the candidate vehicle set that can be allocated is screened from the corresponding supply area and sorted by the score value until the allocation quota of the target area is met or the candidate resources are exhausted. The formula used is as follows: ； Where, represents the vehicle-level scheduling scoring function, 、 and are respectively expressed as the weighted coefficients of each scoring item, Indicates the path distance from the current location of the vehicle to the target area. Indicates the penalty term brought by the vehicle's current battery status, which is used to restrict vehicles with low battery status from participating in scheduling; Shadow vehicle backup scheduling: Specifically, in areas where there is a gap in task demand, the scheduling score of each shadow vehicle in the area is calculated according to the vehicle-level scheduling scoring function, and the scores are allocated in descending order until the task demand is met; if the demand supply area The number of vehicles that should be dispatched has not been reached If the shadow vehicle resources in this area cannot meet the required quota for all tasks, the system will mark the unassigned tasks as failed and will no longer call other regional vehicles to participate in the backup scheduling, thus obtaining the regional shadow vehicle allocation result; The dispatching strategy is generated by integrating the regional core vehicle allocation results, regional alternative vehicle allocation results, and regional shadow vehicle allocation results to obtain the new energy shared heavy-duty truck vehicle dispatching strategy.

7. The digital shared heavy truck transportation management system according to claim 1, characterized in that: The intelligent transportation demand management module specifically allocates vehicle resources and issues tasks in each target area according to the new energy shared heavy-duty truck vehicle scheduling strategy, executes scheduling instructions, dynamically manages vehicle operating status, and realizes intelligent management of shared transportation capacity of new energy heavy-duty trucks.

8. The digital shared heavy truck transportation management system according to claim 1, characterized in that: The original data acquisition module specifically collects the original data required for truck transportation management through the truck transportation management platform to obtain truck transportation management original data; the truck transportation management original data includes transportation demand data and single-vehicle transportation capacity data; the transportation demand data includes transportation order data, operation demand spatiotemporal dimension data, external influencing factor data and task statistics data; the single-vehicle transportation capacity data includes vehicle status data, vehicle energy consumption data, vehicle transportation order data and single-vehicle transportation capacity statistics.

Images