Transportation intelligent assignment scheduling method and system

Abstract

The invention provides an intelligent assignment scheduling method and system for transportation, and relates to the field of logistics management, and the method comprises the steps: obtaining order information and transport capacity resource information, the order information comprises cargo attributes, the own transport capacity resource information comprises vehicle shipping space parameters, and the external transport capacity resource information comprises simplified load parameters; and processing the internal image of the vehicle cabin by using a wavelet transform method so as to determine three-dimensional parameters of the abnormal area and generate accurate vehicle cabin space parameters. Based on the order information and the accurate vehicle shipping space parameters, vehicle and route allocation is carried out through a multi-target scheduling algorithm, and candidate scheduling schemes are generated; and executing three-dimensional loading feasibility verification by utilizing a multi-grid construction module so as to predict the deformation condition of the goods and the vehicle shipping space and mark the ghost volume. And finally outputting an optimal scheduling scheme in response to verification passing. According to the method, the refinement level of logistics transportation management is effectively improved, space conflicts and overload risks are avoided, and the transport capacity resource utilization efficiency and the transportation safety are improved.

Description

Technical Field

[0001] The present disclosure relates to the field of logistics management, and in particular, to a method and system for intelligent transportation dispatching. Background Art

[0002] In the field of modern logistics and transportation, with the rapid growth of the number of orders and transportation demand, how to efficiently use transportation resources to achieve intelligent scheduling and refined management of orders has become an important issue that enterprises need to solve urgently. At present, the logistics and transportation industry is gradually transforming and upgrading from the traditional extensive management mode to the direction of digitalization, intelligence and refinement. However, in the actual operation process, the traditional transportation scheduling method is often still dominated by manual experience or simple rules for vehicle assignment and route planning, usually only based on relatively simple and single parameters such as vehicle load, volume or basic route requirements, and lacks more comprehensive and accurate consideration. However, this rough matching mode has many disadvantages: first, it is impossible to effectively identify the spatial adaptability between the cargo and the vehicle compartment, which is easy to cause waste or insufficient utilization of vehicle space, directly reducing transportation efficiency; second, during the transportation process, due to the lack of refined loading planning, especially in the transportation scenarios involving special cargo such as dangerous goods, cold chain cargo, fragile goods and precision instruments and equipment, it is more likely to cause spatial conflicts, cargo damage, overloading or violation of relevant safety compliance standards, which greatly increases transportation risks and costs.

[0003] In addition, with the increasing competition in the supply chain market, customers are increasingly demanding the timeliness, stability and safety of logistics services. Simply relying on traditional extensive scheduling methods can no longer meet the diversified, personalized and refined market needs. Especially in complex application scenarios such as transportation route optimization, precise matching of goods and vehicles, and multi-vehicle collaborative operations, the shortcomings of existing technical means are more prominent. Therefore, there is an urgent need for a more efficient, accurate and intelligent scheduling management method to achieve refined control of the entire process of logistics transportation, so as to meet the current actual needs of the development of the logistics industry. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present application provides a method and system for intelligent transportation dispatching.

[0005] In a first aspect, the present application provides a method for intelligent transportation dispatching, comprising: Acquire order information and transport resource information, wherein the order information includes: cargo attributes; the transport resource information includes: vehicle space parameters of self-owned transport capacity and simplified load parameters of external transport capacity; the vehicle space parameters include: an image of the interior of the vehicle cabin; The vehicle cabin parameters of the self-owned transport capacity are processed by wavelet transform method to determine the three-dimensional parameters of the abnormal area; based on the three-dimensional parameters of the abnormal area and the vehicle cabin parameters, the accurate vehicle cabin parameters of the self-owned transport capacity are generated.

[0006] Based on the order information, the precise vehicle space parameters of the self-owned transport capacity, and the simplified load parameters of the external transport capacity, a multi-objective scheduling algorithm is used to allocate vehicles and routes to the pre-processed order set to generate candidate scheduling solutions; Using a preset multi-grid construction module, performing a three-dimensional loading feasibility check on the candidate scheduling scheme to generate a feasibility check result; in response to the feasibility check failing, falling back to the multi-objective scheduling algorithm to correct the scheduling scheme; In response to the feasibility checks all being passed, outputting a final scheduling plan; The three-dimensional loading feasibility check includes: determining predicted deformations for cargo and vehicle spaces based on the order information and the transport resource information, and mapping the predicted deformations into ghost volumes in the multi-grid construction module to mark them in coordinate units corresponding to different grid layers; In response to the fact that the overall size of the cargo and the coordinate units of the ghost volume do not overlap, a fit comparison is performed on the local surface of the cargo and the local units of the ghost volume at the grid layer. In response to the existence of a fit conflict, the loading is marked as infeasible and falls back to the multi-objective scheduling algorithm. In response to the absence of a fit conflict, the feasibility check is output as passed.

[0007] In a second aspect, the present application provides a transportation intelligent dispatching and scheduling system, comprising: The receiving module is used to obtain order information and transport resource information, wherein the order information includes: cargo attributes; the transport resource information includes: vehicle space parameters of self-owned transport capacity and simplified load parameters of external transport capacity; the vehicle space parameters include: an image of the interior of the vehicle cabin; The receiving module is also used to process the interior image of the cabin using the wavelet transform method for the vehicle cabin parameters of the self-owned transport capacity to determine the three-dimensional parameters of the abnormal area; based on the three-dimensional parameters of the abnormal area and the vehicle cabin parameters, generate accurate vehicle cabin parameters of the self-owned transport capacity.

[0008] A scheduling module, for allocating vehicles and routes to the pre-processed order set using a multi-objective scheduling algorithm based on the order information, the precise vehicle space parameters of the self-owned transport capacity, and the simplified load parameters of the external transport capacity, and generating candidate scheduling solutions; A detection module, configured to perform a three-dimensional loading feasibility check on the candidate scheduling scheme using a preset multi-grid construction module, and generate a feasibility check result; in response to failure of the feasibility check, fall back to the multi-objective scheduling algorithm to correct the scheduling scheme; An output module, configured to output a final scheduling plan in response to the feasibility checks being passed; The three-dimensional loading feasibility check includes: determining predicted deformations for cargo and vehicle spaces based on the order information and the transport resource information, and mapping the predicted deformations into ghost volumes in the multi-grid construction module to mark them in coordinate units corresponding to different grid layers; In response to the fact that the overall size of the cargo and the coordinate units of the ghost volume do not overlap, a fit comparison is performed on the local surface of the cargo and the local units of the ghost volume at the grid layer. In response to the existence of a fit conflict, the loading is marked as infeasible and falls back to the multi-objective scheduling algorithm. In response to the absence of a fit conflict, the feasibility check is output as passed.

[0009] Compared with the prior art, the present invention uses the wavelet transform method to process the internal image of the vehicle, effectively identifies and extracts the three-dimensional parameters of the abnormal area of ​​the cabin, thereby generating accurate vehicle cabin parameters, significantly improving the accuracy of the spatial matching between the cargo and the vehicle, and avoiding the space waste or mismatch problem caused by the fuzzy parameters of the traditional method. Secondly, the present invention adopts a multi-objective scheduling algorithm to perform refined vehicle and route allocation on the pre-processed order set, and cooperates with the three-dimensional loading feasibility verification mechanism to effectively ensure the feasibility and compliance of the scheduling scheme. In particular, when transporting dangerous goods, cold chain goods and other special goods, by predicting the deformation between the goods and the vehicle cabin and constructing the ghost volume, accurate local surface fit verification is achieved in multiple grids, which greatly reduces the risk of spatial conflict, overloading and cargo damage during transportation. Thirdly, the present invention realizes efficient coordination between self-owned transportation capacity and external transportation capacity, and can perform optimal matching and scheduling according to the refined parameters of different types of transportation capacity, significantly improving the overall utilization efficiency of enterprise transportation resources and the safety and stability of transportation. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A flow chart of a method for intelligent transportation dispatching provided in an embodiment of the present application; Figure 2 A transport resource information structure diagram provided for an embodiment of the present application; Figure 3 A schematic diagram of the effect of an abnormal area on the placement of goods provided in an embodiment of the present application; Figure 4 A schematic diagram of a transportation intelligent assignment and scheduling system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0011] The technical solutions in the embodiments of the present invention will be described clearly and completely 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.

[0012] Research has found that the existing common order scheduling methods generally adopt manual or semi-automatic processing methods, mainly based on simple parameters of the vehicle (such as load, volume, etc.) to roughly match the goods, usually lacking refined management and unified coordination of its own and external transportation capacity.

[0013] Therefore, considering the impact of self-owned transport capacity and external transport capacity on order scheduling, this embodiment provides a refined transportation intelligent assignment and scheduling method, which can achieve efficient and automated order processing through the following settings: Automatic assignment settings: This embodiment automatically determines the matching degree of the transportation task according to the contract agreement by setting condition parameters of multiple dimensions such as transportation routes, provinces, cities, vehicle types, order types, and cargo types. When the order task meets the preset conditions, the system can automatically assign the task to the corresponding external carrier, realizing automatic review of the order and efficient and accurate automatic assignment.

[0014] Automatic dispatch settings: This embodiment takes into account the characteristics and advantages of self-owned transport capacity and provides an automatic dispatch solution for self-owned vehicles. Specifically, it includes automatic splitting of orders (for example, splitting orders according to cargo category, quantity, number of boxes or pallets) and merging of small orders. Through the configuration of parameters such as transportation routes, provinces, cities, vehicle types, order types and cargo types, the system automatically assigns tasks and plans routes for self-owned vehicles and drivers when conditions are met, thereby achieving the purpose of automatic review and efficient automatic dispatch.

[0015] Automatic dispatch and scheduling settings: For scenarios that require coordination with external carriers and their vehicles, this embodiment also considers factors such as transportation routes, provinces, cities, vehicle types, order types, and cargo types, and sets up more accurate automatic dispatch and scheduling functions. When a transportation task meets specific conditions, the system can not only automatically assign the task to the designated external carrier, but also directly dispatch the transportation task to the external vehicle and driver designated by the carrier on behalf of the carrier, thereby achieving comprehensive control over external transportation capacity and ensuring a high degree of adaptability and efficient use of external transportation capacity and tasks.

[0016] See also Figure 1 FIG. 1 is a flow chart of a method for intelligent transportation dispatching provided by an embodiment of the present application, wherein the method comprises steps S101 to S104, wherein: S101: Acquire order information and transport resource information, wherein the order information includes: cargo attributes; the transport resource information includes: vehicle space parameters of self-owned transport capacity and simplified load parameters of external transport capacity; the vehicle space parameters include: an image of the interior of the cabin; for the vehicle space parameters of self-owned transport capacity, process the image of the interior of the cabin using the wavelet transform method to determine the three-dimensional parameters of the abnormal area; based on the three-dimensional parameters of the abnormal area and the vehicle space parameters, generate accurate vehicle space parameters of self-owned transport capacity; S102: Based on the order information, the precise vehicle space parameters of the self-owned transport capacity, and the simplified load parameters of the external transport capacity, a multi-objective scheduling algorithm is used to allocate vehicles and routes to the pre-processed order set, and a candidate scheduling solution is generated; S103: using a preset multi-grid construction module, performing a three-dimensional loading feasibility check on the candidate scheduling scheme to generate a feasibility check result; in response to failure of the feasibility check, returning to the multi-objective scheduling algorithm to modify the scheduling scheme; S104: In response to the feasibility checks being passed, outputting a final scheduling plan.

[0017] In large-scale, multi-type transportation scenarios, this application obtains basic data including cargo attributes and transportation resource information, uses a multi-objective scheduling algorithm to allocate vehicles and routes for order execution, and subsequently combines the precise vehicle space parameters of the company's own transportation capacity and the simplified load parameters of the external transportation capacity to perform a three-dimensional loading feasibility check, so as to achieve automated and accurate transportation assignment for various types of goods including dangerous goods or cold chain.

[0018] For each order, this application first generates candidate dispatch plans by comprehensively considering factors such as freight cost, transportation time, vehicle empty driving rate, and special cargo compliance requirements, and then verifies whether the cargo can be loaded or docked in the designated vehicle's cabin at a fine-grained three-dimensional loading level. If the verification result is feasible, the dispatch plan is directly output; if the verification fails, it is returned to the algorithm end for plan correction, so as to continuously iterate until the optimal assignment that takes into account efficiency and safety compliance is obtained.

[0019] Regarding S101 above: In the specific implementation, the order information and transportation resource information are first obtained, wherein the order information may include but is not limited to: cargo attributes, shipping location, receiving location and time requirements.

[0020] Among them, cargo attributes may include descriptions of size, weight, temperature control requirements or dangerous goods attributes; time requirements indicate when the cargo needs to be shipped or delivered.

[0021] See also Figure 2 , Figure 2A transport resource information structure diagram provided for an embodiment of the present application, wherein the transport resource information is divided into two categories: self-owned transport capacity and external transport capacity, such as self-owned transport capacity A, B, C, etc., and external transport capacity a, b, c, etc. The precise vehicle compartment parameters of self-owned transport capacity include the three-dimensional space information in the cabin, the compartment structure information, and the protection configuration information, which are used to perform a more accurate loading inspection on the interior of the vehicle.

[0022] The so-called three-dimensional spatial information within the cabin refers to data such as the length, width, height, and even the concave and convex structure and compartment layout inside the vehicle; the compartment structure information can distinguish between the cold chain area, ordinary cabin area or explosion-proof cabin area, etc.; the protection configuration information is used to indicate whether the vehicle has additional dangerous goods transportation capabilities, explosion-proof measures or temperature control functions.

[0023] The simplified load parameters of external transportation capacity include the maximum carryable weight, the maximum carryable volume and transportation qualification information, which are usually provided by external carriers to quickly determine whether the goods are overloaded or whether there are dangerous goods or insufficient cold chain qualification matching.

[0024] This application divides the transport capacity information into precise vehicle space parameters and simplified load parameters. In the subsequent loading process, the system can flexibly connect to different types of vehicles and be compatible with heterogeneous transport capacity environments.

[0025] After completing the acquisition of the above information, the order set can be pre-processed, such as merging the same shipping location and receiving location, splitting large orders, or checking missing data to form a pre-processed order set that is more suitable for algorithm processing.

[0026] Regarding S102 above: A multi-objective scheduling algorithm is used to allocate vehicles and routes to the order set and generate several candidate scheduling solutions.

[0027] As an optional implementation, the multi-objective scheduling algorithm includes: Establishing an objective function in the multi-objective scheduling algorithm; wherein the optimization indicators in the objective function include: freight cost, transportation time efficiency, vehicle empty driving rate and special cargo compliance requirements; A hard constraint rule is set in the multi-objective scheduling algorithm; the hard constraint rule includes: In response to the fact that the cargo attributes include dangerous goods, it can only be assigned to the self-owned transport capacity with corresponding protection configuration or external transport capacity with dangerous goods transportation qualification; in response to the fact that the cargo requires cold chain, it can only be assigned to the transport capacity with corresponding refrigeration function; Soft constraint rules are set in the multi-objective scheduling algorithm; the soft constraint rules include: In response to the size or weight of the ordered cargo exceeding the precise vehicle space parameters of the owned transport capacity or the simplified load parameters of the external transport capacity, a penalty value is added to the assignment relationship in the objective function to reduce the fitness of the assignment relationship in the iterative search.

[0028] As an optional implementation manner, generating a candidate scheduling solution includes: Performing iterative solving on the preprocessed order set to generate multiple candidate solutions, each candidate solution including an assignment relationship for own transportation capacity and / or external transportation capacity, a driving route, and an estimated arrival time; In response to detecting an allocation combination that does not satisfy the constraint rule during the scheduling iteration process, the candidate solution is eliminated or mutated during the algorithm iteration until a candidate scheduling solution that satisfies the objective function requirement is output.

[0029] In specific implementation, the multi-objective scheduling algorithm conducts comprehensive evaluation and iterative search on multiple indicators such as freight cost, transportation time, vehicle empty driving rate, and compliance requirements for special goods. If it is dangerous goods, the algorithm will give priority to assigning it to vehicles with corresponding explosion-proof space or dangerous goods transportation qualifications. If it is cold chain goods, it will be assigned to vehicles with corresponding refrigeration functions. If the size or weight of the goods far exceeds the basic load or volume limit of the vehicle, a penalty value will be imposed on the assignment in the objective function, so that the combination will be gradually eliminated in the iteration. In this way, the multi-objective allocation of vehicles and routes can be completed at a large level, so that the scheduling layer can provide candidate solutions in a relatively short time.

[0030] Regarding S103 and S104 above: When the candidate scheduling plan is generated, the system will perform a three-dimensional loading feasibility check on it to verify whether each vehicle can accommodate the assigned goods according to the previous assignment. If the vehicle is self-owned and has accurate vehicle compartment parameters, the three-dimensional stacking or collision detection method can be called to combine the three-dimensional information in the cabin, the compartment structure and the protection configuration information to simulate the length, width and height of the goods or more fine-grained shape data to determine whether the vehicle has enough space, corresponding compartment function and protection level to accommodate the goods.

[0031] If the vehicle is an external transport capacity, it can only make a simple compliance judgment based on the maximum loadable weight, volume and transportation qualification information. At this time, if the system finds that the goods are incompatible with the vehicle space, the dangerous goods or cold chain requirements cannot be met, the space is indeed insufficient, or the size is seriously mismatched during the refined verification, it will generate a result of loading verification failure. The result will be immediately fed back to the multi-objective scheduling algorithm, so that it can update the assignment plan or replace the vehicle in the next round of iteration; if the loading verification passes, the candidate scheduling plan will be marked as feasible. Finally, after all goods and vehicles are successfully matched at the three-dimensional loading level, the system will output the final scheduling plan, which not only ensures the achievement of the macro scheduling goals, but also minimizes the number of rollbacks at the loading level.

[0032] For example, in a typical intelligent transportation dispatching and scheduling system, the software platform usually consists of an order management module, a transportation capacity management module, a multi-objective scheduling module, a three-dimensional loading verification module, and a data storage and interaction interface.

[0033] First, the order management module receives order data pushed by external systems (such as enterprise ERP or online ordering platforms) and stores it in the order table in the database. Each order record contains at least the following fields: Cargo attributes: including length, width and height data or volume / weight marking, dangerous goods identification (if any), temperature range requirements (if cold chain), etc. Place of shipment and place of delivery: can be marked as source and destination coordinates in the geocoding library; Time requirements: such as the latest departure time, the earliest arrival time window, etc., are used to generate feasible time period allocation in the subsequent scheduling algorithm.

[0034] Then, the capacity management module provides capacity resource information for owned vehicles and external vehicles: For self-owned vehicles, accurate vehicle cabin parameters are retained in the "self-owned transport capacity table" of the database, namely the three-dimensional space information in the cabin (such as length, width, height and local concave and convex description), compartment structure information (represented by partition ID or partition coordinates) and protection configuration information (cold chain unit model, explosion-proof level, sensor, etc.); For external vehicles, only the maximum carryable weight, maximum carryable volume and transportation qualification fields (marking whether cold chain, dangerous goods, etc. can be transported) are stored in the database's "external transport capacity table" in a simplified form for subsequent scheduling calls.

[0035] After completing the order and capacity information acquisition, the system enters the multi-objective scheduling module to perform pre-scheduling pre-processing. This may include: Order merging or splitting: If multiple orders have similar shipping locations, delivery locations or time windows, they will be merged into larger units; if a single large order exceeds the load / volume of any vehicle, it will be split.

[0036] Basic verification: For example, remove expired or unavailable vehicles from the external carrier's available time table and vehicle table, and update the remaining dispatchable resources.

[0037] After the preprocessed order set is generated, the multi-objective scheduling module starts to perform vehicle and route allocation. This module includes the following key steps: For the definition of the objective function, the weights of freight cost, transportation time, vehicle empty rate and special cargo compliance requirements can be recorded in an extensible data structure; For constraint settings, you can enter hard rules (such as dangerous goods can only be assigned to vehicles with protective configurations or qualifications) and soft constraint rules (such as imposing a penalty value when the weight or volume of the goods exceeds the vehicle limit); For iterative solution, several candidate solutions can be initialized, each candidate solution describes the "order ID The algorithm then performs iterative search using genetic algorithms, ant colony algorithms, etc., calculates the comprehensive fitness (including freight, timeliness, empty driving rate, compliance, etc.) of the candidate solutions for each iteration, and filters or punishes unreasonable assignments based on hard / soft constraints. After multiple iterations, the algorithm outputs a batch of suitable candidate scheduling solutions.

[0038] After generating candidate dispatch plans, the system passes these plans to the 3D loading verification module for feasibility verification. In this module, the "order-vehicle allocation" in each candidate plan is verified as follows: If an external vehicle is assigned, the feasibility is determined only based on its maximum load-bearing weight, volume, and transportation qualifications. If the limit is exceeded or there is no corresponding qualification, the plan is marked as infeasible and returned to the scheduling module for correction. If a self-owned vehicle is assigned, the 3D information (such as grid or point cloud form) stored in the data structure, the compartment structure, and the protection configuration of the self-owned vehicle cabin are called, combined with the length, width, and height or 3D point cloud of the goods in the order, to perform specific placement or collision detection: First, check the compartment identification: if it is cold chain goods, match it to the cold chain compartment area; if it is dangerous goods, match it to the explosion-proof compartment area; Then, a three-dimensional stacking simulation is performed in the corresponding cabin area to determine whether the cargo can occupy a certain section of coordinate space in the cabin area without collision. If sufficient space cannot be found or the cabin shape seriously conflicts with the cargo shape, the loading check is marked as failed.

[0039] The system reads the above loading verification results in real time: if the verification fails, the vehicle assignment relationship will be marked as "invalid" and returned to the scheduling module for rollback or update; if the verification passes, other assignments will continue to be verified in the same way until all orders pass the loading verification. Finally, if each vehicle-cargo assignment in the candidate solution meets the loading requirements, the system will output the scheduling solution as the final result. During this process, for self-owned vehicles that are prone to cabin modifications or additional accessories, the platform can periodically perform lidar scanning to update its cabin three-dimensional data to avoid misjudgment due to cabin deformation or installation of accessories.

[0040] Through this hierarchical method of scheduling first, then loading and checking, backing off and correcting if the check fails, and outputting the solution if the check passes, this implementation method can effectively take into account the global optimization requirements of large-scale scheduling algorithms for timeliness and vehicle empty driving rate, and can also match the refined demands of goods for three-dimensional space, protection requirements and temperature control requirements in micro-loading details. In particular, for the precise vehicle compartment parameters of self-owned transportation capacity, a higher-precision three-dimensional simulation can be used to ensure accurate placement judgment of dangerous goods, cold chain or oversized goods; for external transportation capacity, only the maximum load and volume are used for simplified processing to accommodate the situation where social carriers lack transportation capacity information. In this way, a complete closed loop is formed from multi-objective scheduling to three-dimensional loading, and then to backing off and correcting and outputting results. This method realizes transportation assignment scheduling that takes into account efficiency, cost and safety and compliance requirements when facing complex orders and multiple transportation capacities, reducing frequent failures and manual work burdens caused by mismatching or ignoring the compartment structure.

[0041] See also Figure 3 , Figure 3 A schematic diagram of the effect of an abnormal area on the placement of goods provided in an embodiment of the present application. Figure 3 (a) is the cabin space without abnormal areas. Figure 3 (b) is the cargo hold space with abnormal area. Figure 3 In (b), the abnormal area has a significant impact on the placement of the goods. Therefore, as an optional implementation method, obtaining the precise vehicle space parameters of the self-owned transport capacity includes: Acquire an image of the interior of the vehicle cabin including key areas; wherein the key areas include: walls, floors, corners and door frames of the vehicle cabin; Performing 2- to 3-layer 2D discrete wavelet transform on the cabin interior image to decompose low-frequency sub-bands and high-frequency sub-bands; The low-frequency sub-band is used to characterize the overall structure and background information of the interior image of the vehicle cabin; the high-frequency sub-band is used to characterize the edge, texture and detail information of the interior image of the vehicle cabin; The amplitude of the wavelet coefficients is calculated in the high frequency sub-band and an adaptive threshold process is applied to obtain abnormal features.

[0042] The abnormal features are mapped back to the original image space using inverse wavelet transform, and adjacent abnormal points are merged and isolated noise points are removed using morphological closing operation to obtain abnormal area location information; Use the pre-trained deep learning model to predict the depth information of the abnormal area from a single image, generate a depth map, and convert the depth map into 3D parameters of the abnormal area by combining the camera internal parameters; Based on the three-dimensional parameters of the abnormal area, the volume of the cabin space occupied by the abnormal area is calculated, and the volume is deducted from the original design size of the cabin to generate accurate vehicle space parameters of the self-owned transport capacity.

[0043] This embodiment aims to detect and calculate the volume of abnormal areas through image processing and deep learning methods in response to possible deformation or addition of accessories inside the cabin after repeated use, thereby providing more real-time and accurate technical support for obtaining precise vehicle cabin parameters of owned transportation capacity in the above S101.

[0044] The basic principle is: first use two-dimensional discrete wavelet transform to decompose the interior image of the key area of ​​the cabin into low-frequency sub-bands and high-frequency sub-bands, then perform adaptive threshold processing on the wavelet coefficients with larger amplitudes in the high-frequency sub-bands to extract the "abnormal features" reflecting bumps or protrusions of accessories, and then predict the three-dimensional parameters of the anomaly through morphology and depth estimation, and deduct the abnormal volume from the original cabin design size to generate accurate vehicle cabin parameters.

[0045] In practice, several key areas can be identified based on vehicle manufacturers or internal maintenance records, such as the walls, floors, corners, and door frames of the cabin, which are often more prone to local bulges or depressions due to impacts, loading of special goods, or installation of cold chain units, explosion-proof panels, etc. First, obtain the internal image of each key area. The image can be taken by a multi-angle camera or regularly obtained by the cabin internal monitoring system to ensure that the latest cabin changes are captured. Perform a two-dimensional discrete wavelet transform on the obtained cabin interior image, and the number of decomposition layers can be selected to be two to three layers.

[0046] After the decomposition is completed, low-frequency sub-bands and high-frequency sub-bands can be obtained. The low-frequency sub-bands here are used to characterize the overall structure and background information of the cabin, retaining the regional brightness or chromaticity distribution to reflect the main outline and main plane of the cabin wall or floor; the high-frequency sub-bands are used to reveal the details, texture and edge information of the image, and can amplify the characteristics of certain local anomalies (such as protrusions, cracks or attachment seams).

[0047] In the high-frequency subband, the amplitude of the wavelet coefficient is calculated to characterize the degree of local change, and the coefficients whose amplitude exceeds a certain range in several segmented thresholds are subjected to adaptive threshold processing, so as to distinguish random noise and weak texture from real obvious bulges or depressions. If the high-frequency coefficient only slightly exceeds the threshold, it is regarded as a general texture, and if the high-frequency coefficient far exceeds the threshold, it is marked as an "abnormal feature". These marked high-frequency coefficients are then mapped back to the original image space during the inverse wavelet transform to determine the distribution of potential abnormal areas in pixel coordinates.

[0048] Typically, these high-frequency anomalies may be caused by newly installed brackets in the cabin, protruding parts of cold chain units, or other sunken floors. In order to remove isolated noise points that may still remain and merge adjacent areas, a morphological closing operation can be performed on the abnormal area obtained by the inverse transformation to fill small gaps, merge adjacent pixel clusters, and finally obtain a connected abnormal area segmentation result.

[0049] After obtaining the two-dimensional segmentation of the abnormal area, in order to further quantify its three-dimensional occupied volume, it is necessary to use the mapping from image to depth information.

[0050] In this embodiment, a deep learning model pre-trained on a vehicle or indoor scene can be used to estimate the depth of the abnormal area. The model combines the texture cues, geometric priors and camera intrinsic parameters (including focal length and distortion coefficient, etc.) of a single image to predict the depth map of each pixel.

[0051] Then, by traversing and 3D projecting the depth map in the abnormal area, the point cloud or 3D contour of the abnormal area in the 3D coordinate system is obtained, and the abnormal volume is calculated using voxel accumulation or polygon construction. Because the model is trained for indoor or vehicle cabin environments, it can better identify structures such as cabin supports and metal accessories than ordinary general models. The 3D parameters of these abnormal areas represent the newly appeared or increased or decreased protrusions and local space occupancy in the cabin.

[0052] The volume of the abnormal area is deducted from the original design size of the cabin to update the actual remaining available space in the cabin. If there are multiple abnormalities in several key areas, the process is repeated to calculate the volume of each abnormal block and merge or superimpose and deduct them, and finally generate three-dimensional cabin information that is closer to reality, thereby forming the so-called "precise vehicle cabin parameters of self-owned capacity" in this implementation method.

[0053] This parameter will be used in subsequent three-dimensional stacking or collision detection of goods to determine whether the goods can be placed in the actual available space, rather than relying on outdated original CAD or manually filled load data. This significantly improves the ability to automatically identify and update the available space volume after deformation or attachment inside the cabin, providing more reliable basic space data for intelligent transportation assignment and scheduling.

[0054] In this way, the present application addresses the technical problem that "abnormal areas" such as unknown protrusions, depressions, and additional supports are easily formed during multiple uses or additional assembly of the vehicle and are difficult to be accurately identified by traditional measurement methods. By acquiring images of the interior of the cabin, extracting high-frequency sub-bands, and adaptively thresholding to detect significant protrusions or depressions, and then using a deep learning model for three-dimensional reconstruction, this application can quickly and automatically deduct these additional or deformed volumes from the original design dimensions of the vehicle, thereby generating more accurate vehicle cabin parameters that are closer to reality.

[0055] On the one hand, traditional manual measurement or static CAD models often cannot update the temporary brackets, cold chain unit protrusions or sunken areas that may appear inside the vehicle in real time; on the other hand, it is difficult to distinguish real protrusions from noise or texture patterns by simply filtering the image. However, this application uses multi-level denoising and feature point extraction of two-dimensional discrete wavelet transforms plus morphological closing operations to accurately locate the parts that are indeed protrusions or depressions in the high-frequency sub-band, and convert them into quantifiable three-dimensional volumes through depth estimation, ultimately making the update of cabin data more dynamic and accurate.

[0056] If other means are tried to solve similar "temporary or dynamic attachments in the vehicle cabin" identification problems, common methods include relying on manual field measurements, based on fixed laser ranging scripts, or simply using simple Gaussian filtering to remove texture noise, but these methods are difficult to distinguish between real attachments and random noise, and often lack the ability to automatically estimate the depth of local protrusions. In contrast, the present application focuses on obvious mutation points in high-frequency sub-bands, excludes non-critical texture noise through adaptive thresholds, and uses morphology and pre-trained deep model predictions to accurately quantify the three-dimensional position and volume of abnormal areas, which can effectively ignore minor interference and retain the main protruding components that affect safety and load, giving full play to the dual advantages of wavelet multi-scale denoising and deep learning scene recognition. This not only reduces the reliance on manual measurement, but also avoids the limitations of simple thresholds or static CAD models, and has significant real-time and accuracy improvements in transportation environments with high data update frequencies.

[0057] In addition, compared to directly using a pure prediction model (for example, end-to-end image semantic segmentation or depth estimation network) to detect and quantify abnormal areas such as bulges and depressions inside the cabin, the present application first introduces an adaptive threshold in the high-frequency sub-band to locate abnormal features with significant amplitudes, and removes a large amount of irrelevant textures in the wavelet domain, which can make the input of the subsequent prediction model "purer" and reduce the dependence on deep models and the demanding requirements on training scale.

[0058] Secondly, when directly relying on the end-to-end network to process cabin images, if the cabin has diverse materials, reflections, or new accessories, without sufficient training samples, the model may treat certain key protrusions as background or noise, or judge normal cabin walls with detailed textures as abnormal. By performing adaptive threshold detection in the wavelet domain, high-frequency responses that actually exceed the threshold are marked as "suspicious protrusions or depressions", and then using morphological operations to merge connected areas, the subsequent deep model only needs to focus on the real suspected areas, thereby significantly reducing the end-to-end network's reliance on large-scale and diverse scenes.

[0059] Finally, in business scenarios, when the transportation environment, cabin lighting or image perspective changes frequently, a single prediction model often requires continuous iterative tuning or incremental training; while the solution that combines wavelet analysis and morphological methods can filter out most of the external lighting and low-level texture interference in a more robust manner, maintaining a stronger robust detection capability for "significant deformation".

[0060] For cabin scenes that require real-time updates or frequent scanning, this method can achieve a high anomaly detection rate without relying heavily on manually annotated databases, and only calls deep predictions in suspected areas, greatly reducing deployment and maintenance costs. In contrast, if you rely entirely on a single deep learning model, once new accessory forms or rare concave-convex structures appear inside the cabin, the model accuracy will drop significantly if there is a lack of supporting data for iterative training.

[0061] For example, after a self-owned transport vehicle has been used for several months and modified several times, an additional cold chain unit and some metal brackets have been installed inside, resulting in changes in the height or shape of some areas inside the cabin, and a large error between the original CAD data and the actual situation. In order to maintain accurate control of the space inside the vehicle cabin, the system performs the following operations to update the precise vehicle cabin parameters of the self-owned transport capacity: When the vehicle is returned for maintenance, the maintenance staff installs a movable camera assembly in key areas of the cabin (including bulkheads, floors, corners, and door frames) and obtains multiple images of the cabin interior for each key area. Since the camera has multi-angle shooting capabilities, it can cover most of the protruding parts due to metal brackets or cold chain units, and ensure that the image can capture bulkheads that are easily deformed or vulnerable to impact.

[0062] For each acquired interior image of the vehicle cabin, the system first performs a two-dimensional discrete wavelet transform, and the number of decomposition layers is set to three to obtain the corresponding low-frequency sub-bands and three layers of high-frequency sub-bands.

[0063] In terms of implementation, the software module uses the Daubechies series of wavelet bases, which can better balance image detail capture and noise robustness. The low-frequency sub-bands obtained by decomposition mainly retain the brightness gradient of large-area structures in the cabin (such as the bulkhead plane and the main floor plane), while the high-frequency sub-bands highlight texture details such as the edges of metal brackets, the connections of cold chain units, and the dents caused by certain impacts.

[0064] The system then scans the coefficient amplitudes of all high-frequency subbands, and excludes extremely small coefficients as texture noise based on the set multi-segment adaptive thresholds, marks high-frequency coefficients exceeding the first threshold as general protrusions, and marks high-frequency coefficients exceeding the second threshold as significant abnormal feature points. For example, the high-frequency coefficient amplitudes of the joints between metal brackets and cold chain units are large, so they are marked as "significantly abnormal" at this stage. During the inverse wavelet transform, these marked high-frequency coefficients will be mapped back to the coordinates of the original image, allowing the algorithm to determine the distribution of abnormal areas in the pixel domain.

[0065] In order to remove isolated noise points and merge adjacent or connected abnormal areas, the system performs a morphological closing operation on the "abnormal segmentation result" obtained after the inverse transformation, deletes small noise points based on pixel connectivity, merges close areas into a more complete outline, and finally obtains a two-dimensional abnormal segmentation map. At this time, it can be seen that one or more large connected areas are formed near the metal bracket and the protruding position of the cold chain unit.

[0066] After that, the system calls a depth estimation network pre-trained in indoor or vehicle cabin scenarios to perform depth prediction on the 2D abnormal segmentation map. The network combines the camera internal parameters (including focal length, distortion coefficient, etc.) with the depth prior to perform single image depth reasoning and output a depth map corresponding to the abnormal area. The software module further projects each pixel of the depth map into a three-dimensional coordinate system, thereby generating a point cloud data of suspicious protrusions or depressions in the algorithm memory. The point cloud is constructed in a voxel or polygonal manner, and the three-dimensional volume size of this abnormal area is finally calculated.

[0067] Because the vehicle originally has a CAD design size or the "initial cabin 3D volume value" recorded at the factory, the system deducts the multiple abnormal volumes calculated above one by one: on the one hand, if the accessories are protruding, they will occupy the original available space inside the cabin; on the other hand, if they are sunken or damaged (such as floor collapse), they will also change the distribution of available space. In this way, the cabin 3D space data that is closer to reality than the original CAD data can be updated, and a new "precise vehicle space parameters for self-owned transportation capacity" can be formed, including a specific description of the space occupied by the protrusions of the cold chain unit and the metal bracket.

[0068] In the subsequent transport assignment process, the scheduling algorithm can accurately avoid metal brackets or protruding parts when simulating cargo stacking by reading the updated three-dimensional cabin parameters, thus avoiding invalid scheduling. If a certain cargo needs to be placed specially or with a safe gap (such as the requirements of explosion-proof areas for dangerous goods), a more detailed placement simulation can also be performed based on the accurate cabin structure information, which greatly reduces loading conflicts or safety hazards caused by insufficient knowledge of cabin deformation.

[0069] In this way, most of the texture noise and small interference are eliminated in the wavelet domain first, and the real protrusions / depressions are concentratedly marked for deep model processing. Even if there are diverse materials and reflective conditions inside the cabin, a high abnormality detection rate and three-dimensional positioning accuracy can be maintained, effectively supporting the accuracy and safety requirements of large-scale transportation intelligent assignment and scheduling at the micro-loading level.

[0070] As an optional implementation, the cargo attributes include: three-dimensional shape data of the ordered cargo; performing three-dimensional loading feasibility verification on the candidate scheduling scheme includes: Based on the cabin interior image processing result, obtaining three-dimensional shape data of the vehicle cabin; The three-dimensional shape data includes: the overall bulkhead structure and basic information of compartments represented by the low-frequency sub-band, and the three-dimensional parameters of the abnormal area obtained by combining the inverse transformation of the high-frequency sub-band; Inputting the three-dimensional shape data of the vehicle compartment and the three-dimensional shape data of the ordered goods into a multi-grid construction module respectively, and sequentially generating a hierarchical discrete expression of a coarse grid layer and a fine grid layer; At the coarse grid layer, global collision screening is performed on vehicle compartments and cargo based on the voxels of the preset units; In response to detecting conflict information between cargo and bulkhead or abnormal area, marking loading as infeasible and falling back to the multi-objective scheduling algorithm; In response to no conflict information being detected, the local concave-convex data of the cargo surface and the vehicle compartment are compared one by one at the fine grid layer for overlay simulation. In response to any unit collision found at the fine grid layer, the loading is marked as infeasible and falls back to the multi-objective scheduling algorithm. In response to no unit collision found at the fine grid layer, the feasibility check is passed.

[0071] In the specific implementation, in order to process the 3D data of cargo and vehicle space at the same time in the 3D loading feasibility verification link, the order management module will maintain a "cargo 3D data record" for each order, which can be named "cargo 3D model". If the cargo has a regular shape, it can be directly described by length, width, height or simplified volume; if the cargo is an irregular object, its outer surface point cloud or mesh information can be obtained through multi-angle scanning and stored in the "cargo 3D model" entry. The record includes the cargo's geometric shape, size and (if any) convex or curved surface features, which are used for subsequent collision detection or stacking simulation.

[0072] At the same time, the capacity management module will store the "vehicle cabin three-dimensional model" in the database, including the "bulkhead basic information" and "abnormal area information" obtained above.

[0073] Specifically, the "bulkhead basic information" corresponds to the overall structure and compartment parameters represented by the low-frequency sub-band; the "abnormal area information" comes from the protruding or depressed parts of the accessories detected after the inverse transformation of the high-frequency sub-band. The system manages the internal coordinate distribution of its own vehicles, the location of the bulkhead, the compartment function, and the "abnormal area coordinates" such as possible metal brackets and cold chain units in the form of a "vehicle three-dimensional data table" in the database.

[0074] When performing a three-dimensional loading feasibility check on a candidate scheduling scheme, the scheduling module sends the "vehicle compartment three-dimensional model" and the "cargo three-dimensional model" to the "multi-grid construction module" respectively. The module will first generate a "coarse grid layer" according to the preset spatial segmentation rules (for example, the voxel size is 10 cm), and then generate a "fine grid layer" with smaller units (for example, 2 cm or 1 cm) within the range that requires fine simulation, thereby forming a "double-layer or multi-layer hierarchical discrete expression".

[0075] At the coarse grid layer, the system performs a "global collision screening" on the vehicle compartment and cargo, dividing the vehicle compartment into several large-sized cubic units (called "bold voxels"); the basic information of the bulkhead or the abnormal area information is marked. If a cubic unit is regarded as a "bulkhead unit" or "abnormal unit", its coordinate range is unavailable; on the contrary, "available unit" means that cargo can be loaded. The three-dimensional model of the cargo is also divided in equal proportions, and the shape of the cargo is mapped into several corresponding bold voxels. The system compares the coordinate intervals of each cubic unit of the cargo and the vehicle one by one. If it is found that the cargo unit overlaps with the vehicle's "bulkhead unit" or "abnormal unit", the conflict is marked at this level, and the output is "loading is not feasible" and a fallback instruction is issued to the multi-objective scheduling algorithm; if all comparisons are without overlap, the coarse grid layer preliminarily determines that it is feasible, allowing entry into the fine grid layer for more accurate simulation.

[0076] After entering the fine grid layer, the system divides the vehicle compartment and cargo more finely, for example, further subdividing the original coarse voxels into smaller-sized voxels or polygonal grid units. At this point, the "abnormal area information" of the vehicle compartment can further restore the true outline of the metal bracket or the sunken floor, and the local curved surface in the cargo shape is also discretized with high resolution. The system compares the surface coordinates of the cargo and the coordinates of the vehicle compartment unit by unit: if any fine unit overlaps or intersperses, it is marked as a collision, and the loading is determined to be infeasible; only when the fine grid layer does not detect a collision, it is finally determined that the cargo can be placed safely and without interference in the vehicle compartment, and the loading is feasible. The result is output.

[0077] During the implementation process, an "occupancy mark" or "free mark" can be set for each voxel unit, and its coordinates or numbers can be recorded; when the cargo unit and the vehicle unit have the same coordinate number and one is "occupied", a conflict occurs; if there is no such situation, the next stage of comparison will be continued. When the vehicle contains a cold chain compartment or a dangerous goods compartment, the fine grid layer will also add a "compartment type mark" to the corresponding unit. If the cargo has cold chain or dangerous goods attributes, it must match them, otherwise it will be judged as a non-compliant conflict.

[0078] Through this double-layer discretization method, the coarse grid screening is first used to quickly filter out large-scale spatial mismatches, and then the fine grid layer is used to detect the fit between the local bumps and the compartment configuration, which greatly reduces the amount of high-complexity full-volume fine collision calculations, and provides a feasible and fast loading judgment mechanism for large-scale orders and multi-type vehicle scenarios. If the fine grid still detects a conflict, it will immediately return "loading is not feasible" to the multi-objective scheduling algorithm to re-correct the plan; if there is no conflict in the fine grid layer, it means that it can be safely loaded at a higher precision, and finally a "feasibility check passed" result is given.

[0079] It should be emphasized that in this implementation, if the three-dimensional shape data of the ordered goods is only simple length, width and height, the module can quickly construct it into an approximate rectangular voxel; if there is a more refined shape (such as round metal equipment), the point cloud or STL grid file can be used for fine grid discretization at the cargo end; if the goods are hazardous chemicals, the corresponding safety gap is left in the grid structure or special marking is made according to the requirements of explosion-proof compartments, so as to strictly avoid approaching ordinary goods or vehicle compartment units without explosion-proof configuration when the local surface is fitted. With this multiple grid construction and collision detection, the whole process ensures the safety and feasibility of the micro-loading link of intelligent transportation assignment and scheduling in a relatively high-precision and efficient manner.

[0080] As an optional implementation manner, performing three-dimensional loading feasibility verification on the candidate scheduling scheme further includes: In response to data on predicted deformation or additional attachment of a vehicle compartment or cargo, the predicted volume or deformation area is mapped into a ghost volume during multi-grid construction and marked in corresponding coordinate units of a coarse grid layer and a fine grid layer; In the global collision screening, in response to the overall size of the cargo overlapping with the coordinate unit of the ghost volume, it is determined that there is no available space after the subsequent deformation occurs, the loading is marked as infeasible and the multi-objective scheduling algorithm is returned; In response to the fact that the overall size of the cargo does not overlap with the coordinate units of the ghost volume, a fit comparison is further performed on the local surface of the cargo and the local units of the ghost volume at the fine grid layer. In response to the existence of a fit conflict, the loading is marked as infeasible and falls back to the multi-objective scheduling algorithm. In response to the absence of a fit conflict, the feasibility check is output as passed.

[0081] In this embodiment, in order to make advance reservations or conflicts predictions for "predicted deformation or additional accessories" in the three-dimensional loading feasibility verification stage, the system introduces the concept of "ghost volume" in the multi-grid construction process, and marks it in the coordinate units corresponding to the coarse grid layer and the fine grid layer respectively, to simulate the space that may be occupied in the vehicle cabin or the volume that may expand the cargo in the subsequent journey.

[0082] In the specific implementation, there are "data on predicted deformation or additional accessories" entries on the cargo side and the vehicle compartment side respectively. If the vehicle side predicts that a cold chain unit, bracket or other accessories will be added in the next stage, or that metal fatigue will cause a bulge or depression somewhere, then that part will be assigned an "accessory addition mark" and a coordinate area; if the cargo side predicts that it will expand or extend in shape in a certain proportion under environmental conditions such as temperature and humidity, then the system will generate "expanded size" or "expanded volume" information in the three-dimensional data of the cargo. In order to ensure that this prediction is responded to during scheduling and loading, the system maps these predicted volumes (or local coordinate areas) as "ghost volumes" when constructing multiple grids, and loads them together in the grid coordinate system.

[0083] Subsequently, the system first completes a "global collision screening" at the coarse grid layer: if any voxel unit of the overall size of the cargo after discretization through the coarse grid overlaps with the voxel unit of the ghost volume, it can be determined that there will be no available space for the cargo during subsequent deformation or addition of accessories. The system immediately determines that loading is not feasible and sends a fallback signal to the multi-objective scheduling algorithm to avoid investing more computing power in subsequent fine grid layers for detailed judgments; if no overlap is detected at the coarse grid layer, it means that there is no obvious conflict between the cargo and the ghost volume on the overall scale, and the fine grid layer can be entered for a more precise local fit check.

[0084] At the fine grid layer, the system divides the three-dimensional shape of the vehicle cabin and cargo into smaller units (for example, the upper 10 cm voxel is further subdivided into 2 cm or 1 cm voxels), and the coordinate range of the "ghost volume" is also marked in the corresponding coordinate area. At this time, a "fit comparison" or "collision detection" will be performed on the local surface of the cargo and the ghost volume: if the cargo coordinates of any fine grid unit overlap with the ghost volume unit, it means that the cargo will inevitably collide or occupy the additional attachment in the future period or when the deformation is expected to occur, and the system marks the loading as infeasible; only when there is no overlap in all unit comparisons of the fine grid layer, it is considered that the cargo can still coexist safely with the interior space of the cabin after the predicted deformation or the addition of accessories, so the output feasibility check passes.

[0085] Through this double-layer grid and "ghost volume" hierarchical marking method, this implementation method can not only quickly eliminate most of the solutions with no available space at the coarse grid layer, but also perform more accurate conflict checks on the cargo and future occupied space (such as cold chain units, raised panels, or cargo expansion areas) at the fine grid layer, reducing the space compression caused by temporary insertion or deformation in subsequent trips. If any conflict is detected, it will immediately fall back to the multi-objective scheduling algorithm to make assignment corrections or re-split the cargo; if the test passes, it will be marked as feasible, so that the system can still obtain stable and effective loading and distribution results when facing the dynamically changing internal environment of the cabin and cargo that may deform.

[0086] For example, after the initial coarse grid processing, the system has divided the vehicle compartment area into "voxels" or "small unit grids", and each unit is recorded in the "vehicle compartment fine grid table".

[0087] If there is a predicted deformation or additional attachment to the vehicle compartment, the system will first write the "ghost volume mark" and the corresponding "occupied" or "need to leave empty" mark on the corresponding coordinate unit in the "vehicle compartment fine grid table", and may attach a safety gap value (such as a 3cm distance).

[0088] For the bulkhead body or abnormal protruding areas (such as metal brackets), the corresponding cells in the "Vehicle Compartment Fine Grid Table" are also marked as "unavailable" or "pending inspection" to distinguish them from the available space cells.

[0089] The system also uses the same fine grid resolution in the "cargo 3D model", dividing it into several small units coordinate by coordinate and recording them in the "cargo fine grid table"; each cargo unit contains a "grid ID" (such as 3D index i, j, k), "boundary coordinates" (such as minimum / maximum xyz), etc.

[0090] If the cargo is predicted to expand itself, its shape will be directly expanded according to the "expanded contour" during the fine grid generation stage, so that the segmented voxel units cover the part that will grow in the future, so as to make the cargo shape larger during subsequent collision detection.

[0091] At this point, the "Cargo Fine Grid Table" and the "Vehicle Space Fine Grid Table" have a comparable coordinate system (usually the same unit length or corresponding coordinate system).

[0092] The system traverses each cell (denoted as C_g) in the "cargo fine grid table" and extracts its center coordinates or spatial boundaries; Then, the corresponding coordinate unit (denoted as C_v) is located in the "vehicle compartment fine grid table" according to the center coordinate or boundary; if the unit boundary of the cargo overlaps with the "ghost volume identifier" unit or "unavailable" unit of the vehicle compartment, it is recorded as a collision.

[0093] The Axis-Aligned Bounding Box (AABB) principle can be used to determine overlap: if the projections of two cubes have overlapping intervals in each of the X, Y, and Z dimensions, it can be determined that a collision has occurred.

[0094] If a collision is detected, the system immediately determines that the loading is not feasible and rolls back. If no collision is found during the entire traversal, it means that the cargo is compatible with the vehicle space at this resolution.

[0095] If a certain safety gap needs to be left, for example, 3 cm around the accessories of the cold chain unit, the system will expand the relevant ghost volume unit outward by several grids in the vehicle compartment fine grid to form a "safety zone" and mark it as "untouchable".

[0096] For example, in the vehicle compartment fine grid table, around the coordinate cells belonging to the ghost volume or the additional attachment, an additional mark "safe zone radius = ”; When detecting collisions, the system does not simply look at the overlap of coordinates, but also checks whether the distance between the cargo unit and the center of the "safe zone" unit is less than ; If so, it is considered a conflict.

[0097] After the system has traversed all units and found no conflicts or insufficient fits, it will output "feasibility check passed", end the inspection and confirm the available solutions.

[0098] In addition, to reduce over-calculation, fine grid calculations are performed only after the coarse grid has eliminated large-area conflicts. If the cargo is large in size, an octree or eight-level partitioning method can be used to improve search efficiency.

[0099] In practical applications, this fine grid combined with the safety gap check can significantly reduce the risk of crushing or damage during subsequent driving for irregular cargo (such as large mechanical parts) or chemical containers that will expand at extreme temperatures.

[0100] As an optional implementation manner, performing three-dimensional loading feasibility verification on the candidate scheduling scheme further includes: At the coarse grid layer, a global collision screening is performed between cargo and vehicle compartments based on the ghost volume; In response to any coarse grid cell detecting a voxel overlap in the global collision screening, marking the loading infeasible and falling back to a multi-objective scheduling algorithm; In response to no voxel overlap being detected in the global collision screening by the coarse grid unit, the local surface of the cargo under the ghost volume is compared with the concave and convex area of ​​the vehicle compartment grid by grid at the fine grid layer; in response to the occurrence of a fit conflict, the loading is marked as infeasible and the method falls back to the multi-objective scheduling algorithm; in response to no fit conflict at the fine grid layer, it is confirmed that the cargo can still be loaded smoothly after its expansion, and the feasibility check is passed.

[0101] In this implementation, in order to further improve the hierarchical screening of the three-dimensional loading feasibility verification in the face of the "ghost volume" scenario, the system performs collision or fit judgment at the coarse grid layer and the fine grid layer respectively to ensure that the cargo can be safely loaded with high accuracy even after expansion or the addition of accessories to the vehicle compartment. The purpose is to quickly screen out large-scale spatial conflicts through simple global collision detection, and then use fine fit verification to accurately determine the microscopic relationship between the outer surface of the cargo and the concave and convex parts of the vehicle compartment, thereby reducing the problem of crowding caused by deformation during transportation.

[0102] In terms of specific principles, when the system detects the presence of cargo expansion prediction information or additional identification of vehicle compartment accessories, it will apply a "ghost volume" identification to the corresponding voxel unit in the coordinate system during the multi-grid construction stage; if the cargo will also increase in volume under conditions such as temperature and humidity, its three-dimensional model will expand its shape before discretization and record it in a larger range in the grid table. Therefore, the coarse grid layer can use larger-sized voxels (such as 10 cm cube) for global collision screening: if any cargo voxel unit overlaps with the ghost volume unit, the system immediately determines that there is no available space in the subsequent trip, marks the loading as infeasible and falls back to the scheduling algorithm. This operation corresponds to "global collision screening of cargo and vehicle compartments based on ghost volume at the coarse grid layer". If any coarse grid unit detects voxel overlap, it will fall back directly to avoid subsequent calculation waste.

[0103] When there is no conflict in the coarse grid, the system enters the fine grid layer and discretizes the same area into smaller voxels or grids (such as 2 cm or 1 cm). At this time, the coordinates of the vehicle compartment under the ghost volume are compared with the local surface of the cargo one by one; because the fine grid can capture factors such as protrusions, depressions or safety gaps, the system will perform higher-precision fit or collision detection at this level. If any fine grid unit overlaps at this time, that is, a fit conflict occurs, the system marks the loading as infeasible and rolls back the scheduling; only when there is no conflict in the traversal of the fine grid layer, it is confirmed that the cargo can still be placed in the vehicle compartment after expansion. In this way, after performing the local verification of the fine grid, due to the absence of fit conflict, the system outputs "feasibility verification passed", allowing the cargo to be loaded smoothly. In engineering applications, this "double-layer screening" combined with "ghost volume identification" has significant advantages: the coarse grid stage can quickly eliminate large-scale mismatches, and the fine grid ensures the microscopic safety space for predicting deformation.

[0104] In one example, if the vehicle compartment predicts that a bulge will be installed in the next stage of the cold chain unit, the coordinate range is marked as a ghost volume unit; the cargo is estimated to expand by 5% due to its chemical properties. In the coarse grid stage, a global collision is performed using 10 cm cubic voxels. If the overall coordinates of the cargo overlap with the ghost volume, it is immediately determined to be infeasible; if there is no overlap, the surface boundary of the cargo and the ghost volume coordinates are compared one by one using 2 cm voxels in the fine grid stage; if a fit conflict or insufficient safety gap is detected in any unit, it is determined to be infeasible and backed off, otherwise it is ultimately determined to be feasible. Through this process, appropriate space can still be reserved under the dual interference of predicted accessory occupancy and cargo expansion to prevent the risk of crowding or collision during driving.

[0105] In this way, rapid filtering of the coarse grid can avoid a large number of unnecessary fine calculations, and precise fit detection of the fine grid can strictly control the "space encroachment" problem caused by temperature expansion or the installation of accessories, ensuring transportation safety and effective utilization.

[0106] As an optional implementation manner, the intelligent transportation dispatching method further includes: At the vehicle compartment end, the preset cabin deformation prediction model is periodically called to perform time series inference based on the vehicle usage time, cabin metal fatigue coefficient and historical modification records, and output the predicted deformation or accessory addition data of each free vehicle in the next stage of its journey; The predicted deformation or accessory addition data is marked with an accessory addition mark or a metal plate bending mark, and the coordinate range is stored in a deformation data table; At the cargo end, based on the order information, the volume expansion rate or structural deformation probability information based on the material or chemical properties of the cargo is extracted, combined with the predicted values ​​of the temperature, humidity and pressure conditions during transportation, and the possible additional outer contour dimensions of the cargo are calculated through the cargo deformation prediction module, and recorded in the form of expanded volume or expanded size in the deformation data table; In the multi-grid construction stage, based on the vehicle compartment accessory addition identification, vehicle compartment metal plate bending, cargo expansion volume and cargo expansion size registered in the deformation data table, the predicted deformation area or expansion form is written into the corresponding ghost volume coordinate unit.

[0107] In this embodiment, in order to truly utilize the "predicted deformation / accessory addition data at the vehicle compartment end or the cargo end" in the loading feasibility verification, the system establishes a set of "deformation data table" and "ghost volume writing" mechanisms between the database and the multi-grid construction process.

[0108] In the specific implementation, at the vehicle compartment end, the "cabin deformation prediction model" is periodically called. The model can be run on the server end or the local maintenance end, and reads by time series analysis or machine learning: The vehicle’s mileage or duration of use (such as obtained from vehicle sensors or mileage data sheets), the metal fatigue coefficient of the cabin (obtained from historical sensor logs or metal stress tests), and historical modification records (such as adding new brackets, modifying cold chain modules, etc.).

[0109] Based on these input parameters, the model uses algorithms such as ARIMA (autoregressive integrated moving average) or LSTM (long short-term memory network) to predict the sheet metal deformation variables of the cabin in the next stage (for example, the next 2,000 kilometers or the next maintenance cycle), and at the same time identify whether there are planned coordinates for additional accessories (such as the height and installation location of the cold chain unit). If it is predicted that a certain area may be dented, convex, or occupied by accessories, its approximate spatial coordinate range (usually in the form of a three-dimensional cuboid or point cloud) and "accessory addition type" are output, and stored in the "vehicle cabin deformation data table" in this embodiment. The system annotates each record in the table with information such as "vehicle ID", "coordinate start and end", "specific deformation variable or accessory size"; if the metal plate may bend, the "bending amplitude or center point" is also recorded.

[0110] Secondly, on the cargo side, the system has a "cargo deformation prediction module" that uses the "cargo material, chemical properties, and temperature sensitivity" marked in the order information and combines environmental prediction data such as the average temperature, humidity, or altitude pressure of the route to infer the volume expansion or deformation that may occur during transportation. If the cargo is a chemical barrel, the module will retrieve the internal "material-expansion rate mapping table" and calculate the coordinate distribution of the "maximum expansion volume" or "local surface expansion" in combination with the route temperature curve.

[0111] For example, if the top of the barrel bulges 5 cm as the temperature rises, the module records the additional bulge as "extended coordinates" and stores it in the "cargo deformation data table." The system also records the potential shape changes of each batch of cargo, allowing space to be reserved in advance during the subsequent loading phase.

[0112] After the above two stages are completed, the information in the "Vehicle Compartment Deformation Data Table" and the "Cargo Deformation Data Table" will be integrated into the "Deformation Data Table" through the data exchange interface. This table can contain the following fields: VehicleID / OrderID: Indicates which vehicle or cargo it is; ShapeType: indicates whether it is "addition of accessories" or "bending of sheet metal" or "inflation of cargo"; CoordRanges: three-dimensional coordinate interval, such as x1x2, y1y2, z1~z2; Margin: If a safety gap is required, a minimum margin value will be recorded.

[0113] TimeWindow: Optional field if the prediction only occurs during a certain time period or temperature.

[0114] When the system enters the "multi-grid construction phase" to discretize the vehicle compartments and cargo (see the multi-grid construction described above), it will first read all the records in the "deformation data table" and write them into the coordinate layer one by one. For example: If the record indicates that the vehicle ID = V123 will install the accessory "Unit A" at (X:5070, Y:1020, Z:0~30), then in the vehicle compartment grid coordinates, the corresponding voxel unit will be marked as "Ghost Volume - Additional Accessory". If Margin=3cm, then a 3 cm range will be expanded around it to form a "safe zone" unit, which will be uniformly marked as "untouchable" or "empty zone".

[0115] If the record shows that the cargo ID = G456 is increased by 5 cm in height at the top, then when discretizing the grid, the cargo coordinate Z: (H-5) ~ H+0 section is expanded or a new grid unit is added to mark the "expansion area". Similarly, it is stored as a ghost volume unit so that the cargo shape is larger during subsequent collision detection.

[0116] After the system finishes writing each deformation record, the corresponding ghost volume unit will be added to the coordinate table of the vehicle compartment grid and the cargo grid. In this way, when the three-dimensional loading feasibility check enters the collision detection stage (including global screening of the coarse grid layer and fit judgment of the fine grid layer), any overlap of these ghost volume coordinate units will be regarded as "no available space in the subsequent journey" or "risk of collision after expansion", thereby triggering loading infeasibility fallback. If no conflict is detected in the double-layer grid in the end, it means that even after the vehicle compartment is installed with accessories or the cargo expands, it can still be loaded safely.

[0117] In this way, the system first calculates the "possible deformation or attachment coordinates" in the prediction stage, writes them into the "ghost volume coordinate unit" when the multiple grids are discretized, and finally uses these coordinates to make spatial conflict judgments in the collision detection stage. The beneficial effect is to reduce unexpected crowding or conflicts during driving, help the scheduling algorithm to more stably output safe and feasible assignment plans, and improve the adaptability to complex temperature-sensitive goods or frequent vehicle modifications. Especially in large-scale order scenarios, this process can save the number of subsequent iterative scheduling and improve overall operational efficiency and safety.

[0118] As an optional implementation, the multi-objective scheduling algorithm further includes: Based on the data of the predicted deformation or the additional attachment, an additional preset penalty value is added to the allocation relationship for assigning the vehicle in the objective function; Based on the expanded volume or extended size of the cargo, the expanded or extended size is regarded as the reference volume of the cargo during the scheduling stage, and a preset high penalty value is set for the allocation exceeding the reference volume.

[0119] In this embodiment, in order to reduce the risk of three-dimensional loading being unfeasible due to deformation of the vehicle compartment or expansion of the cargo volume in the future trip, the system further introduces two "penalty value setting" steps within the multi-objective scheduling algorithm, so that the algorithm can prioritize or reduce the assignment of those high-risk combinations before the vehicle and cargo enter the real collision detection stage. Specifically, a "vehicle prediction deformation information table" and a "cargo expansion information table" are added to the system's scheduling data structure, and several configurable "preset penalty values" are added in the objective function or constraint rules.

[0120] In specific implementation, if a vehicle has been determined by the vehicle-side prediction model to be deformed or installed with accessories during the next stage of use (for example, a new cold chain unit is added to the left bulkhead, or a local bulge caused by metal fatigue appears), the system will record the "deformation volume (or area)" and its corresponding coordinate interval in the "vehicle prediction deformation information table". When the scheduling algorithm initializes or updates the candidate solution, once it finds that the order allocation attempts to assign the goods to the vehicle, it will add a "vehicle deformation penalty value" to the allocation relationship in the objective function. This value can be calculated based on the deformation volume, the vehicle's available space margin, and the business-side priority. For example, the system can use: Deformation penalty = α × expected deformation volume (or volume occupied by attachments) Among them, α is a set of adjustable penalty coefficients. If the vehicle's deformation variable is larger or the future attachment space is more obvious, the vehicle will be more penalized during the algorithm search process, making its assignment relationship with the goods tend to be lower in the overall fitness ranking, thereby reducing the chance of being selected in the iteration process. However, if there is indeed no other vehicle with the corresponding transportation capacity or timeliness advantage, the algorithm can still retain this assignment combination, but its initial fitness is lower.

[0121] Furthermore, for the expansion or extended dimensions of the cargo side, the estimated expansion rate or additional external dimensions of each cargo are also recorded in the "Cargo Expansion Information Table".

[0122] For example, if the volume of a chemical barrel increases by 10% when the temperature is above 30°C, the system will directly update the cargo volume by (1+expansion rate) times to the "base volume after expansion" during dispatch, so that when calculating "size or load constraints" or even freight or energy consumption, this larger size will be used for calculation. If the expansion deformation of a certain cargo does not absolutely prohibit allocation, but it is a high-risk scenario, a "cargo expansion penalty value" can also be added to the objective function for this combination, such as: Expansion penalty = β × (expanded volume - original volume) β is another adjustable coefficient. If the expansion is higher, it means that the risk of occupying space is greater, and the scheduling algorithm tends to avoid allocating this cargo to vehicles with less space. In actual calculations, this inflation penalty value, together with indicators such as freight, timeliness, and empty driving rate, constitutes the objective function, allowing the algorithm to automatically eliminate those candidate solutions that cause deep conflicts due to cargo inflation during evolution or iterative search.

[0123] At the algorithm operation level, every time the algorithm generates or improves an "order-vehicle" mapping relationship, it queries the "vehicle predicted deformation information table" and the "cargo expansion information table" to determine whether the vehicle is in an "estimated deformation" state and whether the cargo has "volume expansion". If either party meets the requirements, the corresponding penalty value is added to the fitness of the mapping relationship, and feasibility screening or selection / mutation operations are performed. In this way, before the actual execution of the three-dimensional loading collision detection, the system has already eliminated or downgraded these potential high-risk and high-conflict combinations from the candidate solutions step by step through the penalty items, thereby reducing the rollback rate in the subsequent three-dimensional loading stage and improving the overall iteration efficiency and solution quality.

[0124] In this way, the system can learn about possible dynamic changes in the vehicle compartment or cargo end in advance without having to wait until the three-dimensional collision detection stage to make a discovery; on the other hand, the algorithm imposes additional penalties on combinations involving high deformation or high expansion during multiple rounds of iterations, so that the scheduling solution is more in line with the actual needs of future trips at the macro level, greatly reducing the number of repeated rollbacks and resource waste in the entire process. At the same time, for special cases, such as dangerous goods with high expansion rates, multiple penalties can be superimposed in the scheduling objective function to ensure that safety requirements take precedence over ordinary operating indicators. Therefore, this method is particularly suitable for highly dynamic transportation scenarios (such as long-distance cold chain, chemical transportation, or frequent vehicle modifications), and can effectively alleviate the safety and efficiency problems caused by the traditional scheduling-loading split.

[0125] Based on the same inventive concept, the embodiment of the present disclosure also provides a transport intelligent assignment and scheduling system corresponding to a transport intelligent assignment and scheduling method. Since the principle of solving the problem by the system in the embodiment of the present disclosure is similar to the above-mentioned transport intelligent assignment and scheduling method in the embodiment of the present disclosure, the implementation of the system can refer to the implementation of the method, and the repeated parts will not be repeated.

[0126] Reference Figure 4 FIG. 1 is a schematic diagram of a transportation intelligent dispatching and scheduling system provided in an embodiment of the present application, wherein the system includes: The receiving module 10 is used to obtain order information and transport resource information, wherein the order information includes: cargo attributes; the transport resource information includes: vehicle space parameters of self-owned transport capacity and simplified load parameters of external transport capacity; the vehicle space parameters include: an image of the interior of the vehicle cabin; The receiving module 10 is also used to process the interior image of the cabin using the wavelet transform method for the vehicle cabin parameters of the self-owned transport capacity to determine the three-dimensional parameters of the abnormal area; based on the three-dimensional parameters of the abnormal area and the vehicle cabin parameters, generate accurate vehicle cabin parameters of the self-owned transport capacity.

[0127] The scheduling module 20 is used to allocate vehicles and routes to the pre-processed order set based on the order information, the precise vehicle space parameters of the self-owned transport capacity and the simplified load parameters of the external transport capacity using a multi-objective scheduling algorithm to generate candidate scheduling solutions; The detection module 30 is used to perform a three-dimensional loading feasibility check on the candidate scheduling scheme using a preset multi-grid construction module to generate a feasibility check result; in response to the feasibility check failing, fall back to the multi-objective scheduling algorithm to correct the scheduling scheme; An output module 40 is used to output a final scheduling plan in response to the feasibility checks being passed; The three-dimensional loading feasibility check includes: determining predicted deformations for cargo and vehicle spaces based on the order information and the transport resource information, and mapping the predicted deformations into ghost volumes in the multi-grid construction module to mark them in coordinate units corresponding to different grid layers; In response to the fact that the overall size of the cargo and the coordinate units of the ghost volume do not overlap, a fit comparison is performed on the local surface of the cargo and the local units of the ghost volume at the grid layer. In response to the existence of a fit conflict, the loading is marked as infeasible and falls back to the multi-objective scheduling algorithm. In response to the absence of a fit conflict, the feasibility check is output as passed.

[0128] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in the present invention can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

Claims

1. A method for intelligent transportation dispatching, characterized in that: include: Acquire order information and transport resource information, wherein the order information includes: cargo attributes; the transport resource information includes: vehicle space parameters of self-owned transport capacity and simplified load parameters of external transport capacity; the vehicle space parameters include: an image of the interior of the vehicle cabin; For the vehicle cabin parameters of the self-owned transport capacity, the interior image of the vehicle cabin is processed by wavelet transform method to determine the three-dimensional parameters of the abnormal area; based on the three-dimensional parameters of the abnormal area and the vehicle cabin parameters, accurate vehicle cabin parameters of the self-owned transport capacity are generated; Based on the order information, the precise vehicle space parameters of the self-owned transport capacity, and the simplified load parameters of the external transport capacity, a multi-objective scheduling algorithm is used to allocate vehicles and routes to the pre-processed order set to generate candidate scheduling solutions; Using a preset multi-grid construction module, performing a three-dimensional loading feasibility check on the candidate scheduling scheme to generate a feasibility check result; in response to the feasibility check failing, falling back to the multi-objective scheduling algorithm to correct the scheduling scheme; In response to the feasibility checks all being passed, outputting a final scheduling plan; The three-dimensional loading feasibility check includes: determining predicted deformations for cargo and vehicle spaces based on the order information and the transport resource information, and mapping the predicted deformations into ghost volumes in the multi-grid construction module to mark them in coordinate units corresponding to different grid layers; In response to the fact that the overall size of the cargo and the coordinate units of the ghost volume do not overlap, a fit comparison is performed on the local surface of the cargo and the local units of the ghost volume at the grid layer. In response to the existence of a fit conflict, the loading is marked as infeasible and falls back to the multi-objective scheduling algorithm. In response to the absence of a fit conflict, the feasibility check is output as passed.

2. The method according to claim 1, characterized in that The cabin interior image includes a cabin interior image of a key area; the key area includes: a wall, a floor, a corner, and a door frame of the cabin; The precise vehicle space parameters for generating self-owned transport capacity include: Performing 2- to 3-layer 2D discrete wavelet transform on the cabin interior image to decompose low-frequency sub-bands and high-frequency sub-bands; The low-frequency sub-band is used to characterize the overall structure and background information of the interior image of the vehicle cabin; the high-frequency sub-band is used to characterize the edge, texture and detail information of the interior image of the vehicle cabin; calculating the amplitude of the wavelet coefficients in the high frequency sub-band and applying adaptive threshold processing to obtain abnormal features; The abnormal features are mapped back to the original image space using an inverse wavelet transform, and adjacent abnormal points are merged and isolated noise points are removed using a morphological closing operation to obtain abnormal area location information; Use the pre-trained deep learning model to predict the depth information of the abnormal area from a single image, generate a depth map, and convert the depth map into 3D parameters of the abnormal area by combining the camera internal parameters; Based on the three-dimensional parameters of the abnormal area, the volume of the cabin space occupied by the abnormal area is calculated, and the volume is deducted from the original design size of the cabin to generate accurate vehicle space parameters of the self-owned transport capacity.

3. The method according to claim 2, characterized in that The cargo attributes include: three-dimensional shape data of the ordered cargo; performing three-dimensional loading feasibility verification on the candidate scheduling scheme includes: Based on the precise vehicle space parameters of the self-owned transport capacity, obtaining three-dimensional shape data of the vehicle space; The three-dimensional shape data includes: the overall bulkhead structure and basic information of compartments represented by the low-frequency sub-band, and the three-dimensional parameters of the abnormal area obtained by combining the inverse transformation of the high-frequency sub-band; Inputting the three-dimensional shape data of the vehicle compartment and the three-dimensional shape data of the ordered goods into a multi-grid construction module respectively, and sequentially generating a hierarchical discrete expression of a coarse grid layer and a fine grid layer; At the coarse grid layer, global collision screening is performed on vehicle compartments and cargo based on the voxels of the preset units; In response to detecting conflict information between cargo and bulkhead or abnormal area, marking loading as infeasible and falling back to the multi-objective scheduling algorithm; In response to no conflict information being detected, the local concave-convex data of the cargo surface and the vehicle compartment are compared one by one at the fine grid layer for overlay simulation. In response to any unit collision found at the fine grid layer, the loading is marked as infeasible and falls back to the multi-objective scheduling algorithm. In response to no unit collision found at the fine grid layer, the feasibility check is passed.

4. The method according to claim 3, characterized in that Performing a three-dimensional loading feasibility check on the candidate scheduling scheme also includes: In response to data on predicted deformation or additional attachment of a vehicle compartment or cargo, the predicted volume or deformation area is mapped into a ghost volume during multi-grid construction and marked in corresponding coordinate units of a coarse grid layer and a fine grid layer; In the global collision screening, in response to the overall size of the cargo overlapping with the coordinate unit of the ghost volume, it is determined that there is no available space after the subsequent deformation occurs, the loading is marked as infeasible and the multi-objective scheduling algorithm is returned; In response to the fact that the overall size of the cargo does not overlap with the coordinate units of the ghost volume, a fit comparison is further performed on the local surface of the cargo and the local units of the ghost volume at the fine grid layer. In response to the existence of a fit conflict, the loading is marked as infeasible and falls back to the multi-objective scheduling algorithm. In response to the absence of a fit conflict, the feasibility check is output as passed.

5. The method according to claim 4, characterized in that Performing a three-dimensional loading feasibility check on the candidate scheduling scheme also includes: At the coarse grid layer, a global collision screening is performed between cargo and vehicle compartments based on the ghost volume; In response to any coarse grid cell detecting a voxel overlap in the global collision screening, marking the loading infeasible and falling back to a multi-objective scheduling algorithm; In response to no voxel overlap being detected in the global collision screening by the coarse grid unit, the local surface of the cargo under the ghost volume is compared with the concave and convex area of ​​the vehicle compartment grid by grid at the fine grid layer; in response to the occurrence of a fit conflict, the loading is marked as infeasible and the method falls back to the multi-objective scheduling algorithm; in response to no fit conflict at the fine grid layer, it is confirmed that the cargo can still be loaded smoothly after its expansion, and the feasibility check is passed.

6. The method according to claim 5, characterized in that Also includes: At the vehicle compartment end, the preset cabin deformation prediction model is periodically called to perform time series inference based on the vehicle usage time, cabin metal fatigue coefficient and historical modification records, and output the predicted deformation or accessory addition data of each free vehicle in the next stage of its journey; The predicted deformation or accessory addition data is marked with an accessory addition mark or a metal plate bending mark, and the coordinate range is stored in a deformation data table; At the cargo end, based on the order information, the volume expansion rate or structural deformation probability information based on the material or chemical properties of the cargo is extracted, combined with the predicted values ​​of the temperature, humidity and pressure conditions during transportation, and the possible additional outer contour dimensions of the cargo are calculated through the cargo deformation prediction module, and recorded in the form of expanded volume or expanded size in the deformation data table; In the multi-grid construction stage, based on the vehicle compartment accessory addition identification, vehicle compartment metal plate bending, cargo expansion volume and cargo expansion size registered in the deformation data table, the predicted deformation area or expansion form is written into the corresponding ghost volume coordinate unit.

7. The method according to claim 6, characterized in that The simplified load parameters include: maximum loadable weight, maximum transportable volume and transport qualification information; the multi-objective scheduling algorithm includes: Establishing an objective function in the multi-objective scheduling algorithm; wherein the optimization indicators in the objective function include: freight cost, transportation time efficiency, vehicle empty driving rate and special cargo compliance requirements; A hard constraint rule is set in the multi-objective scheduling algorithm; the hard constraint rule includes: In response to the fact that the cargo attributes include dangerous goods, it can only be assigned to the self-owned transport capacity with corresponding protection configuration or external transport capacity with dangerous goods transportation qualification; in response to the fact that the cargo requires cold chain, it can only be assigned to the transport capacity with corresponding refrigeration function; Soft constraint rules are set in the multi-objective scheduling algorithm; the soft constraint rules include: In response to the size or weight of the ordered cargo exceeding the precise vehicle space parameters of the owned transport capacity or the simplified load parameters of the external transport capacity, a penalty value is added to the current assignment relationship in the objective function to reduce the fitness of the assignment relationship in the iterative search.

8. The method according to claim 7, characterized in that Generating a candidate scheduling solution includes: Performing iterative solving on the preprocessed order set to generate multiple candidate solutions, each candidate solution including an assignment relationship for own transportation capacity and / or external transportation capacity, a driving route, and an estimated arrival time; In response to detecting an allocation combination that does not satisfy the constraint rule during the scheduling iteration process, the candidate solution is eliminated or mutated during the algorithm iteration until a candidate scheduling solution that satisfies the objective function requirement is output.

9. The method according to claim 8, characterized in that The multi-objective scheduling algorithm also includes: Based on the data of the predicted deformation or the additional attachment, an additional preset penalty value is added to the allocation relationship for assigning the vehicle in the objective function; Based on the expanded volume or extended size of the cargo, the expanded or extended size is regarded as the reference volume of the cargo during the scheduling stage, and a preset high penalty value is set for the allocation exceeding the reference volume.

10. A transportation intelligent dispatching system, characterized in that: include: The receiving module is used to obtain order information and transport resource information, wherein the order information includes: cargo attributes; the transport resource information includes: vehicle space parameters of self-owned transport capacity and simplified load parameters of external transport capacity; the vehicle space parameters include: an image of the interior of the vehicle cabin; The receiving module is further used to process the interior image of the vehicle cabin using a wavelet transform method for the vehicle cabin parameters of the self-owned transport capacity to determine the three-dimensional parameters of the abnormal area; based on the three-dimensional parameters of the abnormal area and the vehicle cabin parameters, generate accurate vehicle cabin parameters of the self-owned transport capacity; A scheduling module, for allocating vehicles and routes to the pre-processed order set using a multi-objective scheduling algorithm based on the order information, the precise vehicle space parameters of the self-owned transport capacity, and the simplified load parameters of the external transport capacity, and generating candidate scheduling solutions; A detection module, configured to perform a three-dimensional loading feasibility check on the candidate scheduling scheme using a preset multi-grid construction module, and generate a feasibility check result; in response to failure of the feasibility check, fall back to the multi-objective scheduling algorithm to correct the scheduling scheme; An output module, configured to output a final scheduling plan in response to the feasibility checks being passed; The three-dimensional loading feasibility check includes: determining predicted deformations for cargo and vehicle spaces based on the order information and the transport resource information, and mapping the predicted deformations into ghost volumes in the multi-grid construction module to mark them in coordinate units corresponding to different grid layers; In response to the fact that the overall size of the cargo and the coordinate units of the ghost volume do not overlap, a fit comparison is performed on the local surface of the cargo and the local units of the ghost volume at the grid layer. In response to the existence of a fit conflict, the loading is marked as infeasible and falls back to the multi-objective scheduling algorithm. In response to the absence of a fit conflict, the feasibility check is output as passed.

Images