Cargo loading management method based on artificial intelligence
By constructing a deep learning model for intelligent cargo loading management, the problems of low efficiency and unstable scheduling in traditional loading methods are solved, achieving efficient and stable vehicle scheduling and route planning, which is suitable for modern logistics systems.
Patent Information
- Application Number
- CN202511095438.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-06
- Publication Date
- 2025-11-14
AI Technical Summary
Traditional methods of loading goods, which rely on manual experience, suffer from problems such as low allocation efficiency, slow scheduling response, and low resource utilization when faced with complex vehicle resources, dynamic order tasks, and loading area restrictions. This is especially true during peak hours or in scenarios where multiple points are loaded simultaneously, which can easily lead to loading congestion, route conflicts, and traffic chaos.
An AI-based cargo loading management method is adopted. By constructing a deep learning model and combining vehicle information and delivery order information, intelligent matching and scheduling management are carried out, including route planning and emergency dispatch. The vehicle loading time window and route are dynamically adjusted to optimize vehicle allocation and avoid loading peak conflicts.
It improves the scientific nature and operational efficiency of vehicle dispatching decisions, avoids peak loading conflicts and route bottlenecks, and ensures the stability and responsiveness of vehicle dispatching. It is suitable for time-sensitive logistics and emergency material transportation.
Smart Images
Figure CN120952650A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of cargo management technology, specifically to a cargo loading management method based on artificial intelligence. Background Technology
[0002] In the modern logistics and transportation system, the loading and dispatching of goods is a key link connecting warehousing and distribution. Its efficiency and accuracy directly affect the timeliness and cost control capabilities of the entire supply chain. With the development of e-commerce, intelligent manufacturing and multi-level distribution models, the order structure is showing characteristics of small batches, high frequency and diversification. Traditional loading methods that rely on manual experience have gradually revealed problems such as low allocation efficiency, slow scheduling response, and low resource utilization when faced with complex vehicle resources, dynamic order tasks, and loading area restrictions. During peak hours or in scenarios where multiple vehicles are loaded simultaneously, if vehicles arrive at the same area in a concentrated manner, it can easily cause loading congestion, route conflicts, or even traffic chaos, affecting the continuity and safety of operations. To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention
[0003] To address the shortcomings of existing technologies, this invention provides an artificial intelligence-based cargo loading management method.
[0004] To achieve the above objectives, the present invention provides the following technical solution: a cargo loading management method based on artificial intelligence, comprising: S1. Obtain the Vehicle Information Table (VTL) and the Delivery Order Information Table (ODR); S2. Construct a training set based on historical scheduling records (HST), train a pre-built deep learning model, and obtain an allocation model (MOD). The allocation model (MOD) includes an input layer, multiple hidden layers for extracting features, and an output layer for outputting results. S3. Input the collected vehicle information table VTL and delivery order information table ODR into the allocation model MOD, and output the vehicle ID, loading area ZON and time window TW corresponding to the order. S4. Based on vehicle ID, loading zone (ZON), and recommended loading time period (TW), scheduling management is carried out through scheduling management methods, including using vehicle application path planning algorithms to generate optimal driving routes and emergency scheduling plans.
[0005] The historical dispatch record (HST) includes a vehicle information table (VTL) and a delivery order information table (ODR). The vehicle information table (VTL) includes vehicle ID, vehicle type, loading volume (VOL) and loading weight (WGT), vehicle status (STA), current location (POS), and historical dispatch record (HST). The delivery order information table (ODR) includes order ID, cargo type (CAT), loading volume (VOL) and loading weight (WGT), loading requirement (REQ), loading zone (ZON), and time window (TW). The historical dispatch record (HST) is obtained from this information.
[0006] Perform range checks on the loading volume (VOL) and loading weight (WGT) in the vehicle information table (VTL), remove outliers that exceed the preset upper and lower thresholds, perform legality checks on the vehicle status (STA), loading requirements (REQ), and cargo type (CAT), and remove data items that are not in the predefined category set.
[0007] The vehicle information set is obtained by retrieving the vehicle information table (VTL). This set includes the vehicle's serial number, years of use, number of repairs, and mileage. A vehicle score (Sve) is calculated for each vehicle in the set using the following formula:
[0008] Wherein, F1 is the service life of the delivery vehicle, f2 is the number of times the delivery vehicle has been repaired, and f3 is the mileage of the delivery vehicle; w1 is the weighting coefficient of the service life of the delivery vehicle, w2 is the weighting coefficient of the number of times the delivery vehicle has been repaired, and w3 is the weighting coefficient of the mileage of the delivery vehicle.
[0009] The allocation model MOD includes an input layer, multiple hidden layers, and an output layer. The input layer contains 14 feature nodes, which correspond to vehicle type, loading capacity, vehicle age F1, vehicle rating Sve, loading weight WGT, vehicle status STA, cargo type CAT, loading volume VOL, cargo loading weight WGT, loading requirements REQ, loading area ZON, order priority PRI, historical average scheduling waiting time AVG, and historical matching success rate HMR.
[0010] The hidden layers include a first hidden layer, a second hidden layer, a third hidden layer, and a fourth hidden layer. The first hidden layer contains 128 nodes, the second hidden layer contains 256 nodes, the third hidden layer contains 128 nodes, and the fourth hidden layer contains 64 nodes. Each hidden layer uses the ReLU activation function. A BatchNormalization normalization layer is set between the second and third hidden layers. Dropout layers are set after the third and fourth hidden layers, respectively, with a dropout rate of 0.3. The output layer contains three output nodes: the optimal vehicle ID, the specified loading area number (ZON), and the loading time window (TW).
[0011] The scheduling management method includes: S11. Set a corresponding time window TW for each loading zone ZON, and preset the maximum allowed number of vehicles to be loaded at the same time for each ZON, and the loading capacity MAX. S12. When it is detected that the number of vehicles allocated to loading area ZON within time window TW exceeds the loading capacity MAX, the time adjustment mechanism is automatically triggered to adjust the loading time of some vehicles to the adjacent time window TW′. Within time window TW, the number of vehicles allocated to loading area ZON does not exceed the loading capacity MAX. S13. Combining the current position POS of each vehicle with the position of its assigned loading zone ZON, the path planning algorithm ROU is invoked to generate the optimal driving route for each vehicle, and the optimal driving route is sent to the scheduling terminal. S14. Collect and monitor the current location (POS) and loading status (STA) of each vehicle through GPS positioning. When a vehicle concentration, route congestion, or abnormal delay is detected in the target zone (ZON), adjust the vehicle's time window (TW) based on the scheduling rules.
[0012] The scheduling rules include: S141. By comparing the number of vehicles allocated to the same ZON within the same TW with the preset loading capacity MAX in real time, the congestion index CND is calculated. When CND≥1, a congestion event is triggered. S142. Based on the order priority (PRI) and estimated queuing time (WAT) of each vehicle, prioritize the vehicles that triggered the congestion event and generate a list of vehicles to be adjusted. S143. In accordance with the PRI from low to high, postpone the time window TW of the vehicles in the list of vehicles to be adjusted to the next time window TW′; S144. Based on the time window TW′, re-invoke the path planning algorithm ROU to calculate the optimal driving route, and send the time window TW′ to the corresponding vehicle through the scheduling terminal. S145. Repeat steps S141 to S144 until the number of vehicles assigned to the same loading zone ZON within the time window TW returns to no more than the preset loading capacity MAX.
[0013] The emergency dispatch plan includes the following steps: S21. Monitor the delivery orders in the delivery order information table ODR. When a delivery order is detected to have a change or emergency order tag, mark the delivery order as a high priority (PRI). S22. Input the data from the delivery order with high priority PRI into the allocation model MOD, recalculate and obtain the optimal vehicle ID, loading zone ZON and loading time window TW corresponding to the delivery order; S23. Based on the calculated vehicle ID, loading zone ZON and loading time window TW, assess whether the number of vehicles in the corresponding loading zone ZON exceeds the loading capacity MAX. If it does, trigger the time window adjustment mechanism to postpone the loading time window TW of non-high priority orders. S24. Based on the changed vehicle allocation results, call the route planning algorithm ROU to recalculate the driving route and send it to the dispatch terminal of the corresponding vehicle.
[0014] This invention provides an artificial intelligence-based method for cargo loading management. Compared with existing technologies, it has the following advantages: This invention uses a constructed deep learning allocation model MOD to jointly model the vehicle information table (VTL) and the delivery order information table (ODR), thereby achieving intelligent matching of orders and vehicles and outputting optimized vehicle IDs and loading zones (ZONs), thus improving the scientific nature of vehicle scheduling decisions and operational efficiency.
[0015] This invention incorporates an emergency dispatch mechanism that dynamically adjusts the vehicle's loading time window (TW) and route route (ROU) based on dispatch rules when order changes occur. It also provides real-time feedback on the execution status via GPS, thus avoiding loading peak conflicts and route bottlenecks and ensuring the stability of vehicle dispatch. Attached Figure Description
[0016] Figure 1 This is a schematic diagram of the principle framework of the present invention. Detailed Implementation
[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0018] Please see Figure 1 This application provides an artificial intelligence-based cargo loading management method, including: S1. Obtain the Vehicle Information Table (VTL) and the Delivery Order Information Table (ODR); S2. Construct a training set based on historical scheduling records (HST), train a pre-built deep learning model, and obtain an allocation model (MOD). The allocation model (MOD) includes an input layer, multiple hidden layers for extracting features, and an output layer for outputting results. S3. Input the collected vehicle information table VTL and delivery order information table ODR into the allocation model MOD, and output the vehicle ID, loading area ZON and time window TW corresponding to the order. S4. Based on vehicle ID, loading zone (ZON), and recommended loading time period (TW), scheduling management is carried out through scheduling management methods, including using vehicle application path planning algorithms to generate optimal driving routes and emergency scheduling plans.
[0019] Historical dispatch records (HST) include a vehicle information table (VTL) and a delivery order information table (ODR). The vehicle information table (VTL) includes vehicle ID, vehicle type, loading volume (VOL) and loading weight (WGT), vehicle status (STA), current location (POS), and historical dispatch records (HST). The delivery order information table (ODR) includes order ID, cargo type (CAT), loading volume (VOL) and loading weight (WGT), loading requirements (REQ), loading zone (ZON), and time window (TW). Historical dispatch records (HST) are then retrieved.
[0020] Perform range checks on the loading volume (VOL) and loading weight (WGT) in the vehicle information table (VTL), remove outliers that exceed the preset upper and lower thresholds, perform legality checks on the vehicle status (STA), loading requirements (REQ), and cargo type (CAT), and remove data items that are not in the predefined category set.
[0021] To improve the quality of training data, data cleaning was performed on the Vehicle Information Table (VTL). Specifically, the Loading Volume (VOL) and Loading Weight (WGT) fields were checked for numerical range. If a field value exceeded a preset upper or lower threshold (e.g., VOL exceeding 200 m³ or WGT below 0.5 tons), the data was marked as abnormal and removed. Furthermore, the validity of category fields such as Vehicle Status (STA), Loading Requirements (REQ), and Cargo Type (CAT) was validated. If a field value was found to be outside the predefined category set, the data was considered invalid and removed.
[0022] The vehicle information set is obtained by retrieving the vehicle information table (VTL). This set includes the vehicle's serial number, years of use, number of repairs, and mileage. A vehicle score (Sve) is calculated for each vehicle in the set using the following formula:
[0023] Wherein, F1 is the service life of the delivery vehicle, f2 is the number of times the delivery vehicle has been repaired, and f3 is the mileage of the delivery vehicle; w1 is the weighting coefficient of the service life of the delivery vehicle, w2 is the weighting coefficient of the number of times the delivery vehicle has been repaired, and w3 is the weighting coefficient of the mileage of the delivery vehicle.
[0024] The vehicle rating (SVE) not only reflects the basic health status and usage level of a single vehicle, but also serves as an important reference indicator when performing multi-vehicle matching and scheduling. In specific applications, the initial values of the weighting coefficients w1, w2, and w3 can be set based on logistics operation experience, and are 0.4, 0.3, and 0.3 respectively, to reflect the different degrees of impact of vehicle age, maintenance frequency, and cumulative mileage on operational reliability.
[0025] When executing order allocation, SVE, as one of the structured feature nodes in the model input layer, participates in modeling along with vehicle loading volume (VOL), loading weight (WGT), and vehicle status (STA). This allows the model to consider not only the capacity matching degree of vehicles but also comprehensive factors such as reliability, usage risk, and maintenance pressure when making vehicle matching decisions. For example, when there are two candidate vehicles with the same loading capacity, the model will prioritize recommending the vehicle with the better SVE score to reduce the risk of possible failures or scheduling interruptions during subsequent operations.
[0026] The allocation model MOD consists of an input layer, multiple hidden layers, and an output layer. The input layer contains 14 feature nodes, corresponding to vehicle type, loading capacity, vehicle age (F1), vehicle rating (Sve), loading weight (WGT), vehicle status (STA), cargo type (CAT), loading volume (VOL), cargo loading weight (WGT), loading requirements (REQ), loading zone (ZON), order priority (PRI), historical average scheduling waiting time (AVG), and historical matching success rate (HMR).
[0027] The hidden layers consist of a first hidden layer, a second hidden layer, a third hidden layer, and a fourth hidden layer. The first hidden layer contains 128 nodes, the second hidden layer contains 256 nodes, the third hidden layer contains 128 nodes, and the fourth hidden layer contains 64 nodes. Each hidden layer uses the ReLU activation function. A BatchNormalization normalization layer is placed between the second and third hidden layers. Dropout layers are placed after the third and fourth hidden layers, respectively, with a dropout rate of 0.3. The output layer contains three output nodes: the optimal vehicle ID, the specified loading area number (ZON), and the loading time window (TW).
[0028] In this specific embodiment, the allocation model MOD is constructed through a combination of pre-training and online updating. The initial model is trained offline based on a training set constructed from historical scheduling records (HST). This training set includes multiple real matching records of completed orders, covering combinations of different vehicle types, cargo types, delivery routes, loading times, and loading areas. During the model training phase, a supervised learning strategy is employed. Vehicle ID, loading area ZON, and loading time window TW are set as label outputs. The input layer receives standardized feature vectors, which are then nonlinearly mapped by the hidden layer to generate the predicted output.
[0029] At the feature encoding level, in order to balance structural expressiveness and training efficiency, some discrete features (such as vehicle type, cargo type CAT, loading requirements REQ, vehicle status STA, etc.) are converted into dense vector form by independent encoding or embedded vector mapping. Continuous features (such as loading volume VOL, loading weight WGT, scheduling waiting time AVG, etc.) are preprocessed by max-min normalization or Z-score normalization to ensure that each feature is input into the model on the same numerical scale.
[0030] The model training process incorporates a cross-validation mechanism, dividing the training set into training and validation data in an 8:2 ratio. The Adam optimizer is used to update parameters, and an early stopping strategy is implemented to monitor changes in the validation set loss value and prevent overfitting. Dropout and BatchNormalization layers work synergistically in the model structure, enhancing both the model's generalization ability and its robustness under different task inputs.
[0031] In addition, the loading time window (TW) in the output results is not only used to control the arrival and loading time of vehicles, but also combined with scheduling rules to limit the vehicle density of the loading zone (ZON) during the time period, so as to prevent congestion and task conflicts caused by concentrated loading.
[0032] Scheduling management methods include: S11. Set a corresponding time window TW for each loading zone ZON, and preset the maximum allowed number of vehicles to be loaded at the same time for each ZON, and the loading capacity MAX. S12. When it is detected that the number of vehicles allocated to loading area ZON within time window TW exceeds the loading capacity MAX, the time adjustment mechanism is automatically triggered to adjust the loading time of some vehicles to the adjacent time window TW′. Within time window TW, the number of vehicles allocated to loading area ZON does not exceed the loading capacity MAX. S13. Combining the current position POS of each vehicle with the position of its assigned loading zone ZON, the path planning algorithm ROU is invoked to generate the optimal driving route for each vehicle, and the optimal driving route is sent to the scheduling terminal. S14. Collect and monitor the current location (POS) and loading status (STA) of each vehicle through GPS positioning. When a vehicle concentration, route congestion, or abnormal delay is detected in the target zone (ZON), adjust the vehicle's time window (TW) based on the scheduling rules.
[0033] In this embodiment, the scheduling management method is used to control the time scheduling and route planning of each vehicle during the loading process, so as to avoid congestion, conflict or route interference in the loading zone (ZON) within a specific time period, and to ensure the orderliness and real-time response capability of the scheduling process.
[0034] First, set a corresponding loading time window (TW) for each loading zone (ZON), and pre-set the maximum number of vehicles that can be loaded simultaneously, i.e., the loading capacity (MAX) (the unit can be vehicles / time period). The TW and MAX values for each ZON can be set based on the historical average loading volume, the number of loading platforms, and the average operation time.
[0035] When the scheduler detects that the number of vehicles assigned to the same loading zone (ZON) within a certain time window (TW) exceeds its set loading capacity (MAX), it automatically triggers a time adjustment mechanism. The loading tasks of some vehicles are postponed to the adjacent available time window (TW′) to ensure that the number of vehicles arriving at the target ZON within the same TW does not exceed MAX, thereby preventing the saturation of regional loading resources.
[0036] Next, by combining the current POS of each vehicle with the geographical coordinates of its corresponding loading zone ZON, a route planning algorithm ROU (such as Dijkstra's or A*) is invoked to generate the optimal driving route. The optimal route comprehensively considers factors such as shortest distance, lowest congestion probability, and traffic restrictions, and the ROU result is sent to the vehicle dispatch terminal in the form of a dispatch command to realize route execution control.
[0037] During the execution of the above scheduling and management method, to ensure the real-time performance and stability of the response, an event monitoring mechanism is set up to continuously monitor the task queuing status and vehicle deployment status within the loading zone (ZON). The vehicle deployment status is constructed based on real-time GPS positioning data and a site resource model, combined with vehicle arrival timestamps and estimated loading time, to form a visualized loading task schedule for dynamically assessing the load risk of each loading zone.
[0038] To improve the efficiency of the scheduling strategy, a conflict constraint optimization mechanism is introduced when adjusting loading time (S143). This includes, but is not limited to: the same vehicle cannot have overlapping scheduling between two time windows TW and TW′; the same vehicle cannot cross two loading areas at the same time; and the same loading area cannot repeatedly accept multiple vehicle scheduling requests from the same fleet within a unit of time. This mechanism solves the constraints through linear programming constraints or heuristic scheduling algorithms to ensure that the adjustment results are both feasible and physically reasonable.
[0039] Meanwhile, for vehicles that are unable to arrive at the target ZON on time due to long route distances or dynamic changes in traffic, arrival time prediction (ETA) is performed by combining historical traffic speed models and real-time traffic API. When the prediction results show that a vehicle cannot arrive as planned at TW, an early warning adjustment mechanism is automatically triggered to shift its scheduling plan to TW′ or the nearby backup loading area AZN in advance, so as to achieve pre-schedule intervention.
[0040] The scheduling rules include: S141. By comparing the number of vehicles allocated to the same ZON within the same TW in real time with the preset loading capacity MAX, the congestion index CND is calculated. When CND≥1, a congestion event is triggered. S142. Based on the order priority (PRI) and estimated queuing time (WAT) of each vehicle, prioritize the vehicles that triggered the congestion event and generate a list of vehicles to be adjusted. S143. In accordance with the PRI from low to high, postpone the time window TW of the vehicles in the list of vehicles to be adjusted to the next time window TW′; S144. Based on the time window TW′, re-invoke the path planning algorithm ROU to calculate the optimal driving route, and send the time window TW′ to the corresponding vehicle through the scheduling terminal. S145. Repeat steps S141 to S144 until the number of vehicles assigned to the same loading zone ZON within the time window TW returns to no more than the preset loading capacity MAX.
[0041] Furthermore, to reduce the cascading impact of TW adjustments on the vehicle queue, the "minimum disturbance principle" is introduced. Priority is given to adjusting vehicles at the end of the queue or those with the lowest scheduling priority (PRI), while maintaining the scheduling plans of higher-priority vehicles within the original TW time window as much as possible, in order to ensure the scheduling stability of urgent orders or important customer tasks.
[0042] The aforementioned scheduling rules can also be dynamically configured or switched according to the actual operation scenario. For example, during peak hours, the CND sensitivity threshold can be enhanced to prepare for congestion in advance; while during off-peak hours or in areas with sparse traffic, a "resource balancing strategy" can be activated to further optimize the utilization rate of the time window and the balance of vehicle routes on the basis of meeting the maximum loading capacity.
[0043] The emergency dispatch plan includes the following steps: S21. Monitor the delivery orders in the delivery order information table ODR. When a delivery order is detected to have a change or emergency order tag, mark the delivery order as a high priority (PRI). S22. Input the data from the delivery order with high priority PRI into the allocation model MOD, recalculate and obtain the optimal vehicle ID, loading zone ZON and loading time window TW corresponding to the delivery order; S23. Based on the calculated vehicle ID, loading zone ZON and loading time window TW, assess whether the number of vehicles in the corresponding loading zone ZON exceeds the loading capacity MAX. If it does, trigger the time window adjustment mechanism to postpone the loading time window TW of non-high priority orders. S24. Based on the changed vehicle allocation results, call the route planning algorithm ROU to recalculate the driving route and send it to the dispatch terminal of the corresponding vehicle.
[0044] In this embodiment, an emergency scheduling mechanism is provided to cope with real-time events such as dynamic changes in order information, insertion of emergency tasks, or sudden scheduling conflicts. It can dynamically generate new vehicle matching schemes based on the allocation model MOD and adjust the scheduling strategy in real time to ensure high responsiveness and scheduling stability even in high-frequency changing scenarios.
[0045] Specifically, by monitoring the order status field in the delivery order information table (ODR), the system can identify in real time whether there have been changes to order details (such as adjustments to loading area or delivery time) or the insertion of new urgent orders. When such events are identified, the order is immediately marked as a high-priority (PRI) order to trigger subsequent emergency dispatch procedures.
[0046] Subsequently, the updated order data is fed back into the allocation model MOD as input. The model outputs the optimal vehicle ID, recommended loading zone (ZON), and time window (TW) corresponding to the emergency order or change order based on the real-time vehicle information table (VTL), order characteristics, and current scheduling status.
[0047] Based on the above calculations, the vehicle scheduling density within the target time window TW in the current loading area ZON is determined. If the number of vehicles is expected to exceed the preset loading capacity MAX after this order is added, the time window adjustment mechanism is automatically activated. This mechanism prioritizes scheduled orders that are not high-priority PRI, extending their original loading time window TW to the next available time window TW′, thereby freeing up scheduling space for urgent orders and avoiding regional overloading or on-site conflicts.
[0048] After completing the new vehicle allocation and time window adjustment, the Route Planning Algorithm ROU (using Dijkstra's algorithm) is invoked to recalculate the optimal driving path for the target vehicle from its current location POS to the target loading area ZON. Fine-tuning of the path is then performed based on actual traffic conditions, estimated arrival time, and scheduling priority. Finally, the scheduling instruction, along with the updated time window TW′ and route information ROU, is sent to the scheduling terminal or vehicle operation interface to guide dispatchers or the autonomous driving system in executing the new loading task.
[0049] In this embodiment, the emergency dispatch mechanism has high-frequency processing capabilities and model-driven support. It can effectively embed sudden order requests without disrupting the original large-scale dispatch plan, thereby achieving priority protection of critical tasks and rapid resource reconstruction. It is suitable for time-sensitive logistics, emergency material transportation, and multi-task concurrent dispatch scenarios.
[0050] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired or wireless network. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.
[0051] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.
[0052] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.
Claims
1. A cargo loading management method based on artificial intelligence, characterized in that, include: S1. Obtain the Vehicle Information Table (VTL) and the Delivery Order Information Table (ODR); S2. Construct a training set based on historical scheduling records (HST), train a pre-built deep learning model, and obtain an allocation model (MOD). The allocation model (MOD) includes an input layer, multiple hidden layers for extracting features, and an output layer for outputting results. S3. Input the collected vehicle information table VTL and delivery order information table ODR into the allocation model MOD, and output the vehicle ID, loading area ZON and time window TW corresponding to the order. S4. Based on vehicle ID, loading zone (ZON), and recommended loading time period (TW), scheduling management is carried out through scheduling management methods, including using vehicle application path planning algorithms to generate optimal driving routes and emergency scheduling plans.
2. The cargo loading management method based on artificial intelligence according to claim 1, characterized in that, The historical dispatch record (HST) includes a vehicle information table (VTL) and a delivery order information table (ODR). The vehicle information table (VTL) includes vehicle ID, vehicle type, loading volume (VOL) and loading weight (WGT), vehicle status (STA), current location (POS), and historical dispatch record (HST). The delivery order information table (ODR) includes order ID, cargo type (CAT), loading volume (VOL) and loading weight (WGT), loading requirement (REQ), loading zone (ZON), and time window (TW). The historical dispatch record (HST) is obtained from this information.
3. The cargo loading management method based on artificial intelligence according to claim 1, characterized in that, Perform range checks on the loading volume (VOL) and loading weight (WGT) in the vehicle information table (VTL), remove outliers that exceed the preset upper and lower thresholds, perform legality checks on the vehicle status (STA), loading requirements (REQ), and cargo type (CAT), and remove data items that are not in the predefined category set.
4. The cargo loading management method based on artificial intelligence according to claim 1, characterized in that, The vehicle information set is obtained by retrieving the vehicle information table (VTL). This set includes the vehicle's serial number, years of use, number of repairs, and mileage. A vehicle score (Sve) is calculated for each vehicle in the set using the following formula: ； Wherein, F1 is the service life of the delivery vehicle, f2 is the number of times the delivery vehicle has been repaired, and f3 is the mileage of the delivery vehicle; w1 is the weighting coefficient of the service life of the delivery vehicle, w2 is the weighting coefficient of the number of times the delivery vehicle has been repaired, and w3 is the weighting coefficient of the mileage of the delivery vehicle.
5. The cargo loading management method based on artificial intelligence according to claim 1, characterized in that, The allocation model MOD includes an input layer, multiple hidden layers, and an output layer. The input layer contains 14 feature nodes, which correspond to vehicle type, loading capacity, vehicle age F1, vehicle rating Sve, loading weight WGT, vehicle status STA, cargo type CAT, loading volume VOL, cargo loading weight WGT, loading requirements REQ, loading area ZON, order priority PRI, historical average scheduling waiting time AVG, and historical matching success rate HMR.
6. The cargo loading management method based on artificial intelligence according to claim 1, characterized in that, The hidden layers include a first hidden layer, a second hidden layer, a third hidden layer, and a fourth hidden layer. The first hidden layer contains 128 nodes, the second hidden layer contains 256 nodes, the third hidden layer contains 128 nodes, and the fourth hidden layer contains 64 nodes. Each hidden layer uses the ReLU activation function. A BatchNormalization normalization layer is set between the second and third hidden layers. Dropout layers are set after the third and fourth hidden layers, respectively, with a dropout rate of 0.
3. The output layer contains three output nodes: the optimal vehicle ID, the specified loading area number (ZON), and the loading time window (TW).
7. The cargo loading management method based on artificial intelligence according to claim 1, characterized in that, The method as described in claim 1, wherein the scheduling management method comprises: S11. Set a corresponding time window TW for each loading zone ZON, and preset the maximum allowed number of vehicles to be loaded at the same time threshold loading capacity MAX for each ZON. S12. When it is detected that the number of vehicles allocated to loading area ZON within time window TW exceeds the loading capacity MAX, the time adjustment mechanism is automatically triggered to adjust the loading time of some vehicles to the adjacent time window TW′. Within time window TW, the number of vehicles allocated to loading area ZON does not exceed the loading capacity MAX. S13. Combining the current position POS of each vehicle with the position of its assigned loading zone ZON, the path planning algorithm ROU is invoked to generate the optimal driving route for each vehicle, and the optimal driving route is sent to the scheduling terminal. S14. Collect and monitor the current location (POS) and loading status (STA) of each vehicle through GPS positioning. When a vehicle concentration, route congestion, or abnormal delay is detected in the target zone (ZON), adjust the vehicle's time window (TW) based on the scheduling rules.
8. The cargo loading management method based on artificial intelligence according to claim 7, characterized in that, The scheduling rules include: S141. By comparing the number of vehicles allocated to the same ZON within the same TW in real time with the preset loading capacity MAX, the congestion index CND is calculated. When CND≥1, a congestion event is triggered. S142. Based on the order priority (PRI) and estimated queuing time (WAT) of each vehicle, prioritize the vehicles that triggered the congestion event and generate a list of vehicles to be adjusted. S143. In accordance with the PRI from low to high, postpone the time window TW of the vehicles in the list of vehicles to be adjusted to the next time window TW′; S144. Based on the time window TW′, re-invoke the path planning algorithm ROU to calculate the optimal driving route, and send the time window TW′ to the corresponding vehicle through the scheduling terminal. S145. Repeat steps S141 to S144 until the number of vehicles assigned to the same loading zone ZON within the time window TW returns to no more than the preset loading capacity MAX.
9. The cargo loading management method based on artificial intelligence according to claim 1, characterized in that, The emergency dispatch plan includes the following steps: S21. Monitor the delivery orders in the delivery order information table ODR. When a delivery order is detected to have a change or emergency order tag, mark the delivery order as a high priority (PRI). S22. Input the data from the delivery order with high priority (PRI) into the allocation model (MOD), recalculate and obtain the optimal vehicle ID, loading zone (ZON), and loading time window (TW) corresponding to the delivery order; S23. Based on the calculated vehicle ID, loading zone ZON and loading time window TW, assess whether the number of vehicles in the corresponding loading zone ZON exceeds the loading capacity MAX. If it does, trigger the time window adjustment mechanism to postpone the loading time window TW of non-high priority orders. S24. Based on the changed vehicle allocation results, call the route planning algorithm ROU to recalculate the driving route and send it to the dispatch terminal of the corresponding vehicle.