Transport vehicle loading point planning method and system based on automatic adjustment mechanism

Abstract

The invention relates to the technical field of intelligent logistics, in particular to a transport vehicle loading point planning method and system based on an automatic adjustment mechanism, and the method comprises the steps: collecting and fusing vehicle dynamic data, loading point state data, order data, environment data and other multi-source data in real time; performing prediction through a pre-trained machine learning model based on the data, and generating short-term prediction data of the accurate arrival time of the vehicle, the loading operation duration and the future load of the loading point; constructing a dynamic weight scoring model, calculating a comprehensive priority score for the combination of the to-be-distributed vehicle and the candidate loading point based on the short-term prediction data, and adaptively adjusting the weight of each evaluation index in the scoring model according to the overall operation target of the system; automatically distributing an optimal loading point for the to-be-distributed vehicle according to the comprehensive priority score, and generating a corresponding guide instruction; and optimizing and updating the machine learning model and the dynamic weight scoring model based on comparison feedback of actual operation data and prediction data after instruction execution.

Description

Technical Field

[0001] This application relates to the field of intelligent logistics technology, specifically to a method and system for planning loading points of transport vehicles based on an automatic adjustment mechanism. Background Technology

[0002] In the current logistics and transportation sector, the planning of loading points for transport vehicles at distribution centers or warehouses generally relies on static schedules or dispatch systems based on fixed rules. These methods are particularly rigid when facing complex real-world operating environments, failing to effectively respond to dynamic disturbances such as traffic congestion, vehicle breakdowns, temporary order changes, or sudden weather changes. Their inherent limitations lead to severe uneven utilization of loading point resources; some points experience excessively long waiting times for vehicles, while others remain idle, thus lowering overall loading and unloading efficiency and site throughput.

[0003] Furthermore, most existing systems lack the ability to predict short-term future trends and cannot make forward-looking resource allocations. This makes scheduling decisions often lag behind changes in the actual situation on site, and they rely heavily on manual intervention. This not only increases operating costs but also leads to unstable and uncontrollable decision quality.

[0004] Application content

[0005] The purpose of this application is to provide a method and system for planning loading points of transport vehicles based on an automatic adjustment mechanism.

[0006] In a first aspect, embodiments of this application provide a method for planning loading points of transport vehicles based on an automatic adjustment mechanism, including:

[0007] S1: Real-time collection and fusion of multi-source data, including at least vehicle dynamic data, loading point status data, order data, and environmental data;

[0008] S2: Based on multi-source data, a pre-trained machine learning model is used to make predictions and generate short-term forecast data including the precise arrival time of vehicles, loading operation duration, and future load at loading points.

[0009] S3: Construct a dynamic weighted scoring model, calculate the comprehensive priority score for the combination of vehicles to be assigned and candidate loading points based on short-term prediction data, and adaptively adjust the weight of each evaluation indicator in the scoring model according to the overall system operation goal.

[0010] S4: Based on the overall priority score, automatically assign the optimal loading point to the vehicle to be assigned and generate the corresponding guidance instructions;

[0011] S5: Based on the actual operational data after the command is executed, compare it with the corresponding predicted data to generate feedback data, and use the feedback data to optimize and update the machine learning model in S2 and the dynamic weight scoring model in S3.

[0012] In some embodiments, vehicle dynamic data includes at least real-time location, estimated arrival time, vehicle specifications, and cargo capacity; loading point status data includes at least occupancy status, estimated idle time, and equipment type; order data includes at least cargo type, volume, weight, and priority identifier; and environmental data includes at least real-time traffic conditions and weather conditions.

[0013] In some embodiments, step S2 includes:

[0014] S21: Based on real-time vehicle location, traffic conditions, and historical data, predict the precise time and probability distribution of the vehicle's arrival at the loading point.

[0015] S22: Based on the order cargo attributes, loading point equipment type, and historical operation data, predict the estimated duration of a single loading operation;

[0016] S23: Based on the predicted arrival time, predicted operation duration, and current loading point status of all vehicles, the queue length and waiting time of each loading point in the future are predicted through simulation.

[0017] In some embodiments, constructing the dynamic weighted scoring model in step S3 includes:

[0018] S31: Set multiple evaluation indicators, including at least efficiency indicators, load balancing indicators, distance indicators, and order priority indicators;

[0019] S32: Assign dynamic weights to each evaluation indicator. The dynamic weights are adjusted periodically or event-triggered based on changes in system-level key performance indicators.

[0020] S33: For each vehicle to be assigned and each available loading point, calculate the standardized score of each evaluation indicator, and use the current dynamic weights to perform a weighted summation to obtain the comprehensive priority score.

[0021] In some embodiments, the adjustment of dynamic weights in step S32 includes:

[0022] When the system detects that the average waiting time exceeds a preset threshold, it automatically increases the weight of efficiency metrics; and / or,

[0023] When the system detects that the load difference between loading points exceeds a preset threshold, it automatically increases the weight of the load balancing metric; and / or,

[0024] When the system identifies a high-priority order, it automatically increases the weight of the order priority metric.

[0025] In some embodiments, the adaptive adjustment of the weights of each evaluation index in the scoring model in step S3 includes:

[0026] Using system-level key performance indicators as optimization targets, and based on the system's performance within historical time windows, the system automatically calculates and outputs the optimal weight combination of each evaluation indicator for the next time window. Among these, the system-level key performance indicators are maximizing the total system throughput, minimizing the average waiting time, or minimizing the average processing time of high-priority orders.

[0027] In some embodiments, generating the corresponding boot instructions in step S4 includes:

[0028] Send the target loading point information and navigation route to the vehicle-mounted terminal and / or driver's mobile terminal of the vehicle to be assigned, and at the same time send the vehicle identification information to the on-site display device of the optimal loading point.

[0029] In some embodiments, the optimization and updating of the machine learning model in S2 and the dynamic weight scoring model in S3 using feedback data in step S5 includes:

[0030] S51: Collect data on actual vehicle arrival time and actual working time, compare them with the corresponding predicted values, and calculate the prediction deviation;

[0031] S52: Use new data containing prediction bias to perform incremental learning or periodic retraining on the prediction model;

[0032] S53: Based on the changes in key performance indicators of the system caused by historical decisions, evaluate the effectiveness of different weighted strategies and optimize the decision-making logic.

[0033] The second aspect of this application provides a vehicle loading point planning system based on an automatic adjustment mechanism, comprising:

[0034] The data acquisition module is used to collect and integrate multi-source data in real time. The multi-source data includes at least vehicle dynamic data, loading point status data, order data, and environmental data.

[0035] The processing module is used to generate short-term forecast data, including the precise arrival time of vehicles, loading operation duration, and future load at loading points, based on multi-source data and a pre-trained machine learning model.

[0036] The planning module is used to build a dynamic weighted scoring model. Based on short-term forecast data, it calculates a comprehensive priority score for the combination of vehicles to be assigned and candidate loading points, and adaptively adjusts the weights of each evaluation indicator in the scoring model according to the overall system operation goals.

[0037] The execution module is used to automatically assign the optimal loading point to the vehicle to be assigned based on the comprehensive priority score, and generate the corresponding guidance instructions;

[0038] The feedback module is used to compare the actual operational data after the command is executed with the corresponding predicted data, generate feedback data, and use the feedback data to optimize and update the machine learning model in S2 and the dynamic weight scoring model in S3.

[0039] In some embodiments, the planning module includes an upper-level weight controller, which is used to automatically calculate and output the optimal weight combination of each evaluation indicator in the next time window based on the system's performance in the historical time window, with system-level key performance indicators as the optimization target.

[0040] The beneficial effects of this application are as follows: By introducing a dynamic data-driven and adaptive learning mechanism, this invention can respond to changes in the internal and external environment in real time, achieving the effect of dynamic planning. Through a dynamic weighted scoring model and multi-objective optimization, it achieves the effect of seeking the global optimal solution among objectives such as efficiency, cost, and fairness, thereby greatly improving the utilization rate of loading points and the total throughput of the system. Among them, the short-term prediction function can anticipate potential congestion and proactively intervene, achieving the effect of proactive management. In addition, the closed-loop feedback and self-learning mechanism enable the system to continuously evolve from historical data and decision consequences. The prediction model and decision-making strategy are continuously optimized as the running time increases, ultimately realizing fully automated scheduling from vehicle entry to the start of loading, significantly reducing the dependence on experienced scheduling personnel, effectively saving labor costs, and significantly improving operational efficiency and reliability. Attached Figure Description

[0041] Figure 1 This is a flowchart of the planning method used in this application;

[0042] Figure 2 This is a flowchart of step S2 in this application;

[0043] Figure 3 A flowchart for constructing the dynamic weighted scoring model in this application;

[0044] Figure 4 This is a flowchart of step S5 in this application;

[0045] Figure 5 This is a module diagram of the planning system in this application. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0047] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0048] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0049] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0050] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0051] In the current logistics and transportation sector, the planning of loading points for transport vehicles at distribution centers or warehouses generally relies on static schedules or dispatch systems based on fixed rules. These methods are particularly rigid when facing complex real-world operating environments, failing to effectively respond to dynamic disturbances such as traffic congestion, vehicle breakdowns, temporary order changes, or sudden weather changes. Their inherent limitations lead to severe uneven utilization of loading point resources; some points experience excessively long waiting times for vehicles, while others remain idle, thus lowering overall loading and unloading efficiency and site throughput.

[0052] Furthermore, most existing systems lack the ability to predict short-term future trends and cannot make forward-looking resource allocations. This makes scheduling decisions often lag behind changes in the actual situation on site, and they rely heavily on manual intervention. This not only increases operating costs but also leads to unstable and uncontrollable decision quality.

[0053] Example 1

[0054] In view of this, the first aspect of this application provides a method for planning loading points of transport vehicles based on an automatic adjustment mechanism, referring to... Figure 1 ,include:

[0055] S1: Real-time collection and fusion of multi-source data, including at least vehicle dynamic data, loading point status data, order data, and environmental data.

[0056] Deployed in the cloud, the distributed data acquisition gateway connects to GNAS vehicle terminals, transportation management systems, warehouse management systems, and third-party traffic data service platforms via standardized application programming interfaces (APIs). It continuously receives and parses streaming data from these heterogeneous data sources at a frequency of at least once per second. After data cleaning, format standardization, and timestamp alignment, the collected raw data is injected into the real-time data bus. Dynamic data objects containing full-dimensional attributes are constructed for each vehicle on the road and each physical loading point, forming a unified data view for subsequent analysis and decision-making.

[0057] Create a dynamic data object for each entity (vehicle, loading point).

[0058] Vehicle-Object: Includes vehicle ID, real-time location, speed, ETA (time to first entry point), vehicle specifications, order list, order priority, and current status (en route, waiting, loading).

[0059] Dock-Object: Contains the load point ID, current status (idle, occupied, faulty), currently serviced vehicle ID, estimated release time, and supported device type (e.g., hydraulic tailgate, forklift).

[0060] S2: Based on multi-source data, it uses a pre-trained machine learning model to generate short-term forecast data including the precise arrival time of vehicles, loading operation duration, and future load at loading points.

[0061] The prediction process is executed by an integrated cluster of prediction microservices. The precise arrival time prediction of vehicles adopts a gradient boosting decision tree model. Its input feature vector includes the instantaneous latitude and longitude coordinates of the vehicle, instantaneous speed, historical average travel time of the road segment during the same period, real-time traffic congestion index, future short-term weather forecast data, and simulated queue length at the park entrance. The model output is the probability distribution of the time when the vehicle arrives at the designated loading point, for example, the expected arrival time and variance are presented in the form of a Gaussian distribution.

[0062] Loading operation duration prediction uses a random forest regression model. The model input features include the total number of pallets, total volume, total weight of the order goods, the standardization level of the goods packaging, the automation level of the loading and unloading equipment configured at the target loading point, and the historical average efficiency coefficient of the pre-assigned operator shifts. Future load prediction at the loading point is based on the aforementioned vehicle arrival time prediction and loading operation duration prediction results. Combined with the currently assigned but not yet started task queues, discrete event simulation technology is used to extrapolate the possible vehicle queue sequence before each loading point within a 15-30 minute time window with second-level accuracy. The predicted queue length and predicted waiting time at each time point are calculated, thus outputting the probability distribution of vehicles arriving at the designated loading point. For example, P(arriving at 10:05) = 60%, P(arriving at 10:06) = 30%.

[0063] S3: Construct a dynamic weighted scoring model, calculate the comprehensive priority score for the combination of vehicles to be assigned and candidate loading points based on short-term prediction data, and adaptively adjust the weight of each evaluation indicator in the scoring model according to the overall system operation goal.

[0064] All loading points with a status of "idle" or "soon to be idle" (e.g., released within 5 minutes) are selected to form a candidate loading point set, Docks-Candidate. The dynamic weighted scoring model first pairs each transport vehicle to be assigned with each candidate loading point currently in an "idle" or "expected to be idle" status, creating a vehicle-loading point pairing to be evaluated. For each pairing, the model calculates standardized scores for a set of predefined evaluation metrics, which at least cover efficiency, load balancing, distance, and order priority dimensions.

[0065] The efficiency dimension index is mainly based on the waiting time of the vehicle at the loading point predicted in S2; the load balancing dimension index is quantified by calculating the Gini coefficient or standard deviation of the expected number of vehicles processed by all loading points in the entire loading yard in the future after the vehicle is assigned to the loading point; the distance dimension index is based on the estimated path length or time of the vehicle from its current location to the candidate loading point; and the order priority dimension is assigned a value based on the preset urgency label in the order data. The standardized score of each evaluation index will be processed to the dimensionless [0, 1] interval through max-min normalization or Z-score normalization methods.

[0066] The overall priority score is obtained by weighting the standardized scores of each indicator with their corresponding dynamic weights. The adaptive adjustment of the dynamic weights is handled by an independent upper-level weight controller. This controller aims to optimize system-level key performance indicators such as maximizing total system throughput, minimizing average system waiting time, or minimizing the average processing time of high-priority orders. Based on historical operational efficiency data within a sliding time window, it uses a multi-armed gambling machine algorithm or a metaheuristic optimization algorithm to periodically or be triggered by specific events to recalculate and output the optimal weight combination of each evaluation indicator for the next time period.

[0067] S4: Based on the overall priority score, automatically assign the optimal loading point to the vehicle to be assigned and generate the corresponding guidance instructions.

[0068] The decision engine iterates through the calculation results of all candidate vehicle-loading point pairings and selects the loading point with the highest comprehensive priority score as the optimal loading point for each vehicle to be assigned. After the allocation decision is generated, the execution instructions are asynchronously issued through a highly available message middleware. Instruction issuance to the vehicle includes pushing a structured data packet containing the target loading point identifier, optimal driving route coordinates, and estimated waiting time to the driver's smart mobile terminal application or in-vehicle smart terminal via the mobile data network; instruction issuance to the site includes sending the vehicle identification information to be served to the industrial-grade electronic Kanban board or display screen configured at the optimal loading point for visual prompts via the enterprise intranet, and simultaneously sending an asynchronous message to the warehouse management system via the enterprise service bus to trigger the picking and pre-delivery scheduling task of the goods for the vehicle's order to the designated loading point.

[0069] S5: Based on the actual operational data after the command is executed, compare it with the corresponding predicted data to generate feedback data, and use the feedback data to optimize and update the machine learning model in S2 and the dynamic weight scoring model in S3.

[0070] The system continuously runs a closed-loop feedback pipeline that monitors and records the actual arrival timestamps of vehicles and the actual start and end timestamps of loading operations. These actual operational data are precisely compared with the corresponding predicted data generated in step S2 to calculate the deviations in predicted arrival times and predicted operation durations, forming a feedback data stream with timestamps and context labels.

[0071] For machine learning models in S2, this feedback data stream is used to incrementally update the model using online learning algorithms, or a batch training task is started daily during off-peak business hours to add the newly added data accumulated that day to the training set for full retraining of the model in order to continuously improve prediction accuracy.

[0072] For the dynamic weighted scoring model in S3, the feedback module analyzes the causal relationship between the different weight combination strategies adopted in the historical period and the actual changes in the system's key performance indicators. It evaluates the long-term benefits of each strategy through a reinforcement learning framework, thereby optimizing the decision logic of the upper-level weight controller and enabling the system to learn autonomously from historical experience and improve its decision-making strategies.

[0073] Example 2

[0074] In some embodiments, vehicle dynamic data includes at least the vehicle's latitude and longitude coordinates acquired and transmitted in real time by the vehicle-mounted Internet of Things terminal via the Global Navigation Satellite System, the dynamically estimated arrival time calculated based on the route planning algorithm and real-time traffic conditions, the rated load and cargo volume specifications indicated on the vehicle nameplate, and the current waybill cargo volume list obtained by querying the relevant data from the transportation management system.

[0075] The loading point status data includes at least the real-time occupancy or vacancy status detected by the geomagnetic sensor or radio frequency identification access control system installed on the loading platform, the estimated vacancy timestamp based on the current work progress, and the type of fixed equipment registered in the loading point file, such as the model and capacity parameters of the hydraulic lifting platform, belt conveyor or telescopic forklift.

[0076] Order data should include at least the master data of goods extracted from the enterprise resource planning system, such as category, fragility label, total volume and total weight, and priority labels set by the customer or business system, such as "normal," "urgent," or "express." Environmental data should include at least real-time road traffic speed data in JSON or Protocol Buffers format obtained from an authoritative third-party service provider's application programming interface (API), and detailed weather forecast information for the next two hours, including precipitation probability and wind speed.

[0077] Example 3

[0078] In some embodiments, step S2 includes:

[0079] S21: Based on real-time vehicle location, traffic conditions, and historical data, predict the precise time of vehicle arrival at the loading point and its probability distribution.

[0080] By integrating GNSS trajectory point sequences reported by vehicles, path topology provided by digital maps, real-time traffic event data, and historical travel time databases from the same period, a high-dimensional feature space is constructed. This space is then input into a pre-trained LightGBM regression model, and the output includes not only the estimated arrival time of points, but more importantly, the probability density function of the arrival time. This probability distribution provides an uncertainty measure for subsequent robust optimization decisions.

[0081] S22: Based on the order cargo attributes, loading point equipment type, and historical operation data, predict the estimated duration of a single loading operation.

[0082] The model takes the physical attribute vector of the goods in the order and the resource attribute vector of the loading point as input. The physical attribute vector includes the unitization form of the goods, the total number of pieces, and the weight distribution of each piece. The resource attribute vector includes the power of loading and unloading equipment and the number of operators. By executing the random forest regression algorithm, the model outputs a point estimate of the duration of a single loading task and its confidence interval. This model can capture the complex nonlinear relationship between goods attributes, resource capabilities and operational efficiency.

[0083] S23: Based on the predicted arrival time, predicted operation duration, and current loading point status of all vehicles, the queue length and waiting time of each loading point in the future are predicted through simulation.

[0084] Through an agent-based simulation system, each vehicle on the way and each loading point is modeled as an agent with state. Based on the probabilistic prediction results provided by S21 and S22, these agents are driven to interact according to preset behavior rules on a virtual timeline, thereby simulating the dynamic operation of the entire loading station in a specific future time period and accurately outputting the predicted queue state of each loading point in each future time slice.

[0085] Example 4

[0086] In some embodiments, constructing the dynamic weighted scoring model in step S3 includes:

[0087] S31: Set multiple evaluation indicators, including at least efficiency indicators, load balancing indicators, distance indicators, and order priority indicators.

[0088] Identify the core decision dimensions that need to be quantified. Efficiency indicators are usually quantified as the inverse function of predicted waiting time; load balancing indicators are quantified as the degree of imbalance in the system load distribution, such as using the Herfindahl-Hirschman index; distance indicators are quantified as the Euclidean distance from the vehicle to the loading point or the actual route planning distance; and order priority indicators are mapped to discrete weight multipliers according to business rules.

[0089] For the vehicle to be assigned, Vj, and the candidate loading point, Di, calculate a set of index values ​​F-ij normalized to [0, 1]:

[0090] F-efficiency (efficiency metric): Based on predicted waiting time. The shorter the waiting time, the higher the score. F-efficiency = 1 / (1 + predicted waiting time).

[0091] F-distance (distance metric): Based on the estimated time it will take for the vehicle to travel from its current location to the loading point. The shorter the time, the higher the score.

[0092] F-balance (load balancing metric): Calculates the standard deviation of the expected number of vehicles processed by all loading points in the next half hour after Vj is assigned to Di. The smaller the standard deviation, the more balanced the load, and the higher the F-balance score.

[0093] F-priority (priority index): If Vj contains high-priority orders and Di is a green channel specifically for handling expedited orders, then this score is very high.

[0094] F-compatibility: This indicator checks whether the type of cargo carried by the vehicle matches the equipment at the loading point (e.g., refrigerated cargo and refrigerated warehouse doors). A match is 1, and a mismatch is 0.

[0095] S32: Assign dynamic weights to each evaluation indicator. The dynamic weights are adjusted periodically or event-triggered based on changes in system-level key performance indicators.

[0096] The system maintains a global upper-level weight controller, which continuously monitors macro-level key performance indicators such as overall system average wait time, throughput, and resource utilization. When these KPIs deviate from the preset target range or specific business events are detected, the controller initiates an optimization calculation process. This process aims to maximize system-level KPIs (such as total completed trips and average wait time), using multi-armed gamblers or metaheuristic algorithms to dynamically adjust the weights Wk of each indicator in the lower-level scoring model, searching for the weight combination that can most effectively improve the target KPIs, and applying this new weight combination to the next decision-making cycle.

[0097] For example:

[0098] Scenario A (off-peak period): The system operates smoothly, and the weight controller may provide [W-efficiency=0.4, W-balance=0.3, W-distance=0.3], emphasizing efficiency and balance.

[0099] Scenario B (Peak Period): The system experiences widespread congestion, causing a surge in average waiting time. Upon detection, the weight controller automatically adjusts the settings to [W-efficiency=0.6, W-balance=0.2, W-distance=0.2], prioritizing smooth operation.

[0100] The adjustment of dynamic weights in step S32 includes:

[0101] When the system monitoring screen displays that the average vehicle waiting time at the entire station exceeds five minutes for a consecutive period, which is higher than the preset service level agreement threshold, the rule engine of the upper-level weight controller will automatically trigger a weight adjustment event. This event will significantly increase the weight coefficient of the efficiency index in the comprehensive score, for example, temporarily increasing its basic weight from 0.3 to 0.5, forcing the system to prioritize the loading point that can start operations the fastest in subsequent allocation decisions, so as to quickly alleviate the congestion situation.

[0102] When the load balancer detects that the standard deviation of the predicted workload at each loading point for the next half hour is consistently higher than the critical value for stable system operation, the upper-level weight controller will receive a load imbalance alarm signal, and then start the optimization algorithm to solve for a set of weight configurations that can most effectively distribute the load, and instruct the weight of the load balancing index to guide vehicles to idle or less loaded loading points.

[0103] When a high-priority order with the "urgent" tag is identified in the order flow, the system will generate a high-priority scheduling event. This event will directly intervene in the decision-making logic of the upper-level weight controller, temporarily and significantly increasing the weight of the order priority indicator, and may even make it the dominant decision factor for a certain period of time, ensuring that high-value or urgent orders can be processed without delay.

[0104] S33: For each vehicle to be assigned and each available loading point, calculate the standardized score of each evaluation indicator, and use the current dynamic weights to perform a weighted summation to obtain the comprehensive priority score.

[0105] For each <vehicle, loading point> pair, the algorithm first extracts the necessary raw values ​​from the real-time data layer and the prediction data layer, then calculates the raw value of each evaluation index according to the formula defined in S31, then transforms these raw values ​​to a uniform and comparable scale through standardization, and finally uses the current authority weight vector output in stage S32 to calculate a linear weighted sum to obtain the final comprehensive priority score used for ranking.

[0106] Example 5

[0107] In some embodiments, the adaptive adjustment of the weights of each evaluation index in the scoring model in step S3 includes:

[0108] Define system-level key performance indicators (KPIs) as optimization goals. These goals are typically high-order business metrics, such as maximizing the total number of transport vehicles processed within a single day's work cycle, minimizing the average waiting time of all vehicles at the loading point, or specifically minimizing the average processing time from arrival to loading completion for VIP customer orders. For example, when an extremely high-priority order is detected entering the system, the controller temporarily and significantly increases the weight of W-priority to ensure that VIP orders are processed with priority and speed.

[0109] The optimization engine constructs a utility function model based on the system's performance data within the most recent configurable historical time window, such as a dataset showing the correspondence between throughput and weight settings over the past hour. By solving for the maximum or minimum value of this utility function model, it automatically calculates and outputs the optimal weight combination of evaluation indicators that is expected to bring the best system-level key performance indicators in the next decision-making time window. This process elevates the planning approach from passive, reactive adjustments to proactive, prediction-based optimization.

[0110] Example 6

[0111] In some embodiments, generating the corresponding boot instructions in step S4 includes:

[0112] The target loading point information and navigation route sent to the vehicle-mounted terminal and / or driver's mobile terminal of the vehicle to be assigned are the final decision results calculated by the load optimization algorithm. This information is encapsulated in the form of a structured data message, which typically includes a unique number of the target loading point, the coordinates of the optimal driving route sequence generated based on a high-precision park map, and the estimated service start time provided based on S2 prediction data. This data message is encrypted and transmitted to the terminal application via HTTPS protocol. After parsing, the application guides the driver through a combination of graphical interface and voice prompts. At the same time, the vehicle identification information of the vehicle to be served, which usually includes key identification information such as the license plate number and the last four digits of the order number, is sent to the on-site display device at the optimal loading point. This is pushed with low latency via WebSocket or MQTT protocol to ensure that the on-site screen can be updated in real time, providing clear work preparation signals to the site operators. The entire instruction distribution process ensures information synchronization and collaborative operation between on-site and off-site work units.

[0113] Example 7

[0114] In some embodiments, the optimization and updating of the machine learning model in S2 and the dynamic weight scoring model in S3 using feedback data in step S5 includes:

[0115] S51: Collect data on actual vehicle arrival time and actual working time, compare them with the corresponding predicted values, and calculate the prediction deviation.

[0116] The system log service persistently records each predicted event and its corresponding actual occurrence. Using the query and aggregation functions of the time-series database, it batch calculates the difference between the actual arrival time and the predicted arrival time of vehicles, as well as the difference between the actual loading operation duration and the predicted operation duration, forming a prediction bias dataset with complete context.

[0117] S52: Use new data containing prediction bias to perform incremental learning or periodic retraining on the prediction model.

[0118] For machine learning models in S2, the system supports two update modes. One is an online learning mode, which uses stochastic gradient descent to fine-tune model parameters using a single sample of prediction bias that comes in in real time. The other is an offline batch learning mode, which starts a distributed computing task after the end of each business day to retrain the prediction model using all the data accumulated over the past 24 hours, generating a new generation of model versions and deploying them online.

[0119] S53: Based on the changes in key performance indicators of the system caused by historical decisions, evaluate the effectiveness of different weighted strategies and optimize the decision-making logic.

[0120] The core of the dynamic weight scoring model in S3 lies in optimizing the policy of the upper-level weight controller. The system constructs a policy evaluation dataset, recording every weight adjustment decision in history and the changes in system KPIs over a subsequent period. By applying reinforcement learning algorithms, such as Q-learning or policy gradient methods, it analyzes which weight policy can obtain the maximum long-term cumulative reward under different system states, thereby continuously refining the decision network of the weight controller, making its weight adjustments more intelligent and forward-looking.

[0121] Example 8

[0122] The second aspect of this application provides a vehicle loading point planning system based on an automatic adjustment mechanism. The system's physical architecture consists of a set of microservices deployed on a cloud computing platform or private data center. These services communicate via a high-speed network and message queues. The planning system includes:

[0123] The data acquisition module is a high-concurrency, horizontally scalable data access service used to collect and integrate multi-source data streams from vehicle GPS sensors, warehouse management databases, traffic information APIs, etc. in real time, and to verify, clean, and unify the format of the data. The multi-source data includes at least vehicle dynamic data, loading point status data, order data, and environmental data.

[0124] The processing module is an inference service cluster that carries machine learning models trained on historical data. It is used to generate short-term prediction data, including the precise arrival time of vehicles, loading operation duration, and future load at loading points, by calling these pre-trained models based on multi-source data and performing parallel prediction calculations. The results are then written to a cache for downstream modules to consume.

[0125] The planning module is the intelligent decision-making center of the system. It contains a dynamic weight scoring model calculation engine and an upper-level weight controller. It is used to build a dynamic weight scoring model, calculate a comprehensive priority score for the combination of vehicles to be assigned and candidate loading points based on short-term prediction data, and adaptively adjust the weight of each evaluation indicator in the scoring model according to the overall operation goal of the system.

[0126] The execution module is an instruction routing and distribution center. It is used to automatically assign the optimal loading point to the vehicles to be assigned based on the comprehensive priority score, and generate corresponding guidance instructions. The instructions are accurately delivered to the driver's mobile APP, the site electronic signboard, and the back-end warehouse management system through different communication channels.

[0127] The feedback module is a data processing pipeline responsible for system performance monitoring and self-evolution. It compares the actual operational data after command execution with the corresponding predicted data to generate feedback data. The feedback data is then used to continuously optimize and update the machine learning model in S2 and the dynamic weight scoring model in S3 through online learning or batch retraining.

[0128] In some embodiments, the planning module includes an upper-level weight controller, which is used to automatically calculate and output the optimal weight combination of each evaluation indicator in the next time window based on the system's performance in the historical time window, with system-level key performance indicators as the optimization target.

[0129] The upper-level weight controller within the planning module is an independent software component with optimization computation capabilities. This controller is designed with system-level key performance indicators (KPIs) as the ultimate optimization objective, such as maximizing total system throughput. It continuously monitors the system's performance within a sliding historical time window, collecting performance data such as actual throughput and latency under different weight configurations. Based on this data, the controller's built-in optimization algorithms, such as Bayesian optimization or multi-objective evolutionary algorithms, automatically search, calculate, and output the optimal weight combination of evaluation indicators expected to most effectively improve the target KPIs within the next time window. This mechanism enables the entire planning system to dynamically adapt to changes in business traffic and site conditions, rather than relying on fixed, pre-set parameters, achieving truly intelligent adaptive planning.

[0130] The above are merely preferred embodiments of this application and are not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0131] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within this application. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for planning loading points of transport vehicles based on an automatic adjustment mechanism, characterized in that, include: S1: Real-time collection and fusion of multi-source data, wherein the multi-source data includes at least vehicle dynamic data, loading point status data, order data and environmental data; S2: Based on the multi-source data, a pre-trained machine learning model is used to make predictions and generate short-term prediction data including the vehicle's precise arrival time, loading operation duration, and future load at the loading point. S3: Construct a dynamic weighted scoring model, calculate a comprehensive priority score for the combination of vehicles to be assigned and candidate loading points based on the short-term prediction data, and adaptively adjust the weights of each evaluation index in the scoring model according to the overall system operation objectives. S4: Based on the comprehensive priority score, automatically assign the optimal loading point to the vehicle to be assigned and generate the corresponding guidance instruction; S5: Based on the actual operational data after the instruction is executed, compare it with the corresponding predicted data to generate feedback data, and use the feedback data to optimize and update the machine learning model in S2 and the dynamic weight scoring model in S3.

2. The planning method according to claim 1, characterized in that, The vehicle dynamic data includes at least real-time location, estimated arrival time, vehicle specifications, and cargo capacity; the loading point status data includes at least occupancy status, estimated idle time, and equipment type; the order data includes at least cargo type, volume, weight, and priority identifier; and the environmental data includes at least real-time traffic conditions and weather conditions.

3. The method according to claim 1, characterized in that, Step S2 includes: S21: Based on real-time vehicle location, traffic conditions, and historical data, predict the precise time and probability distribution of the vehicle's arrival at the loading point. S22: Based on the order cargo attributes, loading point equipment type, and historical operation data, predict the estimated duration of a single loading operation; S23: Based on the predicted arrival time, predicted operation duration, and current loading point status of all vehicles, the queue length and waiting time of each loading point in the future are predicted through simulation.

4. The method according to claim 1, characterized in that, Step S3 involves constructing a dynamic weighted scoring model, including: S31: Set multiple evaluation indicators, which include at least efficiency indicators, load balancing indicators, distance indicators, and order priority indicators; S32: Assign dynamic weights to each evaluation indicator, and adjust the dynamic weights periodically or event-triggeredly based on changes in system-level key performance indicators; S33: For each vehicle to be assigned and each available loading point, calculate the standardized score of each evaluation indicator, and use the current dynamic weights to perform a weighted summation to obtain the comprehensive priority score.

5. The method according to claim 4, characterized in that, The adjustment of dynamic weights in step S32 includes: When the system detects that the average waiting time exceeds a preset threshold, it automatically increases the weight of the efficiency indicator; and / or, When the system detects that the load difference between loading points exceeds a preset threshold, it automatically increases the weight of the load balancing metric; and / or, When the system identifies a high-priority order, it automatically increases the weight of the order priority indicator.

6. The method according to claim 5, characterized in that, Step S3, which adaptively adjusts the weights of each evaluation indicator in the scoring model, includes: Using system-level key performance indicators as optimization targets, and based on the system's performance within historical time windows, the system automatically calculates and outputs the optimal weight combination of each evaluation indicator for the next time window. The system-level key performance indicators are maximizing the total system throughput, minimizing the average waiting time, or minimizing the average processing time of high-priority orders.

7. The method according to claim 1, characterized in that, Step S4 generates the corresponding boot instructions, including: The system sends the target loading point information and navigation route to the vehicle-mounted terminal and / or driver's mobile terminal of the vehicle to be assigned, and simultaneously sends the vehicle identification information to the on-site display device of the optimal loading point.

8. The method according to claim 1, characterized in that, Step S5, which uses the feedback data to optimize and update the machine learning model in S2 and the dynamic weight scoring model in S3, includes: S51: Collect data on actual vehicle arrival time and actual working time, compare them with the corresponding predicted values, and calculate the prediction deviation; S52: Using the new data containing the prediction bias, perform incremental learning or periodic retraining on the prediction model; S53: Based on the changes in key performance indicators of the system caused by historical decisions, evaluate the effectiveness of different weighted strategies and optimize the decision-making logic.

9. A vehicle loading point planning system based on an automatic adjustment mechanism, characterized in that, include: The data acquisition module is used to collect and integrate multi-source data in real time, wherein the multi-source data includes at least vehicle dynamic data, loading point status data, order data and environmental data; The processing module is used to generate short-term prediction data, including the precise arrival time of the vehicle, the loading operation duration, and the future load at the loading point, based on the multi-source data and a pre-trained machine learning model. The planning module is used to construct a dynamic weighted scoring model. Based on the short-term prediction data, it calculates a comprehensive priority score for the combination of vehicles to be assigned and candidate loading points, and adaptively adjusts the weights of each evaluation indicator in the scoring model according to the overall system operation goals. The execution module is used to automatically assign the optimal loading point to the vehicle to be assigned based on the comprehensive priority score, and generate corresponding guidance instructions; The feedback module is used to compare the actual operational data after the instruction is executed with the corresponding predicted data, generate feedback data, and use the feedback data to optimize and update the machine learning model in S2 and the dynamic weight scoring model in S3.

10. The system according to claim 9, characterized in that, The planning module includes an upper-level weight controller, which is used to automatically calculate and output the optimal weight combination of each evaluation indicator in the next time window based on the system's performance in the historical time window, with system-level key performance indicators as the optimization target.

Images